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 <linux/atomic.h> 9 #include <linux/bpf.h> 10 #include <linux/refcount.h> 11 #include <linux/compat.h> 12 #include <linux/skbuff.h> 13 #include <linux/linkage.h> 14 #include <linux/printk.h> 15 #include <linux/workqueue.h> 16 #include <linux/sched.h> 17 #include <linux/sched/clock.h> 18 #include <linux/capability.h> 19 #include <linux/set_memory.h> 20 #include <linux/kallsyms.h> 21 #include <linux/if_vlan.h> 22 #include <linux/vmalloc.h> 23 #include <linux/sockptr.h> 24 #include <crypto/sha1.h> 25 #include <linux/u64_stats_sync.h> 26 27 #include <net/sch_generic.h> 28 29 #include <asm/byteorder.h> 30 #include <uapi/linux/filter.h> 31 32 struct sk_buff; 33 struct sock; 34 struct seccomp_data; 35 struct bpf_prog_aux; 36 struct xdp_rxq_info; 37 struct xdp_buff; 38 struct sock_reuseport; 39 struct ctl_table; 40 struct ctl_table_header; 41 42 /* ArgX, context and stack frame pointer register positions. Note, 43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 44 * calls in BPF_CALL instruction. 45 */ 46 #define BPF_REG_ARG1 BPF_REG_1 47 #define BPF_REG_ARG2 BPF_REG_2 48 #define BPF_REG_ARG3 BPF_REG_3 49 #define BPF_REG_ARG4 BPF_REG_4 50 #define BPF_REG_ARG5 BPF_REG_5 51 #define BPF_REG_CTX BPF_REG_6 52 #define BPF_REG_FP BPF_REG_10 53 54 /* Additional register mappings for converted user programs. */ 55 #define BPF_REG_A BPF_REG_0 56 #define BPF_REG_X BPF_REG_7 57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 60 61 /* Kernel hidden auxiliary/helper register. */ 62 #define BPF_REG_AX MAX_BPF_REG 63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1) 64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG 65 66 /* unused opcode to mark special call to bpf_tail_call() helper */ 67 #define BPF_TAIL_CALL 0xf0 68 69 /* unused opcode to mark special load instruction. Same as BPF_ABS */ 70 #define BPF_PROBE_MEM 0x20 71 72 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */ 73 #define BPF_PROBE_MEMSX 0x40 74 75 /* unused opcode to mark special load instruction. Same as BPF_MSH */ 76 #define BPF_PROBE_MEM32 0xa0 77 78 /* unused opcode to mark special atomic instruction */ 79 #define BPF_PROBE_ATOMIC 0xe0 80 81 /* unused opcode to mark call to interpreter with arguments */ 82 #define BPF_CALL_ARGS 0xe0 83 84 /* unused opcode to mark speculation barrier for mitigating 85 * Speculative Store Bypass 86 */ 87 #define BPF_NOSPEC 0xc0 88 89 /* As per nm, we expose JITed images as text (code) section for 90 * kallsyms. That way, tools like perf can find it to match 91 * addresses. 92 */ 93 #define BPF_SYM_ELF_TYPE 't' 94 95 /* BPF program can access up to 512 bytes of stack space. */ 96 #define MAX_BPF_STACK 512 97 98 /* Helper macros for filter block array initializers. */ 99 100 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 101 102 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \ 103 ((struct bpf_insn) { \ 104 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 105 .dst_reg = DST, \ 106 .src_reg = SRC, \ 107 .off = OFF, \ 108 .imm = 0 }) 109 110 #define BPF_ALU64_REG(OP, DST, SRC) \ 111 BPF_ALU64_REG_OFF(OP, DST, SRC, 0) 112 113 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \ 114 ((struct bpf_insn) { \ 115 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 116 .dst_reg = DST, \ 117 .src_reg = SRC, \ 118 .off = OFF, \ 119 .imm = 0 }) 120 121 #define BPF_ALU32_REG(OP, DST, SRC) \ 122 BPF_ALU32_REG_OFF(OP, DST, SRC, 0) 123 124 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 125 126 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \ 127 ((struct bpf_insn) { \ 128 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 129 .dst_reg = DST, \ 130 .src_reg = 0, \ 131 .off = OFF, \ 132 .imm = IMM }) 133 #define BPF_ALU64_IMM(OP, DST, IMM) \ 134 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0) 135 136 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \ 137 ((struct bpf_insn) { \ 138 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 139 .dst_reg = DST, \ 140 .src_reg = 0, \ 141 .off = OFF, \ 142 .imm = IMM }) 143 #define BPF_ALU32_IMM(OP, DST, IMM) \ 144 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0) 145 146 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 147 148 #define BPF_ENDIAN(TYPE, DST, LEN) \ 149 ((struct bpf_insn) { \ 150 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 151 .dst_reg = DST, \ 152 .src_reg = 0, \ 153 .off = 0, \ 154 .imm = LEN }) 155 156 /* Byte Swap, bswap16/32/64 */ 157 158 #define BPF_BSWAP(DST, LEN) \ 159 ((struct bpf_insn) { \ 160 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \ 161 .dst_reg = DST, \ 162 .src_reg = 0, \ 163 .off = 0, \ 164 .imm = LEN }) 165 166 /* Short form of mov, dst_reg = src_reg */ 167 168 #define BPF_MOV64_REG(DST, SRC) \ 169 ((struct bpf_insn) { \ 170 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 171 .dst_reg = DST, \ 172 .src_reg = SRC, \ 173 .off = 0, \ 174 .imm = 0 }) 175 176 #define BPF_MOV32_REG(DST, SRC) \ 177 ((struct bpf_insn) { \ 178 .code = BPF_ALU | BPF_MOV | BPF_X, \ 179 .dst_reg = DST, \ 180 .src_reg = SRC, \ 181 .off = 0, \ 182 .imm = 0 }) 183 184 /* Special (internal-only) form of mov, used to resolve per-CPU addrs: 185 * dst_reg = src_reg + <percpu_base_off> 186 * BPF_ADDR_PERCPU is used as a special insn->off value. 187 */ 188 #define BPF_ADDR_PERCPU (-1) 189 190 #define BPF_MOV64_PERCPU_REG(DST, SRC) \ 191 ((struct bpf_insn) { \ 192 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 193 .dst_reg = DST, \ 194 .src_reg = SRC, \ 195 .off = BPF_ADDR_PERCPU, \ 196 .imm = 0 }) 197 198 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn) 199 { 200 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU; 201 } 202 203 /* Short form of mov, dst_reg = imm32 */ 204 205 #define BPF_MOV64_IMM(DST, IMM) \ 206 ((struct bpf_insn) { \ 207 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 208 .dst_reg = DST, \ 209 .src_reg = 0, \ 210 .off = 0, \ 211 .imm = IMM }) 212 213 #define BPF_MOV32_IMM(DST, IMM) \ 214 ((struct bpf_insn) { \ 215 .code = BPF_ALU | BPF_MOV | BPF_K, \ 216 .dst_reg = DST, \ 217 .src_reg = 0, \ 218 .off = 0, \ 219 .imm = IMM }) 220 221 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 222 223 #define BPF_MOVSX64_REG(DST, SRC, OFF) \ 224 ((struct bpf_insn) { \ 225 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 226 .dst_reg = DST, \ 227 .src_reg = SRC, \ 228 .off = OFF, \ 229 .imm = 0 }) 230 231 #define BPF_MOVSX32_REG(DST, SRC, OFF) \ 232 ((struct bpf_insn) { \ 233 .code = BPF_ALU | BPF_MOV | BPF_X, \ 234 .dst_reg = DST, \ 235 .src_reg = SRC, \ 236 .off = OFF, \ 237 .imm = 0 }) 238 239 /* Special form of mov32, used for doing explicit zero extension on dst. */ 240 #define BPF_ZEXT_REG(DST) \ 241 ((struct bpf_insn) { \ 242 .code = BPF_ALU | BPF_MOV | BPF_X, \ 243 .dst_reg = DST, \ 244 .src_reg = DST, \ 245 .off = 0, \ 246 .imm = 1 }) 247 248 static inline bool insn_is_zext(const struct bpf_insn *insn) 249 { 250 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 251 } 252 253 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers 254 * to pointers in user vma. 255 */ 256 static inline bool insn_is_cast_user(const struct bpf_insn *insn) 257 { 258 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && 259 insn->off == BPF_ADDR_SPACE_CAST && 260 insn->imm == 1U << 16; 261 } 262 263 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 264 #define BPF_LD_IMM64(DST, IMM) \ 265 BPF_LD_IMM64_RAW(DST, 0, IMM) 266 267 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 268 ((struct bpf_insn) { \ 269 .code = BPF_LD | BPF_DW | BPF_IMM, \ 270 .dst_reg = DST, \ 271 .src_reg = SRC, \ 272 .off = 0, \ 273 .imm = (__u32) (IMM) }), \ 274 ((struct bpf_insn) { \ 275 .code = 0, /* zero is reserved opcode */ \ 276 .dst_reg = 0, \ 277 .src_reg = 0, \ 278 .off = 0, \ 279 .imm = ((__u64) (IMM)) >> 32 }) 280 281 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 282 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 283 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 284 285 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 286 287 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 288 ((struct bpf_insn) { \ 289 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 290 .dst_reg = DST, \ 291 .src_reg = SRC, \ 292 .off = 0, \ 293 .imm = IMM }) 294 295 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 296 ((struct bpf_insn) { \ 297 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 298 .dst_reg = DST, \ 299 .src_reg = SRC, \ 300 .off = 0, \ 301 .imm = IMM }) 302 303 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 304 305 #define BPF_LD_ABS(SIZE, IMM) \ 306 ((struct bpf_insn) { \ 307 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 308 .dst_reg = 0, \ 309 .src_reg = 0, \ 310 .off = 0, \ 311 .imm = IMM }) 312 313 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 314 315 #define BPF_LD_IND(SIZE, SRC, IMM) \ 316 ((struct bpf_insn) { \ 317 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 318 .dst_reg = 0, \ 319 .src_reg = SRC, \ 320 .off = 0, \ 321 .imm = IMM }) 322 323 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 324 325 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 326 ((struct bpf_insn) { \ 327 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 328 .dst_reg = DST, \ 329 .src_reg = SRC, \ 330 .off = OFF, \ 331 .imm = 0 }) 332 333 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 334 335 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 336 ((struct bpf_insn) { \ 337 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 338 .dst_reg = DST, \ 339 .src_reg = SRC, \ 340 .off = OFF, \ 341 .imm = 0 }) 342 343 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 344 345 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 346 ((struct bpf_insn) { \ 347 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 348 .dst_reg = DST, \ 349 .src_reg = SRC, \ 350 .off = OFF, \ 351 .imm = 0 }) 352 353 354 /* 355 * Atomic operations: 356 * 357 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 358 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 359 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 360 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 361 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 362 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 363 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 364 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 365 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 366 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 367 */ 368 369 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 370 ((struct bpf_insn) { \ 371 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 372 .dst_reg = DST, \ 373 .src_reg = SRC, \ 374 .off = OFF, \ 375 .imm = OP }) 376 377 /* Legacy alias */ 378 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 379 380 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 381 382 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 383 ((struct bpf_insn) { \ 384 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 385 .dst_reg = DST, \ 386 .src_reg = 0, \ 387 .off = OFF, \ 388 .imm = IMM }) 389 390 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 391 392 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 393 ((struct bpf_insn) { \ 394 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 395 .dst_reg = DST, \ 396 .src_reg = SRC, \ 397 .off = OFF, \ 398 .imm = 0 }) 399 400 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 401 402 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 403 ((struct bpf_insn) { \ 404 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 405 .dst_reg = DST, \ 406 .src_reg = 0, \ 407 .off = OFF, \ 408 .imm = IMM }) 409 410 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 411 412 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 413 ((struct bpf_insn) { \ 414 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 415 .dst_reg = DST, \ 416 .src_reg = SRC, \ 417 .off = OFF, \ 418 .imm = 0 }) 419 420 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 421 422 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 423 ((struct bpf_insn) { \ 424 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 425 .dst_reg = DST, \ 426 .src_reg = 0, \ 427 .off = OFF, \ 428 .imm = IMM }) 429 430 /* Unconditional jumps, goto pc + off16 */ 431 432 #define BPF_JMP_A(OFF) \ 433 ((struct bpf_insn) { \ 434 .code = BPF_JMP | BPF_JA, \ 435 .dst_reg = 0, \ 436 .src_reg = 0, \ 437 .off = OFF, \ 438 .imm = 0 }) 439 440 /* Relative call */ 441 442 #define BPF_CALL_REL(TGT) \ 443 ((struct bpf_insn) { \ 444 .code = BPF_JMP | BPF_CALL, \ 445 .dst_reg = 0, \ 446 .src_reg = BPF_PSEUDO_CALL, \ 447 .off = 0, \ 448 .imm = TGT }) 449 450 /* Convert function address to BPF immediate */ 451 452 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 453 454 #define BPF_EMIT_CALL(FUNC) \ 455 ((struct bpf_insn) { \ 456 .code = BPF_JMP | BPF_CALL, \ 457 .dst_reg = 0, \ 458 .src_reg = 0, \ 459 .off = 0, \ 460 .imm = BPF_CALL_IMM(FUNC) }) 461 462 /* Raw code statement block */ 463 464 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 465 ((struct bpf_insn) { \ 466 .code = CODE, \ 467 .dst_reg = DST, \ 468 .src_reg = SRC, \ 469 .off = OFF, \ 470 .imm = IMM }) 471 472 /* Program exit */ 473 474 #define BPF_EXIT_INSN() \ 475 ((struct bpf_insn) { \ 476 .code = BPF_JMP | BPF_EXIT, \ 477 .dst_reg = 0, \ 478 .src_reg = 0, \ 479 .off = 0, \ 480 .imm = 0 }) 481 482 /* Speculation barrier */ 483 484 #define BPF_ST_NOSPEC() \ 485 ((struct bpf_insn) { \ 486 .code = BPF_ST | BPF_NOSPEC, \ 487 .dst_reg = 0, \ 488 .src_reg = 0, \ 489 .off = 0, \ 490 .imm = 0 }) 491 492 /* Internal classic blocks for direct assignment */ 493 494 #define __BPF_STMT(CODE, K) \ 495 ((struct sock_filter) BPF_STMT(CODE, K)) 496 497 #define __BPF_JUMP(CODE, K, JT, JF) \ 498 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 499 500 #define bytes_to_bpf_size(bytes) \ 501 ({ \ 502 int bpf_size = -EINVAL; \ 503 \ 504 if (bytes == sizeof(u8)) \ 505 bpf_size = BPF_B; \ 506 else if (bytes == sizeof(u16)) \ 507 bpf_size = BPF_H; \ 508 else if (bytes == sizeof(u32)) \ 509 bpf_size = BPF_W; \ 510 else if (bytes == sizeof(u64)) \ 511 bpf_size = BPF_DW; \ 512 \ 513 bpf_size; \ 514 }) 515 516 #define bpf_size_to_bytes(bpf_size) \ 517 ({ \ 518 int bytes = -EINVAL; \ 519 \ 520 if (bpf_size == BPF_B) \ 521 bytes = sizeof(u8); \ 522 else if (bpf_size == BPF_H) \ 523 bytes = sizeof(u16); \ 524 else if (bpf_size == BPF_W) \ 525 bytes = sizeof(u32); \ 526 else if (bpf_size == BPF_DW) \ 527 bytes = sizeof(u64); \ 528 \ 529 bytes; \ 530 }) 531 532 #define BPF_SIZEOF(type) \ 533 ({ \ 534 const int __size = bytes_to_bpf_size(sizeof(type)); \ 535 BUILD_BUG_ON(__size < 0); \ 536 __size; \ 537 }) 538 539 #define BPF_FIELD_SIZEOF(type, field) \ 540 ({ \ 541 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 542 BUILD_BUG_ON(__size < 0); \ 543 __size; \ 544 }) 545 546 #define BPF_LDST_BYTES(insn) \ 547 ({ \ 548 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 549 WARN_ON(__size < 0); \ 550 __size; \ 551 }) 552 553 #define __BPF_MAP_0(m, v, ...) v 554 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 555 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 556 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 557 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 558 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 559 560 #define __BPF_REG_0(...) __BPF_PAD(5) 561 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 562 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 563 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 564 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 565 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 566 567 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 568 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 569 570 #define __BPF_CAST(t, a) \ 571 (__force t) \ 572 (__force \ 573 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 574 (unsigned long)0, (t)0))) a 575 #define __BPF_V void 576 #define __BPF_N 577 578 #define __BPF_DECL_ARGS(t, a) t a 579 #define __BPF_DECL_REGS(t, a) u64 a 580 581 #define __BPF_PAD(n) \ 582 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 583 u64, __ur_3, u64, __ur_4, u64, __ur_5) 584 585 #define BPF_CALL_x(x, attr, name, ...) \ 586 static __always_inline \ 587 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 588 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 589 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 590 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 591 { \ 592 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 593 } \ 594 static __always_inline \ 595 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 596 597 #define __NOATTR 598 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 599 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 600 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 601 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 602 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 603 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 604 605 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 606 607 #define bpf_ctx_range(TYPE, MEMBER) \ 608 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 609 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 610 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 611 #if BITS_PER_LONG == 64 612 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 613 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 614 #else 615 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 616 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 617 #endif /* BITS_PER_LONG == 64 */ 618 619 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 620 ({ \ 621 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 622 *(PTR_SIZE) = (SIZE); \ 623 offsetof(TYPE, MEMBER); \ 624 }) 625 626 /* A struct sock_filter is architecture independent. */ 627 struct compat_sock_fprog { 628 u16 len; 629 compat_uptr_t filter; /* struct sock_filter * */ 630 }; 631 632 struct sock_fprog_kern { 633 u16 len; 634 struct sock_filter *filter; 635 }; 636 637 /* Some arches need doubleword alignment for their instructions and/or data */ 638 #define BPF_IMAGE_ALIGNMENT 8 639 640 struct bpf_binary_header { 641 u32 size; 642 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 643 }; 644 645 struct bpf_prog_stats { 646 u64_stats_t cnt; 647 u64_stats_t nsecs; 648 u64_stats_t misses; 649 struct u64_stats_sync syncp; 650 } __aligned(2 * sizeof(u64)); 651 652 struct sk_filter { 653 refcount_t refcnt; 654 struct rcu_head rcu; 655 struct bpf_prog *prog; 656 }; 657 658 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 659 660 extern struct mutex nf_conn_btf_access_lock; 661 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 662 const struct bpf_reg_state *reg, 663 int off, int size); 664 665 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 666 const struct bpf_insn *insnsi, 667 unsigned int (*bpf_func)(const void *, 668 const struct bpf_insn *)); 669 670 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 671 const void *ctx, 672 bpf_dispatcher_fn dfunc) 673 { 674 u32 ret; 675 676 cant_migrate(); 677 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 678 struct bpf_prog_stats *stats; 679 u64 duration, start = sched_clock(); 680 unsigned long flags; 681 682 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 683 684 duration = sched_clock() - start; 685 stats = this_cpu_ptr(prog->stats); 686 flags = u64_stats_update_begin_irqsave(&stats->syncp); 687 u64_stats_inc(&stats->cnt); 688 u64_stats_add(&stats->nsecs, duration); 689 u64_stats_update_end_irqrestore(&stats->syncp, flags); 690 } else { 691 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 692 } 693 return ret; 694 } 695 696 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 697 { 698 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 699 } 700 701 /* 702 * Use in preemptible and therefore migratable context to make sure that 703 * the execution of the BPF program runs on one CPU. 704 * 705 * This uses migrate_disable/enable() explicitly to document that the 706 * invocation of a BPF program does not require reentrancy protection 707 * against a BPF program which is invoked from a preempting task. 708 */ 709 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 710 const void *ctx) 711 { 712 u32 ret; 713 714 migrate_disable(); 715 ret = bpf_prog_run(prog, ctx); 716 migrate_enable(); 717 return ret; 718 } 719 720 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 721 722 struct bpf_skb_data_end { 723 struct qdisc_skb_cb qdisc_cb; 724 void *data_meta; 725 void *data_end; 726 }; 727 728 struct bpf_nh_params { 729 u32 nh_family; 730 union { 731 u32 ipv4_nh; 732 struct in6_addr ipv6_nh; 733 }; 734 }; 735 736 /* flags for bpf_redirect_info kern_flags */ 737 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 738 #define BPF_RI_F_RI_INIT BIT(1) 739 #define BPF_RI_F_CPU_MAP_INIT BIT(2) 740 #define BPF_RI_F_DEV_MAP_INIT BIT(3) 741 #define BPF_RI_F_XSK_MAP_INIT BIT(4) 742 743 struct bpf_redirect_info { 744 u64 tgt_index; 745 void *tgt_value; 746 struct bpf_map *map; 747 u32 flags; 748 u32 map_id; 749 enum bpf_map_type map_type; 750 struct bpf_nh_params nh; 751 u32 kern_flags; 752 }; 753 754 struct bpf_net_context { 755 struct bpf_redirect_info ri; 756 struct list_head cpu_map_flush_list; 757 struct list_head dev_map_flush_list; 758 struct list_head xskmap_map_flush_list; 759 }; 760 761 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx) 762 { 763 struct task_struct *tsk = current; 764 765 if (tsk->bpf_net_context != NULL) 766 return NULL; 767 bpf_net_ctx->ri.kern_flags = 0; 768 769 tsk->bpf_net_context = bpf_net_ctx; 770 return bpf_net_ctx; 771 } 772 773 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx) 774 { 775 if (bpf_net_ctx) 776 current->bpf_net_context = NULL; 777 } 778 779 static inline struct bpf_net_context *bpf_net_ctx_get(void) 780 { 781 return current->bpf_net_context; 782 } 783 784 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void) 785 { 786 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 787 788 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) { 789 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh)); 790 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT; 791 } 792 793 return &bpf_net_ctx->ri; 794 } 795 796 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void) 797 { 798 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 799 800 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) { 801 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list); 802 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT; 803 } 804 805 return &bpf_net_ctx->cpu_map_flush_list; 806 } 807 808 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void) 809 { 810 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 811 812 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) { 813 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list); 814 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT; 815 } 816 817 return &bpf_net_ctx->dev_map_flush_list; 818 } 819 820 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void) 821 { 822 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 823 824 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) { 825 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list); 826 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT; 827 } 828 829 return &bpf_net_ctx->xskmap_map_flush_list; 830 } 831 832 /* Compute the linear packet data range [data, data_end) which 833 * will be accessed by various program types (cls_bpf, act_bpf, 834 * lwt, ...). Subsystems allowing direct data access must (!) 835 * ensure that cb[] area can be written to when BPF program is 836 * invoked (otherwise cb[] save/restore is necessary). 837 */ 838 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 839 { 840 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 841 842 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 843 cb->data_meta = skb->data - skb_metadata_len(skb); 844 cb->data_end = skb->data + skb_headlen(skb); 845 } 846 847 /* Similar to bpf_compute_data_pointers(), except that save orginal 848 * data in cb->data and cb->meta_data for restore. 849 */ 850 static inline void bpf_compute_and_save_data_end( 851 struct sk_buff *skb, void **saved_data_end) 852 { 853 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 854 855 *saved_data_end = cb->data_end; 856 cb->data_end = skb->data + skb_headlen(skb); 857 } 858 859 /* Restore data saved by bpf_compute_and_save_data_end(). */ 860 static inline void bpf_restore_data_end( 861 struct sk_buff *skb, void *saved_data_end) 862 { 863 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 864 865 cb->data_end = saved_data_end; 866 } 867 868 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 869 { 870 /* eBPF programs may read/write skb->cb[] area to transfer meta 871 * data between tail calls. Since this also needs to work with 872 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 873 * 874 * In some socket filter cases, the cb unfortunately needs to be 875 * saved/restored so that protocol specific skb->cb[] data won't 876 * be lost. In any case, due to unpriviledged eBPF programs 877 * attached to sockets, we need to clear the bpf_skb_cb() area 878 * to not leak previous contents to user space. 879 */ 880 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 881 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 882 sizeof_field(struct qdisc_skb_cb, data)); 883 884 return qdisc_skb_cb(skb)->data; 885 } 886 887 /* Must be invoked with migration disabled */ 888 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 889 const void *ctx) 890 { 891 const struct sk_buff *skb = ctx; 892 u8 *cb_data = bpf_skb_cb(skb); 893 u8 cb_saved[BPF_SKB_CB_LEN]; 894 u32 res; 895 896 if (unlikely(prog->cb_access)) { 897 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 898 memset(cb_data, 0, sizeof(cb_saved)); 899 } 900 901 res = bpf_prog_run(prog, skb); 902 903 if (unlikely(prog->cb_access)) 904 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 905 906 return res; 907 } 908 909 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 910 struct sk_buff *skb) 911 { 912 u32 res; 913 914 migrate_disable(); 915 res = __bpf_prog_run_save_cb(prog, skb); 916 migrate_enable(); 917 return res; 918 } 919 920 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 921 struct sk_buff *skb) 922 { 923 u8 *cb_data = bpf_skb_cb(skb); 924 u32 res; 925 926 if (unlikely(prog->cb_access)) 927 memset(cb_data, 0, BPF_SKB_CB_LEN); 928 929 res = bpf_prog_run_pin_on_cpu(prog, skb); 930 return res; 931 } 932 933 DECLARE_BPF_DISPATCHER(xdp) 934 935 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 936 937 u32 xdp_master_redirect(struct xdp_buff *xdp); 938 939 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 940 941 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 942 { 943 return prog->len * sizeof(struct bpf_insn); 944 } 945 946 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 947 { 948 return round_up(bpf_prog_insn_size(prog) + 949 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 950 } 951 952 static inline unsigned int bpf_prog_size(unsigned int proglen) 953 { 954 return max(sizeof(struct bpf_prog), 955 offsetof(struct bpf_prog, insns[proglen])); 956 } 957 958 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 959 { 960 /* When classic BPF programs have been loaded and the arch 961 * does not have a classic BPF JIT (anymore), they have been 962 * converted via bpf_migrate_filter() to eBPF and thus always 963 * have an unspec program type. 964 */ 965 return prog->type == BPF_PROG_TYPE_UNSPEC; 966 } 967 968 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 969 { 970 const u32 size_machine = sizeof(unsigned long); 971 972 if (size > size_machine && size % size_machine == 0) 973 size = size_machine; 974 975 return size; 976 } 977 978 static inline bool 979 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 980 { 981 return size <= size_default && (size & (size - 1)) == 0; 982 } 983 984 static inline u8 985 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 986 { 987 u8 access_off = off & (size_default - 1); 988 989 #ifdef __LITTLE_ENDIAN 990 return access_off; 991 #else 992 return size_default - (access_off + size); 993 #endif 994 } 995 996 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 997 (size == sizeof(__u64) && \ 998 off >= offsetof(type, field) && \ 999 off + sizeof(__u64) <= offsetofend(type, field) && \ 1000 off % sizeof(__u64) == 0) 1001 1002 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 1003 1004 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) 1005 { 1006 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 1007 if (!fp->jited) { 1008 set_vm_flush_reset_perms(fp); 1009 return set_memory_ro((unsigned long)fp, fp->pages); 1010 } 1011 #endif 1012 return 0; 1013 } 1014 1015 static inline int __must_check 1016 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 1017 { 1018 set_vm_flush_reset_perms(hdr); 1019 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 1020 } 1021 1022 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 1023 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 1024 { 1025 return sk_filter_trim_cap(sk, skb, 1); 1026 } 1027 1028 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 1029 void bpf_prog_free(struct bpf_prog *fp); 1030 1031 bool bpf_opcode_in_insntable(u8 code); 1032 1033 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 1034 const u32 *insn_to_jit_off); 1035 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 1036 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 1037 1038 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 1039 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 1040 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 1041 gfp_t gfp_extra_flags); 1042 void __bpf_prog_free(struct bpf_prog *fp); 1043 1044 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 1045 { 1046 __bpf_prog_free(fp); 1047 } 1048 1049 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 1050 unsigned int flen); 1051 1052 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 1053 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1054 bpf_aux_classic_check_t trans, bool save_orig); 1055 void bpf_prog_destroy(struct bpf_prog *fp); 1056 1057 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1058 int sk_attach_bpf(u32 ufd, struct sock *sk); 1059 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1060 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 1061 void sk_reuseport_prog_free(struct bpf_prog *prog); 1062 int sk_detach_filter(struct sock *sk); 1063 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 1064 1065 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 1066 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 1067 1068 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1069 #define __bpf_call_base_args \ 1070 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 1071 (void *)__bpf_call_base) 1072 1073 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 1074 void bpf_jit_compile(struct bpf_prog *prog); 1075 bool bpf_jit_needs_zext(void); 1076 bool bpf_jit_inlines_helper_call(s32 imm); 1077 bool bpf_jit_supports_subprog_tailcalls(void); 1078 bool bpf_jit_supports_percpu_insn(void); 1079 bool bpf_jit_supports_kfunc_call(void); 1080 bool bpf_jit_supports_far_kfunc_call(void); 1081 bool bpf_jit_supports_exceptions(void); 1082 bool bpf_jit_supports_ptr_xchg(void); 1083 bool bpf_jit_supports_arena(void); 1084 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena); 1085 u64 bpf_arch_uaddress_limit(void); 1086 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 1087 bool bpf_helper_changes_pkt_data(void *func); 1088 1089 static inline bool bpf_dump_raw_ok(const struct cred *cred) 1090 { 1091 /* Reconstruction of call-sites is dependent on kallsyms, 1092 * thus make dump the same restriction. 1093 */ 1094 return kallsyms_show_value(cred); 1095 } 1096 1097 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 1098 const struct bpf_insn *patch, u32 len); 1099 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 1100 1101 static inline bool xdp_return_frame_no_direct(void) 1102 { 1103 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1104 1105 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 1106 } 1107 1108 static inline void xdp_set_return_frame_no_direct(void) 1109 { 1110 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1111 1112 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 1113 } 1114 1115 static inline void xdp_clear_return_frame_no_direct(void) 1116 { 1117 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1118 1119 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1120 } 1121 1122 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1123 unsigned int pktlen) 1124 { 1125 unsigned int len; 1126 1127 if (unlikely(!(fwd->flags & IFF_UP))) 1128 return -ENETDOWN; 1129 1130 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1131 if (pktlen > len) 1132 return -EMSGSIZE; 1133 1134 return 0; 1135 } 1136 1137 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1138 * same cpu context. Further for best results no more than a single map 1139 * for the do_redirect/do_flush pair should be used. This limitation is 1140 * because we only track one map and force a flush when the map changes. 1141 * This does not appear to be a real limitation for existing software. 1142 */ 1143 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1144 struct xdp_buff *xdp, struct bpf_prog *prog); 1145 int xdp_do_redirect(struct net_device *dev, 1146 struct xdp_buff *xdp, 1147 struct bpf_prog *prog); 1148 int xdp_do_redirect_frame(struct net_device *dev, 1149 struct xdp_buff *xdp, 1150 struct xdp_frame *xdpf, 1151 struct bpf_prog *prog); 1152 void xdp_do_flush(void); 1153 1154 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1155 1156 #ifdef CONFIG_INET 1157 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1158 struct bpf_prog *prog, struct sk_buff *skb, 1159 struct sock *migrating_sk, 1160 u32 hash); 1161 #else 1162 static inline struct sock * 1163 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1164 struct bpf_prog *prog, struct sk_buff *skb, 1165 struct sock *migrating_sk, 1166 u32 hash) 1167 { 1168 return NULL; 1169 } 1170 #endif 1171 1172 #ifdef CONFIG_BPF_JIT 1173 extern int bpf_jit_enable; 1174 extern int bpf_jit_harden; 1175 extern int bpf_jit_kallsyms; 1176 extern long bpf_jit_limit; 1177 extern long bpf_jit_limit_max; 1178 1179 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1180 1181 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1182 1183 struct bpf_binary_header * 1184 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1185 unsigned int alignment, 1186 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1187 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1188 u64 bpf_jit_alloc_exec_limit(void); 1189 void *bpf_jit_alloc_exec(unsigned long size); 1190 void bpf_jit_free_exec(void *addr); 1191 void bpf_jit_free(struct bpf_prog *fp); 1192 struct bpf_binary_header * 1193 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1194 1195 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1196 void bpf_prog_pack_free(void *ptr, u32 size); 1197 1198 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1199 { 1200 return list_empty(&fp->aux->ksym.lnode) || 1201 fp->aux->ksym.lnode.prev == LIST_POISON2; 1202 } 1203 1204 struct bpf_binary_header * 1205 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1206 unsigned int alignment, 1207 struct bpf_binary_header **rw_hdr, 1208 u8 **rw_image, 1209 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1210 int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, 1211 struct bpf_binary_header *ro_header, 1212 struct bpf_binary_header *rw_header); 1213 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1214 struct bpf_binary_header *rw_header); 1215 1216 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1217 struct bpf_jit_poke_descriptor *poke); 1218 1219 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1220 const struct bpf_insn *insn, bool extra_pass, 1221 u64 *func_addr, bool *func_addr_fixed); 1222 1223 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1224 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1225 1226 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1227 u32 pass, void *image) 1228 { 1229 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1230 proglen, pass, image, current->comm, task_pid_nr(current)); 1231 1232 if (image) 1233 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1234 16, 1, image, proglen, false); 1235 } 1236 1237 static inline bool bpf_jit_is_ebpf(void) 1238 { 1239 # ifdef CONFIG_HAVE_EBPF_JIT 1240 return true; 1241 # else 1242 return false; 1243 # endif 1244 } 1245 1246 static inline bool ebpf_jit_enabled(void) 1247 { 1248 return bpf_jit_enable && bpf_jit_is_ebpf(); 1249 } 1250 1251 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1252 { 1253 return fp->jited && bpf_jit_is_ebpf(); 1254 } 1255 1256 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1257 { 1258 /* These are the prerequisites, should someone ever have the 1259 * idea to call blinding outside of them, we make sure to 1260 * bail out. 1261 */ 1262 if (!bpf_jit_is_ebpf()) 1263 return false; 1264 if (!prog->jit_requested) 1265 return false; 1266 if (!bpf_jit_harden) 1267 return false; 1268 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1269 return false; 1270 1271 return true; 1272 } 1273 1274 static inline bool bpf_jit_kallsyms_enabled(void) 1275 { 1276 /* There are a couple of corner cases where kallsyms should 1277 * not be enabled f.e. on hardening. 1278 */ 1279 if (bpf_jit_harden) 1280 return false; 1281 if (!bpf_jit_kallsyms) 1282 return false; 1283 if (bpf_jit_kallsyms == 1) 1284 return true; 1285 1286 return false; 1287 } 1288 1289 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size, 1290 unsigned long *off, char *sym); 1291 bool is_bpf_text_address(unsigned long addr); 1292 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1293 char *sym); 1294 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1295 1296 static inline const char * 1297 bpf_address_lookup(unsigned long addr, unsigned long *size, 1298 unsigned long *off, char **modname, char *sym) 1299 { 1300 const char *ret = __bpf_address_lookup(addr, size, off, sym); 1301 1302 if (ret && modname) 1303 *modname = NULL; 1304 return ret; 1305 } 1306 1307 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1308 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1309 1310 #else /* CONFIG_BPF_JIT */ 1311 1312 static inline bool ebpf_jit_enabled(void) 1313 { 1314 return false; 1315 } 1316 1317 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1318 { 1319 return false; 1320 } 1321 1322 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1323 { 1324 return false; 1325 } 1326 1327 static inline int 1328 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1329 struct bpf_jit_poke_descriptor *poke) 1330 { 1331 return -ENOTSUPP; 1332 } 1333 1334 static inline void bpf_jit_free(struct bpf_prog *fp) 1335 { 1336 bpf_prog_unlock_free(fp); 1337 } 1338 1339 static inline bool bpf_jit_kallsyms_enabled(void) 1340 { 1341 return false; 1342 } 1343 1344 static inline const char * 1345 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1346 unsigned long *off, char *sym) 1347 { 1348 return NULL; 1349 } 1350 1351 static inline bool is_bpf_text_address(unsigned long addr) 1352 { 1353 return false; 1354 } 1355 1356 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1357 char *type, char *sym) 1358 { 1359 return -ERANGE; 1360 } 1361 1362 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1363 { 1364 return NULL; 1365 } 1366 1367 static inline const char * 1368 bpf_address_lookup(unsigned long addr, unsigned long *size, 1369 unsigned long *off, char **modname, char *sym) 1370 { 1371 return NULL; 1372 } 1373 1374 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1375 { 1376 } 1377 1378 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1379 { 1380 } 1381 1382 #endif /* CONFIG_BPF_JIT */ 1383 1384 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1385 1386 #define BPF_ANC BIT(15) 1387 1388 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1389 { 1390 switch (first->code) { 1391 case BPF_RET | BPF_K: 1392 case BPF_LD | BPF_W | BPF_LEN: 1393 return false; 1394 1395 case BPF_LD | BPF_W | BPF_ABS: 1396 case BPF_LD | BPF_H | BPF_ABS: 1397 case BPF_LD | BPF_B | BPF_ABS: 1398 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1399 return true; 1400 return false; 1401 1402 default: 1403 return true; 1404 } 1405 } 1406 1407 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1408 { 1409 BUG_ON(ftest->code & BPF_ANC); 1410 1411 switch (ftest->code) { 1412 case BPF_LD | BPF_W | BPF_ABS: 1413 case BPF_LD | BPF_H | BPF_ABS: 1414 case BPF_LD | BPF_B | BPF_ABS: 1415 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1416 return BPF_ANC | SKF_AD_##CODE 1417 switch (ftest->k) { 1418 BPF_ANCILLARY(PROTOCOL); 1419 BPF_ANCILLARY(PKTTYPE); 1420 BPF_ANCILLARY(IFINDEX); 1421 BPF_ANCILLARY(NLATTR); 1422 BPF_ANCILLARY(NLATTR_NEST); 1423 BPF_ANCILLARY(MARK); 1424 BPF_ANCILLARY(QUEUE); 1425 BPF_ANCILLARY(HATYPE); 1426 BPF_ANCILLARY(RXHASH); 1427 BPF_ANCILLARY(CPU); 1428 BPF_ANCILLARY(ALU_XOR_X); 1429 BPF_ANCILLARY(VLAN_TAG); 1430 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1431 BPF_ANCILLARY(PAY_OFFSET); 1432 BPF_ANCILLARY(RANDOM); 1433 BPF_ANCILLARY(VLAN_TPID); 1434 } 1435 fallthrough; 1436 default: 1437 return ftest->code; 1438 } 1439 } 1440 1441 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1442 int k, unsigned int size); 1443 1444 static inline int bpf_tell_extensions(void) 1445 { 1446 return SKF_AD_MAX; 1447 } 1448 1449 struct bpf_sock_addr_kern { 1450 struct sock *sk; 1451 struct sockaddr *uaddr; 1452 /* Temporary "register" to make indirect stores to nested structures 1453 * defined above. We need three registers to make such a store, but 1454 * only two (src and dst) are available at convert_ctx_access time 1455 */ 1456 u64 tmp_reg; 1457 void *t_ctx; /* Attach type specific context. */ 1458 u32 uaddrlen; 1459 }; 1460 1461 struct bpf_sock_ops_kern { 1462 struct sock *sk; 1463 union { 1464 u32 args[4]; 1465 u32 reply; 1466 u32 replylong[4]; 1467 }; 1468 struct sk_buff *syn_skb; 1469 struct sk_buff *skb; 1470 void *skb_data_end; 1471 u8 op; 1472 u8 is_fullsock; 1473 u8 remaining_opt_len; 1474 u64 temp; /* temp and everything after is not 1475 * initialized to 0 before calling 1476 * the BPF program. New fields that 1477 * should be initialized to 0 should 1478 * be inserted before temp. 1479 * temp is scratch storage used by 1480 * sock_ops_convert_ctx_access 1481 * as temporary storage of a register. 1482 */ 1483 }; 1484 1485 struct bpf_sysctl_kern { 1486 struct ctl_table_header *head; 1487 const struct ctl_table *table; 1488 void *cur_val; 1489 size_t cur_len; 1490 void *new_val; 1491 size_t new_len; 1492 int new_updated; 1493 int write; 1494 loff_t *ppos; 1495 /* Temporary "register" for indirect stores to ppos. */ 1496 u64 tmp_reg; 1497 }; 1498 1499 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1500 struct bpf_sockopt_buf { 1501 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1502 }; 1503 1504 struct bpf_sockopt_kern { 1505 struct sock *sk; 1506 u8 *optval; 1507 u8 *optval_end; 1508 s32 level; 1509 s32 optname; 1510 s32 optlen; 1511 /* for retval in struct bpf_cg_run_ctx */ 1512 struct task_struct *current_task; 1513 /* Temporary "register" for indirect stores to ppos. */ 1514 u64 tmp_reg; 1515 }; 1516 1517 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1518 1519 struct bpf_sk_lookup_kern { 1520 u16 family; 1521 u16 protocol; 1522 __be16 sport; 1523 u16 dport; 1524 struct { 1525 __be32 saddr; 1526 __be32 daddr; 1527 } v4; 1528 struct { 1529 const struct in6_addr *saddr; 1530 const struct in6_addr *daddr; 1531 } v6; 1532 struct sock *selected_sk; 1533 u32 ingress_ifindex; 1534 bool no_reuseport; 1535 }; 1536 1537 extern struct static_key_false bpf_sk_lookup_enabled; 1538 1539 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1540 * 1541 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1542 * SK_DROP. Their meaning is as follows: 1543 * 1544 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1545 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1546 * SK_DROP : terminate lookup with -ECONNREFUSED 1547 * 1548 * This macro aggregates return values and selected sockets from 1549 * multiple BPF programs according to following rules in order: 1550 * 1551 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1552 * macro result is SK_PASS and last ctx.selected_sk is used. 1553 * 2. If any program returned SK_DROP return value, 1554 * macro result is SK_DROP. 1555 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1556 * 1557 * Caller must ensure that the prog array is non-NULL, and that the 1558 * array as well as the programs it contains remain valid. 1559 */ 1560 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1561 ({ \ 1562 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1563 struct bpf_prog_array_item *_item; \ 1564 struct sock *_selected_sk = NULL; \ 1565 bool _no_reuseport = false; \ 1566 struct bpf_prog *_prog; \ 1567 bool _all_pass = true; \ 1568 u32 _ret; \ 1569 \ 1570 migrate_disable(); \ 1571 _item = &(array)->items[0]; \ 1572 while ((_prog = READ_ONCE(_item->prog))) { \ 1573 /* restore most recent selection */ \ 1574 _ctx->selected_sk = _selected_sk; \ 1575 _ctx->no_reuseport = _no_reuseport; \ 1576 \ 1577 _ret = func(_prog, _ctx); \ 1578 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1579 /* remember last non-NULL socket */ \ 1580 _selected_sk = _ctx->selected_sk; \ 1581 _no_reuseport = _ctx->no_reuseport; \ 1582 } else if (_ret == SK_DROP && _all_pass) { \ 1583 _all_pass = false; \ 1584 } \ 1585 _item++; \ 1586 } \ 1587 _ctx->selected_sk = _selected_sk; \ 1588 _ctx->no_reuseport = _no_reuseport; \ 1589 migrate_enable(); \ 1590 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1591 }) 1592 1593 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol, 1594 const __be32 saddr, const __be16 sport, 1595 const __be32 daddr, const u16 dport, 1596 const int ifindex, struct sock **psk) 1597 { 1598 struct bpf_prog_array *run_array; 1599 struct sock *selected_sk = NULL; 1600 bool no_reuseport = false; 1601 1602 rcu_read_lock(); 1603 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1604 if (run_array) { 1605 struct bpf_sk_lookup_kern ctx = { 1606 .family = AF_INET, 1607 .protocol = protocol, 1608 .v4.saddr = saddr, 1609 .v4.daddr = daddr, 1610 .sport = sport, 1611 .dport = dport, 1612 .ingress_ifindex = ifindex, 1613 }; 1614 u32 act; 1615 1616 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1617 if (act == SK_PASS) { 1618 selected_sk = ctx.selected_sk; 1619 no_reuseport = ctx.no_reuseport; 1620 } else { 1621 selected_sk = ERR_PTR(-ECONNREFUSED); 1622 } 1623 } 1624 rcu_read_unlock(); 1625 *psk = selected_sk; 1626 return no_reuseport; 1627 } 1628 1629 #if IS_ENABLED(CONFIG_IPV6) 1630 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol, 1631 const struct in6_addr *saddr, 1632 const __be16 sport, 1633 const struct in6_addr *daddr, 1634 const u16 dport, 1635 const int ifindex, struct sock **psk) 1636 { 1637 struct bpf_prog_array *run_array; 1638 struct sock *selected_sk = NULL; 1639 bool no_reuseport = false; 1640 1641 rcu_read_lock(); 1642 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1643 if (run_array) { 1644 struct bpf_sk_lookup_kern ctx = { 1645 .family = AF_INET6, 1646 .protocol = protocol, 1647 .v6.saddr = saddr, 1648 .v6.daddr = daddr, 1649 .sport = sport, 1650 .dport = dport, 1651 .ingress_ifindex = ifindex, 1652 }; 1653 u32 act; 1654 1655 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1656 if (act == SK_PASS) { 1657 selected_sk = ctx.selected_sk; 1658 no_reuseport = ctx.no_reuseport; 1659 } else { 1660 selected_sk = ERR_PTR(-ECONNREFUSED); 1661 } 1662 } 1663 rcu_read_unlock(); 1664 *psk = selected_sk; 1665 return no_reuseport; 1666 } 1667 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1668 1669 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1670 u64 flags, const u64 flag_mask, 1671 void *lookup_elem(struct bpf_map *map, u32 key)) 1672 { 1673 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1674 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1675 1676 /* Lower bits of the flags are used as return code on lookup failure */ 1677 if (unlikely(flags & ~(action_mask | flag_mask))) 1678 return XDP_ABORTED; 1679 1680 ri->tgt_value = lookup_elem(map, index); 1681 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1682 /* If the lookup fails we want to clear out the state in the 1683 * redirect_info struct completely, so that if an eBPF program 1684 * performs multiple lookups, the last one always takes 1685 * precedence. 1686 */ 1687 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1688 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1689 return flags & action_mask; 1690 } 1691 1692 ri->tgt_index = index; 1693 ri->map_id = map->id; 1694 ri->map_type = map->map_type; 1695 1696 if (flags & BPF_F_BROADCAST) { 1697 WRITE_ONCE(ri->map, map); 1698 ri->flags = flags; 1699 } else { 1700 WRITE_ONCE(ri->map, NULL); 1701 ri->flags = 0; 1702 } 1703 1704 return XDP_REDIRECT; 1705 } 1706 1707 #ifdef CONFIG_NET 1708 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1709 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1710 u32 len, u64 flags); 1711 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1712 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1713 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1714 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1715 void *buf, unsigned long len, bool flush); 1716 #else /* CONFIG_NET */ 1717 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1718 void *to, u32 len) 1719 { 1720 return -EOPNOTSUPP; 1721 } 1722 1723 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1724 const void *from, u32 len, u64 flags) 1725 { 1726 return -EOPNOTSUPP; 1727 } 1728 1729 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1730 void *buf, u32 len) 1731 { 1732 return -EOPNOTSUPP; 1733 } 1734 1735 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1736 void *buf, u32 len) 1737 { 1738 return -EOPNOTSUPP; 1739 } 1740 1741 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1742 { 1743 return NULL; 1744 } 1745 1746 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1747 unsigned long len, bool flush) 1748 { 1749 } 1750 #endif /* CONFIG_NET */ 1751 1752 #endif /* __LINUX_FILTER_H__ */ 1753