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 <stdarg.h> 9 10 #include <linux/atomic.h> 11 #include <linux/refcount.h> 12 #include <linux/compat.h> 13 #include <linux/skbuff.h> 14 #include <linux/linkage.h> 15 #include <linux/printk.h> 16 #include <linux/workqueue.h> 17 #include <linux/sched.h> 18 #include <linux/capability.h> 19 #include <linux/cryptohash.h> 20 #include <linux/set_memory.h> 21 #include <linux/kallsyms.h> 22 #include <linux/if_vlan.h> 23 24 #include <net/sch_generic.h> 25 26 #include <uapi/linux/filter.h> 27 #include <uapi/linux/bpf.h> 28 29 struct sk_buff; 30 struct sock; 31 struct seccomp_data; 32 struct bpf_prog_aux; 33 struct xdp_rxq_info; 34 struct xdp_buff; 35 struct sock_reuseport; 36 37 /* ArgX, context and stack frame pointer register positions. Note, 38 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 39 * calls in BPF_CALL instruction. 40 */ 41 #define BPF_REG_ARG1 BPF_REG_1 42 #define BPF_REG_ARG2 BPF_REG_2 43 #define BPF_REG_ARG3 BPF_REG_3 44 #define BPF_REG_ARG4 BPF_REG_4 45 #define BPF_REG_ARG5 BPF_REG_5 46 #define BPF_REG_CTX BPF_REG_6 47 #define BPF_REG_FP BPF_REG_10 48 49 /* Additional register mappings for converted user programs. */ 50 #define BPF_REG_A BPF_REG_0 51 #define BPF_REG_X BPF_REG_7 52 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 53 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 54 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 55 56 /* Kernel hidden auxiliary/helper register. */ 57 #define BPF_REG_AX MAX_BPF_REG 58 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 59 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 60 61 /* unused opcode to mark special call to bpf_tail_call() helper */ 62 #define BPF_TAIL_CALL 0xf0 63 64 /* unused opcode to mark call to interpreter with arguments */ 65 #define BPF_CALL_ARGS 0xe0 66 67 /* As per nm, we expose JITed images as text (code) section for 68 * kallsyms. That way, tools like perf can find it to match 69 * addresses. 70 */ 71 #define BPF_SYM_ELF_TYPE 't' 72 73 /* BPF program can access up to 512 bytes of stack space. */ 74 #define MAX_BPF_STACK 512 75 76 /* Helper macros for filter block array initializers. */ 77 78 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 79 80 #define BPF_ALU64_REG(OP, DST, SRC) \ 81 ((struct bpf_insn) { \ 82 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 83 .dst_reg = DST, \ 84 .src_reg = SRC, \ 85 .off = 0, \ 86 .imm = 0 }) 87 88 #define BPF_ALU32_REG(OP, DST, SRC) \ 89 ((struct bpf_insn) { \ 90 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 91 .dst_reg = DST, \ 92 .src_reg = SRC, \ 93 .off = 0, \ 94 .imm = 0 }) 95 96 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 97 98 #define BPF_ALU64_IMM(OP, DST, IMM) \ 99 ((struct bpf_insn) { \ 100 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 101 .dst_reg = DST, \ 102 .src_reg = 0, \ 103 .off = 0, \ 104 .imm = IMM }) 105 106 #define BPF_ALU32_IMM(OP, DST, IMM) \ 107 ((struct bpf_insn) { \ 108 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 109 .dst_reg = DST, \ 110 .src_reg = 0, \ 111 .off = 0, \ 112 .imm = IMM }) 113 114 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 115 116 #define BPF_ENDIAN(TYPE, DST, LEN) \ 117 ((struct bpf_insn) { \ 118 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 119 .dst_reg = DST, \ 120 .src_reg = 0, \ 121 .off = 0, \ 122 .imm = LEN }) 123 124 /* Short form of mov, dst_reg = src_reg */ 125 126 #define BPF_MOV64_REG(DST, SRC) \ 127 ((struct bpf_insn) { \ 128 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 129 .dst_reg = DST, \ 130 .src_reg = SRC, \ 131 .off = 0, \ 132 .imm = 0 }) 133 134 #define BPF_MOV32_REG(DST, SRC) \ 135 ((struct bpf_insn) { \ 136 .code = BPF_ALU | BPF_MOV | BPF_X, \ 137 .dst_reg = DST, \ 138 .src_reg = SRC, \ 139 .off = 0, \ 140 .imm = 0 }) 141 142 /* Short form of mov, dst_reg = imm32 */ 143 144 #define BPF_MOV64_IMM(DST, IMM) \ 145 ((struct bpf_insn) { \ 146 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 147 .dst_reg = DST, \ 148 .src_reg = 0, \ 149 .off = 0, \ 150 .imm = IMM }) 151 152 #define BPF_MOV32_IMM(DST, IMM) \ 153 ((struct bpf_insn) { \ 154 .code = BPF_ALU | BPF_MOV | BPF_K, \ 155 .dst_reg = DST, \ 156 .src_reg = 0, \ 157 .off = 0, \ 158 .imm = IMM }) 159 160 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 161 #define BPF_LD_IMM64(DST, IMM) \ 162 BPF_LD_IMM64_RAW(DST, 0, IMM) 163 164 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 165 ((struct bpf_insn) { \ 166 .code = BPF_LD | BPF_DW | BPF_IMM, \ 167 .dst_reg = DST, \ 168 .src_reg = SRC, \ 169 .off = 0, \ 170 .imm = (__u32) (IMM) }), \ 171 ((struct bpf_insn) { \ 172 .code = 0, /* zero is reserved opcode */ \ 173 .dst_reg = 0, \ 174 .src_reg = 0, \ 175 .off = 0, \ 176 .imm = ((__u64) (IMM)) >> 32 }) 177 178 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 179 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 180 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 181 182 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 183 184 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 185 ((struct bpf_insn) { \ 186 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 187 .dst_reg = DST, \ 188 .src_reg = SRC, \ 189 .off = 0, \ 190 .imm = IMM }) 191 192 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 193 ((struct bpf_insn) { \ 194 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 195 .dst_reg = DST, \ 196 .src_reg = SRC, \ 197 .off = 0, \ 198 .imm = IMM }) 199 200 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 201 202 #define BPF_LD_ABS(SIZE, IMM) \ 203 ((struct bpf_insn) { \ 204 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 205 .dst_reg = 0, \ 206 .src_reg = 0, \ 207 .off = 0, \ 208 .imm = IMM }) 209 210 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 211 212 #define BPF_LD_IND(SIZE, SRC, IMM) \ 213 ((struct bpf_insn) { \ 214 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 215 .dst_reg = 0, \ 216 .src_reg = SRC, \ 217 .off = 0, \ 218 .imm = IMM }) 219 220 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 221 222 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 223 ((struct bpf_insn) { \ 224 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 225 .dst_reg = DST, \ 226 .src_reg = SRC, \ 227 .off = OFF, \ 228 .imm = 0 }) 229 230 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 231 232 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 233 ((struct bpf_insn) { \ 234 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 235 .dst_reg = DST, \ 236 .src_reg = SRC, \ 237 .off = OFF, \ 238 .imm = 0 }) 239 240 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */ 241 242 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \ 243 ((struct bpf_insn) { \ 244 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \ 245 .dst_reg = DST, \ 246 .src_reg = SRC, \ 247 .off = OFF, \ 248 .imm = 0 }) 249 250 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 251 252 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 253 ((struct bpf_insn) { \ 254 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 255 .dst_reg = DST, \ 256 .src_reg = 0, \ 257 .off = OFF, \ 258 .imm = IMM }) 259 260 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 261 262 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 263 ((struct bpf_insn) { \ 264 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 265 .dst_reg = DST, \ 266 .src_reg = SRC, \ 267 .off = OFF, \ 268 .imm = 0 }) 269 270 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 271 272 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 273 ((struct bpf_insn) { \ 274 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 275 .dst_reg = DST, \ 276 .src_reg = 0, \ 277 .off = OFF, \ 278 .imm = IMM }) 279 280 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 281 282 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 283 ((struct bpf_insn) { \ 284 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 285 .dst_reg = DST, \ 286 .src_reg = SRC, \ 287 .off = OFF, \ 288 .imm = 0 }) 289 290 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 291 292 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 293 ((struct bpf_insn) { \ 294 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 295 .dst_reg = DST, \ 296 .src_reg = 0, \ 297 .off = OFF, \ 298 .imm = IMM }) 299 300 /* Unconditional jumps, goto pc + off16 */ 301 302 #define BPF_JMP_A(OFF) \ 303 ((struct bpf_insn) { \ 304 .code = BPF_JMP | BPF_JA, \ 305 .dst_reg = 0, \ 306 .src_reg = 0, \ 307 .off = OFF, \ 308 .imm = 0 }) 309 310 /* Relative call */ 311 312 #define BPF_CALL_REL(TGT) \ 313 ((struct bpf_insn) { \ 314 .code = BPF_JMP | BPF_CALL, \ 315 .dst_reg = 0, \ 316 .src_reg = BPF_PSEUDO_CALL, \ 317 .off = 0, \ 318 .imm = TGT }) 319 320 /* Function call */ 321 322 #define BPF_CAST_CALL(x) \ 323 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 324 325 #define BPF_EMIT_CALL(FUNC) \ 326 ((struct bpf_insn) { \ 327 .code = BPF_JMP | BPF_CALL, \ 328 .dst_reg = 0, \ 329 .src_reg = 0, \ 330 .off = 0, \ 331 .imm = ((FUNC) - __bpf_call_base) }) 332 333 /* Raw code statement block */ 334 335 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 336 ((struct bpf_insn) { \ 337 .code = CODE, \ 338 .dst_reg = DST, \ 339 .src_reg = SRC, \ 340 .off = OFF, \ 341 .imm = IMM }) 342 343 /* Program exit */ 344 345 #define BPF_EXIT_INSN() \ 346 ((struct bpf_insn) { \ 347 .code = BPF_JMP | BPF_EXIT, \ 348 .dst_reg = 0, \ 349 .src_reg = 0, \ 350 .off = 0, \ 351 .imm = 0 }) 352 353 /* Internal classic blocks for direct assignment */ 354 355 #define __BPF_STMT(CODE, K) \ 356 ((struct sock_filter) BPF_STMT(CODE, K)) 357 358 #define __BPF_JUMP(CODE, K, JT, JF) \ 359 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 360 361 #define bytes_to_bpf_size(bytes) \ 362 ({ \ 363 int bpf_size = -EINVAL; \ 364 \ 365 if (bytes == sizeof(u8)) \ 366 bpf_size = BPF_B; \ 367 else if (bytes == sizeof(u16)) \ 368 bpf_size = BPF_H; \ 369 else if (bytes == sizeof(u32)) \ 370 bpf_size = BPF_W; \ 371 else if (bytes == sizeof(u64)) \ 372 bpf_size = BPF_DW; \ 373 \ 374 bpf_size; \ 375 }) 376 377 #define bpf_size_to_bytes(bpf_size) \ 378 ({ \ 379 int bytes = -EINVAL; \ 380 \ 381 if (bpf_size == BPF_B) \ 382 bytes = sizeof(u8); \ 383 else if (bpf_size == BPF_H) \ 384 bytes = sizeof(u16); \ 385 else if (bpf_size == BPF_W) \ 386 bytes = sizeof(u32); \ 387 else if (bpf_size == BPF_DW) \ 388 bytes = sizeof(u64); \ 389 \ 390 bytes; \ 391 }) 392 393 #define BPF_SIZEOF(type) \ 394 ({ \ 395 const int __size = bytes_to_bpf_size(sizeof(type)); \ 396 BUILD_BUG_ON(__size < 0); \ 397 __size; \ 398 }) 399 400 #define BPF_FIELD_SIZEOF(type, field) \ 401 ({ \ 402 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ 403 BUILD_BUG_ON(__size < 0); \ 404 __size; \ 405 }) 406 407 #define BPF_LDST_BYTES(insn) \ 408 ({ \ 409 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 410 WARN_ON(__size < 0); \ 411 __size; \ 412 }) 413 414 #define __BPF_MAP_0(m, v, ...) v 415 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 416 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 417 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 418 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 419 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 420 421 #define __BPF_REG_0(...) __BPF_PAD(5) 422 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 423 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 424 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 425 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 426 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 427 428 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 429 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 430 431 #define __BPF_CAST(t, a) \ 432 (__force t) \ 433 (__force \ 434 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 435 (unsigned long)0, (t)0))) a 436 #define __BPF_V void 437 #define __BPF_N 438 439 #define __BPF_DECL_ARGS(t, a) t a 440 #define __BPF_DECL_REGS(t, a) u64 a 441 442 #define __BPF_PAD(n) \ 443 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 444 u64, __ur_3, u64, __ur_4, u64, __ur_5) 445 446 #define BPF_CALL_x(x, name, ...) \ 447 static __always_inline \ 448 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 449 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 450 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 451 { \ 452 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 453 } \ 454 static __always_inline \ 455 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 456 457 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 458 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 459 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 460 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 461 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 462 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 463 464 #define bpf_ctx_range(TYPE, MEMBER) \ 465 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 466 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 467 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 468 #if BITS_PER_LONG == 64 469 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 470 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 471 #else 472 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 473 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 474 #endif /* BITS_PER_LONG == 64 */ 475 476 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 477 ({ \ 478 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \ 479 *(PTR_SIZE) = (SIZE); \ 480 offsetof(TYPE, MEMBER); \ 481 }) 482 483 #ifdef CONFIG_COMPAT 484 /* A struct sock_filter is architecture independent. */ 485 struct compat_sock_fprog { 486 u16 len; 487 compat_uptr_t filter; /* struct sock_filter * */ 488 }; 489 #endif 490 491 struct sock_fprog_kern { 492 u16 len; 493 struct sock_filter *filter; 494 }; 495 496 struct bpf_binary_header { 497 u32 pages; 498 /* Some arches need word alignment for their instructions */ 499 u8 image[] __aligned(4); 500 }; 501 502 struct bpf_prog { 503 u16 pages; /* Number of allocated pages */ 504 u16 jited:1, /* Is our filter JIT'ed? */ 505 jit_requested:1,/* archs need to JIT the prog */ 506 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */ 507 gpl_compatible:1, /* Is filter GPL compatible? */ 508 cb_access:1, /* Is control block accessed? */ 509 dst_needed:1, /* Do we need dst entry? */ 510 blinded:1, /* Was blinded */ 511 is_func:1, /* program is a bpf function */ 512 kprobe_override:1, /* Do we override a kprobe? */ 513 has_callchain_buf:1; /* callchain buffer allocated? */ 514 enum bpf_prog_type type; /* Type of BPF program */ 515 enum bpf_attach_type expected_attach_type; /* For some prog types */ 516 u32 len; /* Number of filter blocks */ 517 u32 jited_len; /* Size of jited insns in bytes */ 518 u8 tag[BPF_TAG_SIZE]; 519 struct bpf_prog_aux *aux; /* Auxiliary fields */ 520 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 521 unsigned int (*bpf_func)(const void *ctx, 522 const struct bpf_insn *insn); 523 /* Instructions for interpreter */ 524 union { 525 struct sock_filter insns[0]; 526 struct bpf_insn insnsi[0]; 527 }; 528 }; 529 530 struct sk_filter { 531 refcount_t refcnt; 532 struct rcu_head rcu; 533 struct bpf_prog *prog; 534 }; 535 536 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 537 538 #define BPF_PROG_RUN(prog, ctx) ({ \ 539 u32 ret; \ 540 cant_sleep(); \ 541 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ 542 struct bpf_prog_stats *stats; \ 543 u64 start = sched_clock(); \ 544 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \ 545 stats = this_cpu_ptr(prog->aux->stats); \ 546 u64_stats_update_begin(&stats->syncp); \ 547 stats->cnt++; \ 548 stats->nsecs += sched_clock() - start; \ 549 u64_stats_update_end(&stats->syncp); \ 550 } else { \ 551 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \ 552 } \ 553 ret; }) 554 555 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 556 557 struct bpf_skb_data_end { 558 struct qdisc_skb_cb qdisc_cb; 559 void *data_meta; 560 void *data_end; 561 }; 562 563 struct bpf_redirect_info { 564 u32 ifindex; 565 u32 flags; 566 struct bpf_map *map; 567 struct bpf_map *map_to_flush; 568 u32 kern_flags; 569 }; 570 571 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 572 573 /* flags for bpf_redirect_info kern_flags */ 574 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 575 576 /* Compute the linear packet data range [data, data_end) which 577 * will be accessed by various program types (cls_bpf, act_bpf, 578 * lwt, ...). Subsystems allowing direct data access must (!) 579 * ensure that cb[] area can be written to when BPF program is 580 * invoked (otherwise cb[] save/restore is necessary). 581 */ 582 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 583 { 584 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 585 586 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb)); 587 cb->data_meta = skb->data - skb_metadata_len(skb); 588 cb->data_end = skb->data + skb_headlen(skb); 589 } 590 591 /* Similar to bpf_compute_data_pointers(), except that save orginal 592 * data in cb->data and cb->meta_data for restore. 593 */ 594 static inline void bpf_compute_and_save_data_end( 595 struct sk_buff *skb, void **saved_data_end) 596 { 597 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 598 599 *saved_data_end = cb->data_end; 600 cb->data_end = skb->data + skb_headlen(skb); 601 } 602 603 /* Restore data saved by bpf_compute_data_pointers(). */ 604 static inline void bpf_restore_data_end( 605 struct sk_buff *skb, void *saved_data_end) 606 { 607 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 608 609 cb->data_end = saved_data_end; 610 } 611 612 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 613 { 614 /* eBPF programs may read/write skb->cb[] area to transfer meta 615 * data between tail calls. Since this also needs to work with 616 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 617 * 618 * In some socket filter cases, the cb unfortunately needs to be 619 * saved/restored so that protocol specific skb->cb[] data won't 620 * be lost. In any case, due to unpriviledged eBPF programs 621 * attached to sockets, we need to clear the bpf_skb_cb() area 622 * to not leak previous contents to user space. 623 */ 624 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 625 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != 626 FIELD_SIZEOF(struct qdisc_skb_cb, data)); 627 628 return qdisc_skb_cb(skb)->data; 629 } 630 631 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 632 struct sk_buff *skb) 633 { 634 u8 *cb_data = bpf_skb_cb(skb); 635 u8 cb_saved[BPF_SKB_CB_LEN]; 636 u32 res; 637 638 if (unlikely(prog->cb_access)) { 639 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 640 memset(cb_data, 0, sizeof(cb_saved)); 641 } 642 643 res = BPF_PROG_RUN(prog, skb); 644 645 if (unlikely(prog->cb_access)) 646 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 647 648 return res; 649 } 650 651 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 652 struct sk_buff *skb) 653 { 654 u32 res; 655 656 preempt_disable(); 657 res = __bpf_prog_run_save_cb(prog, skb); 658 preempt_enable(); 659 return res; 660 } 661 662 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 663 struct sk_buff *skb) 664 { 665 u8 *cb_data = bpf_skb_cb(skb); 666 u32 res; 667 668 if (unlikely(prog->cb_access)) 669 memset(cb_data, 0, BPF_SKB_CB_LEN); 670 671 preempt_disable(); 672 res = BPF_PROG_RUN(prog, skb); 673 preempt_enable(); 674 return res; 675 } 676 677 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 678 struct xdp_buff *xdp) 679 { 680 /* Caller needs to hold rcu_read_lock() (!), otherwise program 681 * can be released while still running, or map elements could be 682 * freed early while still having concurrent users. XDP fastpath 683 * already takes rcu_read_lock() when fetching the program, so 684 * it's not necessary here anymore. 685 */ 686 return BPF_PROG_RUN(prog, xdp); 687 } 688 689 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 690 { 691 return prog->len * sizeof(struct bpf_insn); 692 } 693 694 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 695 { 696 return round_up(bpf_prog_insn_size(prog) + 697 sizeof(__be64) + 1, SHA_MESSAGE_BYTES); 698 } 699 700 static inline unsigned int bpf_prog_size(unsigned int proglen) 701 { 702 return max(sizeof(struct bpf_prog), 703 offsetof(struct bpf_prog, insns[proglen])); 704 } 705 706 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 707 { 708 /* When classic BPF programs have been loaded and the arch 709 * does not have a classic BPF JIT (anymore), they have been 710 * converted via bpf_migrate_filter() to eBPF and thus always 711 * have an unspec program type. 712 */ 713 return prog->type == BPF_PROG_TYPE_UNSPEC; 714 } 715 716 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 717 { 718 const u32 size_machine = sizeof(unsigned long); 719 720 if (size > size_machine && size % size_machine == 0) 721 size = size_machine; 722 723 return size; 724 } 725 726 static inline bool 727 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 728 { 729 return size <= size_default && (size & (size - 1)) == 0; 730 } 731 732 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 733 734 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 735 { 736 fp->undo_set_mem = 1; 737 set_memory_ro((unsigned long)fp, fp->pages); 738 } 739 740 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) 741 { 742 if (fp->undo_set_mem) 743 set_memory_rw((unsigned long)fp, fp->pages); 744 } 745 746 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 747 { 748 set_memory_ro((unsigned long)hdr, hdr->pages); 749 } 750 751 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) 752 { 753 set_memory_rw((unsigned long)hdr, hdr->pages); 754 } 755 756 static inline struct bpf_binary_header * 757 bpf_jit_binary_hdr(const struct bpf_prog *fp) 758 { 759 unsigned long real_start = (unsigned long)fp->bpf_func; 760 unsigned long addr = real_start & PAGE_MASK; 761 762 return (void *)addr; 763 } 764 765 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 766 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 767 { 768 return sk_filter_trim_cap(sk, skb, 1); 769 } 770 771 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 772 void bpf_prog_free(struct bpf_prog *fp); 773 774 bool bpf_opcode_in_insntable(u8 code); 775 776 void bpf_prog_free_linfo(struct bpf_prog *prog); 777 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 778 const u32 *insn_to_jit_off); 779 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 780 void bpf_prog_free_jited_linfo(struct bpf_prog *prog); 781 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog); 782 783 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 784 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 785 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 786 gfp_t gfp_extra_flags); 787 void __bpf_prog_free(struct bpf_prog *fp); 788 789 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 790 { 791 bpf_prog_unlock_ro(fp); 792 __bpf_prog_free(fp); 793 } 794 795 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 796 unsigned int flen); 797 798 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 799 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 800 bpf_aux_classic_check_t trans, bool save_orig); 801 void bpf_prog_destroy(struct bpf_prog *fp); 802 803 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 804 int sk_attach_bpf(u32 ufd, struct sock *sk); 805 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 806 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 807 void sk_reuseport_prog_free(struct bpf_prog *prog); 808 int sk_detach_filter(struct sock *sk); 809 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 810 unsigned int len); 811 812 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 813 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 814 815 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 816 #define __bpf_call_base_args \ 817 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 818 __bpf_call_base) 819 820 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 821 void bpf_jit_compile(struct bpf_prog *prog); 822 bool bpf_helper_changes_pkt_data(void *func); 823 824 static inline bool bpf_dump_raw_ok(void) 825 { 826 /* Reconstruction of call-sites is dependent on kallsyms, 827 * thus make dump the same restriction. 828 */ 829 return kallsyms_show_value() == 1; 830 } 831 832 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 833 const struct bpf_insn *patch, u32 len); 834 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 835 836 void bpf_clear_redirect_map(struct bpf_map *map); 837 838 static inline bool xdp_return_frame_no_direct(void) 839 { 840 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 841 842 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 843 } 844 845 static inline void xdp_set_return_frame_no_direct(void) 846 { 847 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 848 849 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 850 } 851 852 static inline void xdp_clear_return_frame_no_direct(void) 853 { 854 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 855 856 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 857 } 858 859 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 860 unsigned int pktlen) 861 { 862 unsigned int len; 863 864 if (unlikely(!(fwd->flags & IFF_UP))) 865 return -ENETDOWN; 866 867 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 868 if (pktlen > len) 869 return -EMSGSIZE; 870 871 return 0; 872 } 873 874 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the 875 * same cpu context. Further for best results no more than a single map 876 * for the do_redirect/do_flush pair should be used. This limitation is 877 * because we only track one map and force a flush when the map changes. 878 * This does not appear to be a real limitation for existing software. 879 */ 880 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 881 struct xdp_buff *xdp, struct bpf_prog *prog); 882 int xdp_do_redirect(struct net_device *dev, 883 struct xdp_buff *xdp, 884 struct bpf_prog *prog); 885 void xdp_do_flush_map(void); 886 887 void bpf_warn_invalid_xdp_action(u32 act); 888 889 #ifdef CONFIG_INET 890 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 891 struct bpf_prog *prog, struct sk_buff *skb, 892 u32 hash); 893 #else 894 static inline struct sock * 895 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 896 struct bpf_prog *prog, struct sk_buff *skb, 897 u32 hash) 898 { 899 return NULL; 900 } 901 #endif 902 903 #ifdef CONFIG_BPF_JIT 904 extern int bpf_jit_enable; 905 extern int bpf_jit_harden; 906 extern int bpf_jit_kallsyms; 907 extern long bpf_jit_limit; 908 909 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 910 911 struct bpf_binary_header * 912 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 913 unsigned int alignment, 914 bpf_jit_fill_hole_t bpf_fill_ill_insns); 915 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 916 u64 bpf_jit_alloc_exec_limit(void); 917 void *bpf_jit_alloc_exec(unsigned long size); 918 void bpf_jit_free_exec(void *addr); 919 void bpf_jit_free(struct bpf_prog *fp); 920 921 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 922 const struct bpf_insn *insn, bool extra_pass, 923 u64 *func_addr, bool *func_addr_fixed); 924 925 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 926 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 927 928 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 929 u32 pass, void *image) 930 { 931 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 932 proglen, pass, image, current->comm, task_pid_nr(current)); 933 934 if (image) 935 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 936 16, 1, image, proglen, false); 937 } 938 939 static inline bool bpf_jit_is_ebpf(void) 940 { 941 # ifdef CONFIG_HAVE_EBPF_JIT 942 return true; 943 # else 944 return false; 945 # endif 946 } 947 948 static inline bool ebpf_jit_enabled(void) 949 { 950 return bpf_jit_enable && bpf_jit_is_ebpf(); 951 } 952 953 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 954 { 955 return fp->jited && bpf_jit_is_ebpf(); 956 } 957 958 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 959 { 960 /* These are the prerequisites, should someone ever have the 961 * idea to call blinding outside of them, we make sure to 962 * bail out. 963 */ 964 if (!bpf_jit_is_ebpf()) 965 return false; 966 if (!prog->jit_requested) 967 return false; 968 if (!bpf_jit_harden) 969 return false; 970 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 971 return false; 972 973 return true; 974 } 975 976 static inline bool bpf_jit_kallsyms_enabled(void) 977 { 978 /* There are a couple of corner cases where kallsyms should 979 * not be enabled f.e. on hardening. 980 */ 981 if (bpf_jit_harden) 982 return false; 983 if (!bpf_jit_kallsyms) 984 return false; 985 if (bpf_jit_kallsyms == 1) 986 return true; 987 988 return false; 989 } 990 991 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 992 unsigned long *off, char *sym); 993 bool is_bpf_text_address(unsigned long addr); 994 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 995 char *sym); 996 997 static inline const char * 998 bpf_address_lookup(unsigned long addr, unsigned long *size, 999 unsigned long *off, char **modname, char *sym) 1000 { 1001 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1002 1003 if (ret && modname) 1004 *modname = NULL; 1005 return ret; 1006 } 1007 1008 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1009 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1010 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym); 1011 1012 #else /* CONFIG_BPF_JIT */ 1013 1014 static inline bool ebpf_jit_enabled(void) 1015 { 1016 return false; 1017 } 1018 1019 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1020 { 1021 return false; 1022 } 1023 1024 static inline void bpf_jit_free(struct bpf_prog *fp) 1025 { 1026 bpf_prog_unlock_free(fp); 1027 } 1028 1029 static inline bool bpf_jit_kallsyms_enabled(void) 1030 { 1031 return false; 1032 } 1033 1034 static inline const char * 1035 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1036 unsigned long *off, char *sym) 1037 { 1038 return NULL; 1039 } 1040 1041 static inline bool is_bpf_text_address(unsigned long addr) 1042 { 1043 return false; 1044 } 1045 1046 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1047 char *type, char *sym) 1048 { 1049 return -ERANGE; 1050 } 1051 1052 static inline const char * 1053 bpf_address_lookup(unsigned long addr, unsigned long *size, 1054 unsigned long *off, char **modname, char *sym) 1055 { 1056 return NULL; 1057 } 1058 1059 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1060 { 1061 } 1062 1063 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1064 { 1065 } 1066 1067 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) 1068 { 1069 sym[0] = '\0'; 1070 } 1071 1072 #endif /* CONFIG_BPF_JIT */ 1073 1074 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp); 1075 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1076 1077 #define BPF_ANC BIT(15) 1078 1079 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1080 { 1081 switch (first->code) { 1082 case BPF_RET | BPF_K: 1083 case BPF_LD | BPF_W | BPF_LEN: 1084 return false; 1085 1086 case BPF_LD | BPF_W | BPF_ABS: 1087 case BPF_LD | BPF_H | BPF_ABS: 1088 case BPF_LD | BPF_B | BPF_ABS: 1089 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1090 return true; 1091 return false; 1092 1093 default: 1094 return true; 1095 } 1096 } 1097 1098 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1099 { 1100 BUG_ON(ftest->code & BPF_ANC); 1101 1102 switch (ftest->code) { 1103 case BPF_LD | BPF_W | BPF_ABS: 1104 case BPF_LD | BPF_H | BPF_ABS: 1105 case BPF_LD | BPF_B | BPF_ABS: 1106 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1107 return BPF_ANC | SKF_AD_##CODE 1108 switch (ftest->k) { 1109 BPF_ANCILLARY(PROTOCOL); 1110 BPF_ANCILLARY(PKTTYPE); 1111 BPF_ANCILLARY(IFINDEX); 1112 BPF_ANCILLARY(NLATTR); 1113 BPF_ANCILLARY(NLATTR_NEST); 1114 BPF_ANCILLARY(MARK); 1115 BPF_ANCILLARY(QUEUE); 1116 BPF_ANCILLARY(HATYPE); 1117 BPF_ANCILLARY(RXHASH); 1118 BPF_ANCILLARY(CPU); 1119 BPF_ANCILLARY(ALU_XOR_X); 1120 BPF_ANCILLARY(VLAN_TAG); 1121 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1122 BPF_ANCILLARY(PAY_OFFSET); 1123 BPF_ANCILLARY(RANDOM); 1124 BPF_ANCILLARY(VLAN_TPID); 1125 } 1126 /* Fallthrough. */ 1127 default: 1128 return ftest->code; 1129 } 1130 } 1131 1132 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1133 int k, unsigned int size); 1134 1135 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, 1136 unsigned int size, void *buffer) 1137 { 1138 if (k >= 0) 1139 return skb_header_pointer(skb, k, size, buffer); 1140 1141 return bpf_internal_load_pointer_neg_helper(skb, k, size); 1142 } 1143 1144 static inline int bpf_tell_extensions(void) 1145 { 1146 return SKF_AD_MAX; 1147 } 1148 1149 struct bpf_sock_addr_kern { 1150 struct sock *sk; 1151 struct sockaddr *uaddr; 1152 /* Temporary "register" to make indirect stores to nested structures 1153 * defined above. We need three registers to make such a store, but 1154 * only two (src and dst) are available at convert_ctx_access time 1155 */ 1156 u64 tmp_reg; 1157 void *t_ctx; /* Attach type specific context. */ 1158 }; 1159 1160 struct bpf_sock_ops_kern { 1161 struct sock *sk; 1162 u32 op; 1163 union { 1164 u32 args[4]; 1165 u32 reply; 1166 u32 replylong[4]; 1167 }; 1168 u32 is_fullsock; 1169 u64 temp; /* temp and everything after is not 1170 * initialized to 0 before calling 1171 * the BPF program. New fields that 1172 * should be initialized to 0 should 1173 * be inserted before temp. 1174 * temp is scratch storage used by 1175 * sock_ops_convert_ctx_access 1176 * as temporary storage of a register. 1177 */ 1178 }; 1179 1180 #endif /* __LINUX_FILTER_H__ */ 1181