1 /* 2 * Linux Socket Filter - Kernel level socket filtering 3 * 4 * Based on the design of the Berkeley Packet Filter. The new 5 * internal format has been designed by PLUMgrid: 6 * 7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com 8 * 9 * Authors: 10 * 11 * Jay Schulist <jschlst@samba.org> 12 * Alexei Starovoitov <ast@plumgrid.com> 13 * Daniel Borkmann <dborkman@redhat.com> 14 * 15 * This program is free software; you can redistribute it and/or 16 * modify it under the terms of the GNU General Public License 17 * as published by the Free Software Foundation; either version 18 * 2 of the License, or (at your option) any later version. 19 * 20 * Andi Kleen - Fix a few bad bugs and races. 21 * Kris Katterjohn - Added many additional checks in bpf_check_classic() 22 */ 23 24 #include <linux/module.h> 25 #include <linux/types.h> 26 #include <linux/mm.h> 27 #include <linux/fcntl.h> 28 #include <linux/socket.h> 29 #include <linux/sock_diag.h> 30 #include <linux/in.h> 31 #include <linux/inet.h> 32 #include <linux/netdevice.h> 33 #include <linux/if_packet.h> 34 #include <linux/if_arp.h> 35 #include <linux/gfp.h> 36 #include <net/ip.h> 37 #include <net/protocol.h> 38 #include <net/netlink.h> 39 #include <linux/skbuff.h> 40 #include <net/sock.h> 41 #include <net/flow_dissector.h> 42 #include <linux/errno.h> 43 #include <linux/timer.h> 44 #include <linux/uaccess.h> 45 #include <asm/unaligned.h> 46 #include <linux/filter.h> 47 #include <linux/ratelimit.h> 48 #include <linux/seccomp.h> 49 #include <linux/if_vlan.h> 50 #include <linux/bpf.h> 51 #include <net/sch_generic.h> 52 #include <net/cls_cgroup.h> 53 #include <net/dst_metadata.h> 54 #include <net/dst.h> 55 #include <net/sock_reuseport.h> 56 #include <net/busy_poll.h> 57 #include <net/tcp.h> 58 59 /** 60 * sk_filter_trim_cap - run a packet through a socket filter 61 * @sk: sock associated with &sk_buff 62 * @skb: buffer to filter 63 * @cap: limit on how short the eBPF program may trim the packet 64 * 65 * Run the eBPF program and then cut skb->data to correct size returned by 66 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 67 * than pkt_len we keep whole skb->data. This is the socket level 68 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should 69 * be accepted or -EPERM if the packet should be tossed. 70 * 71 */ 72 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 73 { 74 int err; 75 struct sk_filter *filter; 76 77 /* 78 * If the skb was allocated from pfmemalloc reserves, only 79 * allow SOCK_MEMALLOC sockets to use it as this socket is 80 * helping free memory 81 */ 82 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 83 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 84 return -ENOMEM; 85 } 86 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 87 if (err) 88 return err; 89 90 err = security_sock_rcv_skb(sk, skb); 91 if (err) 92 return err; 93 94 rcu_read_lock(); 95 filter = rcu_dereference(sk->sk_filter); 96 if (filter) { 97 struct sock *save_sk = skb->sk; 98 unsigned int pkt_len; 99 100 skb->sk = sk; 101 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 102 skb->sk = save_sk; 103 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 104 } 105 rcu_read_unlock(); 106 107 return err; 108 } 109 EXPORT_SYMBOL(sk_filter_trim_cap); 110 111 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb) 112 { 113 return skb_get_poff(skb); 114 } 115 116 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 117 { 118 struct nlattr *nla; 119 120 if (skb_is_nonlinear(skb)) 121 return 0; 122 123 if (skb->len < sizeof(struct nlattr)) 124 return 0; 125 126 if (a > skb->len - sizeof(struct nlattr)) 127 return 0; 128 129 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 130 if (nla) 131 return (void *) nla - (void *) skb->data; 132 133 return 0; 134 } 135 136 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 137 { 138 struct nlattr *nla; 139 140 if (skb_is_nonlinear(skb)) 141 return 0; 142 143 if (skb->len < sizeof(struct nlattr)) 144 return 0; 145 146 if (a > skb->len - sizeof(struct nlattr)) 147 return 0; 148 149 nla = (struct nlattr *) &skb->data[a]; 150 if (nla->nla_len > skb->len - a) 151 return 0; 152 153 nla = nla_find_nested(nla, x); 154 if (nla) 155 return (void *) nla - (void *) skb->data; 156 157 return 0; 158 } 159 160 BPF_CALL_0(__get_raw_cpu_id) 161 { 162 return raw_smp_processor_id(); 163 } 164 165 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = { 166 .func = __get_raw_cpu_id, 167 .gpl_only = false, 168 .ret_type = RET_INTEGER, 169 }; 170 171 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 172 struct bpf_insn *insn_buf) 173 { 174 struct bpf_insn *insn = insn_buf; 175 176 switch (skb_field) { 177 case SKF_AD_MARK: 178 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); 179 180 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 181 offsetof(struct sk_buff, mark)); 182 break; 183 184 case SKF_AD_PKTTYPE: 185 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); 186 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 187 #ifdef __BIG_ENDIAN_BITFIELD 188 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 189 #endif 190 break; 191 192 case SKF_AD_QUEUE: 193 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); 194 195 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 196 offsetof(struct sk_buff, queue_mapping)); 197 break; 198 199 case SKF_AD_VLAN_TAG: 200 case SKF_AD_VLAN_TAG_PRESENT: 201 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); 202 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); 203 204 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 205 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 206 offsetof(struct sk_buff, vlan_tci)); 207 if (skb_field == SKF_AD_VLAN_TAG) { 208 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 209 ~VLAN_TAG_PRESENT); 210 } else { 211 /* dst_reg >>= 12 */ 212 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12); 213 /* dst_reg &= 1 */ 214 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); 215 } 216 break; 217 } 218 219 return insn - insn_buf; 220 } 221 222 static bool convert_bpf_extensions(struct sock_filter *fp, 223 struct bpf_insn **insnp) 224 { 225 struct bpf_insn *insn = *insnp; 226 u32 cnt; 227 228 switch (fp->k) { 229 case SKF_AD_OFF + SKF_AD_PROTOCOL: 230 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); 231 232 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 233 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 234 offsetof(struct sk_buff, protocol)); 235 /* A = ntohs(A) [emitting a nop or swap16] */ 236 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 237 break; 238 239 case SKF_AD_OFF + SKF_AD_PKTTYPE: 240 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 241 insn += cnt - 1; 242 break; 243 244 case SKF_AD_OFF + SKF_AD_IFINDEX: 245 case SKF_AD_OFF + SKF_AD_HATYPE: 246 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); 247 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); 248 249 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 250 BPF_REG_TMP, BPF_REG_CTX, 251 offsetof(struct sk_buff, dev)); 252 /* if (tmp != 0) goto pc + 1 */ 253 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 254 *insn++ = BPF_EXIT_INSN(); 255 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 256 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 257 offsetof(struct net_device, ifindex)); 258 else 259 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 260 offsetof(struct net_device, type)); 261 break; 262 263 case SKF_AD_OFF + SKF_AD_MARK: 264 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 265 insn += cnt - 1; 266 break; 267 268 case SKF_AD_OFF + SKF_AD_RXHASH: 269 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); 270 271 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 272 offsetof(struct sk_buff, hash)); 273 break; 274 275 case SKF_AD_OFF + SKF_AD_QUEUE: 276 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 277 insn += cnt - 1; 278 break; 279 280 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 281 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 282 BPF_REG_A, BPF_REG_CTX, insn); 283 insn += cnt - 1; 284 break; 285 286 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 287 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 288 BPF_REG_A, BPF_REG_CTX, insn); 289 insn += cnt - 1; 290 break; 291 292 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 293 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); 294 295 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 296 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 297 offsetof(struct sk_buff, vlan_proto)); 298 /* A = ntohs(A) [emitting a nop or swap16] */ 299 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 300 break; 301 302 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 303 case SKF_AD_OFF + SKF_AD_NLATTR: 304 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 305 case SKF_AD_OFF + SKF_AD_CPU: 306 case SKF_AD_OFF + SKF_AD_RANDOM: 307 /* arg1 = CTX */ 308 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 309 /* arg2 = A */ 310 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 311 /* arg3 = X */ 312 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 313 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 314 switch (fp->k) { 315 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 316 *insn = BPF_EMIT_CALL(__skb_get_pay_offset); 317 break; 318 case SKF_AD_OFF + SKF_AD_NLATTR: 319 *insn = BPF_EMIT_CALL(__skb_get_nlattr); 320 break; 321 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 322 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest); 323 break; 324 case SKF_AD_OFF + SKF_AD_CPU: 325 *insn = BPF_EMIT_CALL(__get_raw_cpu_id); 326 break; 327 case SKF_AD_OFF + SKF_AD_RANDOM: 328 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 329 bpf_user_rnd_init_once(); 330 break; 331 } 332 break; 333 334 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 335 /* A ^= X */ 336 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 337 break; 338 339 default: 340 /* This is just a dummy call to avoid letting the compiler 341 * evict __bpf_call_base() as an optimization. Placed here 342 * where no-one bothers. 343 */ 344 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 345 return false; 346 } 347 348 *insnp = insn; 349 return true; 350 } 351 352 /** 353 * bpf_convert_filter - convert filter program 354 * @prog: the user passed filter program 355 * @len: the length of the user passed filter program 356 * @new_prog: allocated 'struct bpf_prog' or NULL 357 * @new_len: pointer to store length of converted program 358 * 359 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 360 * style extended BPF (eBPF). 361 * Conversion workflow: 362 * 363 * 1) First pass for calculating the new program length: 364 * bpf_convert_filter(old_prog, old_len, NULL, &new_len) 365 * 366 * 2) 2nd pass to remap in two passes: 1st pass finds new 367 * jump offsets, 2nd pass remapping: 368 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len); 369 */ 370 static int bpf_convert_filter(struct sock_filter *prog, int len, 371 struct bpf_prog *new_prog, int *new_len) 372 { 373 int new_flen = 0, pass = 0, target, i, stack_off; 374 struct bpf_insn *new_insn, *first_insn = NULL; 375 struct sock_filter *fp; 376 int *addrs = NULL; 377 u8 bpf_src; 378 379 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 380 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 381 382 if (len <= 0 || len > BPF_MAXINSNS) 383 return -EINVAL; 384 385 if (new_prog) { 386 first_insn = new_prog->insnsi; 387 addrs = kcalloc(len, sizeof(*addrs), 388 GFP_KERNEL | __GFP_NOWARN); 389 if (!addrs) 390 return -ENOMEM; 391 } 392 393 do_pass: 394 new_insn = first_insn; 395 fp = prog; 396 397 /* Classic BPF related prologue emission. */ 398 if (new_prog) { 399 /* Classic BPF expects A and X to be reset first. These need 400 * to be guaranteed to be the first two instructions. 401 */ 402 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 403 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 404 405 /* All programs must keep CTX in callee saved BPF_REG_CTX. 406 * In eBPF case it's done by the compiler, here we need to 407 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 408 */ 409 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 410 } else { 411 new_insn += 3; 412 } 413 414 for (i = 0; i < len; fp++, i++) { 415 struct bpf_insn tmp_insns[6] = { }; 416 struct bpf_insn *insn = tmp_insns; 417 418 if (addrs) 419 addrs[i] = new_insn - first_insn; 420 421 switch (fp->code) { 422 /* All arithmetic insns and skb loads map as-is. */ 423 case BPF_ALU | BPF_ADD | BPF_X: 424 case BPF_ALU | BPF_ADD | BPF_K: 425 case BPF_ALU | BPF_SUB | BPF_X: 426 case BPF_ALU | BPF_SUB | BPF_K: 427 case BPF_ALU | BPF_AND | BPF_X: 428 case BPF_ALU | BPF_AND | BPF_K: 429 case BPF_ALU | BPF_OR | BPF_X: 430 case BPF_ALU | BPF_OR | BPF_K: 431 case BPF_ALU | BPF_LSH | BPF_X: 432 case BPF_ALU | BPF_LSH | BPF_K: 433 case BPF_ALU | BPF_RSH | BPF_X: 434 case BPF_ALU | BPF_RSH | BPF_K: 435 case BPF_ALU | BPF_XOR | BPF_X: 436 case BPF_ALU | BPF_XOR | BPF_K: 437 case BPF_ALU | BPF_MUL | BPF_X: 438 case BPF_ALU | BPF_MUL | BPF_K: 439 case BPF_ALU | BPF_DIV | BPF_X: 440 case BPF_ALU | BPF_DIV | BPF_K: 441 case BPF_ALU | BPF_MOD | BPF_X: 442 case BPF_ALU | BPF_MOD | BPF_K: 443 case BPF_ALU | BPF_NEG: 444 case BPF_LD | BPF_ABS | BPF_W: 445 case BPF_LD | BPF_ABS | BPF_H: 446 case BPF_LD | BPF_ABS | BPF_B: 447 case BPF_LD | BPF_IND | BPF_W: 448 case BPF_LD | BPF_IND | BPF_H: 449 case BPF_LD | BPF_IND | BPF_B: 450 /* Check for overloaded BPF extension and 451 * directly convert it if found, otherwise 452 * just move on with mapping. 453 */ 454 if (BPF_CLASS(fp->code) == BPF_LD && 455 BPF_MODE(fp->code) == BPF_ABS && 456 convert_bpf_extensions(fp, &insn)) 457 break; 458 459 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 460 break; 461 462 /* Jump transformation cannot use BPF block macros 463 * everywhere as offset calculation and target updates 464 * require a bit more work than the rest, i.e. jump 465 * opcodes map as-is, but offsets need adjustment. 466 */ 467 468 #define BPF_EMIT_JMP \ 469 do { \ 470 if (target >= len || target < 0) \ 471 goto err; \ 472 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 473 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 474 insn->off -= insn - tmp_insns; \ 475 } while (0) 476 477 case BPF_JMP | BPF_JA: 478 target = i + fp->k + 1; 479 insn->code = fp->code; 480 BPF_EMIT_JMP; 481 break; 482 483 case BPF_JMP | BPF_JEQ | BPF_K: 484 case BPF_JMP | BPF_JEQ | BPF_X: 485 case BPF_JMP | BPF_JSET | BPF_K: 486 case BPF_JMP | BPF_JSET | BPF_X: 487 case BPF_JMP | BPF_JGT | BPF_K: 488 case BPF_JMP | BPF_JGT | BPF_X: 489 case BPF_JMP | BPF_JGE | BPF_K: 490 case BPF_JMP | BPF_JGE | BPF_X: 491 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 492 /* BPF immediates are signed, zero extend 493 * immediate into tmp register and use it 494 * in compare insn. 495 */ 496 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 497 498 insn->dst_reg = BPF_REG_A; 499 insn->src_reg = BPF_REG_TMP; 500 bpf_src = BPF_X; 501 } else { 502 insn->dst_reg = BPF_REG_A; 503 insn->imm = fp->k; 504 bpf_src = BPF_SRC(fp->code); 505 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 506 } 507 508 /* Common case where 'jump_false' is next insn. */ 509 if (fp->jf == 0) { 510 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 511 target = i + fp->jt + 1; 512 BPF_EMIT_JMP; 513 break; 514 } 515 516 /* Convert JEQ into JNE when 'jump_true' is next insn. */ 517 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) { 518 insn->code = BPF_JMP | BPF_JNE | bpf_src; 519 target = i + fp->jf + 1; 520 BPF_EMIT_JMP; 521 break; 522 } 523 524 /* Other jumps are mapped into two insns: Jxx and JA. */ 525 target = i + fp->jt + 1; 526 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 527 BPF_EMIT_JMP; 528 insn++; 529 530 insn->code = BPF_JMP | BPF_JA; 531 target = i + fp->jf + 1; 532 BPF_EMIT_JMP; 533 break; 534 535 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 536 case BPF_LDX | BPF_MSH | BPF_B: 537 /* tmp = A */ 538 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A); 539 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 540 *insn++ = BPF_LD_ABS(BPF_B, fp->k); 541 /* A &= 0xf */ 542 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 543 /* A <<= 2 */ 544 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 545 /* X = A */ 546 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 547 /* A = tmp */ 548 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 549 break; 550 551 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 552 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 553 */ 554 case BPF_RET | BPF_A: 555 case BPF_RET | BPF_K: 556 if (BPF_RVAL(fp->code) == BPF_K) 557 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 558 0, fp->k); 559 *insn = BPF_EXIT_INSN(); 560 break; 561 562 /* Store to stack. */ 563 case BPF_ST: 564 case BPF_STX: 565 stack_off = fp->k * 4 + 4; 566 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 567 BPF_ST ? BPF_REG_A : BPF_REG_X, 568 -stack_off); 569 /* check_load_and_stores() verifies that classic BPF can 570 * load from stack only after write, so tracking 571 * stack_depth for ST|STX insns is enough 572 */ 573 if (new_prog && new_prog->aux->stack_depth < stack_off) 574 new_prog->aux->stack_depth = stack_off; 575 break; 576 577 /* Load from stack. */ 578 case BPF_LD | BPF_MEM: 579 case BPF_LDX | BPF_MEM: 580 stack_off = fp->k * 4 + 4; 581 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 582 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 583 -stack_off); 584 break; 585 586 /* A = K or X = K */ 587 case BPF_LD | BPF_IMM: 588 case BPF_LDX | BPF_IMM: 589 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 590 BPF_REG_A : BPF_REG_X, fp->k); 591 break; 592 593 /* X = A */ 594 case BPF_MISC | BPF_TAX: 595 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 596 break; 597 598 /* A = X */ 599 case BPF_MISC | BPF_TXA: 600 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 601 break; 602 603 /* A = skb->len or X = skb->len */ 604 case BPF_LD | BPF_W | BPF_LEN: 605 case BPF_LDX | BPF_W | BPF_LEN: 606 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 607 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 608 offsetof(struct sk_buff, len)); 609 break; 610 611 /* Access seccomp_data fields. */ 612 case BPF_LDX | BPF_ABS | BPF_W: 613 /* A = *(u32 *) (ctx + K) */ 614 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 615 break; 616 617 /* Unknown instruction. */ 618 default: 619 goto err; 620 } 621 622 insn++; 623 if (new_prog) 624 memcpy(new_insn, tmp_insns, 625 sizeof(*insn) * (insn - tmp_insns)); 626 new_insn += insn - tmp_insns; 627 } 628 629 if (!new_prog) { 630 /* Only calculating new length. */ 631 *new_len = new_insn - first_insn; 632 return 0; 633 } 634 635 pass++; 636 if (new_flen != new_insn - first_insn) { 637 new_flen = new_insn - first_insn; 638 if (pass > 2) 639 goto err; 640 goto do_pass; 641 } 642 643 kfree(addrs); 644 BUG_ON(*new_len != new_flen); 645 return 0; 646 err: 647 kfree(addrs); 648 return -EINVAL; 649 } 650 651 /* Security: 652 * 653 * As we dont want to clear mem[] array for each packet going through 654 * __bpf_prog_run(), we check that filter loaded by user never try to read 655 * a cell if not previously written, and we check all branches to be sure 656 * a malicious user doesn't try to abuse us. 657 */ 658 static int check_load_and_stores(const struct sock_filter *filter, int flen) 659 { 660 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 661 int pc, ret = 0; 662 663 BUILD_BUG_ON(BPF_MEMWORDS > 16); 664 665 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 666 if (!masks) 667 return -ENOMEM; 668 669 memset(masks, 0xff, flen * sizeof(*masks)); 670 671 for (pc = 0; pc < flen; pc++) { 672 memvalid &= masks[pc]; 673 674 switch (filter[pc].code) { 675 case BPF_ST: 676 case BPF_STX: 677 memvalid |= (1 << filter[pc].k); 678 break; 679 case BPF_LD | BPF_MEM: 680 case BPF_LDX | BPF_MEM: 681 if (!(memvalid & (1 << filter[pc].k))) { 682 ret = -EINVAL; 683 goto error; 684 } 685 break; 686 case BPF_JMP | BPF_JA: 687 /* A jump must set masks on target */ 688 masks[pc + 1 + filter[pc].k] &= memvalid; 689 memvalid = ~0; 690 break; 691 case BPF_JMP | BPF_JEQ | BPF_K: 692 case BPF_JMP | BPF_JEQ | BPF_X: 693 case BPF_JMP | BPF_JGE | BPF_K: 694 case BPF_JMP | BPF_JGE | BPF_X: 695 case BPF_JMP | BPF_JGT | BPF_K: 696 case BPF_JMP | BPF_JGT | BPF_X: 697 case BPF_JMP | BPF_JSET | BPF_K: 698 case BPF_JMP | BPF_JSET | BPF_X: 699 /* A jump must set masks on targets */ 700 masks[pc + 1 + filter[pc].jt] &= memvalid; 701 masks[pc + 1 + filter[pc].jf] &= memvalid; 702 memvalid = ~0; 703 break; 704 } 705 } 706 error: 707 kfree(masks); 708 return ret; 709 } 710 711 static bool chk_code_allowed(u16 code_to_probe) 712 { 713 static const bool codes[] = { 714 /* 32 bit ALU operations */ 715 [BPF_ALU | BPF_ADD | BPF_K] = true, 716 [BPF_ALU | BPF_ADD | BPF_X] = true, 717 [BPF_ALU | BPF_SUB | BPF_K] = true, 718 [BPF_ALU | BPF_SUB | BPF_X] = true, 719 [BPF_ALU | BPF_MUL | BPF_K] = true, 720 [BPF_ALU | BPF_MUL | BPF_X] = true, 721 [BPF_ALU | BPF_DIV | BPF_K] = true, 722 [BPF_ALU | BPF_DIV | BPF_X] = true, 723 [BPF_ALU | BPF_MOD | BPF_K] = true, 724 [BPF_ALU | BPF_MOD | BPF_X] = true, 725 [BPF_ALU | BPF_AND | BPF_K] = true, 726 [BPF_ALU | BPF_AND | BPF_X] = true, 727 [BPF_ALU | BPF_OR | BPF_K] = true, 728 [BPF_ALU | BPF_OR | BPF_X] = true, 729 [BPF_ALU | BPF_XOR | BPF_K] = true, 730 [BPF_ALU | BPF_XOR | BPF_X] = true, 731 [BPF_ALU | BPF_LSH | BPF_K] = true, 732 [BPF_ALU | BPF_LSH | BPF_X] = true, 733 [BPF_ALU | BPF_RSH | BPF_K] = true, 734 [BPF_ALU | BPF_RSH | BPF_X] = true, 735 [BPF_ALU | BPF_NEG] = true, 736 /* Load instructions */ 737 [BPF_LD | BPF_W | BPF_ABS] = true, 738 [BPF_LD | BPF_H | BPF_ABS] = true, 739 [BPF_LD | BPF_B | BPF_ABS] = true, 740 [BPF_LD | BPF_W | BPF_LEN] = true, 741 [BPF_LD | BPF_W | BPF_IND] = true, 742 [BPF_LD | BPF_H | BPF_IND] = true, 743 [BPF_LD | BPF_B | BPF_IND] = true, 744 [BPF_LD | BPF_IMM] = true, 745 [BPF_LD | BPF_MEM] = true, 746 [BPF_LDX | BPF_W | BPF_LEN] = true, 747 [BPF_LDX | BPF_B | BPF_MSH] = true, 748 [BPF_LDX | BPF_IMM] = true, 749 [BPF_LDX | BPF_MEM] = true, 750 /* Store instructions */ 751 [BPF_ST] = true, 752 [BPF_STX] = true, 753 /* Misc instructions */ 754 [BPF_MISC | BPF_TAX] = true, 755 [BPF_MISC | BPF_TXA] = true, 756 /* Return instructions */ 757 [BPF_RET | BPF_K] = true, 758 [BPF_RET | BPF_A] = true, 759 /* Jump instructions */ 760 [BPF_JMP | BPF_JA] = true, 761 [BPF_JMP | BPF_JEQ | BPF_K] = true, 762 [BPF_JMP | BPF_JEQ | BPF_X] = true, 763 [BPF_JMP | BPF_JGE | BPF_K] = true, 764 [BPF_JMP | BPF_JGE | BPF_X] = true, 765 [BPF_JMP | BPF_JGT | BPF_K] = true, 766 [BPF_JMP | BPF_JGT | BPF_X] = true, 767 [BPF_JMP | BPF_JSET | BPF_K] = true, 768 [BPF_JMP | BPF_JSET | BPF_X] = true, 769 }; 770 771 if (code_to_probe >= ARRAY_SIZE(codes)) 772 return false; 773 774 return codes[code_to_probe]; 775 } 776 777 static bool bpf_check_basics_ok(const struct sock_filter *filter, 778 unsigned int flen) 779 { 780 if (filter == NULL) 781 return false; 782 if (flen == 0 || flen > BPF_MAXINSNS) 783 return false; 784 785 return true; 786 } 787 788 /** 789 * bpf_check_classic - verify socket filter code 790 * @filter: filter to verify 791 * @flen: length of filter 792 * 793 * Check the user's filter code. If we let some ugly 794 * filter code slip through kaboom! The filter must contain 795 * no references or jumps that are out of range, no illegal 796 * instructions, and must end with a RET instruction. 797 * 798 * All jumps are forward as they are not signed. 799 * 800 * Returns 0 if the rule set is legal or -EINVAL if not. 801 */ 802 static int bpf_check_classic(const struct sock_filter *filter, 803 unsigned int flen) 804 { 805 bool anc_found; 806 int pc; 807 808 /* Check the filter code now */ 809 for (pc = 0; pc < flen; pc++) { 810 const struct sock_filter *ftest = &filter[pc]; 811 812 /* May we actually operate on this code? */ 813 if (!chk_code_allowed(ftest->code)) 814 return -EINVAL; 815 816 /* Some instructions need special checks */ 817 switch (ftest->code) { 818 case BPF_ALU | BPF_DIV | BPF_K: 819 case BPF_ALU | BPF_MOD | BPF_K: 820 /* Check for division by zero */ 821 if (ftest->k == 0) 822 return -EINVAL; 823 break; 824 case BPF_ALU | BPF_LSH | BPF_K: 825 case BPF_ALU | BPF_RSH | BPF_K: 826 if (ftest->k >= 32) 827 return -EINVAL; 828 break; 829 case BPF_LD | BPF_MEM: 830 case BPF_LDX | BPF_MEM: 831 case BPF_ST: 832 case BPF_STX: 833 /* Check for invalid memory addresses */ 834 if (ftest->k >= BPF_MEMWORDS) 835 return -EINVAL; 836 break; 837 case BPF_JMP | BPF_JA: 838 /* Note, the large ftest->k might cause loops. 839 * Compare this with conditional jumps below, 840 * where offsets are limited. --ANK (981016) 841 */ 842 if (ftest->k >= (unsigned int)(flen - pc - 1)) 843 return -EINVAL; 844 break; 845 case BPF_JMP | BPF_JEQ | BPF_K: 846 case BPF_JMP | BPF_JEQ | BPF_X: 847 case BPF_JMP | BPF_JGE | BPF_K: 848 case BPF_JMP | BPF_JGE | BPF_X: 849 case BPF_JMP | BPF_JGT | BPF_K: 850 case BPF_JMP | BPF_JGT | BPF_X: 851 case BPF_JMP | BPF_JSET | BPF_K: 852 case BPF_JMP | BPF_JSET | BPF_X: 853 /* Both conditionals must be safe */ 854 if (pc + ftest->jt + 1 >= flen || 855 pc + ftest->jf + 1 >= flen) 856 return -EINVAL; 857 break; 858 case BPF_LD | BPF_W | BPF_ABS: 859 case BPF_LD | BPF_H | BPF_ABS: 860 case BPF_LD | BPF_B | BPF_ABS: 861 anc_found = false; 862 if (bpf_anc_helper(ftest) & BPF_ANC) 863 anc_found = true; 864 /* Ancillary operation unknown or unsupported */ 865 if (anc_found == false && ftest->k >= SKF_AD_OFF) 866 return -EINVAL; 867 } 868 } 869 870 /* Last instruction must be a RET code */ 871 switch (filter[flen - 1].code) { 872 case BPF_RET | BPF_K: 873 case BPF_RET | BPF_A: 874 return check_load_and_stores(filter, flen); 875 } 876 877 return -EINVAL; 878 } 879 880 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 881 const struct sock_fprog *fprog) 882 { 883 unsigned int fsize = bpf_classic_proglen(fprog); 884 struct sock_fprog_kern *fkprog; 885 886 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 887 if (!fp->orig_prog) 888 return -ENOMEM; 889 890 fkprog = fp->orig_prog; 891 fkprog->len = fprog->len; 892 893 fkprog->filter = kmemdup(fp->insns, fsize, 894 GFP_KERNEL | __GFP_NOWARN); 895 if (!fkprog->filter) { 896 kfree(fp->orig_prog); 897 return -ENOMEM; 898 } 899 900 return 0; 901 } 902 903 static void bpf_release_orig_filter(struct bpf_prog *fp) 904 { 905 struct sock_fprog_kern *fprog = fp->orig_prog; 906 907 if (fprog) { 908 kfree(fprog->filter); 909 kfree(fprog); 910 } 911 } 912 913 static void __bpf_prog_release(struct bpf_prog *prog) 914 { 915 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 916 bpf_prog_put(prog); 917 } else { 918 bpf_release_orig_filter(prog); 919 bpf_prog_free(prog); 920 } 921 } 922 923 static void __sk_filter_release(struct sk_filter *fp) 924 { 925 __bpf_prog_release(fp->prog); 926 kfree(fp); 927 } 928 929 /** 930 * sk_filter_release_rcu - Release a socket filter by rcu_head 931 * @rcu: rcu_head that contains the sk_filter to free 932 */ 933 static void sk_filter_release_rcu(struct rcu_head *rcu) 934 { 935 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 936 937 __sk_filter_release(fp); 938 } 939 940 /** 941 * sk_filter_release - release a socket filter 942 * @fp: filter to remove 943 * 944 * Remove a filter from a socket and release its resources. 945 */ 946 static void sk_filter_release(struct sk_filter *fp) 947 { 948 if (refcount_dec_and_test(&fp->refcnt)) 949 call_rcu(&fp->rcu, sk_filter_release_rcu); 950 } 951 952 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 953 { 954 u32 filter_size = bpf_prog_size(fp->prog->len); 955 956 atomic_sub(filter_size, &sk->sk_omem_alloc); 957 sk_filter_release(fp); 958 } 959 960 /* try to charge the socket memory if there is space available 961 * return true on success 962 */ 963 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 964 { 965 u32 filter_size = bpf_prog_size(fp->prog->len); 966 967 /* same check as in sock_kmalloc() */ 968 if (filter_size <= sysctl_optmem_max && 969 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { 970 atomic_add(filter_size, &sk->sk_omem_alloc); 971 return true; 972 } 973 return false; 974 } 975 976 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 977 { 978 bool ret = __sk_filter_charge(sk, fp); 979 if (ret) 980 refcount_inc(&fp->refcnt); 981 return ret; 982 } 983 984 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 985 { 986 struct sock_filter *old_prog; 987 struct bpf_prog *old_fp; 988 int err, new_len, old_len = fp->len; 989 990 /* We are free to overwrite insns et al right here as it 991 * won't be used at this point in time anymore internally 992 * after the migration to the internal BPF instruction 993 * representation. 994 */ 995 BUILD_BUG_ON(sizeof(struct sock_filter) != 996 sizeof(struct bpf_insn)); 997 998 /* Conversion cannot happen on overlapping memory areas, 999 * so we need to keep the user BPF around until the 2nd 1000 * pass. At this time, the user BPF is stored in fp->insns. 1001 */ 1002 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1003 GFP_KERNEL | __GFP_NOWARN); 1004 if (!old_prog) { 1005 err = -ENOMEM; 1006 goto out_err; 1007 } 1008 1009 /* 1st pass: calculate the new program length. */ 1010 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len); 1011 if (err) 1012 goto out_err_free; 1013 1014 /* Expand fp for appending the new filter representation. */ 1015 old_fp = fp; 1016 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1017 if (!fp) { 1018 /* The old_fp is still around in case we couldn't 1019 * allocate new memory, so uncharge on that one. 1020 */ 1021 fp = old_fp; 1022 err = -ENOMEM; 1023 goto out_err_free; 1024 } 1025 1026 fp->len = new_len; 1027 1028 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1029 err = bpf_convert_filter(old_prog, old_len, fp, &new_len); 1030 if (err) 1031 /* 2nd bpf_convert_filter() can fail only if it fails 1032 * to allocate memory, remapping must succeed. Note, 1033 * that at this time old_fp has already been released 1034 * by krealloc(). 1035 */ 1036 goto out_err_free; 1037 1038 /* We are guaranteed to never error here with cBPF to eBPF 1039 * transitions, since there's no issue with type compatibility 1040 * checks on program arrays. 1041 */ 1042 fp = bpf_prog_select_runtime(fp, &err); 1043 1044 kfree(old_prog); 1045 return fp; 1046 1047 out_err_free: 1048 kfree(old_prog); 1049 out_err: 1050 __bpf_prog_release(fp); 1051 return ERR_PTR(err); 1052 } 1053 1054 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1055 bpf_aux_classic_check_t trans) 1056 { 1057 int err; 1058 1059 fp->bpf_func = NULL; 1060 fp->jited = 0; 1061 1062 err = bpf_check_classic(fp->insns, fp->len); 1063 if (err) { 1064 __bpf_prog_release(fp); 1065 return ERR_PTR(err); 1066 } 1067 1068 /* There might be additional checks and transformations 1069 * needed on classic filters, f.e. in case of seccomp. 1070 */ 1071 if (trans) { 1072 err = trans(fp->insns, fp->len); 1073 if (err) { 1074 __bpf_prog_release(fp); 1075 return ERR_PTR(err); 1076 } 1077 } 1078 1079 /* Probe if we can JIT compile the filter and if so, do 1080 * the compilation of the filter. 1081 */ 1082 bpf_jit_compile(fp); 1083 1084 /* JIT compiler couldn't process this filter, so do the 1085 * internal BPF translation for the optimized interpreter. 1086 */ 1087 if (!fp->jited) 1088 fp = bpf_migrate_filter(fp); 1089 1090 return fp; 1091 } 1092 1093 /** 1094 * bpf_prog_create - create an unattached filter 1095 * @pfp: the unattached filter that is created 1096 * @fprog: the filter program 1097 * 1098 * Create a filter independent of any socket. We first run some 1099 * sanity checks on it to make sure it does not explode on us later. 1100 * If an error occurs or there is insufficient memory for the filter 1101 * a negative errno code is returned. On success the return is zero. 1102 */ 1103 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1104 { 1105 unsigned int fsize = bpf_classic_proglen(fprog); 1106 struct bpf_prog *fp; 1107 1108 /* Make sure new filter is there and in the right amounts. */ 1109 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1110 return -EINVAL; 1111 1112 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1113 if (!fp) 1114 return -ENOMEM; 1115 1116 memcpy(fp->insns, fprog->filter, fsize); 1117 1118 fp->len = fprog->len; 1119 /* Since unattached filters are not copied back to user 1120 * space through sk_get_filter(), we do not need to hold 1121 * a copy here, and can spare us the work. 1122 */ 1123 fp->orig_prog = NULL; 1124 1125 /* bpf_prepare_filter() already takes care of freeing 1126 * memory in case something goes wrong. 1127 */ 1128 fp = bpf_prepare_filter(fp, NULL); 1129 if (IS_ERR(fp)) 1130 return PTR_ERR(fp); 1131 1132 *pfp = fp; 1133 return 0; 1134 } 1135 EXPORT_SYMBOL_GPL(bpf_prog_create); 1136 1137 /** 1138 * bpf_prog_create_from_user - create an unattached filter from user buffer 1139 * @pfp: the unattached filter that is created 1140 * @fprog: the filter program 1141 * @trans: post-classic verifier transformation handler 1142 * @save_orig: save classic BPF program 1143 * 1144 * This function effectively does the same as bpf_prog_create(), only 1145 * that it builds up its insns buffer from user space provided buffer. 1146 * It also allows for passing a bpf_aux_classic_check_t handler. 1147 */ 1148 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1149 bpf_aux_classic_check_t trans, bool save_orig) 1150 { 1151 unsigned int fsize = bpf_classic_proglen(fprog); 1152 struct bpf_prog *fp; 1153 int err; 1154 1155 /* Make sure new filter is there and in the right amounts. */ 1156 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1157 return -EINVAL; 1158 1159 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1160 if (!fp) 1161 return -ENOMEM; 1162 1163 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1164 __bpf_prog_free(fp); 1165 return -EFAULT; 1166 } 1167 1168 fp->len = fprog->len; 1169 fp->orig_prog = NULL; 1170 1171 if (save_orig) { 1172 err = bpf_prog_store_orig_filter(fp, fprog); 1173 if (err) { 1174 __bpf_prog_free(fp); 1175 return -ENOMEM; 1176 } 1177 } 1178 1179 /* bpf_prepare_filter() already takes care of freeing 1180 * memory in case something goes wrong. 1181 */ 1182 fp = bpf_prepare_filter(fp, trans); 1183 if (IS_ERR(fp)) 1184 return PTR_ERR(fp); 1185 1186 *pfp = fp; 1187 return 0; 1188 } 1189 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1190 1191 void bpf_prog_destroy(struct bpf_prog *fp) 1192 { 1193 __bpf_prog_release(fp); 1194 } 1195 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1196 1197 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1198 { 1199 struct sk_filter *fp, *old_fp; 1200 1201 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1202 if (!fp) 1203 return -ENOMEM; 1204 1205 fp->prog = prog; 1206 1207 if (!__sk_filter_charge(sk, fp)) { 1208 kfree(fp); 1209 return -ENOMEM; 1210 } 1211 refcount_set(&fp->refcnt, 1); 1212 1213 old_fp = rcu_dereference_protected(sk->sk_filter, 1214 lockdep_sock_is_held(sk)); 1215 rcu_assign_pointer(sk->sk_filter, fp); 1216 1217 if (old_fp) 1218 sk_filter_uncharge(sk, old_fp); 1219 1220 return 0; 1221 } 1222 1223 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk) 1224 { 1225 struct bpf_prog *old_prog; 1226 int err; 1227 1228 if (bpf_prog_size(prog->len) > sysctl_optmem_max) 1229 return -ENOMEM; 1230 1231 if (sk_unhashed(sk) && sk->sk_reuseport) { 1232 err = reuseport_alloc(sk); 1233 if (err) 1234 return err; 1235 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) { 1236 /* The socket wasn't bound with SO_REUSEPORT */ 1237 return -EINVAL; 1238 } 1239 1240 old_prog = reuseport_attach_prog(sk, prog); 1241 if (old_prog) 1242 bpf_prog_destroy(old_prog); 1243 1244 return 0; 1245 } 1246 1247 static 1248 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1249 { 1250 unsigned int fsize = bpf_classic_proglen(fprog); 1251 struct bpf_prog *prog; 1252 int err; 1253 1254 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1255 return ERR_PTR(-EPERM); 1256 1257 /* Make sure new filter is there and in the right amounts. */ 1258 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1259 return ERR_PTR(-EINVAL); 1260 1261 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1262 if (!prog) 1263 return ERR_PTR(-ENOMEM); 1264 1265 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1266 __bpf_prog_free(prog); 1267 return ERR_PTR(-EFAULT); 1268 } 1269 1270 prog->len = fprog->len; 1271 1272 err = bpf_prog_store_orig_filter(prog, fprog); 1273 if (err) { 1274 __bpf_prog_free(prog); 1275 return ERR_PTR(-ENOMEM); 1276 } 1277 1278 /* bpf_prepare_filter() already takes care of freeing 1279 * memory in case something goes wrong. 1280 */ 1281 return bpf_prepare_filter(prog, NULL); 1282 } 1283 1284 /** 1285 * sk_attach_filter - attach a socket filter 1286 * @fprog: the filter program 1287 * @sk: the socket to use 1288 * 1289 * Attach the user's filter code. We first run some sanity checks on 1290 * it to make sure it does not explode on us later. If an error 1291 * occurs or there is insufficient memory for the filter a negative 1292 * errno code is returned. On success the return is zero. 1293 */ 1294 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1295 { 1296 struct bpf_prog *prog = __get_filter(fprog, sk); 1297 int err; 1298 1299 if (IS_ERR(prog)) 1300 return PTR_ERR(prog); 1301 1302 err = __sk_attach_prog(prog, sk); 1303 if (err < 0) { 1304 __bpf_prog_release(prog); 1305 return err; 1306 } 1307 1308 return 0; 1309 } 1310 EXPORT_SYMBOL_GPL(sk_attach_filter); 1311 1312 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1313 { 1314 struct bpf_prog *prog = __get_filter(fprog, sk); 1315 int err; 1316 1317 if (IS_ERR(prog)) 1318 return PTR_ERR(prog); 1319 1320 err = __reuseport_attach_prog(prog, sk); 1321 if (err < 0) { 1322 __bpf_prog_release(prog); 1323 return err; 1324 } 1325 1326 return 0; 1327 } 1328 1329 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1330 { 1331 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1332 return ERR_PTR(-EPERM); 1333 1334 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1335 } 1336 1337 int sk_attach_bpf(u32 ufd, struct sock *sk) 1338 { 1339 struct bpf_prog *prog = __get_bpf(ufd, sk); 1340 int err; 1341 1342 if (IS_ERR(prog)) 1343 return PTR_ERR(prog); 1344 1345 err = __sk_attach_prog(prog, sk); 1346 if (err < 0) { 1347 bpf_prog_put(prog); 1348 return err; 1349 } 1350 1351 return 0; 1352 } 1353 1354 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1355 { 1356 struct bpf_prog *prog = __get_bpf(ufd, sk); 1357 int err; 1358 1359 if (IS_ERR(prog)) 1360 return PTR_ERR(prog); 1361 1362 err = __reuseport_attach_prog(prog, sk); 1363 if (err < 0) { 1364 bpf_prog_put(prog); 1365 return err; 1366 } 1367 1368 return 0; 1369 } 1370 1371 struct bpf_scratchpad { 1372 union { 1373 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1374 u8 buff[MAX_BPF_STACK]; 1375 }; 1376 }; 1377 1378 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1379 1380 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1381 unsigned int write_len) 1382 { 1383 return skb_ensure_writable(skb, write_len); 1384 } 1385 1386 static inline int bpf_try_make_writable(struct sk_buff *skb, 1387 unsigned int write_len) 1388 { 1389 int err = __bpf_try_make_writable(skb, write_len); 1390 1391 bpf_compute_data_end(skb); 1392 return err; 1393 } 1394 1395 static int bpf_try_make_head_writable(struct sk_buff *skb) 1396 { 1397 return bpf_try_make_writable(skb, skb_headlen(skb)); 1398 } 1399 1400 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1401 { 1402 if (skb_at_tc_ingress(skb)) 1403 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1404 } 1405 1406 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1407 { 1408 if (skb_at_tc_ingress(skb)) 1409 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1410 } 1411 1412 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1413 const void *, from, u32, len, u64, flags) 1414 { 1415 void *ptr; 1416 1417 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1418 return -EINVAL; 1419 if (unlikely(offset > 0xffff)) 1420 return -EFAULT; 1421 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1422 return -EFAULT; 1423 1424 ptr = skb->data + offset; 1425 if (flags & BPF_F_RECOMPUTE_CSUM) 1426 __skb_postpull_rcsum(skb, ptr, len, offset); 1427 1428 memcpy(ptr, from, len); 1429 1430 if (flags & BPF_F_RECOMPUTE_CSUM) 1431 __skb_postpush_rcsum(skb, ptr, len, offset); 1432 if (flags & BPF_F_INVALIDATE_HASH) 1433 skb_clear_hash(skb); 1434 1435 return 0; 1436 } 1437 1438 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1439 .func = bpf_skb_store_bytes, 1440 .gpl_only = false, 1441 .ret_type = RET_INTEGER, 1442 .arg1_type = ARG_PTR_TO_CTX, 1443 .arg2_type = ARG_ANYTHING, 1444 .arg3_type = ARG_PTR_TO_MEM, 1445 .arg4_type = ARG_CONST_SIZE, 1446 .arg5_type = ARG_ANYTHING, 1447 }; 1448 1449 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1450 void *, to, u32, len) 1451 { 1452 void *ptr; 1453 1454 if (unlikely(offset > 0xffff)) 1455 goto err_clear; 1456 1457 ptr = skb_header_pointer(skb, offset, len, to); 1458 if (unlikely(!ptr)) 1459 goto err_clear; 1460 if (ptr != to) 1461 memcpy(to, ptr, len); 1462 1463 return 0; 1464 err_clear: 1465 memset(to, 0, len); 1466 return -EFAULT; 1467 } 1468 1469 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1470 .func = bpf_skb_load_bytes, 1471 .gpl_only = false, 1472 .ret_type = RET_INTEGER, 1473 .arg1_type = ARG_PTR_TO_CTX, 1474 .arg2_type = ARG_ANYTHING, 1475 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1476 .arg4_type = ARG_CONST_SIZE, 1477 }; 1478 1479 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1480 { 1481 /* Idea is the following: should the needed direct read/write 1482 * test fail during runtime, we can pull in more data and redo 1483 * again, since implicitly, we invalidate previous checks here. 1484 * 1485 * Or, since we know how much we need to make read/writeable, 1486 * this can be done once at the program beginning for direct 1487 * access case. By this we overcome limitations of only current 1488 * headroom being accessible. 1489 */ 1490 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1491 } 1492 1493 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1494 .func = bpf_skb_pull_data, 1495 .gpl_only = false, 1496 .ret_type = RET_INTEGER, 1497 .arg1_type = ARG_PTR_TO_CTX, 1498 .arg2_type = ARG_ANYTHING, 1499 }; 1500 1501 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1502 u64, from, u64, to, u64, flags) 1503 { 1504 __sum16 *ptr; 1505 1506 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1507 return -EINVAL; 1508 if (unlikely(offset > 0xffff || offset & 1)) 1509 return -EFAULT; 1510 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1511 return -EFAULT; 1512 1513 ptr = (__sum16 *)(skb->data + offset); 1514 switch (flags & BPF_F_HDR_FIELD_MASK) { 1515 case 0: 1516 if (unlikely(from != 0)) 1517 return -EINVAL; 1518 1519 csum_replace_by_diff(ptr, to); 1520 break; 1521 case 2: 1522 csum_replace2(ptr, from, to); 1523 break; 1524 case 4: 1525 csum_replace4(ptr, from, to); 1526 break; 1527 default: 1528 return -EINVAL; 1529 } 1530 1531 return 0; 1532 } 1533 1534 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1535 .func = bpf_l3_csum_replace, 1536 .gpl_only = false, 1537 .ret_type = RET_INTEGER, 1538 .arg1_type = ARG_PTR_TO_CTX, 1539 .arg2_type = ARG_ANYTHING, 1540 .arg3_type = ARG_ANYTHING, 1541 .arg4_type = ARG_ANYTHING, 1542 .arg5_type = ARG_ANYTHING, 1543 }; 1544 1545 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1546 u64, from, u64, to, u64, flags) 1547 { 1548 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1549 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1550 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1551 __sum16 *ptr; 1552 1553 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1554 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1555 return -EINVAL; 1556 if (unlikely(offset > 0xffff || offset & 1)) 1557 return -EFAULT; 1558 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1559 return -EFAULT; 1560 1561 ptr = (__sum16 *)(skb->data + offset); 1562 if (is_mmzero && !do_mforce && !*ptr) 1563 return 0; 1564 1565 switch (flags & BPF_F_HDR_FIELD_MASK) { 1566 case 0: 1567 if (unlikely(from != 0)) 1568 return -EINVAL; 1569 1570 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1571 break; 1572 case 2: 1573 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1574 break; 1575 case 4: 1576 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1577 break; 1578 default: 1579 return -EINVAL; 1580 } 1581 1582 if (is_mmzero && !*ptr) 1583 *ptr = CSUM_MANGLED_0; 1584 return 0; 1585 } 1586 1587 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1588 .func = bpf_l4_csum_replace, 1589 .gpl_only = false, 1590 .ret_type = RET_INTEGER, 1591 .arg1_type = ARG_PTR_TO_CTX, 1592 .arg2_type = ARG_ANYTHING, 1593 .arg3_type = ARG_ANYTHING, 1594 .arg4_type = ARG_ANYTHING, 1595 .arg5_type = ARG_ANYTHING, 1596 }; 1597 1598 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 1599 __be32 *, to, u32, to_size, __wsum, seed) 1600 { 1601 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 1602 u32 diff_size = from_size + to_size; 1603 int i, j = 0; 1604 1605 /* This is quite flexible, some examples: 1606 * 1607 * from_size == 0, to_size > 0, seed := csum --> pushing data 1608 * from_size > 0, to_size == 0, seed := csum --> pulling data 1609 * from_size > 0, to_size > 0, seed := 0 --> diffing data 1610 * 1611 * Even for diffing, from_size and to_size don't need to be equal. 1612 */ 1613 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 1614 diff_size > sizeof(sp->diff))) 1615 return -EINVAL; 1616 1617 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 1618 sp->diff[j] = ~from[i]; 1619 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 1620 sp->diff[j] = to[i]; 1621 1622 return csum_partial(sp->diff, diff_size, seed); 1623 } 1624 1625 static const struct bpf_func_proto bpf_csum_diff_proto = { 1626 .func = bpf_csum_diff, 1627 .gpl_only = false, 1628 .pkt_access = true, 1629 .ret_type = RET_INTEGER, 1630 .arg1_type = ARG_PTR_TO_MEM, 1631 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 1632 .arg3_type = ARG_PTR_TO_MEM, 1633 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 1634 .arg5_type = ARG_ANYTHING, 1635 }; 1636 1637 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 1638 { 1639 /* The interface is to be used in combination with bpf_csum_diff() 1640 * for direct packet writes. csum rotation for alignment as well 1641 * as emulating csum_sub() can be done from the eBPF program. 1642 */ 1643 if (skb->ip_summed == CHECKSUM_COMPLETE) 1644 return (skb->csum = csum_add(skb->csum, csum)); 1645 1646 return -ENOTSUPP; 1647 } 1648 1649 static const struct bpf_func_proto bpf_csum_update_proto = { 1650 .func = bpf_csum_update, 1651 .gpl_only = false, 1652 .ret_type = RET_INTEGER, 1653 .arg1_type = ARG_PTR_TO_CTX, 1654 .arg2_type = ARG_ANYTHING, 1655 }; 1656 1657 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 1658 { 1659 return dev_forward_skb(dev, skb); 1660 } 1661 1662 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 1663 struct sk_buff *skb) 1664 { 1665 int ret = ____dev_forward_skb(dev, skb); 1666 1667 if (likely(!ret)) { 1668 skb->dev = dev; 1669 ret = netif_rx(skb); 1670 } 1671 1672 return ret; 1673 } 1674 1675 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 1676 { 1677 int ret; 1678 1679 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) { 1680 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 1681 kfree_skb(skb); 1682 return -ENETDOWN; 1683 } 1684 1685 skb->dev = dev; 1686 1687 __this_cpu_inc(xmit_recursion); 1688 ret = dev_queue_xmit(skb); 1689 __this_cpu_dec(xmit_recursion); 1690 1691 return ret; 1692 } 1693 1694 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 1695 u32 flags) 1696 { 1697 /* skb->mac_len is not set on normal egress */ 1698 unsigned int mlen = skb->network_header - skb->mac_header; 1699 1700 __skb_pull(skb, mlen); 1701 1702 /* At ingress, the mac header has already been pulled once. 1703 * At egress, skb_pospull_rcsum has to be done in case that 1704 * the skb is originated from ingress (i.e. a forwarded skb) 1705 * to ensure that rcsum starts at net header. 1706 */ 1707 if (!skb_at_tc_ingress(skb)) 1708 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 1709 skb_pop_mac_header(skb); 1710 skb_reset_mac_len(skb); 1711 return flags & BPF_F_INGRESS ? 1712 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 1713 } 1714 1715 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 1716 u32 flags) 1717 { 1718 /* Verify that a link layer header is carried */ 1719 if (unlikely(skb->mac_header >= skb->network_header)) { 1720 kfree_skb(skb); 1721 return -ERANGE; 1722 } 1723 1724 bpf_push_mac_rcsum(skb); 1725 return flags & BPF_F_INGRESS ? 1726 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 1727 } 1728 1729 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 1730 u32 flags) 1731 { 1732 if (dev_is_mac_header_xmit(dev)) 1733 return __bpf_redirect_common(skb, dev, flags); 1734 else 1735 return __bpf_redirect_no_mac(skb, dev, flags); 1736 } 1737 1738 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 1739 { 1740 struct net_device *dev; 1741 struct sk_buff *clone; 1742 int ret; 1743 1744 if (unlikely(flags & ~(BPF_F_INGRESS))) 1745 return -EINVAL; 1746 1747 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 1748 if (unlikely(!dev)) 1749 return -EINVAL; 1750 1751 clone = skb_clone(skb, GFP_ATOMIC); 1752 if (unlikely(!clone)) 1753 return -ENOMEM; 1754 1755 /* For direct write, we need to keep the invariant that the skbs 1756 * we're dealing with need to be uncloned. Should uncloning fail 1757 * here, we need to free the just generated clone to unclone once 1758 * again. 1759 */ 1760 ret = bpf_try_make_head_writable(skb); 1761 if (unlikely(ret)) { 1762 kfree_skb(clone); 1763 return -ENOMEM; 1764 } 1765 1766 return __bpf_redirect(clone, dev, flags); 1767 } 1768 1769 static const struct bpf_func_proto bpf_clone_redirect_proto = { 1770 .func = bpf_clone_redirect, 1771 .gpl_only = false, 1772 .ret_type = RET_INTEGER, 1773 .arg1_type = ARG_PTR_TO_CTX, 1774 .arg2_type = ARG_ANYTHING, 1775 .arg3_type = ARG_ANYTHING, 1776 }; 1777 1778 struct redirect_info { 1779 u32 ifindex; 1780 u32 flags; 1781 }; 1782 1783 static DEFINE_PER_CPU(struct redirect_info, redirect_info); 1784 1785 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 1786 { 1787 struct redirect_info *ri = this_cpu_ptr(&redirect_info); 1788 1789 if (unlikely(flags & ~(BPF_F_INGRESS))) 1790 return TC_ACT_SHOT; 1791 1792 ri->ifindex = ifindex; 1793 ri->flags = flags; 1794 1795 return TC_ACT_REDIRECT; 1796 } 1797 1798 int skb_do_redirect(struct sk_buff *skb) 1799 { 1800 struct redirect_info *ri = this_cpu_ptr(&redirect_info); 1801 struct net_device *dev; 1802 1803 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex); 1804 ri->ifindex = 0; 1805 if (unlikely(!dev)) { 1806 kfree_skb(skb); 1807 return -EINVAL; 1808 } 1809 1810 return __bpf_redirect(skb, dev, ri->flags); 1811 } 1812 1813 static const struct bpf_func_proto bpf_redirect_proto = { 1814 .func = bpf_redirect, 1815 .gpl_only = false, 1816 .ret_type = RET_INTEGER, 1817 .arg1_type = ARG_ANYTHING, 1818 .arg2_type = ARG_ANYTHING, 1819 }; 1820 1821 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 1822 { 1823 return task_get_classid(skb); 1824 } 1825 1826 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 1827 .func = bpf_get_cgroup_classid, 1828 .gpl_only = false, 1829 .ret_type = RET_INTEGER, 1830 .arg1_type = ARG_PTR_TO_CTX, 1831 }; 1832 1833 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 1834 { 1835 return dst_tclassid(skb); 1836 } 1837 1838 static const struct bpf_func_proto bpf_get_route_realm_proto = { 1839 .func = bpf_get_route_realm, 1840 .gpl_only = false, 1841 .ret_type = RET_INTEGER, 1842 .arg1_type = ARG_PTR_TO_CTX, 1843 }; 1844 1845 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 1846 { 1847 /* If skb_clear_hash() was called due to mangling, we can 1848 * trigger SW recalculation here. Later access to hash 1849 * can then use the inline skb->hash via context directly 1850 * instead of calling this helper again. 1851 */ 1852 return skb_get_hash(skb); 1853 } 1854 1855 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 1856 .func = bpf_get_hash_recalc, 1857 .gpl_only = false, 1858 .ret_type = RET_INTEGER, 1859 .arg1_type = ARG_PTR_TO_CTX, 1860 }; 1861 1862 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 1863 { 1864 /* After all direct packet write, this can be used once for 1865 * triggering a lazy recalc on next skb_get_hash() invocation. 1866 */ 1867 skb_clear_hash(skb); 1868 return 0; 1869 } 1870 1871 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 1872 .func = bpf_set_hash_invalid, 1873 .gpl_only = false, 1874 .ret_type = RET_INTEGER, 1875 .arg1_type = ARG_PTR_TO_CTX, 1876 }; 1877 1878 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 1879 { 1880 /* Set user specified hash as L4(+), so that it gets returned 1881 * on skb_get_hash() call unless BPF prog later on triggers a 1882 * skb_clear_hash(). 1883 */ 1884 __skb_set_sw_hash(skb, hash, true); 1885 return 0; 1886 } 1887 1888 static const struct bpf_func_proto bpf_set_hash_proto = { 1889 .func = bpf_set_hash, 1890 .gpl_only = false, 1891 .ret_type = RET_INTEGER, 1892 .arg1_type = ARG_PTR_TO_CTX, 1893 .arg2_type = ARG_ANYTHING, 1894 }; 1895 1896 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 1897 u16, vlan_tci) 1898 { 1899 int ret; 1900 1901 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 1902 vlan_proto != htons(ETH_P_8021AD))) 1903 vlan_proto = htons(ETH_P_8021Q); 1904 1905 bpf_push_mac_rcsum(skb); 1906 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 1907 bpf_pull_mac_rcsum(skb); 1908 1909 bpf_compute_data_end(skb); 1910 return ret; 1911 } 1912 1913 const struct bpf_func_proto bpf_skb_vlan_push_proto = { 1914 .func = bpf_skb_vlan_push, 1915 .gpl_only = false, 1916 .ret_type = RET_INTEGER, 1917 .arg1_type = ARG_PTR_TO_CTX, 1918 .arg2_type = ARG_ANYTHING, 1919 .arg3_type = ARG_ANYTHING, 1920 }; 1921 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto); 1922 1923 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 1924 { 1925 int ret; 1926 1927 bpf_push_mac_rcsum(skb); 1928 ret = skb_vlan_pop(skb); 1929 bpf_pull_mac_rcsum(skb); 1930 1931 bpf_compute_data_end(skb); 1932 return ret; 1933 } 1934 1935 const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 1936 .func = bpf_skb_vlan_pop, 1937 .gpl_only = false, 1938 .ret_type = RET_INTEGER, 1939 .arg1_type = ARG_PTR_TO_CTX, 1940 }; 1941 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto); 1942 1943 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 1944 { 1945 /* Caller already did skb_cow() with len as headroom, 1946 * so no need to do it here. 1947 */ 1948 skb_push(skb, len); 1949 memmove(skb->data, skb->data + len, off); 1950 memset(skb->data + off, 0, len); 1951 1952 /* No skb_postpush_rcsum(skb, skb->data + off, len) 1953 * needed here as it does not change the skb->csum 1954 * result for checksum complete when summing over 1955 * zeroed blocks. 1956 */ 1957 return 0; 1958 } 1959 1960 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 1961 { 1962 /* skb_ensure_writable() is not needed here, as we're 1963 * already working on an uncloned skb. 1964 */ 1965 if (unlikely(!pskb_may_pull(skb, off + len))) 1966 return -ENOMEM; 1967 1968 skb_postpull_rcsum(skb, skb->data + off, len); 1969 memmove(skb->data + len, skb->data, off); 1970 __skb_pull(skb, len); 1971 1972 return 0; 1973 } 1974 1975 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 1976 { 1977 bool trans_same = skb->transport_header == skb->network_header; 1978 int ret; 1979 1980 /* There's no need for __skb_push()/__skb_pull() pair to 1981 * get to the start of the mac header as we're guaranteed 1982 * to always start from here under eBPF. 1983 */ 1984 ret = bpf_skb_generic_push(skb, off, len); 1985 if (likely(!ret)) { 1986 skb->mac_header -= len; 1987 skb->network_header -= len; 1988 if (trans_same) 1989 skb->transport_header = skb->network_header; 1990 } 1991 1992 return ret; 1993 } 1994 1995 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 1996 { 1997 bool trans_same = skb->transport_header == skb->network_header; 1998 int ret; 1999 2000 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 2001 ret = bpf_skb_generic_pop(skb, off, len); 2002 if (likely(!ret)) { 2003 skb->mac_header += len; 2004 skb->network_header += len; 2005 if (trans_same) 2006 skb->transport_header = skb->network_header; 2007 } 2008 2009 return ret; 2010 } 2011 2012 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 2013 { 2014 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2015 u32 off = skb_mac_header_len(skb); 2016 int ret; 2017 2018 ret = skb_cow(skb, len_diff); 2019 if (unlikely(ret < 0)) 2020 return ret; 2021 2022 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 2023 if (unlikely(ret < 0)) 2024 return ret; 2025 2026 if (skb_is_gso(skb)) { 2027 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV4 needs to 2028 * be changed into SKB_GSO_TCPV6. 2029 */ 2030 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 2031 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4; 2032 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6; 2033 } 2034 2035 /* Due to IPv6 header, MSS needs to be downgraded. */ 2036 skb_shinfo(skb)->gso_size -= len_diff; 2037 /* Header must be checked, and gso_segs recomputed. */ 2038 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2039 skb_shinfo(skb)->gso_segs = 0; 2040 } 2041 2042 skb->protocol = htons(ETH_P_IPV6); 2043 skb_clear_hash(skb); 2044 2045 return 0; 2046 } 2047 2048 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 2049 { 2050 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2051 u32 off = skb_mac_header_len(skb); 2052 int ret; 2053 2054 ret = skb_unclone(skb, GFP_ATOMIC); 2055 if (unlikely(ret < 0)) 2056 return ret; 2057 2058 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 2059 if (unlikely(ret < 0)) 2060 return ret; 2061 2062 if (skb_is_gso(skb)) { 2063 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV6 needs to 2064 * be changed into SKB_GSO_TCPV4. 2065 */ 2066 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { 2067 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6; 2068 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4; 2069 } 2070 2071 /* Due to IPv4 header, MSS can be upgraded. */ 2072 skb_shinfo(skb)->gso_size += len_diff; 2073 /* Header must be checked, and gso_segs recomputed. */ 2074 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2075 skb_shinfo(skb)->gso_segs = 0; 2076 } 2077 2078 skb->protocol = htons(ETH_P_IP); 2079 skb_clear_hash(skb); 2080 2081 return 0; 2082 } 2083 2084 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 2085 { 2086 __be16 from_proto = skb->protocol; 2087 2088 if (from_proto == htons(ETH_P_IP) && 2089 to_proto == htons(ETH_P_IPV6)) 2090 return bpf_skb_proto_4_to_6(skb); 2091 2092 if (from_proto == htons(ETH_P_IPV6) && 2093 to_proto == htons(ETH_P_IP)) 2094 return bpf_skb_proto_6_to_4(skb); 2095 2096 return -ENOTSUPP; 2097 } 2098 2099 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 2100 u64, flags) 2101 { 2102 int ret; 2103 2104 if (unlikely(flags)) 2105 return -EINVAL; 2106 2107 /* General idea is that this helper does the basic groundwork 2108 * needed for changing the protocol, and eBPF program fills the 2109 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 2110 * and other helpers, rather than passing a raw buffer here. 2111 * 2112 * The rationale is to keep this minimal and without a need to 2113 * deal with raw packet data. F.e. even if we would pass buffers 2114 * here, the program still needs to call the bpf_lX_csum_replace() 2115 * helpers anyway. Plus, this way we keep also separation of 2116 * concerns, since f.e. bpf_skb_store_bytes() should only take 2117 * care of stores. 2118 * 2119 * Currently, additional options and extension header space are 2120 * not supported, but flags register is reserved so we can adapt 2121 * that. For offloads, we mark packet as dodgy, so that headers 2122 * need to be verified first. 2123 */ 2124 ret = bpf_skb_proto_xlat(skb, proto); 2125 bpf_compute_data_end(skb); 2126 return ret; 2127 } 2128 2129 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 2130 .func = bpf_skb_change_proto, 2131 .gpl_only = false, 2132 .ret_type = RET_INTEGER, 2133 .arg1_type = ARG_PTR_TO_CTX, 2134 .arg2_type = ARG_ANYTHING, 2135 .arg3_type = ARG_ANYTHING, 2136 }; 2137 2138 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 2139 { 2140 /* We only allow a restricted subset to be changed for now. */ 2141 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 2142 !skb_pkt_type_ok(pkt_type))) 2143 return -EINVAL; 2144 2145 skb->pkt_type = pkt_type; 2146 return 0; 2147 } 2148 2149 static const struct bpf_func_proto bpf_skb_change_type_proto = { 2150 .func = bpf_skb_change_type, 2151 .gpl_only = false, 2152 .ret_type = RET_INTEGER, 2153 .arg1_type = ARG_PTR_TO_CTX, 2154 .arg2_type = ARG_ANYTHING, 2155 }; 2156 2157 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 2158 { 2159 switch (skb->protocol) { 2160 case htons(ETH_P_IP): 2161 return sizeof(struct iphdr); 2162 case htons(ETH_P_IPV6): 2163 return sizeof(struct ipv6hdr); 2164 default: 2165 return ~0U; 2166 } 2167 } 2168 2169 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff) 2170 { 2171 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb); 2172 int ret; 2173 2174 ret = skb_cow(skb, len_diff); 2175 if (unlikely(ret < 0)) 2176 return ret; 2177 2178 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 2179 if (unlikely(ret < 0)) 2180 return ret; 2181 2182 if (skb_is_gso(skb)) { 2183 /* Due to header grow, MSS needs to be downgraded. */ 2184 skb_shinfo(skb)->gso_size -= len_diff; 2185 /* Header must be checked, and gso_segs recomputed. */ 2186 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2187 skb_shinfo(skb)->gso_segs = 0; 2188 } 2189 2190 return 0; 2191 } 2192 2193 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff) 2194 { 2195 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb); 2196 int ret; 2197 2198 ret = skb_unclone(skb, GFP_ATOMIC); 2199 if (unlikely(ret < 0)) 2200 return ret; 2201 2202 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 2203 if (unlikely(ret < 0)) 2204 return ret; 2205 2206 if (skb_is_gso(skb)) { 2207 /* Due to header shrink, MSS can be upgraded. */ 2208 skb_shinfo(skb)->gso_size += len_diff; 2209 /* Header must be checked, and gso_segs recomputed. */ 2210 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 2211 skb_shinfo(skb)->gso_segs = 0; 2212 } 2213 2214 return 0; 2215 } 2216 2217 static u32 __bpf_skb_max_len(const struct sk_buff *skb) 2218 { 2219 return skb->dev->mtu + skb->dev->hard_header_len; 2220 } 2221 2222 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff) 2223 { 2224 bool trans_same = skb->transport_header == skb->network_header; 2225 u32 len_cur, len_diff_abs = abs(len_diff); 2226 u32 len_min = bpf_skb_net_base_len(skb); 2227 u32 len_max = __bpf_skb_max_len(skb); 2228 __be16 proto = skb->protocol; 2229 bool shrink = len_diff < 0; 2230 int ret; 2231 2232 if (unlikely(len_diff_abs > 0xfffU)) 2233 return -EFAULT; 2234 if (unlikely(proto != htons(ETH_P_IP) && 2235 proto != htons(ETH_P_IPV6))) 2236 return -ENOTSUPP; 2237 2238 len_cur = skb->len - skb_network_offset(skb); 2239 if (skb_transport_header_was_set(skb) && !trans_same) 2240 len_cur = skb_network_header_len(skb); 2241 if ((shrink && (len_diff_abs >= len_cur || 2242 len_cur - len_diff_abs < len_min)) || 2243 (!shrink && (skb->len + len_diff_abs > len_max && 2244 !skb_is_gso(skb)))) 2245 return -ENOTSUPP; 2246 2247 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) : 2248 bpf_skb_net_grow(skb, len_diff_abs); 2249 2250 bpf_compute_data_end(skb); 2251 return ret; 2252 } 2253 2254 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 2255 u32, mode, u64, flags) 2256 { 2257 if (unlikely(flags)) 2258 return -EINVAL; 2259 if (likely(mode == BPF_ADJ_ROOM_NET)) 2260 return bpf_skb_adjust_net(skb, len_diff); 2261 2262 return -ENOTSUPP; 2263 } 2264 2265 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 2266 .func = bpf_skb_adjust_room, 2267 .gpl_only = false, 2268 .ret_type = RET_INTEGER, 2269 .arg1_type = ARG_PTR_TO_CTX, 2270 .arg2_type = ARG_ANYTHING, 2271 .arg3_type = ARG_ANYTHING, 2272 .arg4_type = ARG_ANYTHING, 2273 }; 2274 2275 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 2276 { 2277 u32 min_len = skb_network_offset(skb); 2278 2279 if (skb_transport_header_was_set(skb)) 2280 min_len = skb_transport_offset(skb); 2281 if (skb->ip_summed == CHECKSUM_PARTIAL) 2282 min_len = skb_checksum_start_offset(skb) + 2283 skb->csum_offset + sizeof(__sum16); 2284 return min_len; 2285 } 2286 2287 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 2288 { 2289 unsigned int old_len = skb->len; 2290 int ret; 2291 2292 ret = __skb_grow_rcsum(skb, new_len); 2293 if (!ret) 2294 memset(skb->data + old_len, 0, new_len - old_len); 2295 return ret; 2296 } 2297 2298 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 2299 { 2300 return __skb_trim_rcsum(skb, new_len); 2301 } 2302 2303 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 2304 u64, flags) 2305 { 2306 u32 max_len = __bpf_skb_max_len(skb); 2307 u32 min_len = __bpf_skb_min_len(skb); 2308 int ret; 2309 2310 if (unlikely(flags || new_len > max_len || new_len < min_len)) 2311 return -EINVAL; 2312 if (skb->encapsulation) 2313 return -ENOTSUPP; 2314 2315 /* The basic idea of this helper is that it's performing the 2316 * needed work to either grow or trim an skb, and eBPF program 2317 * rewrites the rest via helpers like bpf_skb_store_bytes(), 2318 * bpf_lX_csum_replace() and others rather than passing a raw 2319 * buffer here. This one is a slow path helper and intended 2320 * for replies with control messages. 2321 * 2322 * Like in bpf_skb_change_proto(), we want to keep this rather 2323 * minimal and without protocol specifics so that we are able 2324 * to separate concerns as in bpf_skb_store_bytes() should only 2325 * be the one responsible for writing buffers. 2326 * 2327 * It's really expected to be a slow path operation here for 2328 * control message replies, so we're implicitly linearizing, 2329 * uncloning and drop offloads from the skb by this. 2330 */ 2331 ret = __bpf_try_make_writable(skb, skb->len); 2332 if (!ret) { 2333 if (new_len > skb->len) 2334 ret = bpf_skb_grow_rcsum(skb, new_len); 2335 else if (new_len < skb->len) 2336 ret = bpf_skb_trim_rcsum(skb, new_len); 2337 if (!ret && skb_is_gso(skb)) 2338 skb_gso_reset(skb); 2339 } 2340 2341 bpf_compute_data_end(skb); 2342 return ret; 2343 } 2344 2345 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 2346 .func = bpf_skb_change_tail, 2347 .gpl_only = false, 2348 .ret_type = RET_INTEGER, 2349 .arg1_type = ARG_PTR_TO_CTX, 2350 .arg2_type = ARG_ANYTHING, 2351 .arg3_type = ARG_ANYTHING, 2352 }; 2353 2354 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 2355 u64, flags) 2356 { 2357 u32 max_len = __bpf_skb_max_len(skb); 2358 u32 new_len = skb->len + head_room; 2359 int ret; 2360 2361 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 2362 new_len < skb->len)) 2363 return -EINVAL; 2364 2365 ret = skb_cow(skb, head_room); 2366 if (likely(!ret)) { 2367 /* Idea for this helper is that we currently only 2368 * allow to expand on mac header. This means that 2369 * skb->protocol network header, etc, stay as is. 2370 * Compared to bpf_skb_change_tail(), we're more 2371 * flexible due to not needing to linearize or 2372 * reset GSO. Intention for this helper is to be 2373 * used by an L3 skb that needs to push mac header 2374 * for redirection into L2 device. 2375 */ 2376 __skb_push(skb, head_room); 2377 memset(skb->data, 0, head_room); 2378 skb_reset_mac_header(skb); 2379 } 2380 2381 bpf_compute_data_end(skb); 2382 return 0; 2383 } 2384 2385 static const struct bpf_func_proto bpf_skb_change_head_proto = { 2386 .func = bpf_skb_change_head, 2387 .gpl_only = false, 2388 .ret_type = RET_INTEGER, 2389 .arg1_type = ARG_PTR_TO_CTX, 2390 .arg2_type = ARG_ANYTHING, 2391 .arg3_type = ARG_ANYTHING, 2392 }; 2393 2394 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 2395 { 2396 void *data = xdp->data + offset; 2397 2398 if (unlikely(data < xdp->data_hard_start || 2399 data > xdp->data_end - ETH_HLEN)) 2400 return -EINVAL; 2401 2402 xdp->data = data; 2403 2404 return 0; 2405 } 2406 2407 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 2408 .func = bpf_xdp_adjust_head, 2409 .gpl_only = false, 2410 .ret_type = RET_INTEGER, 2411 .arg1_type = ARG_PTR_TO_CTX, 2412 .arg2_type = ARG_ANYTHING, 2413 }; 2414 2415 bool bpf_helper_changes_pkt_data(void *func) 2416 { 2417 if (func == bpf_skb_vlan_push || 2418 func == bpf_skb_vlan_pop || 2419 func == bpf_skb_store_bytes || 2420 func == bpf_skb_change_proto || 2421 func == bpf_skb_change_head || 2422 func == bpf_skb_change_tail || 2423 func == bpf_skb_adjust_room || 2424 func == bpf_skb_pull_data || 2425 func == bpf_clone_redirect || 2426 func == bpf_l3_csum_replace || 2427 func == bpf_l4_csum_replace || 2428 func == bpf_xdp_adjust_head) 2429 return true; 2430 2431 return false; 2432 } 2433 2434 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 2435 unsigned long off, unsigned long len) 2436 { 2437 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 2438 2439 if (unlikely(!ptr)) 2440 return len; 2441 if (ptr != dst_buff) 2442 memcpy(dst_buff, ptr, len); 2443 2444 return 0; 2445 } 2446 2447 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 2448 u64, flags, void *, meta, u64, meta_size) 2449 { 2450 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 2451 2452 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 2453 return -EINVAL; 2454 if (unlikely(skb_size > skb->len)) 2455 return -EFAULT; 2456 2457 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 2458 bpf_skb_copy); 2459 } 2460 2461 static const struct bpf_func_proto bpf_skb_event_output_proto = { 2462 .func = bpf_skb_event_output, 2463 .gpl_only = true, 2464 .ret_type = RET_INTEGER, 2465 .arg1_type = ARG_PTR_TO_CTX, 2466 .arg2_type = ARG_CONST_MAP_PTR, 2467 .arg3_type = ARG_ANYTHING, 2468 .arg4_type = ARG_PTR_TO_MEM, 2469 .arg5_type = ARG_CONST_SIZE, 2470 }; 2471 2472 static unsigned short bpf_tunnel_key_af(u64 flags) 2473 { 2474 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 2475 } 2476 2477 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 2478 u32, size, u64, flags) 2479 { 2480 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 2481 u8 compat[sizeof(struct bpf_tunnel_key)]; 2482 void *to_orig = to; 2483 int err; 2484 2485 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 2486 err = -EINVAL; 2487 goto err_clear; 2488 } 2489 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 2490 err = -EPROTO; 2491 goto err_clear; 2492 } 2493 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 2494 err = -EINVAL; 2495 switch (size) { 2496 case offsetof(struct bpf_tunnel_key, tunnel_label): 2497 case offsetof(struct bpf_tunnel_key, tunnel_ext): 2498 goto set_compat; 2499 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 2500 /* Fixup deprecated structure layouts here, so we have 2501 * a common path later on. 2502 */ 2503 if (ip_tunnel_info_af(info) != AF_INET) 2504 goto err_clear; 2505 set_compat: 2506 to = (struct bpf_tunnel_key *)compat; 2507 break; 2508 default: 2509 goto err_clear; 2510 } 2511 } 2512 2513 to->tunnel_id = be64_to_cpu(info->key.tun_id); 2514 to->tunnel_tos = info->key.tos; 2515 to->tunnel_ttl = info->key.ttl; 2516 2517 if (flags & BPF_F_TUNINFO_IPV6) { 2518 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 2519 sizeof(to->remote_ipv6)); 2520 to->tunnel_label = be32_to_cpu(info->key.label); 2521 } else { 2522 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 2523 } 2524 2525 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 2526 memcpy(to_orig, to, size); 2527 2528 return 0; 2529 err_clear: 2530 memset(to_orig, 0, size); 2531 return err; 2532 } 2533 2534 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 2535 .func = bpf_skb_get_tunnel_key, 2536 .gpl_only = false, 2537 .ret_type = RET_INTEGER, 2538 .arg1_type = ARG_PTR_TO_CTX, 2539 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2540 .arg3_type = ARG_CONST_SIZE, 2541 .arg4_type = ARG_ANYTHING, 2542 }; 2543 2544 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 2545 { 2546 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 2547 int err; 2548 2549 if (unlikely(!info || 2550 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 2551 err = -ENOENT; 2552 goto err_clear; 2553 } 2554 if (unlikely(size < info->options_len)) { 2555 err = -ENOMEM; 2556 goto err_clear; 2557 } 2558 2559 ip_tunnel_info_opts_get(to, info); 2560 if (size > info->options_len) 2561 memset(to + info->options_len, 0, size - info->options_len); 2562 2563 return info->options_len; 2564 err_clear: 2565 memset(to, 0, size); 2566 return err; 2567 } 2568 2569 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 2570 .func = bpf_skb_get_tunnel_opt, 2571 .gpl_only = false, 2572 .ret_type = RET_INTEGER, 2573 .arg1_type = ARG_PTR_TO_CTX, 2574 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 2575 .arg3_type = ARG_CONST_SIZE, 2576 }; 2577 2578 static struct metadata_dst __percpu *md_dst; 2579 2580 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 2581 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 2582 { 2583 struct metadata_dst *md = this_cpu_ptr(md_dst); 2584 u8 compat[sizeof(struct bpf_tunnel_key)]; 2585 struct ip_tunnel_info *info; 2586 2587 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 2588 BPF_F_DONT_FRAGMENT))) 2589 return -EINVAL; 2590 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 2591 switch (size) { 2592 case offsetof(struct bpf_tunnel_key, tunnel_label): 2593 case offsetof(struct bpf_tunnel_key, tunnel_ext): 2594 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 2595 /* Fixup deprecated structure layouts here, so we have 2596 * a common path later on. 2597 */ 2598 memcpy(compat, from, size); 2599 memset(compat + size, 0, sizeof(compat) - size); 2600 from = (const struct bpf_tunnel_key *) compat; 2601 break; 2602 default: 2603 return -EINVAL; 2604 } 2605 } 2606 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 2607 from->tunnel_ext)) 2608 return -EINVAL; 2609 2610 skb_dst_drop(skb); 2611 dst_hold((struct dst_entry *) md); 2612 skb_dst_set(skb, (struct dst_entry *) md); 2613 2614 info = &md->u.tun_info; 2615 info->mode = IP_TUNNEL_INFO_TX; 2616 2617 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 2618 if (flags & BPF_F_DONT_FRAGMENT) 2619 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 2620 2621 info->key.tun_id = cpu_to_be64(from->tunnel_id); 2622 info->key.tos = from->tunnel_tos; 2623 info->key.ttl = from->tunnel_ttl; 2624 2625 if (flags & BPF_F_TUNINFO_IPV6) { 2626 info->mode |= IP_TUNNEL_INFO_IPV6; 2627 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 2628 sizeof(from->remote_ipv6)); 2629 info->key.label = cpu_to_be32(from->tunnel_label) & 2630 IPV6_FLOWLABEL_MASK; 2631 } else { 2632 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 2633 if (flags & BPF_F_ZERO_CSUM_TX) 2634 info->key.tun_flags &= ~TUNNEL_CSUM; 2635 } 2636 2637 return 0; 2638 } 2639 2640 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 2641 .func = bpf_skb_set_tunnel_key, 2642 .gpl_only = false, 2643 .ret_type = RET_INTEGER, 2644 .arg1_type = ARG_PTR_TO_CTX, 2645 .arg2_type = ARG_PTR_TO_MEM, 2646 .arg3_type = ARG_CONST_SIZE, 2647 .arg4_type = ARG_ANYTHING, 2648 }; 2649 2650 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 2651 const u8 *, from, u32, size) 2652 { 2653 struct ip_tunnel_info *info = skb_tunnel_info(skb); 2654 const struct metadata_dst *md = this_cpu_ptr(md_dst); 2655 2656 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 2657 return -EINVAL; 2658 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 2659 return -ENOMEM; 2660 2661 ip_tunnel_info_opts_set(info, from, size); 2662 2663 return 0; 2664 } 2665 2666 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 2667 .func = bpf_skb_set_tunnel_opt, 2668 .gpl_only = false, 2669 .ret_type = RET_INTEGER, 2670 .arg1_type = ARG_PTR_TO_CTX, 2671 .arg2_type = ARG_PTR_TO_MEM, 2672 .arg3_type = ARG_CONST_SIZE, 2673 }; 2674 2675 static const struct bpf_func_proto * 2676 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 2677 { 2678 if (!md_dst) { 2679 /* Race is not possible, since it's called from verifier 2680 * that is holding verifier mutex. 2681 */ 2682 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 2683 METADATA_IP_TUNNEL, 2684 GFP_KERNEL); 2685 if (!md_dst) 2686 return NULL; 2687 } 2688 2689 switch (which) { 2690 case BPF_FUNC_skb_set_tunnel_key: 2691 return &bpf_skb_set_tunnel_key_proto; 2692 case BPF_FUNC_skb_set_tunnel_opt: 2693 return &bpf_skb_set_tunnel_opt_proto; 2694 default: 2695 return NULL; 2696 } 2697 } 2698 2699 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 2700 u32, idx) 2701 { 2702 struct bpf_array *array = container_of(map, struct bpf_array, map); 2703 struct cgroup *cgrp; 2704 struct sock *sk; 2705 2706 sk = skb_to_full_sk(skb); 2707 if (!sk || !sk_fullsock(sk)) 2708 return -ENOENT; 2709 if (unlikely(idx >= array->map.max_entries)) 2710 return -E2BIG; 2711 2712 cgrp = READ_ONCE(array->ptrs[idx]); 2713 if (unlikely(!cgrp)) 2714 return -EAGAIN; 2715 2716 return sk_under_cgroup_hierarchy(sk, cgrp); 2717 } 2718 2719 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 2720 .func = bpf_skb_under_cgroup, 2721 .gpl_only = false, 2722 .ret_type = RET_INTEGER, 2723 .arg1_type = ARG_PTR_TO_CTX, 2724 .arg2_type = ARG_CONST_MAP_PTR, 2725 .arg3_type = ARG_ANYTHING, 2726 }; 2727 2728 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, 2729 unsigned long off, unsigned long len) 2730 { 2731 memcpy(dst_buff, src_buff + off, len); 2732 return 0; 2733 } 2734 2735 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 2736 u64, flags, void *, meta, u64, meta_size) 2737 { 2738 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 2739 2740 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 2741 return -EINVAL; 2742 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data))) 2743 return -EFAULT; 2744 2745 return bpf_event_output(map, flags, meta, meta_size, xdp->data, 2746 xdp_size, bpf_xdp_copy); 2747 } 2748 2749 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 2750 .func = bpf_xdp_event_output, 2751 .gpl_only = true, 2752 .ret_type = RET_INTEGER, 2753 .arg1_type = ARG_PTR_TO_CTX, 2754 .arg2_type = ARG_CONST_MAP_PTR, 2755 .arg3_type = ARG_ANYTHING, 2756 .arg4_type = ARG_PTR_TO_MEM, 2757 .arg5_type = ARG_CONST_SIZE, 2758 }; 2759 2760 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 2761 { 2762 return skb->sk ? sock_gen_cookie(skb->sk) : 0; 2763 } 2764 2765 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 2766 .func = bpf_get_socket_cookie, 2767 .gpl_only = false, 2768 .ret_type = RET_INTEGER, 2769 .arg1_type = ARG_PTR_TO_CTX, 2770 }; 2771 2772 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 2773 { 2774 struct sock *sk = sk_to_full_sk(skb->sk); 2775 kuid_t kuid; 2776 2777 if (!sk || !sk_fullsock(sk)) 2778 return overflowuid; 2779 kuid = sock_net_uid(sock_net(sk), sk); 2780 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 2781 } 2782 2783 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 2784 .func = bpf_get_socket_uid, 2785 .gpl_only = false, 2786 .ret_type = RET_INTEGER, 2787 .arg1_type = ARG_PTR_TO_CTX, 2788 }; 2789 2790 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 2791 int, level, int, optname, char *, optval, int, optlen) 2792 { 2793 struct sock *sk = bpf_sock->sk; 2794 int ret = 0; 2795 int val; 2796 2797 if (!sk_fullsock(sk)) 2798 return -EINVAL; 2799 2800 if (level == SOL_SOCKET) { 2801 if (optlen != sizeof(int)) 2802 return -EINVAL; 2803 val = *((int *)optval); 2804 2805 /* Only some socketops are supported */ 2806 switch (optname) { 2807 case SO_RCVBUF: 2808 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 2809 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 2810 break; 2811 case SO_SNDBUF: 2812 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 2813 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 2814 break; 2815 case SO_MAX_PACING_RATE: 2816 sk->sk_max_pacing_rate = val; 2817 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 2818 sk->sk_max_pacing_rate); 2819 break; 2820 case SO_PRIORITY: 2821 sk->sk_priority = val; 2822 break; 2823 case SO_RCVLOWAT: 2824 if (val < 0) 2825 val = INT_MAX; 2826 sk->sk_rcvlowat = val ? : 1; 2827 break; 2828 case SO_MARK: 2829 sk->sk_mark = val; 2830 break; 2831 default: 2832 ret = -EINVAL; 2833 } 2834 #ifdef CONFIG_INET 2835 } else if (level == SOL_TCP && 2836 sk->sk_prot->setsockopt == tcp_setsockopt) { 2837 if (optname == TCP_CONGESTION) { 2838 char name[TCP_CA_NAME_MAX]; 2839 2840 strncpy(name, optval, min_t(long, optlen, 2841 TCP_CA_NAME_MAX-1)); 2842 name[TCP_CA_NAME_MAX-1] = 0; 2843 ret = tcp_set_congestion_control(sk, name, false); 2844 if (!ret && bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN) 2845 /* replacing an existing ca */ 2846 tcp_reinit_congestion_control(sk, 2847 inet_csk(sk)->icsk_ca_ops); 2848 } else { 2849 struct tcp_sock *tp = tcp_sk(sk); 2850 2851 if (optlen != sizeof(int)) 2852 return -EINVAL; 2853 2854 val = *((int *)optval); 2855 /* Only some options are supported */ 2856 switch (optname) { 2857 case TCP_BPF_IW: 2858 if (val <= 0 || tp->data_segs_out > 0) 2859 ret = -EINVAL; 2860 else 2861 tp->snd_cwnd = val; 2862 break; 2863 case TCP_BPF_SNDCWND_CLAMP: 2864 if (val <= 0) { 2865 ret = -EINVAL; 2866 } else { 2867 tp->snd_cwnd_clamp = val; 2868 tp->snd_ssthresh = val; 2869 } 2870 break; 2871 default: 2872 ret = -EINVAL; 2873 } 2874 } 2875 ret = -EINVAL; 2876 #endif 2877 } else { 2878 ret = -EINVAL; 2879 } 2880 return ret; 2881 } 2882 2883 static const struct bpf_func_proto bpf_setsockopt_proto = { 2884 .func = bpf_setsockopt, 2885 .gpl_only = true, 2886 .ret_type = RET_INTEGER, 2887 .arg1_type = ARG_PTR_TO_CTX, 2888 .arg2_type = ARG_ANYTHING, 2889 .arg3_type = ARG_ANYTHING, 2890 .arg4_type = ARG_PTR_TO_MEM, 2891 .arg5_type = ARG_CONST_SIZE, 2892 }; 2893 2894 static const struct bpf_func_proto * 2895 bpf_base_func_proto(enum bpf_func_id func_id) 2896 { 2897 switch (func_id) { 2898 case BPF_FUNC_map_lookup_elem: 2899 return &bpf_map_lookup_elem_proto; 2900 case BPF_FUNC_map_update_elem: 2901 return &bpf_map_update_elem_proto; 2902 case BPF_FUNC_map_delete_elem: 2903 return &bpf_map_delete_elem_proto; 2904 case BPF_FUNC_get_prandom_u32: 2905 return &bpf_get_prandom_u32_proto; 2906 case BPF_FUNC_get_smp_processor_id: 2907 return &bpf_get_raw_smp_processor_id_proto; 2908 case BPF_FUNC_get_numa_node_id: 2909 return &bpf_get_numa_node_id_proto; 2910 case BPF_FUNC_tail_call: 2911 return &bpf_tail_call_proto; 2912 case BPF_FUNC_ktime_get_ns: 2913 return &bpf_ktime_get_ns_proto; 2914 case BPF_FUNC_trace_printk: 2915 if (capable(CAP_SYS_ADMIN)) 2916 return bpf_get_trace_printk_proto(); 2917 default: 2918 return NULL; 2919 } 2920 } 2921 2922 static const struct bpf_func_proto * 2923 sk_filter_func_proto(enum bpf_func_id func_id) 2924 { 2925 switch (func_id) { 2926 case BPF_FUNC_skb_load_bytes: 2927 return &bpf_skb_load_bytes_proto; 2928 case BPF_FUNC_get_socket_cookie: 2929 return &bpf_get_socket_cookie_proto; 2930 case BPF_FUNC_get_socket_uid: 2931 return &bpf_get_socket_uid_proto; 2932 default: 2933 return bpf_base_func_proto(func_id); 2934 } 2935 } 2936 2937 static const struct bpf_func_proto * 2938 tc_cls_act_func_proto(enum bpf_func_id func_id) 2939 { 2940 switch (func_id) { 2941 case BPF_FUNC_skb_store_bytes: 2942 return &bpf_skb_store_bytes_proto; 2943 case BPF_FUNC_skb_load_bytes: 2944 return &bpf_skb_load_bytes_proto; 2945 case BPF_FUNC_skb_pull_data: 2946 return &bpf_skb_pull_data_proto; 2947 case BPF_FUNC_csum_diff: 2948 return &bpf_csum_diff_proto; 2949 case BPF_FUNC_csum_update: 2950 return &bpf_csum_update_proto; 2951 case BPF_FUNC_l3_csum_replace: 2952 return &bpf_l3_csum_replace_proto; 2953 case BPF_FUNC_l4_csum_replace: 2954 return &bpf_l4_csum_replace_proto; 2955 case BPF_FUNC_clone_redirect: 2956 return &bpf_clone_redirect_proto; 2957 case BPF_FUNC_get_cgroup_classid: 2958 return &bpf_get_cgroup_classid_proto; 2959 case BPF_FUNC_skb_vlan_push: 2960 return &bpf_skb_vlan_push_proto; 2961 case BPF_FUNC_skb_vlan_pop: 2962 return &bpf_skb_vlan_pop_proto; 2963 case BPF_FUNC_skb_change_proto: 2964 return &bpf_skb_change_proto_proto; 2965 case BPF_FUNC_skb_change_type: 2966 return &bpf_skb_change_type_proto; 2967 case BPF_FUNC_skb_adjust_room: 2968 return &bpf_skb_adjust_room_proto; 2969 case BPF_FUNC_skb_change_tail: 2970 return &bpf_skb_change_tail_proto; 2971 case BPF_FUNC_skb_get_tunnel_key: 2972 return &bpf_skb_get_tunnel_key_proto; 2973 case BPF_FUNC_skb_set_tunnel_key: 2974 return bpf_get_skb_set_tunnel_proto(func_id); 2975 case BPF_FUNC_skb_get_tunnel_opt: 2976 return &bpf_skb_get_tunnel_opt_proto; 2977 case BPF_FUNC_skb_set_tunnel_opt: 2978 return bpf_get_skb_set_tunnel_proto(func_id); 2979 case BPF_FUNC_redirect: 2980 return &bpf_redirect_proto; 2981 case BPF_FUNC_get_route_realm: 2982 return &bpf_get_route_realm_proto; 2983 case BPF_FUNC_get_hash_recalc: 2984 return &bpf_get_hash_recalc_proto; 2985 case BPF_FUNC_set_hash_invalid: 2986 return &bpf_set_hash_invalid_proto; 2987 case BPF_FUNC_set_hash: 2988 return &bpf_set_hash_proto; 2989 case BPF_FUNC_perf_event_output: 2990 return &bpf_skb_event_output_proto; 2991 case BPF_FUNC_get_smp_processor_id: 2992 return &bpf_get_smp_processor_id_proto; 2993 case BPF_FUNC_skb_under_cgroup: 2994 return &bpf_skb_under_cgroup_proto; 2995 case BPF_FUNC_get_socket_cookie: 2996 return &bpf_get_socket_cookie_proto; 2997 case BPF_FUNC_get_socket_uid: 2998 return &bpf_get_socket_uid_proto; 2999 default: 3000 return bpf_base_func_proto(func_id); 3001 } 3002 } 3003 3004 static const struct bpf_func_proto * 3005 xdp_func_proto(enum bpf_func_id func_id) 3006 { 3007 switch (func_id) { 3008 case BPF_FUNC_perf_event_output: 3009 return &bpf_xdp_event_output_proto; 3010 case BPF_FUNC_get_smp_processor_id: 3011 return &bpf_get_smp_processor_id_proto; 3012 case BPF_FUNC_xdp_adjust_head: 3013 return &bpf_xdp_adjust_head_proto; 3014 default: 3015 return bpf_base_func_proto(func_id); 3016 } 3017 } 3018 3019 static const struct bpf_func_proto * 3020 lwt_inout_func_proto(enum bpf_func_id func_id) 3021 { 3022 switch (func_id) { 3023 case BPF_FUNC_skb_load_bytes: 3024 return &bpf_skb_load_bytes_proto; 3025 case BPF_FUNC_skb_pull_data: 3026 return &bpf_skb_pull_data_proto; 3027 case BPF_FUNC_csum_diff: 3028 return &bpf_csum_diff_proto; 3029 case BPF_FUNC_get_cgroup_classid: 3030 return &bpf_get_cgroup_classid_proto; 3031 case BPF_FUNC_get_route_realm: 3032 return &bpf_get_route_realm_proto; 3033 case BPF_FUNC_get_hash_recalc: 3034 return &bpf_get_hash_recalc_proto; 3035 case BPF_FUNC_perf_event_output: 3036 return &bpf_skb_event_output_proto; 3037 case BPF_FUNC_get_smp_processor_id: 3038 return &bpf_get_smp_processor_id_proto; 3039 case BPF_FUNC_skb_under_cgroup: 3040 return &bpf_skb_under_cgroup_proto; 3041 default: 3042 return bpf_base_func_proto(func_id); 3043 } 3044 } 3045 3046 static const struct bpf_func_proto * 3047 sock_ops_func_proto(enum bpf_func_id func_id) 3048 { 3049 switch (func_id) { 3050 case BPF_FUNC_setsockopt: 3051 return &bpf_setsockopt_proto; 3052 default: 3053 return bpf_base_func_proto(func_id); 3054 } 3055 } 3056 3057 static const struct bpf_func_proto * 3058 lwt_xmit_func_proto(enum bpf_func_id func_id) 3059 { 3060 switch (func_id) { 3061 case BPF_FUNC_skb_get_tunnel_key: 3062 return &bpf_skb_get_tunnel_key_proto; 3063 case BPF_FUNC_skb_set_tunnel_key: 3064 return bpf_get_skb_set_tunnel_proto(func_id); 3065 case BPF_FUNC_skb_get_tunnel_opt: 3066 return &bpf_skb_get_tunnel_opt_proto; 3067 case BPF_FUNC_skb_set_tunnel_opt: 3068 return bpf_get_skb_set_tunnel_proto(func_id); 3069 case BPF_FUNC_redirect: 3070 return &bpf_redirect_proto; 3071 case BPF_FUNC_clone_redirect: 3072 return &bpf_clone_redirect_proto; 3073 case BPF_FUNC_skb_change_tail: 3074 return &bpf_skb_change_tail_proto; 3075 case BPF_FUNC_skb_change_head: 3076 return &bpf_skb_change_head_proto; 3077 case BPF_FUNC_skb_store_bytes: 3078 return &bpf_skb_store_bytes_proto; 3079 case BPF_FUNC_csum_update: 3080 return &bpf_csum_update_proto; 3081 case BPF_FUNC_l3_csum_replace: 3082 return &bpf_l3_csum_replace_proto; 3083 case BPF_FUNC_l4_csum_replace: 3084 return &bpf_l4_csum_replace_proto; 3085 case BPF_FUNC_set_hash_invalid: 3086 return &bpf_set_hash_invalid_proto; 3087 default: 3088 return lwt_inout_func_proto(func_id); 3089 } 3090 } 3091 3092 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 3093 struct bpf_insn_access_aux *info) 3094 { 3095 const int size_default = sizeof(__u32); 3096 3097 if (off < 0 || off >= sizeof(struct __sk_buff)) 3098 return false; 3099 3100 /* The verifier guarantees that size > 0. */ 3101 if (off % size != 0) 3102 return false; 3103 3104 switch (off) { 3105 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 3106 if (off + size > offsetofend(struct __sk_buff, cb[4])) 3107 return false; 3108 break; 3109 case bpf_ctx_range(struct __sk_buff, data): 3110 case bpf_ctx_range(struct __sk_buff, data_end): 3111 if (size != size_default) 3112 return false; 3113 break; 3114 default: 3115 /* Only narrow read access allowed for now. */ 3116 if (type == BPF_WRITE) { 3117 if (size != size_default) 3118 return false; 3119 } else { 3120 bpf_ctx_record_field_size(info, size_default); 3121 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 3122 return false; 3123 } 3124 } 3125 3126 return true; 3127 } 3128 3129 static bool sk_filter_is_valid_access(int off, int size, 3130 enum bpf_access_type type, 3131 struct bpf_insn_access_aux *info) 3132 { 3133 switch (off) { 3134 case bpf_ctx_range(struct __sk_buff, tc_classid): 3135 case bpf_ctx_range(struct __sk_buff, data): 3136 case bpf_ctx_range(struct __sk_buff, data_end): 3137 return false; 3138 } 3139 3140 if (type == BPF_WRITE) { 3141 switch (off) { 3142 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 3143 break; 3144 default: 3145 return false; 3146 } 3147 } 3148 3149 return bpf_skb_is_valid_access(off, size, type, info); 3150 } 3151 3152 static bool lwt_is_valid_access(int off, int size, 3153 enum bpf_access_type type, 3154 struct bpf_insn_access_aux *info) 3155 { 3156 switch (off) { 3157 case bpf_ctx_range(struct __sk_buff, tc_classid): 3158 return false; 3159 } 3160 3161 if (type == BPF_WRITE) { 3162 switch (off) { 3163 case bpf_ctx_range(struct __sk_buff, mark): 3164 case bpf_ctx_range(struct __sk_buff, priority): 3165 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 3166 break; 3167 default: 3168 return false; 3169 } 3170 } 3171 3172 switch (off) { 3173 case bpf_ctx_range(struct __sk_buff, data): 3174 info->reg_type = PTR_TO_PACKET; 3175 break; 3176 case bpf_ctx_range(struct __sk_buff, data_end): 3177 info->reg_type = PTR_TO_PACKET_END; 3178 break; 3179 } 3180 3181 return bpf_skb_is_valid_access(off, size, type, info); 3182 } 3183 3184 static bool sock_filter_is_valid_access(int off, int size, 3185 enum bpf_access_type type, 3186 struct bpf_insn_access_aux *info) 3187 { 3188 if (type == BPF_WRITE) { 3189 switch (off) { 3190 case offsetof(struct bpf_sock, bound_dev_if): 3191 break; 3192 default: 3193 return false; 3194 } 3195 } 3196 3197 if (off < 0 || off + size > sizeof(struct bpf_sock)) 3198 return false; 3199 /* The verifier guarantees that size > 0. */ 3200 if (off % size != 0) 3201 return false; 3202 if (size != sizeof(__u32)) 3203 return false; 3204 3205 return true; 3206 } 3207 3208 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 3209 const struct bpf_prog *prog) 3210 { 3211 struct bpf_insn *insn = insn_buf; 3212 3213 if (!direct_write) 3214 return 0; 3215 3216 /* if (!skb->cloned) 3217 * goto start; 3218 * 3219 * (Fast-path, otherwise approximation that we might be 3220 * a clone, do the rest in helper.) 3221 */ 3222 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 3223 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 3224 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 3225 3226 /* ret = bpf_skb_pull_data(skb, 0); */ 3227 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 3228 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 3229 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 3230 BPF_FUNC_skb_pull_data); 3231 /* if (!ret) 3232 * goto restore; 3233 * return TC_ACT_SHOT; 3234 */ 3235 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 3236 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, TC_ACT_SHOT); 3237 *insn++ = BPF_EXIT_INSN(); 3238 3239 /* restore: */ 3240 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 3241 /* start: */ 3242 *insn++ = prog->insnsi[0]; 3243 3244 return insn - insn_buf; 3245 } 3246 3247 static bool tc_cls_act_is_valid_access(int off, int size, 3248 enum bpf_access_type type, 3249 struct bpf_insn_access_aux *info) 3250 { 3251 if (type == BPF_WRITE) { 3252 switch (off) { 3253 case bpf_ctx_range(struct __sk_buff, mark): 3254 case bpf_ctx_range(struct __sk_buff, tc_index): 3255 case bpf_ctx_range(struct __sk_buff, priority): 3256 case bpf_ctx_range(struct __sk_buff, tc_classid): 3257 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 3258 break; 3259 default: 3260 return false; 3261 } 3262 } 3263 3264 switch (off) { 3265 case bpf_ctx_range(struct __sk_buff, data): 3266 info->reg_type = PTR_TO_PACKET; 3267 break; 3268 case bpf_ctx_range(struct __sk_buff, data_end): 3269 info->reg_type = PTR_TO_PACKET_END; 3270 break; 3271 } 3272 3273 return bpf_skb_is_valid_access(off, size, type, info); 3274 } 3275 3276 static bool __is_valid_xdp_access(int off, int size) 3277 { 3278 if (off < 0 || off >= sizeof(struct xdp_md)) 3279 return false; 3280 if (off % size != 0) 3281 return false; 3282 if (size != sizeof(__u32)) 3283 return false; 3284 3285 return true; 3286 } 3287 3288 static bool xdp_is_valid_access(int off, int size, 3289 enum bpf_access_type type, 3290 struct bpf_insn_access_aux *info) 3291 { 3292 if (type == BPF_WRITE) 3293 return false; 3294 3295 switch (off) { 3296 case offsetof(struct xdp_md, data): 3297 info->reg_type = PTR_TO_PACKET; 3298 break; 3299 case offsetof(struct xdp_md, data_end): 3300 info->reg_type = PTR_TO_PACKET_END; 3301 break; 3302 } 3303 3304 return __is_valid_xdp_access(off, size); 3305 } 3306 3307 void bpf_warn_invalid_xdp_action(u32 act) 3308 { 3309 WARN_ONCE(1, "Illegal XDP return value %u, expect packet loss\n", act); 3310 } 3311 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 3312 3313 static bool __is_valid_sock_ops_access(int off, int size) 3314 { 3315 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 3316 return false; 3317 /* The verifier guarantees that size > 0. */ 3318 if (off % size != 0) 3319 return false; 3320 if (size != sizeof(__u32)) 3321 return false; 3322 3323 return true; 3324 } 3325 3326 static bool sock_ops_is_valid_access(int off, int size, 3327 enum bpf_access_type type, 3328 struct bpf_insn_access_aux *info) 3329 { 3330 if (type == BPF_WRITE) { 3331 switch (off) { 3332 case offsetof(struct bpf_sock_ops, op) ... 3333 offsetof(struct bpf_sock_ops, replylong[3]): 3334 break; 3335 default: 3336 return false; 3337 } 3338 } 3339 3340 return __is_valid_sock_ops_access(off, size); 3341 } 3342 3343 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 3344 const struct bpf_insn *si, 3345 struct bpf_insn *insn_buf, 3346 struct bpf_prog *prog, u32 *target_size) 3347 { 3348 struct bpf_insn *insn = insn_buf; 3349 int off; 3350 3351 switch (si->off) { 3352 case offsetof(struct __sk_buff, len): 3353 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3354 bpf_target_off(struct sk_buff, len, 4, 3355 target_size)); 3356 break; 3357 3358 case offsetof(struct __sk_buff, protocol): 3359 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3360 bpf_target_off(struct sk_buff, protocol, 2, 3361 target_size)); 3362 break; 3363 3364 case offsetof(struct __sk_buff, vlan_proto): 3365 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3366 bpf_target_off(struct sk_buff, vlan_proto, 2, 3367 target_size)); 3368 break; 3369 3370 case offsetof(struct __sk_buff, priority): 3371 if (type == BPF_WRITE) 3372 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 3373 bpf_target_off(struct sk_buff, priority, 4, 3374 target_size)); 3375 else 3376 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3377 bpf_target_off(struct sk_buff, priority, 4, 3378 target_size)); 3379 break; 3380 3381 case offsetof(struct __sk_buff, ingress_ifindex): 3382 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3383 bpf_target_off(struct sk_buff, skb_iif, 4, 3384 target_size)); 3385 break; 3386 3387 case offsetof(struct __sk_buff, ifindex): 3388 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 3389 si->dst_reg, si->src_reg, 3390 offsetof(struct sk_buff, dev)); 3391 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 3392 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3393 bpf_target_off(struct net_device, ifindex, 4, 3394 target_size)); 3395 break; 3396 3397 case offsetof(struct __sk_buff, hash): 3398 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3399 bpf_target_off(struct sk_buff, hash, 4, 3400 target_size)); 3401 break; 3402 3403 case offsetof(struct __sk_buff, mark): 3404 if (type == BPF_WRITE) 3405 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 3406 bpf_target_off(struct sk_buff, mark, 4, 3407 target_size)); 3408 else 3409 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3410 bpf_target_off(struct sk_buff, mark, 4, 3411 target_size)); 3412 break; 3413 3414 case offsetof(struct __sk_buff, pkt_type): 3415 *target_size = 1; 3416 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 3417 PKT_TYPE_OFFSET()); 3418 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 3419 #ifdef __BIG_ENDIAN_BITFIELD 3420 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 3421 #endif 3422 break; 3423 3424 case offsetof(struct __sk_buff, queue_mapping): 3425 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3426 bpf_target_off(struct sk_buff, queue_mapping, 2, 3427 target_size)); 3428 break; 3429 3430 case offsetof(struct __sk_buff, vlan_present): 3431 case offsetof(struct __sk_buff, vlan_tci): 3432 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); 3433 3434 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3435 bpf_target_off(struct sk_buff, vlan_tci, 2, 3436 target_size)); 3437 if (si->off == offsetof(struct __sk_buff, vlan_tci)) { 3438 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 3439 ~VLAN_TAG_PRESENT); 3440 } else { 3441 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12); 3442 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 3443 } 3444 break; 3445 3446 case offsetof(struct __sk_buff, cb[0]) ... 3447 offsetofend(struct __sk_buff, cb[4]) - 1: 3448 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20); 3449 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 3450 offsetof(struct qdisc_skb_cb, data)) % 3451 sizeof(__u64)); 3452 3453 prog->cb_access = 1; 3454 off = si->off; 3455 off -= offsetof(struct __sk_buff, cb[0]); 3456 off += offsetof(struct sk_buff, cb); 3457 off += offsetof(struct qdisc_skb_cb, data); 3458 if (type == BPF_WRITE) 3459 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 3460 si->src_reg, off); 3461 else 3462 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 3463 si->src_reg, off); 3464 break; 3465 3466 case offsetof(struct __sk_buff, tc_classid): 3467 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2); 3468 3469 off = si->off; 3470 off -= offsetof(struct __sk_buff, tc_classid); 3471 off += offsetof(struct sk_buff, cb); 3472 off += offsetof(struct qdisc_skb_cb, tc_classid); 3473 *target_size = 2; 3474 if (type == BPF_WRITE) 3475 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 3476 si->src_reg, off); 3477 else 3478 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 3479 si->src_reg, off); 3480 break; 3481 3482 case offsetof(struct __sk_buff, data): 3483 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 3484 si->dst_reg, si->src_reg, 3485 offsetof(struct sk_buff, data)); 3486 break; 3487 3488 case offsetof(struct __sk_buff, data_end): 3489 off = si->off; 3490 off -= offsetof(struct __sk_buff, data_end); 3491 off += offsetof(struct sk_buff, cb); 3492 off += offsetof(struct bpf_skb_data_end, data_end); 3493 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 3494 si->src_reg, off); 3495 break; 3496 3497 case offsetof(struct __sk_buff, tc_index): 3498 #ifdef CONFIG_NET_SCHED 3499 if (type == BPF_WRITE) 3500 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 3501 bpf_target_off(struct sk_buff, tc_index, 2, 3502 target_size)); 3503 else 3504 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3505 bpf_target_off(struct sk_buff, tc_index, 2, 3506 target_size)); 3507 #else 3508 if (type == BPF_WRITE) 3509 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 3510 else 3511 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 3512 #endif 3513 break; 3514 3515 case offsetof(struct __sk_buff, napi_id): 3516 #if defined(CONFIG_NET_RX_BUSY_POLL) 3517 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3518 bpf_target_off(struct sk_buff, napi_id, 4, 3519 target_size)); 3520 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 3521 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 3522 #else 3523 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 3524 #endif 3525 break; 3526 } 3527 3528 return insn - insn_buf; 3529 } 3530 3531 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type, 3532 const struct bpf_insn *si, 3533 struct bpf_insn *insn_buf, 3534 struct bpf_prog *prog, u32 *target_size) 3535 { 3536 struct bpf_insn *insn = insn_buf; 3537 3538 switch (si->off) { 3539 case offsetof(struct bpf_sock, bound_dev_if): 3540 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4); 3541 3542 if (type == BPF_WRITE) 3543 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 3544 offsetof(struct sock, sk_bound_dev_if)); 3545 else 3546 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3547 offsetof(struct sock, sk_bound_dev_if)); 3548 break; 3549 3550 case offsetof(struct bpf_sock, family): 3551 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2); 3552 3553 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 3554 offsetof(struct sock, sk_family)); 3555 break; 3556 3557 case offsetof(struct bpf_sock, type): 3558 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3559 offsetof(struct sock, __sk_flags_offset)); 3560 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK); 3561 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT); 3562 break; 3563 3564 case offsetof(struct bpf_sock, protocol): 3565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3566 offsetof(struct sock, __sk_flags_offset)); 3567 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK); 3568 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT); 3569 break; 3570 } 3571 3572 return insn - insn_buf; 3573 } 3574 3575 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 3576 const struct bpf_insn *si, 3577 struct bpf_insn *insn_buf, 3578 struct bpf_prog *prog, u32 *target_size) 3579 { 3580 struct bpf_insn *insn = insn_buf; 3581 3582 switch (si->off) { 3583 case offsetof(struct __sk_buff, ifindex): 3584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 3585 si->dst_reg, si->src_reg, 3586 offsetof(struct sk_buff, dev)); 3587 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3588 bpf_target_off(struct net_device, ifindex, 4, 3589 target_size)); 3590 break; 3591 default: 3592 return bpf_convert_ctx_access(type, si, insn_buf, prog, 3593 target_size); 3594 } 3595 3596 return insn - insn_buf; 3597 } 3598 3599 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 3600 const struct bpf_insn *si, 3601 struct bpf_insn *insn_buf, 3602 struct bpf_prog *prog, u32 *target_size) 3603 { 3604 struct bpf_insn *insn = insn_buf; 3605 3606 switch (si->off) { 3607 case offsetof(struct xdp_md, data): 3608 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 3609 si->dst_reg, si->src_reg, 3610 offsetof(struct xdp_buff, data)); 3611 break; 3612 case offsetof(struct xdp_md, data_end): 3613 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 3614 si->dst_reg, si->src_reg, 3615 offsetof(struct xdp_buff, data_end)); 3616 break; 3617 } 3618 3619 return insn - insn_buf; 3620 } 3621 3622 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 3623 const struct bpf_insn *si, 3624 struct bpf_insn *insn_buf, 3625 struct bpf_prog *prog, 3626 u32 *target_size) 3627 { 3628 struct bpf_insn *insn = insn_buf; 3629 int off; 3630 3631 switch (si->off) { 3632 case offsetof(struct bpf_sock_ops, op) ... 3633 offsetof(struct bpf_sock_ops, replylong[3]): 3634 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) != 3635 FIELD_SIZEOF(struct bpf_sock_ops_kern, op)); 3636 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) != 3637 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply)); 3638 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) != 3639 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong)); 3640 off = si->off; 3641 off -= offsetof(struct bpf_sock_ops, op); 3642 off += offsetof(struct bpf_sock_ops_kern, op); 3643 if (type == BPF_WRITE) 3644 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 3645 off); 3646 else 3647 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 3648 off); 3649 break; 3650 3651 case offsetof(struct bpf_sock_ops, family): 3652 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2); 3653 3654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3655 struct bpf_sock_ops_kern, sk), 3656 si->dst_reg, si->src_reg, 3657 offsetof(struct bpf_sock_ops_kern, sk)); 3658 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 3659 offsetof(struct sock_common, skc_family)); 3660 break; 3661 3662 case offsetof(struct bpf_sock_ops, remote_ip4): 3663 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4); 3664 3665 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3666 struct bpf_sock_ops_kern, sk), 3667 si->dst_reg, si->src_reg, 3668 offsetof(struct bpf_sock_ops_kern, sk)); 3669 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3670 offsetof(struct sock_common, skc_daddr)); 3671 break; 3672 3673 case offsetof(struct bpf_sock_ops, local_ip4): 3674 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4); 3675 3676 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3677 struct bpf_sock_ops_kern, sk), 3678 si->dst_reg, si->src_reg, 3679 offsetof(struct bpf_sock_ops_kern, sk)); 3680 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3681 offsetof(struct sock_common, 3682 skc_rcv_saddr)); 3683 break; 3684 3685 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 3686 offsetof(struct bpf_sock_ops, remote_ip6[3]): 3687 #if IS_ENABLED(CONFIG_IPV6) 3688 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 3689 skc_v6_daddr.s6_addr32[0]) != 4); 3690 3691 off = si->off; 3692 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 3693 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3694 struct bpf_sock_ops_kern, sk), 3695 si->dst_reg, si->src_reg, 3696 offsetof(struct bpf_sock_ops_kern, sk)); 3697 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3698 offsetof(struct sock_common, 3699 skc_v6_daddr.s6_addr32[0]) + 3700 off); 3701 #else 3702 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 3703 #endif 3704 break; 3705 3706 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 3707 offsetof(struct bpf_sock_ops, local_ip6[3]): 3708 #if IS_ENABLED(CONFIG_IPV6) 3709 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 3710 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 3711 3712 off = si->off; 3713 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 3714 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3715 struct bpf_sock_ops_kern, sk), 3716 si->dst_reg, si->src_reg, 3717 offsetof(struct bpf_sock_ops_kern, sk)); 3718 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 3719 offsetof(struct sock_common, 3720 skc_v6_rcv_saddr.s6_addr32[0]) + 3721 off); 3722 #else 3723 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 3724 #endif 3725 break; 3726 3727 case offsetof(struct bpf_sock_ops, remote_port): 3728 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2); 3729 3730 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3731 struct bpf_sock_ops_kern, sk), 3732 si->dst_reg, si->src_reg, 3733 offsetof(struct bpf_sock_ops_kern, sk)); 3734 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 3735 offsetof(struct sock_common, skc_dport)); 3736 #ifndef __BIG_ENDIAN_BITFIELD 3737 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 3738 #endif 3739 break; 3740 3741 case offsetof(struct bpf_sock_ops, local_port): 3742 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2); 3743 3744 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 3745 struct bpf_sock_ops_kern, sk), 3746 si->dst_reg, si->src_reg, 3747 offsetof(struct bpf_sock_ops_kern, sk)); 3748 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 3749 offsetof(struct sock_common, skc_num)); 3750 break; 3751 } 3752 return insn - insn_buf; 3753 } 3754 3755 const struct bpf_verifier_ops sk_filter_prog_ops = { 3756 .get_func_proto = sk_filter_func_proto, 3757 .is_valid_access = sk_filter_is_valid_access, 3758 .convert_ctx_access = bpf_convert_ctx_access, 3759 }; 3760 3761 const struct bpf_verifier_ops tc_cls_act_prog_ops = { 3762 .get_func_proto = tc_cls_act_func_proto, 3763 .is_valid_access = tc_cls_act_is_valid_access, 3764 .convert_ctx_access = tc_cls_act_convert_ctx_access, 3765 .gen_prologue = tc_cls_act_prologue, 3766 .test_run = bpf_prog_test_run_skb, 3767 }; 3768 3769 const struct bpf_verifier_ops xdp_prog_ops = { 3770 .get_func_proto = xdp_func_proto, 3771 .is_valid_access = xdp_is_valid_access, 3772 .convert_ctx_access = xdp_convert_ctx_access, 3773 .test_run = bpf_prog_test_run_xdp, 3774 }; 3775 3776 const struct bpf_verifier_ops cg_skb_prog_ops = { 3777 .get_func_proto = sk_filter_func_proto, 3778 .is_valid_access = sk_filter_is_valid_access, 3779 .convert_ctx_access = bpf_convert_ctx_access, 3780 .test_run = bpf_prog_test_run_skb, 3781 }; 3782 3783 const struct bpf_verifier_ops lwt_inout_prog_ops = { 3784 .get_func_proto = lwt_inout_func_proto, 3785 .is_valid_access = lwt_is_valid_access, 3786 .convert_ctx_access = bpf_convert_ctx_access, 3787 .test_run = bpf_prog_test_run_skb, 3788 }; 3789 3790 const struct bpf_verifier_ops lwt_xmit_prog_ops = { 3791 .get_func_proto = lwt_xmit_func_proto, 3792 .is_valid_access = lwt_is_valid_access, 3793 .convert_ctx_access = bpf_convert_ctx_access, 3794 .gen_prologue = tc_cls_act_prologue, 3795 .test_run = bpf_prog_test_run_skb, 3796 }; 3797 3798 const struct bpf_verifier_ops cg_sock_prog_ops = { 3799 .get_func_proto = bpf_base_func_proto, 3800 .is_valid_access = sock_filter_is_valid_access, 3801 .convert_ctx_access = sock_filter_convert_ctx_access, 3802 }; 3803 3804 const struct bpf_verifier_ops sock_ops_prog_ops = { 3805 .get_func_proto = sock_ops_func_proto, 3806 .is_valid_access = sock_ops_is_valid_access, 3807 .convert_ctx_access = sock_ops_convert_ctx_access, 3808 }; 3809 3810 int sk_detach_filter(struct sock *sk) 3811 { 3812 int ret = -ENOENT; 3813 struct sk_filter *filter; 3814 3815 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 3816 return -EPERM; 3817 3818 filter = rcu_dereference_protected(sk->sk_filter, 3819 lockdep_sock_is_held(sk)); 3820 if (filter) { 3821 RCU_INIT_POINTER(sk->sk_filter, NULL); 3822 sk_filter_uncharge(sk, filter); 3823 ret = 0; 3824 } 3825 3826 return ret; 3827 } 3828 EXPORT_SYMBOL_GPL(sk_detach_filter); 3829 3830 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 3831 unsigned int len) 3832 { 3833 struct sock_fprog_kern *fprog; 3834 struct sk_filter *filter; 3835 int ret = 0; 3836 3837 lock_sock(sk); 3838 filter = rcu_dereference_protected(sk->sk_filter, 3839 lockdep_sock_is_held(sk)); 3840 if (!filter) 3841 goto out; 3842 3843 /* We're copying the filter that has been originally attached, 3844 * so no conversion/decode needed anymore. eBPF programs that 3845 * have no original program cannot be dumped through this. 3846 */ 3847 ret = -EACCES; 3848 fprog = filter->prog->orig_prog; 3849 if (!fprog) 3850 goto out; 3851 3852 ret = fprog->len; 3853 if (!len) 3854 /* User space only enquires number of filter blocks. */ 3855 goto out; 3856 3857 ret = -EINVAL; 3858 if (len < fprog->len) 3859 goto out; 3860 3861 ret = -EFAULT; 3862 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 3863 goto out; 3864 3865 /* Instead of bytes, the API requests to return the number 3866 * of filter blocks. 3867 */ 3868 ret = fprog->len; 3869 out: 3870 release_sock(sk); 3871 return ret; 3872 } 3873