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 38 /* ArgX, context and stack frame pointer register positions. Note, 39 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 40 * calls in BPF_CALL instruction. 41 */ 42 #define BPF_REG_ARG1 BPF_REG_1 43 #define BPF_REG_ARG2 BPF_REG_2 44 #define BPF_REG_ARG3 BPF_REG_3 45 #define BPF_REG_ARG4 BPF_REG_4 46 #define BPF_REG_ARG5 BPF_REG_5 47 #define BPF_REG_CTX BPF_REG_6 48 #define BPF_REG_FP BPF_REG_10 49 50 /* Additional register mappings for converted user programs. */ 51 #define BPF_REG_A BPF_REG_0 52 #define BPF_REG_X BPF_REG_7 53 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 54 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 55 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 56 57 /* Kernel hidden auxiliary/helper register. */ 58 #define BPF_REG_AX MAX_BPF_REG 59 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 60 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 61 62 /* unused opcode to mark special call to bpf_tail_call() helper */ 63 #define BPF_TAIL_CALL 0xf0 64 65 /* unused opcode to mark call to interpreter with arguments */ 66 #define BPF_CALL_ARGS 0xe0 67 68 /* As per nm, we expose JITed images as text (code) section for 69 * kallsyms. That way, tools like perf can find it to match 70 * addresses. 71 */ 72 #define BPF_SYM_ELF_TYPE 't' 73 74 /* BPF program can access up to 512 bytes of stack space. */ 75 #define MAX_BPF_STACK 512 76 77 /* Helper macros for filter block array initializers. */ 78 79 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 80 81 #define BPF_ALU64_REG(OP, DST, SRC) \ 82 ((struct bpf_insn) { \ 83 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 84 .dst_reg = DST, \ 85 .src_reg = SRC, \ 86 .off = 0, \ 87 .imm = 0 }) 88 89 #define BPF_ALU32_REG(OP, DST, SRC) \ 90 ((struct bpf_insn) { \ 91 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 92 .dst_reg = DST, \ 93 .src_reg = SRC, \ 94 .off = 0, \ 95 .imm = 0 }) 96 97 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 98 99 #define BPF_ALU64_IMM(OP, DST, IMM) \ 100 ((struct bpf_insn) { \ 101 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 102 .dst_reg = DST, \ 103 .src_reg = 0, \ 104 .off = 0, \ 105 .imm = IMM }) 106 107 #define BPF_ALU32_IMM(OP, DST, IMM) \ 108 ((struct bpf_insn) { \ 109 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 110 .dst_reg = DST, \ 111 .src_reg = 0, \ 112 .off = 0, \ 113 .imm = IMM }) 114 115 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 116 117 #define BPF_ENDIAN(TYPE, DST, LEN) \ 118 ((struct bpf_insn) { \ 119 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 120 .dst_reg = DST, \ 121 .src_reg = 0, \ 122 .off = 0, \ 123 .imm = LEN }) 124 125 /* Short form of mov, dst_reg = src_reg */ 126 127 #define BPF_MOV64_REG(DST, SRC) \ 128 ((struct bpf_insn) { \ 129 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 130 .dst_reg = DST, \ 131 .src_reg = SRC, \ 132 .off = 0, \ 133 .imm = 0 }) 134 135 #define BPF_MOV32_REG(DST, SRC) \ 136 ((struct bpf_insn) { \ 137 .code = BPF_ALU | BPF_MOV | BPF_X, \ 138 .dst_reg = DST, \ 139 .src_reg = SRC, \ 140 .off = 0, \ 141 .imm = 0 }) 142 143 /* Short form of mov, dst_reg = imm32 */ 144 145 #define BPF_MOV64_IMM(DST, IMM) \ 146 ((struct bpf_insn) { \ 147 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 148 .dst_reg = DST, \ 149 .src_reg = 0, \ 150 .off = 0, \ 151 .imm = IMM }) 152 153 #define BPF_MOV32_IMM(DST, IMM) \ 154 ((struct bpf_insn) { \ 155 .code = BPF_ALU | BPF_MOV | BPF_K, \ 156 .dst_reg = DST, \ 157 .src_reg = 0, \ 158 .off = 0, \ 159 .imm = IMM }) 160 161 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 162 #define BPF_LD_IMM64(DST, IMM) \ 163 BPF_LD_IMM64_RAW(DST, 0, IMM) 164 165 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 166 ((struct bpf_insn) { \ 167 .code = BPF_LD | BPF_DW | BPF_IMM, \ 168 .dst_reg = DST, \ 169 .src_reg = SRC, \ 170 .off = 0, \ 171 .imm = (__u32) (IMM) }), \ 172 ((struct bpf_insn) { \ 173 .code = 0, /* zero is reserved opcode */ \ 174 .dst_reg = 0, \ 175 .src_reg = 0, \ 176 .off = 0, \ 177 .imm = ((__u64) (IMM)) >> 32 }) 178 179 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 180 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 181 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 182 183 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 184 185 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 186 ((struct bpf_insn) { \ 187 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 188 .dst_reg = DST, \ 189 .src_reg = SRC, \ 190 .off = 0, \ 191 .imm = IMM }) 192 193 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 194 ((struct bpf_insn) { \ 195 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 196 .dst_reg = DST, \ 197 .src_reg = SRC, \ 198 .off = 0, \ 199 .imm = IMM }) 200 201 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 202 203 #define BPF_LD_ABS(SIZE, IMM) \ 204 ((struct bpf_insn) { \ 205 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 206 .dst_reg = 0, \ 207 .src_reg = 0, \ 208 .off = 0, \ 209 .imm = IMM }) 210 211 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 212 213 #define BPF_LD_IND(SIZE, SRC, IMM) \ 214 ((struct bpf_insn) { \ 215 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 216 .dst_reg = 0, \ 217 .src_reg = SRC, \ 218 .off = 0, \ 219 .imm = IMM }) 220 221 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 222 223 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 224 ((struct bpf_insn) { \ 225 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 226 .dst_reg = DST, \ 227 .src_reg = SRC, \ 228 .off = OFF, \ 229 .imm = 0 }) 230 231 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 232 233 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 234 ((struct bpf_insn) { \ 235 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 236 .dst_reg = DST, \ 237 .src_reg = SRC, \ 238 .off = OFF, \ 239 .imm = 0 }) 240 241 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */ 242 243 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \ 244 ((struct bpf_insn) { \ 245 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \ 246 .dst_reg = DST, \ 247 .src_reg = SRC, \ 248 .off = OFF, \ 249 .imm = 0 }) 250 251 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 252 253 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 254 ((struct bpf_insn) { \ 255 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 256 .dst_reg = DST, \ 257 .src_reg = 0, \ 258 .off = OFF, \ 259 .imm = IMM }) 260 261 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 262 263 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 264 ((struct bpf_insn) { \ 265 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 266 .dst_reg = DST, \ 267 .src_reg = SRC, \ 268 .off = OFF, \ 269 .imm = 0 }) 270 271 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 272 273 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 274 ((struct bpf_insn) { \ 275 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 276 .dst_reg = DST, \ 277 .src_reg = 0, \ 278 .off = OFF, \ 279 .imm = IMM }) 280 281 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 282 283 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 284 ((struct bpf_insn) { \ 285 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 286 .dst_reg = DST, \ 287 .src_reg = SRC, \ 288 .off = OFF, \ 289 .imm = 0 }) 290 291 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 292 293 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 294 ((struct bpf_insn) { \ 295 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 296 .dst_reg = DST, \ 297 .src_reg = 0, \ 298 .off = OFF, \ 299 .imm = IMM }) 300 301 /* Unconditional jumps, goto pc + off16 */ 302 303 #define BPF_JMP_A(OFF) \ 304 ((struct bpf_insn) { \ 305 .code = BPF_JMP | BPF_JA, \ 306 .dst_reg = 0, \ 307 .src_reg = 0, \ 308 .off = OFF, \ 309 .imm = 0 }) 310 311 /* Relative call */ 312 313 #define BPF_CALL_REL(TGT) \ 314 ((struct bpf_insn) { \ 315 .code = BPF_JMP | BPF_CALL, \ 316 .dst_reg = 0, \ 317 .src_reg = BPF_PSEUDO_CALL, \ 318 .off = 0, \ 319 .imm = TGT }) 320 321 /* Function call */ 322 323 #define BPF_CAST_CALL(x) \ 324 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 325 326 #define BPF_EMIT_CALL(FUNC) \ 327 ((struct bpf_insn) { \ 328 .code = BPF_JMP | BPF_CALL, \ 329 .dst_reg = 0, \ 330 .src_reg = 0, \ 331 .off = 0, \ 332 .imm = ((FUNC) - __bpf_call_base) }) 333 334 /* Raw code statement block */ 335 336 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 337 ((struct bpf_insn) { \ 338 .code = CODE, \ 339 .dst_reg = DST, \ 340 .src_reg = SRC, \ 341 .off = OFF, \ 342 .imm = IMM }) 343 344 /* Program exit */ 345 346 #define BPF_EXIT_INSN() \ 347 ((struct bpf_insn) { \ 348 .code = BPF_JMP | BPF_EXIT, \ 349 .dst_reg = 0, \ 350 .src_reg = 0, \ 351 .off = 0, \ 352 .imm = 0 }) 353 354 /* Internal classic blocks for direct assignment */ 355 356 #define __BPF_STMT(CODE, K) \ 357 ((struct sock_filter) BPF_STMT(CODE, K)) 358 359 #define __BPF_JUMP(CODE, K, JT, JF) \ 360 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 361 362 #define bytes_to_bpf_size(bytes) \ 363 ({ \ 364 int bpf_size = -EINVAL; \ 365 \ 366 if (bytes == sizeof(u8)) \ 367 bpf_size = BPF_B; \ 368 else if (bytes == sizeof(u16)) \ 369 bpf_size = BPF_H; \ 370 else if (bytes == sizeof(u32)) \ 371 bpf_size = BPF_W; \ 372 else if (bytes == sizeof(u64)) \ 373 bpf_size = BPF_DW; \ 374 \ 375 bpf_size; \ 376 }) 377 378 #define bpf_size_to_bytes(bpf_size) \ 379 ({ \ 380 int bytes = -EINVAL; \ 381 \ 382 if (bpf_size == BPF_B) \ 383 bytes = sizeof(u8); \ 384 else if (bpf_size == BPF_H) \ 385 bytes = sizeof(u16); \ 386 else if (bpf_size == BPF_W) \ 387 bytes = sizeof(u32); \ 388 else if (bpf_size == BPF_DW) \ 389 bytes = sizeof(u64); \ 390 \ 391 bytes; \ 392 }) 393 394 #define BPF_SIZEOF(type) \ 395 ({ \ 396 const int __size = bytes_to_bpf_size(sizeof(type)); \ 397 BUILD_BUG_ON(__size < 0); \ 398 __size; \ 399 }) 400 401 #define BPF_FIELD_SIZEOF(type, field) \ 402 ({ \ 403 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ 404 BUILD_BUG_ON(__size < 0); \ 405 __size; \ 406 }) 407 408 #define BPF_LDST_BYTES(insn) \ 409 ({ \ 410 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 411 WARN_ON(__size < 0); \ 412 __size; \ 413 }) 414 415 #define __BPF_MAP_0(m, v, ...) v 416 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 417 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 418 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 419 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 420 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 421 422 #define __BPF_REG_0(...) __BPF_PAD(5) 423 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 424 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 425 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 426 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 427 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 428 429 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 430 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 431 432 #define __BPF_CAST(t, a) \ 433 (__force t) \ 434 (__force \ 435 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 436 (unsigned long)0, (t)0))) a 437 #define __BPF_V void 438 #define __BPF_N 439 440 #define __BPF_DECL_ARGS(t, a) t a 441 #define __BPF_DECL_REGS(t, a) u64 a 442 443 #define __BPF_PAD(n) \ 444 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 445 u64, __ur_3, u64, __ur_4, u64, __ur_5) 446 447 #define BPF_CALL_x(x, name, ...) \ 448 static __always_inline \ 449 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 450 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 451 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 452 { \ 453 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 454 } \ 455 static __always_inline \ 456 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 457 458 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 459 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 460 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 461 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 462 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 463 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 464 465 #define bpf_ctx_range(TYPE, MEMBER) \ 466 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 467 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 468 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 469 #if BITS_PER_LONG == 64 470 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 471 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 472 #else 473 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 474 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 475 #endif /* BITS_PER_LONG == 64 */ 476 477 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 478 ({ \ 479 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \ 480 *(PTR_SIZE) = (SIZE); \ 481 offsetof(TYPE, MEMBER); \ 482 }) 483 484 #ifdef CONFIG_COMPAT 485 /* A struct sock_filter is architecture independent. */ 486 struct compat_sock_fprog { 487 u16 len; 488 compat_uptr_t filter; /* struct sock_filter * */ 489 }; 490 #endif 491 492 struct sock_fprog_kern { 493 u16 len; 494 struct sock_filter *filter; 495 }; 496 497 struct bpf_binary_header { 498 u32 pages; 499 /* Some arches need word alignment for their instructions */ 500 u8 image[] __aligned(4); 501 }; 502 503 struct bpf_prog { 504 u16 pages; /* Number of allocated pages */ 505 u16 jited:1, /* Is our filter JIT'ed? */ 506 jit_requested:1,/* archs need to JIT the prog */ 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 set_vm_flush_reset_perms(fp); 737 set_memory_ro((unsigned long)fp, fp->pages); 738 } 739 740 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 741 { 742 set_vm_flush_reset_perms(hdr); 743 set_memory_ro((unsigned long)hdr, hdr->pages); 744 set_memory_x((unsigned long)hdr, hdr->pages); 745 } 746 747 static inline struct bpf_binary_header * 748 bpf_jit_binary_hdr(const struct bpf_prog *fp) 749 { 750 unsigned long real_start = (unsigned long)fp->bpf_func; 751 unsigned long addr = real_start & PAGE_MASK; 752 753 return (void *)addr; 754 } 755 756 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 757 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 758 { 759 return sk_filter_trim_cap(sk, skb, 1); 760 } 761 762 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 763 void bpf_prog_free(struct bpf_prog *fp); 764 765 bool bpf_opcode_in_insntable(u8 code); 766 767 void bpf_prog_free_linfo(struct bpf_prog *prog); 768 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 769 const u32 *insn_to_jit_off); 770 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 771 void bpf_prog_free_jited_linfo(struct bpf_prog *prog); 772 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog); 773 774 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 775 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 776 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 777 gfp_t gfp_extra_flags); 778 void __bpf_prog_free(struct bpf_prog *fp); 779 780 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 781 { 782 __bpf_prog_free(fp); 783 } 784 785 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 786 unsigned int flen); 787 788 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 789 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 790 bpf_aux_classic_check_t trans, bool save_orig); 791 void bpf_prog_destroy(struct bpf_prog *fp); 792 793 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 794 int sk_attach_bpf(u32 ufd, struct sock *sk); 795 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 796 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 797 void sk_reuseport_prog_free(struct bpf_prog *prog); 798 int sk_detach_filter(struct sock *sk); 799 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 800 unsigned int len); 801 802 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 803 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 804 805 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 806 #define __bpf_call_base_args \ 807 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 808 __bpf_call_base) 809 810 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 811 void bpf_jit_compile(struct bpf_prog *prog); 812 bool bpf_helper_changes_pkt_data(void *func); 813 814 static inline bool bpf_dump_raw_ok(void) 815 { 816 /* Reconstruction of call-sites is dependent on kallsyms, 817 * thus make dump the same restriction. 818 */ 819 return kallsyms_show_value() == 1; 820 } 821 822 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 823 const struct bpf_insn *patch, u32 len); 824 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 825 826 void bpf_clear_redirect_map(struct bpf_map *map); 827 828 static inline bool xdp_return_frame_no_direct(void) 829 { 830 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 831 832 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 833 } 834 835 static inline void xdp_set_return_frame_no_direct(void) 836 { 837 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 838 839 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 840 } 841 842 static inline void xdp_clear_return_frame_no_direct(void) 843 { 844 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 845 846 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 847 } 848 849 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 850 unsigned int pktlen) 851 { 852 unsigned int len; 853 854 if (unlikely(!(fwd->flags & IFF_UP))) 855 return -ENETDOWN; 856 857 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 858 if (pktlen > len) 859 return -EMSGSIZE; 860 861 return 0; 862 } 863 864 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the 865 * same cpu context. Further for best results no more than a single map 866 * for the do_redirect/do_flush pair should be used. This limitation is 867 * because we only track one map and force a flush when the map changes. 868 * This does not appear to be a real limitation for existing software. 869 */ 870 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 871 struct xdp_buff *xdp, struct bpf_prog *prog); 872 int xdp_do_redirect(struct net_device *dev, 873 struct xdp_buff *xdp, 874 struct bpf_prog *prog); 875 void xdp_do_flush_map(void); 876 877 void bpf_warn_invalid_xdp_action(u32 act); 878 879 #ifdef CONFIG_INET 880 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 881 struct bpf_prog *prog, struct sk_buff *skb, 882 u32 hash); 883 #else 884 static inline struct sock * 885 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 886 struct bpf_prog *prog, struct sk_buff *skb, 887 u32 hash) 888 { 889 return NULL; 890 } 891 #endif 892 893 #ifdef CONFIG_BPF_JIT 894 extern int bpf_jit_enable; 895 extern int bpf_jit_harden; 896 extern int bpf_jit_kallsyms; 897 extern long bpf_jit_limit; 898 899 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 900 901 struct bpf_binary_header * 902 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 903 unsigned int alignment, 904 bpf_jit_fill_hole_t bpf_fill_ill_insns); 905 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 906 u64 bpf_jit_alloc_exec_limit(void); 907 void *bpf_jit_alloc_exec(unsigned long size); 908 void bpf_jit_free_exec(void *addr); 909 void bpf_jit_free(struct bpf_prog *fp); 910 911 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 912 const struct bpf_insn *insn, bool extra_pass, 913 u64 *func_addr, bool *func_addr_fixed); 914 915 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 916 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 917 918 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 919 u32 pass, void *image) 920 { 921 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 922 proglen, pass, image, current->comm, task_pid_nr(current)); 923 924 if (image) 925 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 926 16, 1, image, proglen, false); 927 } 928 929 static inline bool bpf_jit_is_ebpf(void) 930 { 931 # ifdef CONFIG_HAVE_EBPF_JIT 932 return true; 933 # else 934 return false; 935 # endif 936 } 937 938 static inline bool ebpf_jit_enabled(void) 939 { 940 return bpf_jit_enable && bpf_jit_is_ebpf(); 941 } 942 943 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 944 { 945 return fp->jited && bpf_jit_is_ebpf(); 946 } 947 948 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 949 { 950 /* These are the prerequisites, should someone ever have the 951 * idea to call blinding outside of them, we make sure to 952 * bail out. 953 */ 954 if (!bpf_jit_is_ebpf()) 955 return false; 956 if (!prog->jit_requested) 957 return false; 958 if (!bpf_jit_harden) 959 return false; 960 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 961 return false; 962 963 return true; 964 } 965 966 static inline bool bpf_jit_kallsyms_enabled(void) 967 { 968 /* There are a couple of corner cases where kallsyms should 969 * not be enabled f.e. on hardening. 970 */ 971 if (bpf_jit_harden) 972 return false; 973 if (!bpf_jit_kallsyms) 974 return false; 975 if (bpf_jit_kallsyms == 1) 976 return true; 977 978 return false; 979 } 980 981 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 982 unsigned long *off, char *sym); 983 bool is_bpf_text_address(unsigned long addr); 984 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 985 char *sym); 986 987 static inline const char * 988 bpf_address_lookup(unsigned long addr, unsigned long *size, 989 unsigned long *off, char **modname, char *sym) 990 { 991 const char *ret = __bpf_address_lookup(addr, size, off, sym); 992 993 if (ret && modname) 994 *modname = NULL; 995 return ret; 996 } 997 998 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 999 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1000 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym); 1001 1002 #else /* CONFIG_BPF_JIT */ 1003 1004 static inline bool ebpf_jit_enabled(void) 1005 { 1006 return false; 1007 } 1008 1009 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1010 { 1011 return false; 1012 } 1013 1014 static inline void bpf_jit_free(struct bpf_prog *fp) 1015 { 1016 bpf_prog_unlock_free(fp); 1017 } 1018 1019 static inline bool bpf_jit_kallsyms_enabled(void) 1020 { 1021 return false; 1022 } 1023 1024 static inline const char * 1025 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1026 unsigned long *off, char *sym) 1027 { 1028 return NULL; 1029 } 1030 1031 static inline bool is_bpf_text_address(unsigned long addr) 1032 { 1033 return false; 1034 } 1035 1036 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1037 char *type, char *sym) 1038 { 1039 return -ERANGE; 1040 } 1041 1042 static inline const char * 1043 bpf_address_lookup(unsigned long addr, unsigned long *size, 1044 unsigned long *off, char **modname, char *sym) 1045 { 1046 return NULL; 1047 } 1048 1049 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1050 { 1051 } 1052 1053 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1054 { 1055 } 1056 1057 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym) 1058 { 1059 sym[0] = '\0'; 1060 } 1061 1062 #endif /* CONFIG_BPF_JIT */ 1063 1064 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp); 1065 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1066 1067 #define BPF_ANC BIT(15) 1068 1069 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1070 { 1071 switch (first->code) { 1072 case BPF_RET | BPF_K: 1073 case BPF_LD | BPF_W | BPF_LEN: 1074 return false; 1075 1076 case BPF_LD | BPF_W | BPF_ABS: 1077 case BPF_LD | BPF_H | BPF_ABS: 1078 case BPF_LD | BPF_B | BPF_ABS: 1079 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1080 return true; 1081 return false; 1082 1083 default: 1084 return true; 1085 } 1086 } 1087 1088 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1089 { 1090 BUG_ON(ftest->code & BPF_ANC); 1091 1092 switch (ftest->code) { 1093 case BPF_LD | BPF_W | BPF_ABS: 1094 case BPF_LD | BPF_H | BPF_ABS: 1095 case BPF_LD | BPF_B | BPF_ABS: 1096 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1097 return BPF_ANC | SKF_AD_##CODE 1098 switch (ftest->k) { 1099 BPF_ANCILLARY(PROTOCOL); 1100 BPF_ANCILLARY(PKTTYPE); 1101 BPF_ANCILLARY(IFINDEX); 1102 BPF_ANCILLARY(NLATTR); 1103 BPF_ANCILLARY(NLATTR_NEST); 1104 BPF_ANCILLARY(MARK); 1105 BPF_ANCILLARY(QUEUE); 1106 BPF_ANCILLARY(HATYPE); 1107 BPF_ANCILLARY(RXHASH); 1108 BPF_ANCILLARY(CPU); 1109 BPF_ANCILLARY(ALU_XOR_X); 1110 BPF_ANCILLARY(VLAN_TAG); 1111 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1112 BPF_ANCILLARY(PAY_OFFSET); 1113 BPF_ANCILLARY(RANDOM); 1114 BPF_ANCILLARY(VLAN_TPID); 1115 } 1116 /* Fallthrough. */ 1117 default: 1118 return ftest->code; 1119 } 1120 } 1121 1122 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1123 int k, unsigned int size); 1124 1125 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, 1126 unsigned int size, void *buffer) 1127 { 1128 if (k >= 0) 1129 return skb_header_pointer(skb, k, size, buffer); 1130 1131 return bpf_internal_load_pointer_neg_helper(skb, k, size); 1132 } 1133 1134 static inline int bpf_tell_extensions(void) 1135 { 1136 return SKF_AD_MAX; 1137 } 1138 1139 struct bpf_sock_addr_kern { 1140 struct sock *sk; 1141 struct sockaddr *uaddr; 1142 /* Temporary "register" to make indirect stores to nested structures 1143 * defined above. We need three registers to make such a store, but 1144 * only two (src and dst) are available at convert_ctx_access time 1145 */ 1146 u64 tmp_reg; 1147 void *t_ctx; /* Attach type specific context. */ 1148 }; 1149 1150 struct bpf_sock_ops_kern { 1151 struct sock *sk; 1152 u32 op; 1153 union { 1154 u32 args[4]; 1155 u32 reply; 1156 u32 replylong[4]; 1157 }; 1158 u32 is_fullsock; 1159 u64 temp; /* temp and everything after is not 1160 * initialized to 0 before calling 1161 * the BPF program. New fields that 1162 * should be initialized to 0 should 1163 * be inserted before temp. 1164 * temp is scratch storage used by 1165 * sock_ops_convert_ctx_access 1166 * as temporary storage of a register. 1167 */ 1168 }; 1169 1170 #endif /* __LINUX_FILTER_H__ */ 1171