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 /* Unconditional jumps, goto pc + off16 */ 281 282 #define BPF_JMP_A(OFF) \ 283 ((struct bpf_insn) { \ 284 .code = BPF_JMP | BPF_JA, \ 285 .dst_reg = 0, \ 286 .src_reg = 0, \ 287 .off = OFF, \ 288 .imm = 0 }) 289 290 /* Relative call */ 291 292 #define BPF_CALL_REL(TGT) \ 293 ((struct bpf_insn) { \ 294 .code = BPF_JMP | BPF_CALL, \ 295 .dst_reg = 0, \ 296 .src_reg = BPF_PSEUDO_CALL, \ 297 .off = 0, \ 298 .imm = TGT }) 299 300 /* Function call */ 301 302 #define BPF_CAST_CALL(x) \ 303 ((u64 (*)(u64, u64, u64, u64, u64))(x)) 304 305 #define BPF_EMIT_CALL(FUNC) \ 306 ((struct bpf_insn) { \ 307 .code = BPF_JMP | BPF_CALL, \ 308 .dst_reg = 0, \ 309 .src_reg = 0, \ 310 .off = 0, \ 311 .imm = ((FUNC) - __bpf_call_base) }) 312 313 /* Raw code statement block */ 314 315 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 316 ((struct bpf_insn) { \ 317 .code = CODE, \ 318 .dst_reg = DST, \ 319 .src_reg = SRC, \ 320 .off = OFF, \ 321 .imm = IMM }) 322 323 /* Program exit */ 324 325 #define BPF_EXIT_INSN() \ 326 ((struct bpf_insn) { \ 327 .code = BPF_JMP | BPF_EXIT, \ 328 .dst_reg = 0, \ 329 .src_reg = 0, \ 330 .off = 0, \ 331 .imm = 0 }) 332 333 /* Internal classic blocks for direct assignment */ 334 335 #define __BPF_STMT(CODE, K) \ 336 ((struct sock_filter) BPF_STMT(CODE, K)) 337 338 #define __BPF_JUMP(CODE, K, JT, JF) \ 339 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 340 341 #define bytes_to_bpf_size(bytes) \ 342 ({ \ 343 int bpf_size = -EINVAL; \ 344 \ 345 if (bytes == sizeof(u8)) \ 346 bpf_size = BPF_B; \ 347 else if (bytes == sizeof(u16)) \ 348 bpf_size = BPF_H; \ 349 else if (bytes == sizeof(u32)) \ 350 bpf_size = BPF_W; \ 351 else if (bytes == sizeof(u64)) \ 352 bpf_size = BPF_DW; \ 353 \ 354 bpf_size; \ 355 }) 356 357 #define bpf_size_to_bytes(bpf_size) \ 358 ({ \ 359 int bytes = -EINVAL; \ 360 \ 361 if (bpf_size == BPF_B) \ 362 bytes = sizeof(u8); \ 363 else if (bpf_size == BPF_H) \ 364 bytes = sizeof(u16); \ 365 else if (bpf_size == BPF_W) \ 366 bytes = sizeof(u32); \ 367 else if (bpf_size == BPF_DW) \ 368 bytes = sizeof(u64); \ 369 \ 370 bytes; \ 371 }) 372 373 #define BPF_SIZEOF(type) \ 374 ({ \ 375 const int __size = bytes_to_bpf_size(sizeof(type)); \ 376 BUILD_BUG_ON(__size < 0); \ 377 __size; \ 378 }) 379 380 #define BPF_FIELD_SIZEOF(type, field) \ 381 ({ \ 382 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \ 383 BUILD_BUG_ON(__size < 0); \ 384 __size; \ 385 }) 386 387 #define BPF_LDST_BYTES(insn) \ 388 ({ \ 389 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 390 WARN_ON(__size < 0); \ 391 __size; \ 392 }) 393 394 #define __BPF_MAP_0(m, v, ...) v 395 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 396 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 397 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 398 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 399 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 400 401 #define __BPF_REG_0(...) __BPF_PAD(5) 402 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 403 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 404 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 405 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 406 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 407 408 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 409 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 410 411 #define __BPF_CAST(t, a) \ 412 (__force t) \ 413 (__force \ 414 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 415 (unsigned long)0, (t)0))) a 416 #define __BPF_V void 417 #define __BPF_N 418 419 #define __BPF_DECL_ARGS(t, a) t a 420 #define __BPF_DECL_REGS(t, a) u64 a 421 422 #define __BPF_PAD(n) \ 423 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 424 u64, __ur_3, u64, __ur_4, u64, __ur_5) 425 426 #define BPF_CALL_x(x, name, ...) \ 427 static __always_inline \ 428 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 429 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 430 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 431 { \ 432 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 433 } \ 434 static __always_inline \ 435 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 436 437 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__) 438 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__) 439 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__) 440 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__) 441 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__) 442 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__) 443 444 #define bpf_ctx_range(TYPE, MEMBER) \ 445 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 446 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 447 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 448 #if BITS_PER_LONG == 64 449 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 450 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 451 #else 452 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 453 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 454 #endif /* BITS_PER_LONG == 64 */ 455 456 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 457 ({ \ 458 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \ 459 *(PTR_SIZE) = (SIZE); \ 460 offsetof(TYPE, MEMBER); \ 461 }) 462 463 #ifdef CONFIG_COMPAT 464 /* A struct sock_filter is architecture independent. */ 465 struct compat_sock_fprog { 466 u16 len; 467 compat_uptr_t filter; /* struct sock_filter * */ 468 }; 469 #endif 470 471 struct sock_fprog_kern { 472 u16 len; 473 struct sock_filter *filter; 474 }; 475 476 struct bpf_binary_header { 477 u32 pages; 478 /* Some arches need word alignment for their instructions */ 479 u8 image[] __aligned(4); 480 }; 481 482 struct bpf_prog { 483 u16 pages; /* Number of allocated pages */ 484 u16 jited:1, /* Is our filter JIT'ed? */ 485 jit_requested:1,/* archs need to JIT the prog */ 486 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */ 487 gpl_compatible:1, /* Is filter GPL compatible? */ 488 cb_access:1, /* Is control block accessed? */ 489 dst_needed:1, /* Do we need dst entry? */ 490 blinded:1, /* Was blinded */ 491 is_func:1, /* program is a bpf function */ 492 kprobe_override:1, /* Do we override a kprobe? */ 493 has_callchain_buf:1; /* callchain buffer allocated? */ 494 enum bpf_prog_type type; /* Type of BPF program */ 495 enum bpf_attach_type expected_attach_type; /* For some prog types */ 496 u32 len; /* Number of filter blocks */ 497 u32 jited_len; /* Size of jited insns in bytes */ 498 u8 tag[BPF_TAG_SIZE]; 499 struct bpf_prog_aux *aux; /* Auxiliary fields */ 500 struct sock_fprog_kern *orig_prog; /* Original BPF program */ 501 unsigned int (*bpf_func)(const void *ctx, 502 const struct bpf_insn *insn); 503 /* Instructions for interpreter */ 504 union { 505 struct sock_filter insns[0]; 506 struct bpf_insn insnsi[0]; 507 }; 508 }; 509 510 struct sk_filter { 511 refcount_t refcnt; 512 struct rcu_head rcu; 513 struct bpf_prog *prog; 514 }; 515 516 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi) 517 518 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 519 520 struct bpf_skb_data_end { 521 struct qdisc_skb_cb qdisc_cb; 522 void *data_meta; 523 void *data_end; 524 }; 525 526 struct bpf_redirect_info { 527 u32 ifindex; 528 u32 flags; 529 struct bpf_map *map; 530 struct bpf_map *map_to_flush; 531 u32 kern_flags; 532 }; 533 534 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 535 536 /* flags for bpf_redirect_info kern_flags */ 537 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 538 539 /* Compute the linear packet data range [data, data_end) which 540 * will be accessed by various program types (cls_bpf, act_bpf, 541 * lwt, ...). Subsystems allowing direct data access must (!) 542 * ensure that cb[] area can be written to when BPF program is 543 * invoked (otherwise cb[] save/restore is necessary). 544 */ 545 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 546 { 547 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 548 549 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb)); 550 cb->data_meta = skb->data - skb_metadata_len(skb); 551 cb->data_end = skb->data + skb_headlen(skb); 552 } 553 554 /* Similar to bpf_compute_data_pointers(), except that save orginal 555 * data in cb->data and cb->meta_data for restore. 556 */ 557 static inline void bpf_compute_and_save_data_end( 558 struct sk_buff *skb, void **saved_data_end) 559 { 560 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 561 562 *saved_data_end = cb->data_end; 563 cb->data_end = skb->data + skb_headlen(skb); 564 } 565 566 /* Restore data saved by bpf_compute_data_pointers(). */ 567 static inline void bpf_restore_data_end( 568 struct sk_buff *skb, void *saved_data_end) 569 { 570 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 571 572 cb->data_end = saved_data_end; 573 } 574 575 static inline u8 *bpf_skb_cb(struct sk_buff *skb) 576 { 577 /* eBPF programs may read/write skb->cb[] area to transfer meta 578 * data between tail calls. Since this also needs to work with 579 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 580 * 581 * In some socket filter cases, the cb unfortunately needs to be 582 * saved/restored so that protocol specific skb->cb[] data won't 583 * be lost. In any case, due to unpriviledged eBPF programs 584 * attached to sockets, we need to clear the bpf_skb_cb() area 585 * to not leak previous contents to user space. 586 */ 587 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 588 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != 589 FIELD_SIZEOF(struct qdisc_skb_cb, data)); 590 591 return qdisc_skb_cb(skb)->data; 592 } 593 594 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 595 struct sk_buff *skb) 596 { 597 u8 *cb_data = bpf_skb_cb(skb); 598 u8 cb_saved[BPF_SKB_CB_LEN]; 599 u32 res; 600 601 if (unlikely(prog->cb_access)) { 602 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 603 memset(cb_data, 0, sizeof(cb_saved)); 604 } 605 606 res = BPF_PROG_RUN(prog, skb); 607 608 if (unlikely(prog->cb_access)) 609 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 610 611 return res; 612 } 613 614 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 615 struct sk_buff *skb) 616 { 617 u8 *cb_data = bpf_skb_cb(skb); 618 619 if (unlikely(prog->cb_access)) 620 memset(cb_data, 0, BPF_SKB_CB_LEN); 621 622 return BPF_PROG_RUN(prog, skb); 623 } 624 625 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 626 struct xdp_buff *xdp) 627 { 628 /* Caller needs to hold rcu_read_lock() (!), otherwise program 629 * can be released while still running, or map elements could be 630 * freed early while still having concurrent users. XDP fastpath 631 * already takes rcu_read_lock() when fetching the program, so 632 * it's not necessary here anymore. 633 */ 634 return BPF_PROG_RUN(prog, xdp); 635 } 636 637 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 638 { 639 return prog->len * sizeof(struct bpf_insn); 640 } 641 642 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 643 { 644 return round_up(bpf_prog_insn_size(prog) + 645 sizeof(__be64) + 1, SHA_MESSAGE_BYTES); 646 } 647 648 static inline unsigned int bpf_prog_size(unsigned int proglen) 649 { 650 return max(sizeof(struct bpf_prog), 651 offsetof(struct bpf_prog, insns[proglen])); 652 } 653 654 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 655 { 656 /* When classic BPF programs have been loaded and the arch 657 * does not have a classic BPF JIT (anymore), they have been 658 * converted via bpf_migrate_filter() to eBPF and thus always 659 * have an unspec program type. 660 */ 661 return prog->type == BPF_PROG_TYPE_UNSPEC; 662 } 663 664 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 665 { 666 const u32 size_machine = sizeof(unsigned long); 667 668 if (size > size_machine && size % size_machine == 0) 669 size = size_machine; 670 671 return size; 672 } 673 674 static inline bool 675 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 676 { 677 return size <= size_default && (size & (size - 1)) == 0; 678 } 679 680 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 681 682 static inline void bpf_prog_lock_ro(struct bpf_prog *fp) 683 { 684 fp->undo_set_mem = 1; 685 set_memory_ro((unsigned long)fp, fp->pages); 686 } 687 688 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp) 689 { 690 if (fp->undo_set_mem) 691 set_memory_rw((unsigned long)fp, fp->pages); 692 } 693 694 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 695 { 696 set_memory_ro((unsigned long)hdr, hdr->pages); 697 } 698 699 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr) 700 { 701 set_memory_rw((unsigned long)hdr, hdr->pages); 702 } 703 704 static inline struct bpf_binary_header * 705 bpf_jit_binary_hdr(const struct bpf_prog *fp) 706 { 707 unsigned long real_start = (unsigned long)fp->bpf_func; 708 unsigned long addr = real_start & PAGE_MASK; 709 710 return (void *)addr; 711 } 712 713 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 714 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 715 { 716 return sk_filter_trim_cap(sk, skb, 1); 717 } 718 719 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 720 void bpf_prog_free(struct bpf_prog *fp); 721 722 bool bpf_opcode_in_insntable(u8 code); 723 724 void bpf_prog_free_linfo(struct bpf_prog *prog); 725 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 726 const u32 *insn_to_jit_off); 727 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 728 void bpf_prog_free_jited_linfo(struct bpf_prog *prog); 729 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog); 730 731 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 732 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 733 gfp_t gfp_extra_flags); 734 void __bpf_prog_free(struct bpf_prog *fp); 735 736 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 737 { 738 bpf_prog_unlock_ro(fp); 739 __bpf_prog_free(fp); 740 } 741 742 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 743 unsigned int flen); 744 745 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 746 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 747 bpf_aux_classic_check_t trans, bool save_orig); 748 void bpf_prog_destroy(struct bpf_prog *fp); 749 750 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 751 int sk_attach_bpf(u32 ufd, struct sock *sk); 752 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 753 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 754 void sk_reuseport_prog_free(struct bpf_prog *prog); 755 int sk_detach_filter(struct sock *sk); 756 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter, 757 unsigned int len); 758 759 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 760 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 761 762 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 763 #define __bpf_call_base_args \ 764 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 765 __bpf_call_base) 766 767 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 768 void bpf_jit_compile(struct bpf_prog *prog); 769 bool bpf_helper_changes_pkt_data(void *func); 770 771 static inline bool bpf_dump_raw_ok(void) 772 { 773 /* Reconstruction of call-sites is dependent on kallsyms, 774 * thus make dump the same restriction. 775 */ 776 return kallsyms_show_value() == 1; 777 } 778 779 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 780 const struct bpf_insn *patch, u32 len); 781 782 void bpf_clear_redirect_map(struct bpf_map *map); 783 784 static inline bool xdp_return_frame_no_direct(void) 785 { 786 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 787 788 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 789 } 790 791 static inline void xdp_set_return_frame_no_direct(void) 792 { 793 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 794 795 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 796 } 797 798 static inline void xdp_clear_return_frame_no_direct(void) 799 { 800 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 801 802 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 803 } 804 805 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 806 unsigned int pktlen) 807 { 808 unsigned int len; 809 810 if (unlikely(!(fwd->flags & IFF_UP))) 811 return -ENETDOWN; 812 813 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 814 if (pktlen > len) 815 return -EMSGSIZE; 816 817 return 0; 818 } 819 820 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the 821 * same cpu context. Further for best results no more than a single map 822 * for the do_redirect/do_flush pair should be used. This limitation is 823 * because we only track one map and force a flush when the map changes. 824 * This does not appear to be a real limitation for existing software. 825 */ 826 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 827 struct xdp_buff *xdp, struct bpf_prog *prog); 828 int xdp_do_redirect(struct net_device *dev, 829 struct xdp_buff *xdp, 830 struct bpf_prog *prog); 831 void xdp_do_flush_map(void); 832 833 void bpf_warn_invalid_xdp_action(u32 act); 834 835 #ifdef CONFIG_INET 836 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 837 struct bpf_prog *prog, struct sk_buff *skb, 838 u32 hash); 839 #else 840 static inline struct sock * 841 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 842 struct bpf_prog *prog, struct sk_buff *skb, 843 u32 hash) 844 { 845 return NULL; 846 } 847 #endif 848 849 #ifdef CONFIG_BPF_JIT 850 extern int bpf_jit_enable; 851 extern int bpf_jit_harden; 852 extern int bpf_jit_kallsyms; 853 extern long bpf_jit_limit; 854 855 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 856 857 struct bpf_binary_header * 858 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 859 unsigned int alignment, 860 bpf_jit_fill_hole_t bpf_fill_ill_insns); 861 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 862 863 void bpf_jit_free(struct bpf_prog *fp); 864 865 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 866 const struct bpf_insn *insn, bool extra_pass, 867 u64 *func_addr, bool *func_addr_fixed); 868 869 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 870 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 871 872 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 873 u32 pass, void *image) 874 { 875 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 876 proglen, pass, image, current->comm, task_pid_nr(current)); 877 878 if (image) 879 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 880 16, 1, image, proglen, false); 881 } 882 883 static inline bool bpf_jit_is_ebpf(void) 884 { 885 # ifdef CONFIG_HAVE_EBPF_JIT 886 return true; 887 # else 888 return false; 889 # endif 890 } 891 892 static inline bool ebpf_jit_enabled(void) 893 { 894 return bpf_jit_enable && bpf_jit_is_ebpf(); 895 } 896 897 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 898 { 899 return fp->jited && bpf_jit_is_ebpf(); 900 } 901 902 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 903 { 904 /* These are the prerequisites, should someone ever have the 905 * idea to call blinding outside of them, we make sure to 906 * bail out. 907 */ 908 if (!bpf_jit_is_ebpf()) 909 return false; 910 if (!prog->jit_requested) 911 return false; 912 if (!bpf_jit_harden) 913 return false; 914 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN)) 915 return false; 916 917 return true; 918 } 919 920 static inline bool bpf_jit_kallsyms_enabled(void) 921 { 922 /* There are a couple of corner cases where kallsyms should 923 * not be enabled f.e. on hardening. 924 */ 925 if (bpf_jit_harden) 926 return false; 927 if (!bpf_jit_kallsyms) 928 return false; 929 if (bpf_jit_kallsyms == 1) 930 return true; 931 932 return false; 933 } 934 935 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 936 unsigned long *off, char *sym); 937 bool is_bpf_text_address(unsigned long addr); 938 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 939 char *sym); 940 941 static inline const char * 942 bpf_address_lookup(unsigned long addr, unsigned long *size, 943 unsigned long *off, char **modname, char *sym) 944 { 945 const char *ret = __bpf_address_lookup(addr, size, off, sym); 946 947 if (ret && modname) 948 *modname = NULL; 949 return ret; 950 } 951 952 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 953 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 954 955 #else /* CONFIG_BPF_JIT */ 956 957 static inline bool ebpf_jit_enabled(void) 958 { 959 return false; 960 } 961 962 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 963 { 964 return false; 965 } 966 967 static inline void bpf_jit_free(struct bpf_prog *fp) 968 { 969 bpf_prog_unlock_free(fp); 970 } 971 972 static inline bool bpf_jit_kallsyms_enabled(void) 973 { 974 return false; 975 } 976 977 static inline const char * 978 __bpf_address_lookup(unsigned long addr, unsigned long *size, 979 unsigned long *off, char *sym) 980 { 981 return NULL; 982 } 983 984 static inline bool is_bpf_text_address(unsigned long addr) 985 { 986 return false; 987 } 988 989 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 990 char *type, char *sym) 991 { 992 return -ERANGE; 993 } 994 995 static inline const char * 996 bpf_address_lookup(unsigned long addr, unsigned long *size, 997 unsigned long *off, char **modname, char *sym) 998 { 999 return NULL; 1000 } 1001 1002 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1003 { 1004 } 1005 1006 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1007 { 1008 } 1009 #endif /* CONFIG_BPF_JIT */ 1010 1011 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp); 1012 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1013 1014 #define BPF_ANC BIT(15) 1015 1016 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1017 { 1018 switch (first->code) { 1019 case BPF_RET | BPF_K: 1020 case BPF_LD | BPF_W | BPF_LEN: 1021 return false; 1022 1023 case BPF_LD | BPF_W | BPF_ABS: 1024 case BPF_LD | BPF_H | BPF_ABS: 1025 case BPF_LD | BPF_B | BPF_ABS: 1026 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1027 return true; 1028 return false; 1029 1030 default: 1031 return true; 1032 } 1033 } 1034 1035 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1036 { 1037 BUG_ON(ftest->code & BPF_ANC); 1038 1039 switch (ftest->code) { 1040 case BPF_LD | BPF_W | BPF_ABS: 1041 case BPF_LD | BPF_H | BPF_ABS: 1042 case BPF_LD | BPF_B | BPF_ABS: 1043 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1044 return BPF_ANC | SKF_AD_##CODE 1045 switch (ftest->k) { 1046 BPF_ANCILLARY(PROTOCOL); 1047 BPF_ANCILLARY(PKTTYPE); 1048 BPF_ANCILLARY(IFINDEX); 1049 BPF_ANCILLARY(NLATTR); 1050 BPF_ANCILLARY(NLATTR_NEST); 1051 BPF_ANCILLARY(MARK); 1052 BPF_ANCILLARY(QUEUE); 1053 BPF_ANCILLARY(HATYPE); 1054 BPF_ANCILLARY(RXHASH); 1055 BPF_ANCILLARY(CPU); 1056 BPF_ANCILLARY(ALU_XOR_X); 1057 BPF_ANCILLARY(VLAN_TAG); 1058 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1059 BPF_ANCILLARY(PAY_OFFSET); 1060 BPF_ANCILLARY(RANDOM); 1061 BPF_ANCILLARY(VLAN_TPID); 1062 } 1063 /* Fallthrough. */ 1064 default: 1065 return ftest->code; 1066 } 1067 } 1068 1069 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1070 int k, unsigned int size); 1071 1072 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k, 1073 unsigned int size, void *buffer) 1074 { 1075 if (k >= 0) 1076 return skb_header_pointer(skb, k, size, buffer); 1077 1078 return bpf_internal_load_pointer_neg_helper(skb, k, size); 1079 } 1080 1081 static inline int bpf_tell_extensions(void) 1082 { 1083 return SKF_AD_MAX; 1084 } 1085 1086 struct bpf_sock_addr_kern { 1087 struct sock *sk; 1088 struct sockaddr *uaddr; 1089 /* Temporary "register" to make indirect stores to nested structures 1090 * defined above. We need three registers to make such a store, but 1091 * only two (src and dst) are available at convert_ctx_access time 1092 */ 1093 u64 tmp_reg; 1094 void *t_ctx; /* Attach type specific context. */ 1095 }; 1096 1097 struct bpf_sock_ops_kern { 1098 struct sock *sk; 1099 u32 op; 1100 union { 1101 u32 args[4]; 1102 u32 reply; 1103 u32 replylong[4]; 1104 }; 1105 u32 is_fullsock; 1106 u64 temp; /* temp and everything after is not 1107 * initialized to 0 before calling 1108 * the BPF program. New fields that 1109 * should be initialized to 0 should 1110 * be inserted before temp. 1111 * temp is scratch storage used by 1112 * sock_ops_convert_ctx_access 1113 * as temporary storage of a register. 1114 */ 1115 }; 1116 1117 #endif /* __LINUX_FILTER_H__ */ 1118