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