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