1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Linux Socket Filter Data Structures 4 */ 5#ifndef __LINUX_FILTER_H__ 6#define __LINUX_FILTER_H__ 7 8#include <linux/atomic.h> 9#include <linux/bpf.h> 10#include <linux/refcount.h> 11#include <linux/compat.h> 12#include <linux/skbuff.h> 13#include <linux/linkage.h> 14#include <linux/printk.h> 15#include <linux/workqueue.h> 16#include <linux/sched.h> 17#include <linux/sched/clock.h> 18#include <linux/capability.h> 19#include <linux/set_memory.h> 20#include <linux/kallsyms.h> 21#include <linux/if_vlan.h> 22#include <linux/vmalloc.h> 23#include <linux/sockptr.h> 24#include <crypto/sha1.h> 25#include <linux/u64_stats_sync.h> 26 27#include <net/sch_generic.h> 28 29#include <asm/byteorder.h> 30#include <uapi/linux/filter.h> 31 32struct sk_buff; 33struct sock; 34struct seccomp_data; 35struct bpf_prog_aux; 36struct xdp_rxq_info; 37struct xdp_buff; 38struct sock_reuseport; 39struct ctl_table; 40struct ctl_table_header; 41 42/* ArgX, context and stack frame pointer register positions. Note, 43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 44 * calls in BPF_CALL instruction. 45 */ 46#define BPF_REG_ARG1 BPF_REG_1 47#define BPF_REG_ARG2 BPF_REG_2 48#define BPF_REG_ARG3 BPF_REG_3 49#define BPF_REG_ARG4 BPF_REG_4 50#define BPF_REG_ARG5 BPF_REG_5 51#define BPF_REG_CTX BPF_REG_6 52#define BPF_REG_FP BPF_REG_10 53 54/* Additional register mappings for converted user programs. */ 55#define BPF_REG_A BPF_REG_0 56#define BPF_REG_X BPF_REG_7 57#define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 58#define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 59#define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 60 61/* Kernel hidden auxiliary/helper register. */ 62#define BPF_REG_AX MAX_BPF_REG 63#define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 64#define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 65 66/* unused opcode to mark special call to bpf_tail_call() helper */ 67#define BPF_TAIL_CALL 0xf0 68 69/* unused opcode to mark special load instruction. Same as BPF_ABS */ 70#define BPF_PROBE_MEM 0x20 71 72/* unused opcode to mark special ldsx instruction. Same as BPF_IND */ 73#define BPF_PROBE_MEMSX 0x40 74 75/* unused opcode to mark special load instruction. Same as BPF_MSH */ 76#define BPF_PROBE_MEM32 0xa0 77 78/* unused opcode to mark call to interpreter with arguments */ 79#define BPF_CALL_ARGS 0xe0 80 81/* unused opcode to mark speculation barrier for mitigating 82 * Speculative Store Bypass 83 */ 84#define BPF_NOSPEC 0xc0 85 86/* As per nm, we expose JITed images as text (code) section for 87 * kallsyms. That way, tools like perf can find it to match 88 * addresses. 89 */ 90#define BPF_SYM_ELF_TYPE 't' 91 92/* BPF program can access up to 512 bytes of stack space. */ 93#define MAX_BPF_STACK 512 94 95/* Helper macros for filter block array initializers. */ 96 97/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 98 99#define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \ 100 ((struct bpf_insn) { \ 101 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 102 .dst_reg = DST, \ 103 .src_reg = SRC, \ 104 .off = OFF, \ 105 .imm = 0 }) 106 107#define BPF_ALU64_REG(OP, DST, SRC) \ 108 BPF_ALU64_REG_OFF(OP, DST, SRC, 0) 109 110#define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \ 111 ((struct bpf_insn) { \ 112 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 113 .dst_reg = DST, \ 114 .src_reg = SRC, \ 115 .off = OFF, \ 116 .imm = 0 }) 117 118#define BPF_ALU32_REG(OP, DST, SRC) \ 119 BPF_ALU32_REG_OFF(OP, DST, SRC, 0) 120 121/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 122 123#define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \ 124 ((struct bpf_insn) { \ 125 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 126 .dst_reg = DST, \ 127 .src_reg = 0, \ 128 .off = OFF, \ 129 .imm = IMM }) 130#define BPF_ALU64_IMM(OP, DST, IMM) \ 131 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0) 132 133#define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \ 134 ((struct bpf_insn) { \ 135 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 136 .dst_reg = DST, \ 137 .src_reg = 0, \ 138 .off = OFF, \ 139 .imm = IMM }) 140#define BPF_ALU32_IMM(OP, DST, IMM) \ 141 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0) 142 143/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 144 145#define BPF_ENDIAN(TYPE, DST, LEN) \ 146 ((struct bpf_insn) { \ 147 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 148 .dst_reg = DST, \ 149 .src_reg = 0, \ 150 .off = 0, \ 151 .imm = LEN }) 152 153/* Byte Swap, bswap16/32/64 */ 154 155#define BPF_BSWAP(DST, LEN) \ 156 ((struct bpf_insn) { \ 157 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \ 158 .dst_reg = DST, \ 159 .src_reg = 0, \ 160 .off = 0, \ 161 .imm = LEN }) 162 163/* Short form of mov, dst_reg = src_reg */ 164 165#define BPF_MOV64_REG(DST, SRC) \ 166 ((struct bpf_insn) { \ 167 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 168 .dst_reg = DST, \ 169 .src_reg = SRC, \ 170 .off = 0, \ 171 .imm = 0 }) 172 173#define BPF_MOV32_REG(DST, SRC) \ 174 ((struct bpf_insn) { \ 175 .code = BPF_ALU | BPF_MOV | BPF_X, \ 176 .dst_reg = DST, \ 177 .src_reg = SRC, \ 178 .off = 0, \ 179 .imm = 0 }) 180 181/* Short form of mov, dst_reg = imm32 */ 182 183#define BPF_MOV64_IMM(DST, IMM) \ 184 ((struct bpf_insn) { \ 185 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 186 .dst_reg = DST, \ 187 .src_reg = 0, \ 188 .off = 0, \ 189 .imm = IMM }) 190 191#define BPF_MOV32_IMM(DST, IMM) \ 192 ((struct bpf_insn) { \ 193 .code = BPF_ALU | BPF_MOV | BPF_K, \ 194 .dst_reg = DST, \ 195 .src_reg = 0, \ 196 .off = 0, \ 197 .imm = IMM }) 198 199/* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 200 201#define BPF_MOVSX64_REG(DST, SRC, OFF) \ 202 ((struct bpf_insn) { \ 203 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 204 .dst_reg = DST, \ 205 .src_reg = SRC, \ 206 .off = OFF, \ 207 .imm = 0 }) 208 209#define BPF_MOVSX32_REG(DST, SRC, OFF) \ 210 ((struct bpf_insn) { \ 211 .code = BPF_ALU | BPF_MOV | BPF_X, \ 212 .dst_reg = DST, \ 213 .src_reg = SRC, \ 214 .off = OFF, \ 215 .imm = 0 }) 216 217/* Special form of mov32, used for doing explicit zero extension on dst. */ 218#define BPF_ZEXT_REG(DST) \ 219 ((struct bpf_insn) { \ 220 .code = BPF_ALU | BPF_MOV | BPF_X, \ 221 .dst_reg = DST, \ 222 .src_reg = DST, \ 223 .off = 0, \ 224 .imm = 1 }) 225 226static inline bool insn_is_zext(const struct bpf_insn *insn) 227{ 228 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 229} 230 231/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 232#define BPF_LD_IMM64(DST, IMM) \ 233 BPF_LD_IMM64_RAW(DST, 0, IMM) 234 235#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 236 ((struct bpf_insn) { \ 237 .code = BPF_LD | BPF_DW | BPF_IMM, \ 238 .dst_reg = DST, \ 239 .src_reg = SRC, \ 240 .off = 0, \ 241 .imm = (__u32) (IMM) }), \ 242 ((struct bpf_insn) { \ 243 .code = 0, /* zero is reserved opcode */ \ 244 .dst_reg = 0, \ 245 .src_reg = 0, \ 246 .off = 0, \ 247 .imm = ((__u64) (IMM)) >> 32 }) 248 249/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 250#define BPF_LD_MAP_FD(DST, MAP_FD) \ 251 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 252 253/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 254 255#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 256 ((struct bpf_insn) { \ 257 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 258 .dst_reg = DST, \ 259 .src_reg = SRC, \ 260 .off = 0, \ 261 .imm = IMM }) 262 263#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 264 ((struct bpf_insn) { \ 265 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 266 .dst_reg = DST, \ 267 .src_reg = SRC, \ 268 .off = 0, \ 269 .imm = IMM }) 270 271/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 272 273#define BPF_LD_ABS(SIZE, IMM) \ 274 ((struct bpf_insn) { \ 275 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 276 .dst_reg = 0, \ 277 .src_reg = 0, \ 278 .off = 0, \ 279 .imm = IMM }) 280 281/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 282 283#define BPF_LD_IND(SIZE, SRC, IMM) \ 284 ((struct bpf_insn) { \ 285 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 286 .dst_reg = 0, \ 287 .src_reg = SRC, \ 288 .off = 0, \ 289 .imm = IMM }) 290 291/* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 292 293#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 294 ((struct bpf_insn) { \ 295 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 296 .dst_reg = DST, \ 297 .src_reg = SRC, \ 298 .off = OFF, \ 299 .imm = 0 }) 300 301/* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 302 303#define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 304 ((struct bpf_insn) { \ 305 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 306 .dst_reg = DST, \ 307 .src_reg = SRC, \ 308 .off = OFF, \ 309 .imm = 0 }) 310 311/* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 312 313#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 314 ((struct bpf_insn) { \ 315 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 316 .dst_reg = DST, \ 317 .src_reg = SRC, \ 318 .off = OFF, \ 319 .imm = 0 }) 320 321 322/* 323 * Atomic operations: 324 * 325 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 326 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 327 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 328 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 329 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 330 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 331 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 332 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 333 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 334 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 335 */ 336 337#define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 338 ((struct bpf_insn) { \ 339 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 340 .dst_reg = DST, \ 341 .src_reg = SRC, \ 342 .off = OFF, \ 343 .imm = OP }) 344 345/* Legacy alias */ 346#define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 347 348/* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 349 350#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 351 ((struct bpf_insn) { \ 352 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 353 .dst_reg = DST, \ 354 .src_reg = 0, \ 355 .off = OFF, \ 356 .imm = IMM }) 357 358/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 359 360#define BPF_JMP_REG(OP, DST, SRC, OFF) \ 361 ((struct bpf_insn) { \ 362 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 363 .dst_reg = DST, \ 364 .src_reg = SRC, \ 365 .off = OFF, \ 366 .imm = 0 }) 367 368/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 369 370#define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 371 ((struct bpf_insn) { \ 372 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 373 .dst_reg = DST, \ 374 .src_reg = 0, \ 375 .off = OFF, \ 376 .imm = IMM }) 377 378/* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 379 380#define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 381 ((struct bpf_insn) { \ 382 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 383 .dst_reg = DST, \ 384 .src_reg = SRC, \ 385 .off = OFF, \ 386 .imm = 0 }) 387 388/* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 389 390#define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 391 ((struct bpf_insn) { \ 392 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 393 .dst_reg = DST, \ 394 .src_reg = 0, \ 395 .off = OFF, \ 396 .imm = IMM }) 397 398/* Unconditional jumps, goto pc + off16 */ 399 400#define BPF_JMP_A(OFF) \ 401 ((struct bpf_insn) { \ 402 .code = BPF_JMP | BPF_JA, \ 403 .dst_reg = 0, \ 404 .src_reg = 0, \ 405 .off = OFF, \ 406 .imm = 0 }) 407 408/* Relative call */ 409 410#define BPF_CALL_REL(TGT) \ 411 ((struct bpf_insn) { \ 412 .code = BPF_JMP | BPF_CALL, \ 413 .dst_reg = 0, \ 414 .src_reg = BPF_PSEUDO_CALL, \ 415 .off = 0, \ 416 .imm = TGT }) 417 418/* Convert function address to BPF immediate */ 419 420#define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 421 422#define BPF_EMIT_CALL(FUNC) \ 423 ((struct bpf_insn) { \ 424 .code = BPF_JMP | BPF_CALL, \ 425 .dst_reg = 0, \ 426 .src_reg = 0, \ 427 .off = 0, \ 428 .imm = BPF_CALL_IMM(FUNC) }) 429 430/* Raw code statement block */ 431 432#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 433 ((struct bpf_insn) { \ 434 .code = CODE, \ 435 .dst_reg = DST, \ 436 .src_reg = SRC, \ 437 .off = OFF, \ 438 .imm = IMM }) 439 440/* Program exit */ 441 442#define BPF_EXIT_INSN() \ 443 ((struct bpf_insn) { \ 444 .code = BPF_JMP | BPF_EXIT, \ 445 .dst_reg = 0, \ 446 .src_reg = 0, \ 447 .off = 0, \ 448 .imm = 0 }) 449 450/* Speculation barrier */ 451 452#define BPF_ST_NOSPEC() \ 453 ((struct bpf_insn) { \ 454 .code = BPF_ST | BPF_NOSPEC, \ 455 .dst_reg = 0, \ 456 .src_reg = 0, \ 457 .off = 0, \ 458 .imm = 0 }) 459 460/* Internal classic blocks for direct assignment */ 461 462#define __BPF_STMT(CODE, K) \ 463 ((struct sock_filter) BPF_STMT(CODE, K)) 464 465#define __BPF_JUMP(CODE, K, JT, JF) \ 466 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 467 468#define bytes_to_bpf_size(bytes) \ 469({ \ 470 int bpf_size = -EINVAL; \ 471 \ 472 if (bytes == sizeof(u8)) \ 473 bpf_size = BPF_B; \ 474 else if (bytes == sizeof(u16)) \ 475 bpf_size = BPF_H; \ 476 else if (bytes == sizeof(u32)) \ 477 bpf_size = BPF_W; \ 478 else if (bytes == sizeof(u64)) \ 479 bpf_size = BPF_DW; \ 480 \ 481 bpf_size; \ 482}) 483 484#define bpf_size_to_bytes(bpf_size) \ 485({ \ 486 int bytes = -EINVAL; \ 487 \ 488 if (bpf_size == BPF_B) \ 489 bytes = sizeof(u8); \ 490 else if (bpf_size == BPF_H) \ 491 bytes = sizeof(u16); \ 492 else if (bpf_size == BPF_W) \ 493 bytes = sizeof(u32); \ 494 else if (bpf_size == BPF_DW) \ 495 bytes = sizeof(u64); \ 496 \ 497 bytes; \ 498}) 499 500#define BPF_SIZEOF(type) \ 501 ({ \ 502 const int __size = bytes_to_bpf_size(sizeof(type)); \ 503 BUILD_BUG_ON(__size < 0); \ 504 __size; \ 505 }) 506 507#define BPF_FIELD_SIZEOF(type, field) \ 508 ({ \ 509 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 510 BUILD_BUG_ON(__size < 0); \ 511 __size; \ 512 }) 513 514#define BPF_LDST_BYTES(insn) \ 515 ({ \ 516 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 517 WARN_ON(__size < 0); \ 518 __size; \ 519 }) 520 521#define __BPF_MAP_0(m, v, ...) v 522#define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 523#define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 524#define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 525#define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 526#define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 527 528#define __BPF_REG_0(...) __BPF_PAD(5) 529#define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 530#define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 531#define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 532#define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 533#define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 534 535#define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 536#define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 537 538#define __BPF_CAST(t, a) \ 539 (__force t) \ 540 (__force \ 541 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 542 (unsigned long)0, (t)0))) a 543#define __BPF_V void 544#define __BPF_N 545 546#define __BPF_DECL_ARGS(t, a) t a 547#define __BPF_DECL_REGS(t, a) u64 a 548 549#define __BPF_PAD(n) \ 550 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 551 u64, __ur_3, u64, __ur_4, u64, __ur_5) 552 553#define BPF_CALL_x(x, attr, name, ...) \ 554 static __always_inline \ 555 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 556 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 557 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 558 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 559 { \ 560 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 561 } \ 562 static __always_inline \ 563 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 564 565#define __NOATTR 566#define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 567#define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 568#define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 569#define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 570#define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 571#define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 572 573#define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 574 575#define bpf_ctx_range(TYPE, MEMBER) \ 576 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 577#define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 578 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 579#if BITS_PER_LONG == 64 580# define bpf_ctx_range_ptr(TYPE, MEMBER) \ 581 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 582#else 583# define bpf_ctx_range_ptr(TYPE, MEMBER) \ 584 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 585#endif /* BITS_PER_LONG == 64 */ 586 587#define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 588 ({ \ 589 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 590 *(PTR_SIZE) = (SIZE); \ 591 offsetof(TYPE, MEMBER); \ 592 }) 593 594/* A struct sock_filter is architecture independent. */ 595struct compat_sock_fprog { 596 u16 len; 597 compat_uptr_t filter; /* struct sock_filter * */ 598}; 599 600struct sock_fprog_kern { 601 u16 len; 602 struct sock_filter *filter; 603}; 604 605/* Some arches need doubleword alignment for their instructions and/or data */ 606#define BPF_IMAGE_ALIGNMENT 8 607 608struct bpf_binary_header { 609 u32 size; 610 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 611}; 612 613struct bpf_prog_stats { 614 u64_stats_t cnt; 615 u64_stats_t nsecs; 616 u64_stats_t misses; 617 struct u64_stats_sync syncp; 618} __aligned(2 * sizeof(u64)); 619 620struct sk_filter { 621 refcount_t refcnt; 622 struct rcu_head rcu; 623 struct bpf_prog *prog; 624}; 625 626DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 627 628extern struct mutex nf_conn_btf_access_lock; 629extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 630 const struct bpf_reg_state *reg, 631 int off, int size); 632 633typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 634 const struct bpf_insn *insnsi, 635 unsigned int (*bpf_func)(const void *, 636 const struct bpf_insn *)); 637 638static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 639 const void *ctx, 640 bpf_dispatcher_fn dfunc) 641{ 642 u32 ret; 643 644 cant_migrate(); 645 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 646 struct bpf_prog_stats *stats; 647 u64 start = sched_clock(); 648 unsigned long flags; 649 650 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 651 stats = this_cpu_ptr(prog->stats); 652 flags = u64_stats_update_begin_irqsave(&stats->syncp); 653 u64_stats_inc(&stats->cnt); 654 u64_stats_add(&stats->nsecs, sched_clock() - start); 655 u64_stats_update_end_irqrestore(&stats->syncp, flags); 656 } else { 657 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 658 } 659 return ret; 660} 661 662static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 663{ 664 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 665} 666 667/* 668 * Use in preemptible and therefore migratable context to make sure that 669 * the execution of the BPF program runs on one CPU. 670 * 671 * This uses migrate_disable/enable() explicitly to document that the 672 * invocation of a BPF program does not require reentrancy protection 673 * against a BPF program which is invoked from a preempting task. 674 */ 675static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 676 const void *ctx) 677{ 678 u32 ret; 679 680 migrate_disable(); 681 ret = bpf_prog_run(prog, ctx); 682 migrate_enable(); 683 return ret; 684} 685 686#define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 687 688struct bpf_skb_data_end { 689 struct qdisc_skb_cb qdisc_cb; 690 void *data_meta; 691 void *data_end; 692}; 693 694struct bpf_nh_params { 695 u32 nh_family; 696 union { 697 u32 ipv4_nh; 698 struct in6_addr ipv6_nh; 699 }; 700}; 701 702struct bpf_redirect_info { 703 u64 tgt_index; 704 void *tgt_value; 705 struct bpf_map *map; 706 u32 flags; 707 u32 kern_flags; 708 u32 map_id; 709 enum bpf_map_type map_type; 710 struct bpf_nh_params nh; 711}; 712 713DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 714 715/* flags for bpf_redirect_info kern_flags */ 716#define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 717 718/* Compute the linear packet data range [data, data_end) which 719 * will be accessed by various program types (cls_bpf, act_bpf, 720 * lwt, ...). Subsystems allowing direct data access must (!) 721 * ensure that cb[] area can be written to when BPF program is 722 * invoked (otherwise cb[] save/restore is necessary). 723 */ 724static inline void bpf_compute_data_pointers(struct sk_buff *skb) 725{ 726 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 727 728 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 729 cb->data_meta = skb->data - skb_metadata_len(skb); 730 cb->data_end = skb->data + skb_headlen(skb); 731} 732 733/* Similar to bpf_compute_data_pointers(), except that save orginal 734 * data in cb->data and cb->meta_data for restore. 735 */ 736static inline void bpf_compute_and_save_data_end( 737 struct sk_buff *skb, void **saved_data_end) 738{ 739 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 740 741 *saved_data_end = cb->data_end; 742 cb->data_end = skb->data + skb_headlen(skb); 743} 744 745/* Restore data saved by bpf_compute_and_save_data_end(). */ 746static inline void bpf_restore_data_end( 747 struct sk_buff *skb, void *saved_data_end) 748{ 749 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 750 751 cb->data_end = saved_data_end; 752} 753 754static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 755{ 756 /* eBPF programs may read/write skb->cb[] area to transfer meta 757 * data between tail calls. Since this also needs to work with 758 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 759 * 760 * In some socket filter cases, the cb unfortunately needs to be 761 * saved/restored so that protocol specific skb->cb[] data won't 762 * be lost. In any case, due to unpriviledged eBPF programs 763 * attached to sockets, we need to clear the bpf_skb_cb() area 764 * to not leak previous contents to user space. 765 */ 766 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 767 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 768 sizeof_field(struct qdisc_skb_cb, data)); 769 770 return qdisc_skb_cb(skb)->data; 771} 772 773/* Must be invoked with migration disabled */ 774static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 775 const void *ctx) 776{ 777 const struct sk_buff *skb = ctx; 778 u8 *cb_data = bpf_skb_cb(skb); 779 u8 cb_saved[BPF_SKB_CB_LEN]; 780 u32 res; 781 782 if (unlikely(prog->cb_access)) { 783 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 784 memset(cb_data, 0, sizeof(cb_saved)); 785 } 786 787 res = bpf_prog_run(prog, skb); 788 789 if (unlikely(prog->cb_access)) 790 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 791 792 return res; 793} 794 795static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 796 struct sk_buff *skb) 797{ 798 u32 res; 799 800 migrate_disable(); 801 res = __bpf_prog_run_save_cb(prog, skb); 802 migrate_enable(); 803 return res; 804} 805 806static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 807 struct sk_buff *skb) 808{ 809 u8 *cb_data = bpf_skb_cb(skb); 810 u32 res; 811 812 if (unlikely(prog->cb_access)) 813 memset(cb_data, 0, BPF_SKB_CB_LEN); 814 815 res = bpf_prog_run_pin_on_cpu(prog, skb); 816 return res; 817} 818 819DECLARE_BPF_DISPATCHER(xdp) 820 821DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 822 823u32 xdp_master_redirect(struct xdp_buff *xdp); 824 825void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 826 827static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 828{ 829 return prog->len * sizeof(struct bpf_insn); 830} 831 832static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 833{ 834 return round_up(bpf_prog_insn_size(prog) + 835 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 836} 837 838static inline unsigned int bpf_prog_size(unsigned int proglen) 839{ 840 return max(sizeof(struct bpf_prog), 841 offsetof(struct bpf_prog, insns[proglen])); 842} 843 844static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 845{ 846 /* When classic BPF programs have been loaded and the arch 847 * does not have a classic BPF JIT (anymore), they have been 848 * converted via bpf_migrate_filter() to eBPF and thus always 849 * have an unspec program type. 850 */ 851 return prog->type == BPF_PROG_TYPE_UNSPEC; 852} 853 854static inline u32 bpf_ctx_off_adjust_machine(u32 size) 855{ 856 const u32 size_machine = sizeof(unsigned long); 857 858 if (size > size_machine && size % size_machine == 0) 859 size = size_machine; 860 861 return size; 862} 863 864static inline bool 865bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 866{ 867 return size <= size_default && (size & (size - 1)) == 0; 868} 869 870static inline u8 871bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 872{ 873 u8 access_off = off & (size_default - 1); 874 875#ifdef __LITTLE_ENDIAN 876 return access_off; 877#else 878 return size_default - (access_off + size); 879#endif 880} 881 882#define bpf_ctx_wide_access_ok(off, size, type, field) \ 883 (size == sizeof(__u64) && \ 884 off >= offsetof(type, field) && \ 885 off + sizeof(__u64) <= offsetofend(type, field) && \ 886 off % sizeof(__u64) == 0) 887 888#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 889 890static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 891{ 892#ifndef CONFIG_BPF_JIT_ALWAYS_ON 893 if (!fp->jited) { 894 set_vm_flush_reset_perms(fp); 895 set_memory_ro((unsigned long)fp, fp->pages); 896 } 897#endif 898} 899 900static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 901{ 902 set_vm_flush_reset_perms(hdr); 903 set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 904} 905 906int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 907static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 908{ 909 return sk_filter_trim_cap(sk, skb, 1); 910} 911 912struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 913void bpf_prog_free(struct bpf_prog *fp); 914 915bool bpf_opcode_in_insntable(u8 code); 916 917void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 918 const u32 *insn_to_jit_off); 919int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 920void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 921 922struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 923struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 924struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 925 gfp_t gfp_extra_flags); 926void __bpf_prog_free(struct bpf_prog *fp); 927 928static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 929{ 930 __bpf_prog_free(fp); 931} 932 933typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 934 unsigned int flen); 935 936int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 937int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 938 bpf_aux_classic_check_t trans, bool save_orig); 939void bpf_prog_destroy(struct bpf_prog *fp); 940 941int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 942int sk_attach_bpf(u32 ufd, struct sock *sk); 943int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 944int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 945void sk_reuseport_prog_free(struct bpf_prog *prog); 946int sk_detach_filter(struct sock *sk); 947int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 948 949bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 950void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 951 952u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 953#define __bpf_call_base_args \ 954 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 955 (void *)__bpf_call_base) 956 957struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 958void bpf_jit_compile(struct bpf_prog *prog); 959bool bpf_jit_needs_zext(void); 960bool bpf_jit_supports_subprog_tailcalls(void); 961bool bpf_jit_supports_kfunc_call(void); 962bool bpf_jit_supports_far_kfunc_call(void); 963bool bpf_jit_supports_exceptions(void); 964bool bpf_jit_supports_ptr_xchg(void); 965bool bpf_jit_supports_arena(void); 966void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 967bool bpf_helper_changes_pkt_data(void *func); 968 969static inline bool bpf_dump_raw_ok(const struct cred *cred) 970{ 971 /* Reconstruction of call-sites is dependent on kallsyms, 972 * thus make dump the same restriction. 973 */ 974 return kallsyms_show_value(cred); 975} 976 977struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 978 const struct bpf_insn *patch, u32 len); 979int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 980 981void bpf_clear_redirect_map(struct bpf_map *map); 982 983static inline bool xdp_return_frame_no_direct(void) 984{ 985 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 986 987 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 988} 989 990static inline void xdp_set_return_frame_no_direct(void) 991{ 992 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 993 994 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 995} 996 997static inline void xdp_clear_return_frame_no_direct(void) 998{ 999 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1000 1001 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1002} 1003 1004static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1005 unsigned int pktlen) 1006{ 1007 unsigned int len; 1008 1009 if (unlikely(!(fwd->flags & IFF_UP))) 1010 return -ENETDOWN; 1011 1012 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1013 if (pktlen > len) 1014 return -EMSGSIZE; 1015 1016 return 0; 1017} 1018 1019/* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1020 * same cpu context. Further for best results no more than a single map 1021 * for the do_redirect/do_flush pair should be used. This limitation is 1022 * because we only track one map and force a flush when the map changes. 1023 * This does not appear to be a real limitation for existing software. 1024 */ 1025int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1026 struct xdp_buff *xdp, struct bpf_prog *prog); 1027int xdp_do_redirect(struct net_device *dev, 1028 struct xdp_buff *xdp, 1029 struct bpf_prog *prog); 1030int xdp_do_redirect_frame(struct net_device *dev, 1031 struct xdp_buff *xdp, 1032 struct xdp_frame *xdpf, 1033 struct bpf_prog *prog); 1034void xdp_do_flush(void); 1035 1036void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1037 1038#ifdef CONFIG_INET 1039struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1040 struct bpf_prog *prog, struct sk_buff *skb, 1041 struct sock *migrating_sk, 1042 u32 hash); 1043#else 1044static inline struct sock * 1045bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1046 struct bpf_prog *prog, struct sk_buff *skb, 1047 struct sock *migrating_sk, 1048 u32 hash) 1049{ 1050 return NULL; 1051} 1052#endif 1053 1054#ifdef CONFIG_BPF_JIT 1055extern int bpf_jit_enable; 1056extern int bpf_jit_harden; 1057extern int bpf_jit_kallsyms; 1058extern long bpf_jit_limit; 1059extern long bpf_jit_limit_max; 1060 1061typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1062 1063void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1064 1065struct bpf_binary_header * 1066bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1067 unsigned int alignment, 1068 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1069void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1070u64 bpf_jit_alloc_exec_limit(void); 1071void *bpf_jit_alloc_exec(unsigned long size); 1072void bpf_jit_free_exec(void *addr); 1073void bpf_jit_free(struct bpf_prog *fp); 1074struct bpf_binary_header * 1075bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1076 1077void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1078void bpf_prog_pack_free(void *ptr, u32 size); 1079 1080static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1081{ 1082 return list_empty(&fp->aux->ksym.lnode) || 1083 fp->aux->ksym.lnode.prev == LIST_POISON2; 1084} 1085 1086struct bpf_binary_header * 1087bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1088 unsigned int alignment, 1089 struct bpf_binary_header **rw_hdr, 1090 u8 **rw_image, 1091 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1092int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1093 struct bpf_binary_header *ro_header, 1094 struct bpf_binary_header *rw_header); 1095void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1096 struct bpf_binary_header *rw_header); 1097 1098int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1099 struct bpf_jit_poke_descriptor *poke); 1100 1101int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1102 const struct bpf_insn *insn, bool extra_pass, 1103 u64 *func_addr, bool *func_addr_fixed); 1104 1105struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1106void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1107 1108static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1109 u32 pass, void *image) 1110{ 1111 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1112 proglen, pass, image, current->comm, task_pid_nr(current)); 1113 1114 if (image) 1115 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1116 16, 1, image, proglen, false); 1117} 1118 1119static inline bool bpf_jit_is_ebpf(void) 1120{ 1121# ifdef CONFIG_HAVE_EBPF_JIT 1122 return true; 1123# else 1124 return false; 1125# endif 1126} 1127 1128static inline bool ebpf_jit_enabled(void) 1129{ 1130 return bpf_jit_enable && bpf_jit_is_ebpf(); 1131} 1132 1133static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1134{ 1135 return fp->jited && bpf_jit_is_ebpf(); 1136} 1137 1138static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1139{ 1140 /* These are the prerequisites, should someone ever have the 1141 * idea to call blinding outside of them, we make sure to 1142 * bail out. 1143 */ 1144 if (!bpf_jit_is_ebpf()) 1145 return false; 1146 if (!prog->jit_requested) 1147 return false; 1148 if (!bpf_jit_harden) 1149 return false; 1150 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1151 return false; 1152 1153 return true; 1154} 1155 1156static inline bool bpf_jit_kallsyms_enabled(void) 1157{ 1158 /* There are a couple of corner cases where kallsyms should 1159 * not be enabled f.e. on hardening. 1160 */ 1161 if (bpf_jit_harden) 1162 return false; 1163 if (!bpf_jit_kallsyms) 1164 return false; 1165 if (bpf_jit_kallsyms == 1) 1166 return true; 1167 1168 return false; 1169} 1170 1171const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1172 unsigned long *off, char *sym); 1173bool is_bpf_text_address(unsigned long addr); 1174int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1175 char *sym); 1176struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1177 1178static inline const char * 1179bpf_address_lookup(unsigned long addr, unsigned long *size, 1180 unsigned long *off, char **modname, char *sym) 1181{ 1182 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1183 1184 if (ret && modname) 1185 *modname = NULL; 1186 return ret; 1187} 1188 1189void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1190void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1191 1192#else /* CONFIG_BPF_JIT */ 1193 1194static inline bool ebpf_jit_enabled(void) 1195{ 1196 return false; 1197} 1198 1199static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1200{ 1201 return false; 1202} 1203 1204static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1205{ 1206 return false; 1207} 1208 1209static inline int 1210bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1211 struct bpf_jit_poke_descriptor *poke) 1212{ 1213 return -ENOTSUPP; 1214} 1215 1216static inline void bpf_jit_free(struct bpf_prog *fp) 1217{ 1218 bpf_prog_unlock_free(fp); 1219} 1220 1221static inline bool bpf_jit_kallsyms_enabled(void) 1222{ 1223 return false; 1224} 1225 1226static inline const char * 1227__bpf_address_lookup(unsigned long addr, unsigned long *size, 1228 unsigned long *off, char *sym) 1229{ 1230 return NULL; 1231} 1232 1233static inline bool is_bpf_text_address(unsigned long addr) 1234{ 1235 return false; 1236} 1237 1238static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1239 char *type, char *sym) 1240{ 1241 return -ERANGE; 1242} 1243 1244static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1245{ 1246 return NULL; 1247} 1248 1249static inline const char * 1250bpf_address_lookup(unsigned long addr, unsigned long *size, 1251 unsigned long *off, char **modname, char *sym) 1252{ 1253 return NULL; 1254} 1255 1256static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1257{ 1258} 1259 1260static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1261{ 1262} 1263 1264#endif /* CONFIG_BPF_JIT */ 1265 1266void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1267 1268#define BPF_ANC BIT(15) 1269 1270static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1271{ 1272 switch (first->code) { 1273 case BPF_RET | BPF_K: 1274 case BPF_LD | BPF_W | BPF_LEN: 1275 return false; 1276 1277 case BPF_LD | BPF_W | BPF_ABS: 1278 case BPF_LD | BPF_H | BPF_ABS: 1279 case BPF_LD | BPF_B | BPF_ABS: 1280 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1281 return true; 1282 return false; 1283 1284 default: 1285 return true; 1286 } 1287} 1288 1289static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1290{ 1291 BUG_ON(ftest->code & BPF_ANC); 1292 1293 switch (ftest->code) { 1294 case BPF_LD | BPF_W | BPF_ABS: 1295 case BPF_LD | BPF_H | BPF_ABS: 1296 case BPF_LD | BPF_B | BPF_ABS: 1297#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1298 return BPF_ANC | SKF_AD_##CODE 1299 switch (ftest->k) { 1300 BPF_ANCILLARY(PROTOCOL); 1301 BPF_ANCILLARY(PKTTYPE); 1302 BPF_ANCILLARY(IFINDEX); 1303 BPF_ANCILLARY(NLATTR); 1304 BPF_ANCILLARY(NLATTR_NEST); 1305 BPF_ANCILLARY(MARK); 1306 BPF_ANCILLARY(QUEUE); 1307 BPF_ANCILLARY(HATYPE); 1308 BPF_ANCILLARY(RXHASH); 1309 BPF_ANCILLARY(CPU); 1310 BPF_ANCILLARY(ALU_XOR_X); 1311 BPF_ANCILLARY(VLAN_TAG); 1312 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1313 BPF_ANCILLARY(PAY_OFFSET); 1314 BPF_ANCILLARY(RANDOM); 1315 BPF_ANCILLARY(VLAN_TPID); 1316 } 1317 fallthrough; 1318 default: 1319 return ftest->code; 1320 } 1321} 1322 1323void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1324 int k, unsigned int size); 1325 1326static inline int bpf_tell_extensions(void) 1327{ 1328 return SKF_AD_MAX; 1329} 1330 1331struct bpf_sock_addr_kern { 1332 struct sock *sk; 1333 struct sockaddr *uaddr; 1334 /* Temporary "register" to make indirect stores to nested structures 1335 * defined above. We need three registers to make such a store, but 1336 * only two (src and dst) are available at convert_ctx_access time 1337 */ 1338 u64 tmp_reg; 1339 void *t_ctx; /* Attach type specific context. */ 1340 u32 uaddrlen; 1341}; 1342 1343struct bpf_sock_ops_kern { 1344 struct sock *sk; 1345 union { 1346 u32 args[4]; 1347 u32 reply; 1348 u32 replylong[4]; 1349 }; 1350 struct sk_buff *syn_skb; 1351 struct sk_buff *skb; 1352 void *skb_data_end; 1353 u8 op; 1354 u8 is_fullsock; 1355 u8 remaining_opt_len; 1356 u64 temp; /* temp and everything after is not 1357 * initialized to 0 before calling 1358 * the BPF program. New fields that 1359 * should be initialized to 0 should 1360 * be inserted before temp. 1361 * temp is scratch storage used by 1362 * sock_ops_convert_ctx_access 1363 * as temporary storage of a register. 1364 */ 1365}; 1366 1367struct bpf_sysctl_kern { 1368 struct ctl_table_header *head; 1369 struct ctl_table *table; 1370 void *cur_val; 1371 size_t cur_len; 1372 void *new_val; 1373 size_t new_len; 1374 int new_updated; 1375 int write; 1376 loff_t *ppos; 1377 /* Temporary "register" for indirect stores to ppos. */ 1378 u64 tmp_reg; 1379}; 1380 1381#define BPF_SOCKOPT_KERN_BUF_SIZE 32 1382struct bpf_sockopt_buf { 1383 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1384}; 1385 1386struct bpf_sockopt_kern { 1387 struct sock *sk; 1388 u8 *optval; 1389 u8 *optval_end; 1390 s32 level; 1391 s32 optname; 1392 s32 optlen; 1393 /* for retval in struct bpf_cg_run_ctx */ 1394 struct task_struct *current_task; 1395 /* Temporary "register" for indirect stores to ppos. */ 1396 u64 tmp_reg; 1397}; 1398 1399int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1400 1401struct bpf_sk_lookup_kern { 1402 u16 family; 1403 u16 protocol; 1404 __be16 sport; 1405 u16 dport; 1406 struct { 1407 __be32 saddr; 1408 __be32 daddr; 1409 } v4; 1410 struct { 1411 const struct in6_addr *saddr; 1412 const struct in6_addr *daddr; 1413 } v6; 1414 struct sock *selected_sk; 1415 u32 ingress_ifindex; 1416 bool no_reuseport; 1417}; 1418 1419extern struct static_key_false bpf_sk_lookup_enabled; 1420 1421/* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1422 * 1423 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1424 * SK_DROP. Their meaning is as follows: 1425 * 1426 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1427 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1428 * SK_DROP : terminate lookup with -ECONNREFUSED 1429 * 1430 * This macro aggregates return values and selected sockets from 1431 * multiple BPF programs according to following rules in order: 1432 * 1433 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1434 * macro result is SK_PASS and last ctx.selected_sk is used. 1435 * 2. If any program returned SK_DROP return value, 1436 * macro result is SK_DROP. 1437 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1438 * 1439 * Caller must ensure that the prog array is non-NULL, and that the 1440 * array as well as the programs it contains remain valid. 1441 */ 1442#define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1443 ({ \ 1444 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1445 struct bpf_prog_array_item *_item; \ 1446 struct sock *_selected_sk = NULL; \ 1447 bool _no_reuseport = false; \ 1448 struct bpf_prog *_prog; \ 1449 bool _all_pass = true; \ 1450 u32 _ret; \ 1451 \ 1452 migrate_disable(); \ 1453 _item = &(array)->items[0]; \ 1454 while ((_prog = READ_ONCE(_item->prog))) { \ 1455 /* restore most recent selection */ \ 1456 _ctx->selected_sk = _selected_sk; \ 1457 _ctx->no_reuseport = _no_reuseport; \ 1458 \ 1459 _ret = func(_prog, _ctx); \ 1460 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1461 /* remember last non-NULL socket */ \ 1462 _selected_sk = _ctx->selected_sk; \ 1463 _no_reuseport = _ctx->no_reuseport; \ 1464 } else if (_ret == SK_DROP && _all_pass) { \ 1465 _all_pass = false; \ 1466 } \ 1467 _item++; \ 1468 } \ 1469 _ctx->selected_sk = _selected_sk; \ 1470 _ctx->no_reuseport = _no_reuseport; \ 1471 migrate_enable(); \ 1472 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1473 }) 1474 1475static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1476 const __be32 saddr, const __be16 sport, 1477 const __be32 daddr, const u16 dport, 1478 const int ifindex, struct sock **psk) 1479{ 1480 struct bpf_prog_array *run_array; 1481 struct sock *selected_sk = NULL; 1482 bool no_reuseport = false; 1483 1484 rcu_read_lock(); 1485 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1486 if (run_array) { 1487 struct bpf_sk_lookup_kern ctx = { 1488 .family = AF_INET, 1489 .protocol = protocol, 1490 .v4.saddr = saddr, 1491 .v4.daddr = daddr, 1492 .sport = sport, 1493 .dport = dport, 1494 .ingress_ifindex = ifindex, 1495 }; 1496 u32 act; 1497 1498 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1499 if (act == SK_PASS) { 1500 selected_sk = ctx.selected_sk; 1501 no_reuseport = ctx.no_reuseport; 1502 } else { 1503 selected_sk = ERR_PTR(-ECONNREFUSED); 1504 } 1505 } 1506 rcu_read_unlock(); 1507 *psk = selected_sk; 1508 return no_reuseport; 1509} 1510 1511#if IS_ENABLED(CONFIG_IPV6) 1512static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1513 const struct in6_addr *saddr, 1514 const __be16 sport, 1515 const struct in6_addr *daddr, 1516 const u16 dport, 1517 const int ifindex, struct sock **psk) 1518{ 1519 struct bpf_prog_array *run_array; 1520 struct sock *selected_sk = NULL; 1521 bool no_reuseport = false; 1522 1523 rcu_read_lock(); 1524 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1525 if (run_array) { 1526 struct bpf_sk_lookup_kern ctx = { 1527 .family = AF_INET6, 1528 .protocol = protocol, 1529 .v6.saddr = saddr, 1530 .v6.daddr = daddr, 1531 .sport = sport, 1532 .dport = dport, 1533 .ingress_ifindex = ifindex, 1534 }; 1535 u32 act; 1536 1537 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1538 if (act == SK_PASS) { 1539 selected_sk = ctx.selected_sk; 1540 no_reuseport = ctx.no_reuseport; 1541 } else { 1542 selected_sk = ERR_PTR(-ECONNREFUSED); 1543 } 1544 } 1545 rcu_read_unlock(); 1546 *psk = selected_sk; 1547 return no_reuseport; 1548} 1549#endif /* IS_ENABLED(CONFIG_IPV6) */ 1550 1551static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1552 u64 flags, const u64 flag_mask, 1553 void *lookup_elem(struct bpf_map *map, u32 key)) 1554{ 1555 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 1556 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1557 1558 /* Lower bits of the flags are used as return code on lookup failure */ 1559 if (unlikely(flags & ~(action_mask | flag_mask))) 1560 return XDP_ABORTED; 1561 1562 ri->tgt_value = lookup_elem(map, index); 1563 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1564 /* If the lookup fails we want to clear out the state in the 1565 * redirect_info struct completely, so that if an eBPF program 1566 * performs multiple lookups, the last one always takes 1567 * precedence. 1568 */ 1569 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1570 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1571 return flags & action_mask; 1572 } 1573 1574 ri->tgt_index = index; 1575 ri->map_id = map->id; 1576 ri->map_type = map->map_type; 1577 1578 if (flags & BPF_F_BROADCAST) { 1579 WRITE_ONCE(ri->map, map); 1580 ri->flags = flags; 1581 } else { 1582 WRITE_ONCE(ri->map, NULL); 1583 ri->flags = 0; 1584 } 1585 1586 return XDP_REDIRECT; 1587} 1588 1589#ifdef CONFIG_NET 1590int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1591int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1592 u32 len, u64 flags); 1593int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1594int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1595void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1596void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1597 void *buf, unsigned long len, bool flush); 1598#else /* CONFIG_NET */ 1599static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1600 void *to, u32 len) 1601{ 1602 return -EOPNOTSUPP; 1603} 1604 1605static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1606 const void *from, u32 len, u64 flags) 1607{ 1608 return -EOPNOTSUPP; 1609} 1610 1611static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1612 void *buf, u32 len) 1613{ 1614 return -EOPNOTSUPP; 1615} 1616 1617static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1618 void *buf, u32 len) 1619{ 1620 return -EOPNOTSUPP; 1621} 1622 1623static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1624{ 1625 return NULL; 1626} 1627 1628static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1629 unsigned long len, bool flush) 1630{ 1631} 1632#endif /* CONFIG_NET */ 1633 1634#endif /* __LINUX_FILTER_H__ */ 1635