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 #include <linux/vmalloc.h> 24 25 #include <net/sch_generic.h> 26 27 #include <uapi/linux/filter.h> 28 #include <uapi/linux/bpf.h> 29 30 struct sk_buff; 31 struct sock; 32 struct seccomp_data; 33 struct bpf_prog_aux; 34 struct xdp_rxq_info; 35 struct xdp_buff; 36 struct sock_reuseport; 37 struct ctl_table; 38 struct ctl_table_header; 39 40 /* ArgX, context and stack frame pointer register positions. Note, 41 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 42 * calls in BPF_CALL instruction. 43 */ 44 #define BPF_REG_ARG1 BPF_REG_1 45 #define BPF_REG_ARG2 BPF_REG_2 46 #define BPF_REG_ARG3 BPF_REG_3 47 #define BPF_REG_ARG4 BPF_REG_4 48 #define BPF_REG_ARG5 BPF_REG_5 49 #define BPF_REG_CTX BPF_REG_6 50 #define BPF_REG_FP BPF_REG_10 51 52 /* Additional register mappings for converted user programs. */ 53 #define BPF_REG_A BPF_REG_0 54 #define BPF_REG_X BPF_REG_7 55 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 56 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 57 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 58 59 /* Kernel hidden auxiliary/helper register. */ 60 #define BPF_REG_AX MAX_BPF_REG 61 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 62 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 63 64 /* unused opcode to mark special call to bpf_tail_call() helper */ 65 #define BPF_TAIL_CALL 0xf0 66 67 /* unused opcode to mark call to interpreter with arguments */ 68 #define BPF_CALL_ARGS 0xe0 69 70 /* As per nm, we expose JITed images as text (code) section for 71 * kallsyms. That way, tools like perf can find it to match 72 * addresses. 73 */ 74 #define BPF_SYM_ELF_TYPE 't' 75 76 /* BPF program can access up to 512 bytes of stack space. */ 77 #define MAX_BPF_STACK 512 78 79 /* Helper macros for filter block array initializers. */ 80 81 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 82 83 #define BPF_ALU64_REG(OP, DST, SRC) \ 84 ((struct bpf_insn) { \ 85 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 86 .dst_reg = DST, \ 87 .src_reg = SRC, \ 88 .off = 0, \ 89 .imm = 0 }) 90 91 #define BPF_ALU32_REG(OP, DST, SRC) \ 92 ((struct bpf_insn) { \ 93 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 94 .dst_reg = DST, \ 95 .src_reg = SRC, \ 96 .off = 0, \ 97 .imm = 0 }) 98 99 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 100 101 #define BPF_ALU64_IMM(OP, DST, IMM) \ 102 ((struct bpf_insn) { \ 103 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 104 .dst_reg = DST, \ 105 .src_reg = 0, \ 106 .off = 0, \ 107 .imm = IMM }) 108 109 #define BPF_ALU32_IMM(OP, DST, IMM) \ 110 ((struct bpf_insn) { \ 111 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 112 .dst_reg = DST, \ 113 .src_reg = 0, \ 114 .off = 0, \ 115 .imm = IMM }) 116 117 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 118 119 #define BPF_ENDIAN(TYPE, DST, LEN) \ 120 ((struct bpf_insn) { \ 121 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 122 .dst_reg = DST, \ 123 .src_reg = 0, \ 124 .off = 0, \ 125 .imm = LEN }) 126 127 /* Short form of mov, dst_reg = src_reg */ 128 129 #define BPF_MOV64_REG(DST, SRC) \ 130 ((struct bpf_insn) { \ 131 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 132 .dst_reg = DST, \ 133 .src_reg = SRC, \ 134 .off = 0, \ 135 .imm = 0 }) 136 137 #define BPF_MOV32_REG(DST, SRC) \ 138 ((struct bpf_insn) { \ 139 .code = BPF_ALU | BPF_MOV | BPF_X, \ 140 .dst_reg = DST, \ 141 .src_reg = SRC, \ 142 .off = 0, \ 143 .imm = 0 }) 144 145 /* Short form of mov, dst_reg = imm32 */ 146 147 #define BPF_MOV64_IMM(DST, IMM) \ 148 ((struct bpf_insn) { \ 149 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 150 .dst_reg = DST, \ 151 .src_reg = 0, \ 152 .off = 0, \ 153 .imm = IMM }) 154 155 #define BPF_MOV32_IMM(DST, IMM) \ 156 ((struct bpf_insn) { \ 157 .code = BPF_ALU | BPF_MOV | BPF_K, \ 158 .dst_reg = DST, \ 159 .src_reg = 0, \ 160 .off = 0, \ 161 .imm = IMM }) 162 163 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 164 #define BPF_LD_IMM64(DST, IMM) \ 165 BPF_LD_IMM64_RAW(DST, 0, IMM) 166 167 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 168 ((struct bpf_insn) { \ 169 .code = BPF_LD | BPF_DW | BPF_IMM, \ 170 .dst_reg = DST, \ 171 .src_reg = SRC, \ 172 .off = 0, \ 173 .imm = (__u32) (IMM) }), \ 174 ((struct bpf_insn) { \ 175 .code = 0, /* zero is reserved opcode */ \ 176 .dst_reg = 0, \ 177 .src_reg = 0, \ 178 .off = 0, \ 179 .imm = ((__u64) (IMM)) >> 32 }) 180 181 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 182 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 183 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 184 185 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 186 187 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 188 ((struct bpf_insn) { \ 189 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 190 .dst_reg = DST, \ 191 .src_reg = SRC, \ 192 .off = 0, \ 193 .imm = IMM }) 194 195 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 196 ((struct bpf_insn) { \ 197 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 198 .dst_reg = DST, \ 199 .src_reg = SRC, \ 200 .off = 0, \ 201 .imm = IMM }) 202 203 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 204 205 #define BPF_LD_ABS(SIZE, IMM) \ 206 ((struct bpf_insn) { \ 207 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 208 .dst_reg = 0, \ 209 .src_reg = 0, \ 210 .off = 0, \ 211 .imm = IMM }) 212 213 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 214 215 #define BPF_LD_IND(SIZE, SRC, IMM) \ 216 ((struct bpf_insn) { \ 217 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 218 .dst_reg = 0, \ 219 .src_reg = SRC, \ 220 .off = 0, \ 221 .imm = IMM }) 222 223 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 224 225 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 226 ((struct bpf_insn) { \ 227 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 228 .dst_reg = DST, \ 229 .src_reg = SRC, \ 230 .off = OFF, \ 231 .imm = 0 }) 232 233 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 234 235 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 236 ((struct bpf_insn) { \ 237 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 238 .dst_reg = DST, \ 239 .src_reg = SRC, \ 240 .off = OFF, \ 241 .imm = 0 }) 242 243 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */ 244 245 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \ 246 ((struct bpf_insn) { \ 247 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \ 248 .dst_reg = DST, \ 249 .src_reg = SRC, \ 250 .off = OFF, \ 251 .imm = 0 }) 252 253 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 254 255 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 256 ((struct bpf_insn) { \ 257 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 258 .dst_reg = DST, \ 259 .src_reg = 0, \ 260 .off = OFF, \ 261 .imm = IMM }) 262 263 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 264 265 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 266 ((struct bpf_insn) { \ 267 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 268 .dst_reg = DST, \ 269 .src_reg = SRC, \ 270 .off = OFF, \ 271 .imm = 0 }) 272 273 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 274 275 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 276 ((struct bpf_insn) { \ 277 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 278 .dst_reg = DST, \ 279 .src_reg = 0, \ 280 .off = OFF, \ 281 .imm = IMM }) 282 283 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 284 285 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 286 ((struct bpf_insn) { \ 287 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 288 .dst_reg = DST, \ 289 .src_reg = SRC, \ 290 .off = OFF, \ 291 .imm = 0 }) 292 293 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 294 295 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 296 ((struct bpf_insn) { \ 297 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 298 .dst_reg = DST, \ 299 .src_reg = 0, \ 300 .off = OFF, \ 301 .imm = IMM }) 302 303 /* Unconditional jumps, goto pc + off16 */ 304 305 #define BPF_JMP_A(OFF) \ 306 ((struct bpf_insn) { \ 307 .code = BPF_JMP | BPF_JA, \ 308 .dst_reg = 0, \ 309 .src_reg = 0, \ 310 .off = OFF, \ 311 .imm = 0 }) 312 313 /* Relative call */ 314 315 #define BPF_CALL_REL(TGT) \ 316 ((struct bpf_insn) { \ 317 .code = BPF_JMP | BPF_CALL, \ 318 .dst_reg = 0, \ 319 .src_reg = BPF_PSEUDO_CALL, \ 320 .off = 0, \ 321 .imm = TGT }) 322 323 /* Function call */ 324 325 #define BPF_CAST_CALL(x) \ 326 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 327 328 #define BPF_EMIT_CALL(FUNC) \ 329 ((struct bpf_insn) { \ 330 .code = BPF_JMP | BPF_CALL, \ 331 .dst_reg = 0, \ 332 .src_reg = 0, \ 333 .off = 0, \ 334 .imm = ((FUNC) - __bpf_call_base) }) 335 336 /* Raw code statement block */ 337 338 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 339 ((struct bpf_insn) { \ 340 .code = CODE, \ 341 .dst_reg = DST, \ 342 .src_reg = SRC, \ 343 .off = OFF, \ 344 .imm = IMM }) 345 346 /* Program exit */ 347 348 #define BPF_EXIT_INSN() \ 349 ((struct bpf_insn) { \ 350 .code = BPF_JMP | BPF_EXIT, \ 351 .dst_reg = 0, \ 352 .src_reg = 0, \ 353 .off = 0, \ 354 .imm = 0 }) 355 356 /* Internal classic blocks for direct assignment */ 357 358 #define __BPF_STMT(CODE, K) \ 359 ((struct sock_filter) BPF_STMT(CODE, K)) 360 361 #define __BPF_JUMP(CODE, K, JT, JF) \ 362 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 363 364 #define bytes_to_bpf_size(bytes) \ 365 ({ \ 366 int bpf_size = -EINVAL; \ 367 \ 368 if (bytes == sizeof(u8)) \ 369 bpf_size = BPF_B; \ 370 else if (bytes == sizeof(u16)) \ 371 bpf_size = BPF_H; \ 372 else if (bytes == sizeof(u32)) \ 373 bpf_size = BPF_W; \ 374 else if (bytes == sizeof(u64)) \ 375 bpf_size = BPF_DW; \ 376 \ 377 bpf_size; \ 378 }) 379 380 #define bpf_size_to_bytes(bpf_size) \ 381 ({ \ 382 int bytes = -EINVAL; \ 383 \ 384 if (bpf_size == BPF_B) \ 385 bytes = sizeof(u8); \ 386 else if (bpf_size == BPF_H) \ 387 bytes = sizeof(u16); \ 388 else if (bpf_size == BPF_W) \ 389 bytes = sizeof(u32); \ 390 else if (bpf_size == BPF_DW) \ 391 bytes = sizeof(u64); \ 392 \ 393 bytes; \ 394 }) 395 396 #define BPF_SIZEOF(type) \ 397 ({ \ 398 const int __size = bytes_to_bpf_size(sizeof(type)); \ 399 BUILD_BUG_ON(__size < 0); \ 400 __size; \ 401 }) 402 403 #define BPF_FIELD_SIZEOF(type, field) \ 404 ({ \ 405 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ 406 BUILD_BUG_ON(__size < 0); \ 407 __size; \ 408 }) 409 410 #define BPF_LDST_BYTES(insn) \ 411 ({ \ 412 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 413 WARN_ON(__size < 0); \ 414 __size; \ 415 }) 416 417 #define __BPF_MAP_0(m, v, ...) v 418 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 419 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 420 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 421 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 422 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 423 424 #define __BPF_REG_0(...) __BPF_PAD(5) 425 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 426 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 427 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 428 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 429 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 430 431 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 432 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 433 434 #define __BPF_CAST(t, a) \ 435 (__force t) \ 436 (__force \ 437 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 438 (unsigned long)0, (t)0))) a 439 #define __BPF_V void 440 #define __BPF_N 441 442 #define __BPF_DECL_ARGS(t, a) t a 443 #define __BPF_DECL_REGS(t, a) u64 a 444 445 #define __BPF_PAD(n) \ 446 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 447 u64, __ur_3, u64, __ur_4, u64, __ur_5) 448 449 #define BPF_CALL_x(x, name, ...) \ 450 static __always_inline \ 451 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 452 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 453 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 454 { \ 455 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 456 } \ 457 static __always_inline \ 458 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 459 460 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 461 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 462 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 463 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 464 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 465 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 466 467 #define bpf_ctx_range(TYPE, MEMBER) \ 468 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 469 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 470 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 471 #if BITS_PER_LONG == 64 472 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 473 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 474 #else 475 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 476 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 477 #endif /* BITS_PER_LONG == 64 */ 478 479 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 480 ({ \ 481 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \ 482 *(PTR_SIZE) = (SIZE); \ 483 offsetof(TYPE, MEMBER); \ 484 }) 485 486 #ifdef CONFIG_COMPAT 487 /* A struct sock_filter is architecture independent. */ 488 struct compat_sock_fprog { 489 u16 len; 490 compat_uptr_t filter; /* struct sock_filter * */ 491 }; 492 #endif 493 494 struct sock_fprog_kern { 495 u16 len; 496 struct sock_filter *filter; 497 }; 498 499 struct bpf_binary_header { 500 u32 pages; 501 /* Some arches need word alignment for their instructions */ 502 u8 image[] __aligned(4); 503 }; 504 505 struct bpf_prog { 506 u16 pages; /* Number of allocated pages */ 507 u16 jited:1, /* Is our filter JIT'ed? */ 508 jit_requested:1,/* archs need to JIT the prog */ 509 gpl_compatible:1, /* Is filter GPL compatible? */ 510 cb_access:1, /* Is control block accessed? */ 511 dst_needed:1, /* Do we need dst entry? */ 512 blinded:1, /* Was blinded */ 513 is_func:1, /* program is a bpf function */ 514 kprobe_override:1, /* Do we override a kprobe? */ 515 has_callchain_buf:1; /* callchain buffer allocated? */ 516 enum bpf_prog_type type; /* Type of BPF program */ 517 enum bpf_attach_type expected_attach_type; /* For some prog types */ 518 u32 len; /* Number of filter blocks */ 519 u32 jited_len; /* Size of jited insns in bytes */ 520 u8 tag[BPF_TAG_SIZE]; 521 struct bpf_prog_aux *aux; /* Auxiliary fields */ 522 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 523 unsigned int (*bpf_func)(const void *ctx, 524 const struct bpf_insn *insn); 525 /* Instructions for interpreter */ 526 union { 527 struct sock_filter insns[0]; 528 struct bpf_insn insnsi[0]; 529 }; 530 }; 531 532 struct sk_filter { 533 refcount_t refcnt; 534 struct rcu_head rcu; 535 struct bpf_prog *prog; 536 }; 537 538 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 539 540 #define BPF_PROG_RUN(prog, ctx) ({ \ 541 u32 ret; \ 542 cant_sleep(); \ 543 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \ 544 struct bpf_prog_stats *stats; \ 545 u64 start = sched_clock(); \ 546 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \ 547 stats = this_cpu_ptr(prog->aux->stats); \ 548 u64_stats_update_begin(&stats->syncp); \ 549 stats->cnt++; \ 550 stats->nsecs += sched_clock() - start; \ 551 u64_stats_update_end(&stats->syncp); \ 552 } else { \ 553 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \ 554 } \ 555 ret; }) 556 557 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 558 559 struct bpf_skb_data_end { 560 struct qdisc_skb_cb qdisc_cb; 561 void *data_meta; 562 void *data_end; 563 }; 564 565 struct bpf_redirect_info { 566 u32 ifindex; 567 u32 flags; 568 struct bpf_map *map; 569 struct bpf_map *map_to_flush; 570 u32 kern_flags; 571 }; 572 573 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 574 575 /* flags for bpf_redirect_info kern_flags */ 576 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 577 578 /* Compute the linear packet data range [data, data_end) which 579 * will be accessed by various program types (cls_bpf, act_bpf, 580 * lwt, ...). Subsystems allowing direct data access must (!) 581 * ensure that cb[] area can be written to when BPF program is 582 * invoked (otherwise cb[] save/restore is necessary). 583 */ 584 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 585 { 586 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 587 588 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb)); 589 cb->data_meta = skb->data - skb_metadata_len(skb); 590 cb->data_end = skb->data + skb_headlen(skb); 591 } 592 593 /* Similar to bpf_compute_data_pointers(), except that save orginal 594 * data in cb->data and cb->meta_data for restore. 595 */ 596 static inline void bpf_compute_and_save_data_end( 597 struct sk_buff *skb, void **saved_data_end) 598 { 599 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 600 601 *saved_data_end = cb->data_end; 602 cb->data_end = skb->data + skb_headlen(skb); 603 } 604 605 /* Restore data saved by bpf_compute_data_pointers(). */ 606 static inline void bpf_restore_data_end( 607 struct sk_buff *skb, void *saved_data_end) 608 { 609 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 610 611 cb->data_end = saved_data_end; 612 } 613 614 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 615 { 616 /* eBPF programs may read/write skb->cb[] area to transfer meta 617 * data between tail calls. Since this also needs to work with 618 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 619 * 620 * In some socket filter cases, the cb unfortunately needs to be 621 * saved/restored so that protocol specific skb->cb[] data won't 622 * be lost. In any case, due to unpriviledged eBPF programs 623 * attached to sockets, we need to clear the bpf_skb_cb() area 624 * to not leak previous contents to user space. 625 */ 626 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 627 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != 628 FIELD_SIZEOF(struct qdisc_skb_cb, data)); 629 630 return qdisc_skb_cb(skb)->data; 631 } 632 633 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 634 struct sk_buff *skb) 635 { 636 u8 *cb_data = bpf_skb_cb(skb); 637 u8 cb_saved[BPF_SKB_CB_LEN]; 638 u32 res; 639 640 if (unlikely(prog->cb_access)) { 641 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 642 memset(cb_data, 0, sizeof(cb_saved)); 643 } 644 645 res = BPF_PROG_RUN(prog, skb); 646 647 if (unlikely(prog->cb_access)) 648 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 649 650 return res; 651 } 652 653 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 654 struct sk_buff *skb) 655 { 656 u32 res; 657 658 preempt_disable(); 659 res = __bpf_prog_run_save_cb(prog, skb); 660 preempt_enable(); 661 return res; 662 } 663 664 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 665 struct sk_buff *skb) 666 { 667 u8 *cb_data = bpf_skb_cb(skb); 668 u32 res; 669 670 if (unlikely(prog->cb_access)) 671 memset(cb_data, 0, BPF_SKB_CB_LEN); 672 673 preempt_disable(); 674 res = BPF_PROG_RUN(prog, skb); 675 preempt_enable(); 676 return res; 677 } 678 679 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 680 struct xdp_buff *xdp) 681 { 682 /* Caller needs to hold rcu_read_lock() (!), otherwise program 683 * can be released while still running, or map elements could be 684 * freed early while still having concurrent users. XDP fastpath 685 * already takes rcu_read_lock() when fetching the program, so 686 * it's not necessary here anymore. 687 */ 688 return BPF_PROG_RUN(prog, xdp); 689 } 690 691 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 692 { 693 return prog->len * sizeof(struct bpf_insn); 694 } 695 696 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 697 { 698 return round_up(bpf_prog_insn_size(prog) + 699 sizeof(__be64) + 1, SHA_MESSAGE_BYTES); 700 } 701 702 static inline unsigned int bpf_prog_size(unsigned int proglen) 703 { 704 return max(sizeof(struct bpf_prog), 705 offsetof(struct bpf_prog, insns[proglen])); 706 } 707 708 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 709 { 710 /* When classic BPF programs have been loaded and the arch 711 * does not have a classic BPF JIT (anymore), they have been 712 * converted via bpf_migrate_filter() to eBPF and thus always 713 * have an unspec program type. 714 */ 715 return prog->type == BPF_PROG_TYPE_UNSPEC; 716 } 717 718 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 719 { 720 const u32 size_machine = sizeof(unsigned long); 721 722 if (size > size_machine && size % size_machine == 0) 723 size = size_machine; 724 725 return size; 726 } 727 728 static inline bool 729 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 730 { 731 return size <= size_default && (size & (size - 1)) == 0; 732 } 733 734 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 735 736 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 737 { 738 set_vm_flush_reset_perms(fp); 739 set_memory_ro((unsigned long)fp, fp->pages); 740 } 741 742 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 743 { 744 set_vm_flush_reset_perms(hdr); 745 set_memory_ro((unsigned long)hdr, hdr->pages); 746 set_memory_x((unsigned long)hdr, hdr->pages); 747 } 748 749 static inline struct bpf_binary_header * 750 bpf_jit_binary_hdr(const struct bpf_prog *fp) 751 { 752 unsigned long real_start = (unsigned long)fp->bpf_func; 753 unsigned long addr = real_start & PAGE_MASK; 754 755 return (void *)addr; 756 } 757 758 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 759 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 760 { 761 return sk_filter_trim_cap(sk, skb, 1); 762 } 763 764 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 765 void bpf_prog_free(struct bpf_prog *fp); 766 767 bool bpf_opcode_in_insntable(u8 code); 768 769 void bpf_prog_free_linfo(struct bpf_prog *prog); 770 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 771 const u32 *insn_to_jit_off); 772 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 773 void bpf_prog_free_jited_linfo(struct bpf_prog *prog); 774 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog); 775 776 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 777 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 778 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 779 gfp_t gfp_extra_flags); 780 void __bpf_prog_free(struct bpf_prog *fp); 781 782 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 783 { 784 __bpf_prog_free(fp); 785 } 786 787 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 788 unsigned int flen); 789 790 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 791 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 792 bpf_aux_classic_check_t trans, bool save_orig); 793 void bpf_prog_destroy(struct bpf_prog *fp); 794 795 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 796 int sk_attach_bpf(u32 ufd, struct sock *sk); 797 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 798 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 799 void sk_reuseport_prog_free(struct bpf_prog *prog); 800 int sk_detach_filter(struct sock *sk); 801 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 802 unsigned int len); 803 804 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 805 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 806 807 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 808 #define __bpf_call_base_args \ 809 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 810 __bpf_call_base) 811 812 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 813 void bpf_jit_compile(struct bpf_prog *prog); 814 bool bpf_helper_changes_pkt_data(void *func); 815 816 static inline bool bpf_dump_raw_ok(void) 817 { 818 /* Reconstruction of call-sites is dependent on kallsyms, 819 * thus make dump the same restriction. 820 */ 821 return kallsyms_show_value() == 1; 822 } 823 824 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 825 const struct bpf_insn *patch, u32 len); 826 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 827 828 void bpf_clear_redirect_map(struct bpf_map *map); 829 830 static inline bool xdp_return_frame_no_direct(void) 831 { 832 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 833 834 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 835 } 836 837 static inline void xdp_set_return_frame_no_direct(void) 838 { 839 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 840 841 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 842 } 843 844 static inline void xdp_clear_return_frame_no_direct(void) 845 { 846 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 847 848 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 849 } 850 851 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 852 unsigned int pktlen) 853 { 854 unsigned int len; 855 856 if (unlikely(!(fwd->flags & IFF_UP))) 857 return -ENETDOWN; 858 859 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 860 if (pktlen > len) 861 return -EMSGSIZE; 862 863 return 0; 864 } 865 866 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the 867 * same cpu context. Further for best results no more than a single map 868 * for the do_redirect/do_flush pair should be used. This limitation is 869 * because we only track one map and force a flush when the map changes. 870 * This does not appear to be a real limitation for existing software. 871 */ 872 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 873 struct xdp_buff *xdp, struct bpf_prog *prog); 874 int xdp_do_redirect(struct net_device *dev, 875 struct xdp_buff *xdp, 876 struct bpf_prog *prog); 877 void xdp_do_flush_map(void); 878 879 void bpf_warn_invalid_xdp_action(u32 act); 880 881 #ifdef CONFIG_INET 882 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 883 struct bpf_prog *prog, struct sk_buff *skb, 884 u32 hash); 885 #else 886 static inline struct sock * 887 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 888 struct bpf_prog *prog, struct sk_buff *skb, 889 u32 hash) 890 { 891 return NULL; 892 } 893 #endif 894 895 #ifdef CONFIG_BPF_JIT 896 extern int bpf_jit_enable; 897 extern int bpf_jit_harden; 898 extern int bpf_jit_kallsyms; 899 extern long bpf_jit_limit; 900 901 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 902 903 struct bpf_binary_header * 904 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 905 unsigned int alignment, 906 bpf_jit_fill_hole_t bpf_fill_ill_insns); 907 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 908 u64 bpf_jit_alloc_exec_limit(void); 909 void *bpf_jit_alloc_exec(unsigned long size); 910 void bpf_jit_free_exec(void *addr); 911 void bpf_jit_free(struct bpf_prog *fp); 912 913 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 914 const struct bpf_insn *insn, bool extra_pass, 915 u64 *func_addr, bool *func_addr_fixed); 916 917 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 918 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 919 920 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 921 u32 pass, void *image) 922 { 923 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 924 proglen, pass, image, current->comm, task_pid_nr(current)); 925 926 if (image) 927 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 928 16, 1, image, proglen, false); 929 } 930 931 static inline bool bpf_jit_is_ebpf(void) 932 { 933 # ifdef CONFIG_HAVE_EBPF_JIT 934 return true; 935 # else 936 return false; 937 # endif 938 } 939 940 static inline bool ebpf_jit_enabled(void) 941 { 942 return bpf_jit_enable && bpf_jit_is_ebpf(); 943 } 944 945 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 946 { 947 return fp->jited && bpf_jit_is_ebpf(); 948 } 949 950 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 951 { 952 /* These are the prerequisites, should someone ever have the 953 * idea to call blinding outside of them, we make sure to 954 * bail out. 955 */ 956 if (!bpf_jit_is_ebpf()) 957 return false; 958 if (!prog->jit_requested) 959 return false; 960 if (!bpf_jit_harden) 961 return false; 962 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 963 return false; 964 965 return true; 966 } 967 968 static inline bool bpf_jit_kallsyms_enabled(void) 969 { 970 /* There are a couple of corner cases where kallsyms should 971 * not be enabled f.e. on hardening. 972 */ 973 if (bpf_jit_harden) 974 return false; 975 if (!bpf_jit_kallsyms) 976 return false; 977 if (bpf_jit_kallsyms == 1) 978 return true; 979 980 return false; 981 } 982 983 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 984 unsigned long *off, char *sym); 985 bool is_bpf_text_address(unsigned long addr); 986 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 987 char *sym); 988 989 static inline const char * 990 bpf_address_lookup(unsigned long addr, unsigned long *size, 991 unsigned long *off, char **modname, char *sym) 992 { 993 const char *ret = __bpf_address_lookup(addr, size, off, sym); 994 995 if (ret && modname) 996 *modname = NULL; 997 return ret; 998 } 999 1000 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1001 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1002 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym); 1003 1004 #else /* CONFIG_BPF_JIT */ 1005 1006 static inline bool ebpf_jit_enabled(void) 1007 { 1008 return false; 1009 } 1010 1011 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1012 { 1013 return false; 1014 } 1015 1016 static inline void bpf_jit_free(struct bpf_prog *fp) 1017 { 1018 bpf_prog_unlock_free(fp); 1019 } 1020 1021 static inline bool bpf_jit_kallsyms_enabled(void) 1022 { 1023 return false; 1024 } 1025 1026 static inline const char * 1027 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1028 unsigned long *off, char *sym) 1029 { 1030 return NULL; 1031 } 1032 1033 static inline bool is_bpf_text_address(unsigned long addr) 1034 { 1035 return false; 1036 } 1037 1038 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1039 char *type, char *sym) 1040 { 1041 return -ERANGE; 1042 } 1043 1044 static inline const char * 1045 bpf_address_lookup(unsigned long addr, unsigned long *size, 1046 unsigned long *off, char **modname, char *sym) 1047 { 1048 return NULL; 1049 } 1050 1051 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1052 { 1053 } 1054 1055 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1056 { 1057 } 1058 1059 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) 1060 { 1061 sym[0] = '\0'; 1062 } 1063 1064 #endif /* CONFIG_BPF_JIT */ 1065 1066 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp); 1067 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1068 1069 #define BPF_ANC BIT(15) 1070 1071 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1072 { 1073 switch (first->code) { 1074 case BPF_RET | BPF_K: 1075 case BPF_LD | BPF_W | BPF_LEN: 1076 return false; 1077 1078 case BPF_LD | BPF_W | BPF_ABS: 1079 case BPF_LD | BPF_H | BPF_ABS: 1080 case BPF_LD | BPF_B | BPF_ABS: 1081 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1082 return true; 1083 return false; 1084 1085 default: 1086 return true; 1087 } 1088 } 1089 1090 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1091 { 1092 BUG_ON(ftest->code & BPF_ANC); 1093 1094 switch (ftest->code) { 1095 case BPF_LD | BPF_W | BPF_ABS: 1096 case BPF_LD | BPF_H | BPF_ABS: 1097 case BPF_LD | BPF_B | BPF_ABS: 1098 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1099 return BPF_ANC | SKF_AD_##CODE 1100 switch (ftest->k) { 1101 BPF_ANCILLARY(PROTOCOL); 1102 BPF_ANCILLARY(PKTTYPE); 1103 BPF_ANCILLARY(IFINDEX); 1104 BPF_ANCILLARY(NLATTR); 1105 BPF_ANCILLARY(NLATTR_NEST); 1106 BPF_ANCILLARY(MARK); 1107 BPF_ANCILLARY(QUEUE); 1108 BPF_ANCILLARY(HATYPE); 1109 BPF_ANCILLARY(RXHASH); 1110 BPF_ANCILLARY(CPU); 1111 BPF_ANCILLARY(ALU_XOR_X); 1112 BPF_ANCILLARY(VLAN_TAG); 1113 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1114 BPF_ANCILLARY(PAY_OFFSET); 1115 BPF_ANCILLARY(RANDOM); 1116 BPF_ANCILLARY(VLAN_TPID); 1117 } 1118 /* Fallthrough. */ 1119 default: 1120 return ftest->code; 1121 } 1122 } 1123 1124 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1125 int k, unsigned int size); 1126 1127 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, 1128 unsigned int size, void *buffer) 1129 { 1130 if (k >= 0) 1131 return skb_header_pointer(skb, k, size, buffer); 1132 1133 return bpf_internal_load_pointer_neg_helper(skb, k, size); 1134 } 1135 1136 static inline int bpf_tell_extensions(void) 1137 { 1138 return SKF_AD_MAX; 1139 } 1140 1141 struct bpf_sock_addr_kern { 1142 struct sock *sk; 1143 struct sockaddr *uaddr; 1144 /* Temporary "register" to make indirect stores to nested structures 1145 * defined above. We need three registers to make such a store, but 1146 * only two (src and dst) are available at convert_ctx_access time 1147 */ 1148 u64 tmp_reg; 1149 void *t_ctx; /* Attach type specific context. */ 1150 }; 1151 1152 struct bpf_sock_ops_kern { 1153 struct sock *sk; 1154 u32 op; 1155 union { 1156 u32 args[4]; 1157 u32 reply; 1158 u32 replylong[4]; 1159 }; 1160 u32 is_fullsock; 1161 u64 temp; /* temp and everything after is not 1162 * initialized to 0 before calling 1163 * the BPF program. New fields that 1164 * should be initialized to 0 should 1165 * be inserted before temp. 1166 * temp is scratch storage used by 1167 * sock_ops_convert_ctx_access 1168 * as temporary storage of a register. 1169 */ 1170 }; 1171 1172 struct bpf_sysctl_kern { 1173 struct ctl_table_header *head; 1174 struct ctl_table *table; 1175 void *cur_val; 1176 size_t cur_len; 1177 void *new_val; 1178 size_t new_len; 1179 int new_updated; 1180 int write; 1181 loff_t *ppos; 1182 /* Temporary "register" for indirect stores to ppos. */ 1183 u64 tmp_reg; 1184 }; 1185 1186 #endif /* __LINUX_FILTER_H__ */ 1187