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