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