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 insn_is_mov_percpu_addr(const struct bpf_insn * insn)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 insn_is_zext(const struct bpf_insn * insn)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 */ insn_is_cast_user(const struct bpf_insn * insn)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 /* Unconditional jumps, gotol pc + imm32 */ 441 442 #define BPF_JMP32_A(IMM) \ 443 ((struct bpf_insn) { \ 444 .code = BPF_JMP32 | BPF_JA, \ 445 .dst_reg = 0, \ 446 .src_reg = 0, \ 447 .off = 0, \ 448 .imm = IMM }) 449 450 /* Relative call */ 451 452 #define BPF_CALL_REL(TGT) \ 453 ((struct bpf_insn) { \ 454 .code = BPF_JMP | BPF_CALL, \ 455 .dst_reg = 0, \ 456 .src_reg = BPF_PSEUDO_CALL, \ 457 .off = 0, \ 458 .imm = TGT }) 459 460 /* Convert function address to BPF immediate */ 461 462 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 463 464 #define BPF_EMIT_CALL(FUNC) \ 465 ((struct bpf_insn) { \ 466 .code = BPF_JMP | BPF_CALL, \ 467 .dst_reg = 0, \ 468 .src_reg = 0, \ 469 .off = 0, \ 470 .imm = BPF_CALL_IMM(FUNC) }) 471 472 /* Raw code statement block */ 473 474 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 475 ((struct bpf_insn) { \ 476 .code = CODE, \ 477 .dst_reg = DST, \ 478 .src_reg = SRC, \ 479 .off = OFF, \ 480 .imm = IMM }) 481 482 /* Program exit */ 483 484 #define BPF_EXIT_INSN() \ 485 ((struct bpf_insn) { \ 486 .code = BPF_JMP | BPF_EXIT, \ 487 .dst_reg = 0, \ 488 .src_reg = 0, \ 489 .off = 0, \ 490 .imm = 0 }) 491 492 /* Speculation barrier */ 493 494 #define BPF_ST_NOSPEC() \ 495 ((struct bpf_insn) { \ 496 .code = BPF_ST | BPF_NOSPEC, \ 497 .dst_reg = 0, \ 498 .src_reg = 0, \ 499 .off = 0, \ 500 .imm = 0 }) 501 502 /* Internal classic blocks for direct assignment */ 503 504 #define __BPF_STMT(CODE, K) \ 505 ((struct sock_filter) BPF_STMT(CODE, K)) 506 507 #define __BPF_JUMP(CODE, K, JT, JF) \ 508 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 509 510 #define bytes_to_bpf_size(bytes) \ 511 ({ \ 512 int bpf_size = -EINVAL; \ 513 \ 514 if (bytes == sizeof(u8)) \ 515 bpf_size = BPF_B; \ 516 else if (bytes == sizeof(u16)) \ 517 bpf_size = BPF_H; \ 518 else if (bytes == sizeof(u32)) \ 519 bpf_size = BPF_W; \ 520 else if (bytes == sizeof(u64)) \ 521 bpf_size = BPF_DW; \ 522 \ 523 bpf_size; \ 524 }) 525 526 #define bpf_size_to_bytes(bpf_size) \ 527 ({ \ 528 int bytes = -EINVAL; \ 529 \ 530 if (bpf_size == BPF_B) \ 531 bytes = sizeof(u8); \ 532 else if (bpf_size == BPF_H) \ 533 bytes = sizeof(u16); \ 534 else if (bpf_size == BPF_W) \ 535 bytes = sizeof(u32); \ 536 else if (bpf_size == BPF_DW) \ 537 bytes = sizeof(u64); \ 538 \ 539 bytes; \ 540 }) 541 542 #define BPF_SIZEOF(type) \ 543 ({ \ 544 const int __size = bytes_to_bpf_size(sizeof(type)); \ 545 BUILD_BUG_ON(__size < 0); \ 546 __size; \ 547 }) 548 549 #define BPF_FIELD_SIZEOF(type, field) \ 550 ({ \ 551 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 552 BUILD_BUG_ON(__size < 0); \ 553 __size; \ 554 }) 555 556 #define BPF_LDST_BYTES(insn) \ 557 ({ \ 558 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 559 WARN_ON(__size < 0); \ 560 __size; \ 561 }) 562 563 #define __BPF_MAP_0(m, v, ...) v 564 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 565 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 566 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 567 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 568 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 569 570 #define __BPF_REG_0(...) __BPF_PAD(5) 571 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 572 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 573 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 574 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 575 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 576 577 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 578 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 579 580 #define __BPF_CAST(t, a) \ 581 (__force t) \ 582 (__force \ 583 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 584 (unsigned long)0, (t)0))) a 585 #define __BPF_V void 586 #define __BPF_N 587 588 #define __BPF_DECL_ARGS(t, a) t a 589 #define __BPF_DECL_REGS(t, a) u64 a 590 591 #define __BPF_PAD(n) \ 592 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 593 u64, __ur_3, u64, __ur_4, u64, __ur_5) 594 595 #define BPF_CALL_x(x, attr, name, ...) \ 596 static __always_inline \ 597 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 598 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 599 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 600 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 601 { \ 602 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 603 } \ 604 static __always_inline \ 605 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 606 607 #define __NOATTR 608 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 609 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 610 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 611 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 612 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 613 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 614 615 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 616 617 #define bpf_ctx_range(TYPE, MEMBER) \ 618 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 619 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 620 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 621 #if BITS_PER_LONG == 64 622 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 623 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 624 #else 625 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 626 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 627 #endif /* BITS_PER_LONG == 64 */ 628 629 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 630 ({ \ 631 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 632 *(PTR_SIZE) = (SIZE); \ 633 offsetof(TYPE, MEMBER); \ 634 }) 635 636 /* A struct sock_filter is architecture independent. */ 637 struct compat_sock_fprog { 638 u16 len; 639 compat_uptr_t filter; /* struct sock_filter * */ 640 }; 641 642 struct sock_fprog_kern { 643 u16 len; 644 struct sock_filter *filter; 645 }; 646 647 /* Some arches need doubleword alignment for their instructions and/or data */ 648 #define BPF_IMAGE_ALIGNMENT 8 649 650 struct bpf_binary_header { 651 u32 size; 652 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 653 }; 654 655 struct bpf_prog_stats { 656 u64_stats_t cnt; 657 u64_stats_t nsecs; 658 u64_stats_t misses; 659 struct u64_stats_sync syncp; 660 } __aligned(2 * sizeof(u64)); 661 662 struct sk_filter { 663 refcount_t refcnt; 664 struct rcu_head rcu; 665 struct bpf_prog *prog; 666 }; 667 668 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 669 670 extern struct mutex nf_conn_btf_access_lock; 671 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 672 const struct bpf_reg_state *reg, 673 int off, int size); 674 675 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 676 const struct bpf_insn *insnsi, 677 unsigned int (*bpf_func)(const void *, 678 const struct bpf_insn *)); 679 __bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)680 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 681 const void *ctx, 682 bpf_dispatcher_fn dfunc) 683 { 684 u32 ret; 685 686 cant_migrate(); 687 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 688 struct bpf_prog_stats *stats; 689 u64 duration, start = sched_clock(); 690 unsigned long flags; 691 692 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 693 694 duration = sched_clock() - start; 695 stats = this_cpu_ptr(prog->stats); 696 flags = u64_stats_update_begin_irqsave(&stats->syncp); 697 u64_stats_inc(&stats->cnt); 698 u64_stats_add(&stats->nsecs, duration); 699 u64_stats_update_end_irqrestore(&stats->syncp, flags); 700 } else { 701 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 702 } 703 return ret; 704 } 705 bpf_prog_run(const struct bpf_prog * prog,const void * ctx)706 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 707 { 708 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 709 } 710 711 /* 712 * Use in preemptible and therefore migratable context to make sure that 713 * the execution of the BPF program runs on one CPU. 714 * 715 * This uses migrate_disable/enable() explicitly to document that the 716 * invocation of a BPF program does not require reentrancy protection 717 * against a BPF program which is invoked from a preempting task. 718 */ bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)719 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 720 const void *ctx) 721 { 722 u32 ret; 723 724 migrate_disable(); 725 ret = bpf_prog_run(prog, ctx); 726 migrate_enable(); 727 return ret; 728 } 729 730 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 731 732 struct bpf_skb_data_end { 733 struct qdisc_skb_cb qdisc_cb; 734 void *data_meta; 735 void *data_end; 736 }; 737 738 struct bpf_nh_params { 739 u32 nh_family; 740 union { 741 u32 ipv4_nh; 742 struct in6_addr ipv6_nh; 743 }; 744 }; 745 746 /* flags for bpf_redirect_info kern_flags */ 747 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 748 #define BPF_RI_F_RI_INIT BIT(1) 749 #define BPF_RI_F_CPU_MAP_INIT BIT(2) 750 #define BPF_RI_F_DEV_MAP_INIT BIT(3) 751 #define BPF_RI_F_XSK_MAP_INIT BIT(4) 752 753 struct bpf_redirect_info { 754 u64 tgt_index; 755 void *tgt_value; 756 struct bpf_map *map; 757 u32 flags; 758 u32 map_id; 759 enum bpf_map_type map_type; 760 struct bpf_nh_params nh; 761 u32 kern_flags; 762 }; 763 764 struct bpf_net_context { 765 struct bpf_redirect_info ri; 766 struct list_head cpu_map_flush_list; 767 struct list_head dev_map_flush_list; 768 struct list_head xskmap_map_flush_list; 769 }; 770 bpf_net_ctx_set(struct bpf_net_context * bpf_net_ctx)771 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx) 772 { 773 struct task_struct *tsk = current; 774 775 if (tsk->bpf_net_context != NULL) 776 return NULL; 777 bpf_net_ctx->ri.kern_flags = 0; 778 779 tsk->bpf_net_context = bpf_net_ctx; 780 return bpf_net_ctx; 781 } 782 bpf_net_ctx_clear(struct bpf_net_context * bpf_net_ctx)783 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx) 784 { 785 if (bpf_net_ctx) 786 current->bpf_net_context = NULL; 787 } 788 bpf_net_ctx_get(void)789 static inline struct bpf_net_context *bpf_net_ctx_get(void) 790 { 791 return current->bpf_net_context; 792 } 793 bpf_net_ctx_get_ri(void)794 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void) 795 { 796 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 797 798 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) { 799 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh)); 800 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT; 801 } 802 803 return &bpf_net_ctx->ri; 804 } 805 bpf_net_ctx_get_cpu_map_flush_list(void)806 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void) 807 { 808 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 809 810 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) { 811 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list); 812 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT; 813 } 814 815 return &bpf_net_ctx->cpu_map_flush_list; 816 } 817 bpf_net_ctx_get_dev_flush_list(void)818 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void) 819 { 820 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 821 822 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) { 823 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list); 824 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT; 825 } 826 827 return &bpf_net_ctx->dev_map_flush_list; 828 } 829 bpf_net_ctx_get_xskmap_flush_list(void)830 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void) 831 { 832 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 833 834 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) { 835 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list); 836 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT; 837 } 838 839 return &bpf_net_ctx->xskmap_map_flush_list; 840 } 841 bpf_net_ctx_get_all_used_flush_lists(struct list_head ** lh_map,struct list_head ** lh_dev,struct list_head ** lh_xsk)842 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map, 843 struct list_head **lh_dev, 844 struct list_head **lh_xsk) 845 { 846 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 847 u32 kern_flags = bpf_net_ctx->ri.kern_flags; 848 struct list_head *lh; 849 850 *lh_map = *lh_dev = *lh_xsk = NULL; 851 852 if (!IS_ENABLED(CONFIG_BPF_SYSCALL)) 853 return; 854 855 lh = &bpf_net_ctx->dev_map_flush_list; 856 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh)) 857 *lh_dev = lh; 858 859 lh = &bpf_net_ctx->cpu_map_flush_list; 860 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh)) 861 *lh_map = lh; 862 863 lh = &bpf_net_ctx->xskmap_map_flush_list; 864 if (IS_ENABLED(CONFIG_XDP_SOCKETS) && 865 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh)) 866 *lh_xsk = lh; 867 } 868 869 /* Compute the linear packet data range [data, data_end) which 870 * will be accessed by various program types (cls_bpf, act_bpf, 871 * lwt, ...). Subsystems allowing direct data access must (!) 872 * ensure that cb[] area can be written to when BPF program is 873 * invoked (otherwise cb[] save/restore is necessary). 874 */ bpf_compute_data_pointers(struct sk_buff * skb)875 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 876 { 877 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 878 879 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 880 cb->data_meta = skb->data - skb_metadata_len(skb); 881 cb->data_end = skb->data + skb_headlen(skb); 882 } 883 884 /* Similar to bpf_compute_data_pointers(), except that save orginal 885 * data in cb->data and cb->meta_data for restore. 886 */ bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)887 static inline void bpf_compute_and_save_data_end( 888 struct sk_buff *skb, void **saved_data_end) 889 { 890 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 891 892 *saved_data_end = cb->data_end; 893 cb->data_end = skb->data + skb_headlen(skb); 894 } 895 896 /* Restore data saved by bpf_compute_and_save_data_end(). */ bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)897 static inline void bpf_restore_data_end( 898 struct sk_buff *skb, void *saved_data_end) 899 { 900 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 901 902 cb->data_end = saved_data_end; 903 } 904 bpf_skb_cb(const struct sk_buff * skb)905 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 906 { 907 /* eBPF programs may read/write skb->cb[] area to transfer meta 908 * data between tail calls. Since this also needs to work with 909 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 910 * 911 * In some socket filter cases, the cb unfortunately needs to be 912 * saved/restored so that protocol specific skb->cb[] data won't 913 * be lost. In any case, due to unpriviledged eBPF programs 914 * attached to sockets, we need to clear the bpf_skb_cb() area 915 * to not leak previous contents to user space. 916 */ 917 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 918 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 919 sizeof_field(struct qdisc_skb_cb, data)); 920 921 return qdisc_skb_cb(skb)->data; 922 } 923 924 /* Must be invoked with migration disabled */ __bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)925 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 926 const void *ctx) 927 { 928 const struct sk_buff *skb = ctx; 929 u8 *cb_data = bpf_skb_cb(skb); 930 u8 cb_saved[BPF_SKB_CB_LEN]; 931 u32 res; 932 933 if (unlikely(prog->cb_access)) { 934 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 935 memset(cb_data, 0, sizeof(cb_saved)); 936 } 937 938 res = bpf_prog_run(prog, skb); 939 940 if (unlikely(prog->cb_access)) 941 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 942 943 return res; 944 } 945 bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)946 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 947 struct sk_buff *skb) 948 { 949 u32 res; 950 951 migrate_disable(); 952 res = __bpf_prog_run_save_cb(prog, skb); 953 migrate_enable(); 954 return res; 955 } 956 bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)957 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 958 struct sk_buff *skb) 959 { 960 u8 *cb_data = bpf_skb_cb(skb); 961 u32 res; 962 963 if (unlikely(prog->cb_access)) 964 memset(cb_data, 0, BPF_SKB_CB_LEN); 965 966 res = bpf_prog_run_pin_on_cpu(prog, skb); 967 return res; 968 } 969 970 DECLARE_BPF_DISPATCHER(xdp) 971 972 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 973 974 u32 xdp_master_redirect(struct xdp_buff *xdp); 975 976 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 977 bpf_prog_insn_size(const struct bpf_prog * prog)978 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 979 { 980 return prog->len * sizeof(struct bpf_insn); 981 } 982 bpf_prog_tag_scratch_size(const struct bpf_prog * prog)983 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog) 984 { 985 return round_up(bpf_prog_insn_size(prog) + 986 sizeof(__be64) + 1, SHA1_BLOCK_SIZE); 987 } 988 bpf_prog_size(unsigned int proglen)989 static inline unsigned int bpf_prog_size(unsigned int proglen) 990 { 991 return max(sizeof(struct bpf_prog), 992 offsetof(struct bpf_prog, insns[proglen])); 993 } 994 bpf_prog_was_classic(const struct bpf_prog * prog)995 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 996 { 997 /* When classic BPF programs have been loaded and the arch 998 * does not have a classic BPF JIT (anymore), they have been 999 * converted via bpf_migrate_filter() to eBPF and thus always 1000 * have an unspec program type. 1001 */ 1002 return prog->type == BPF_PROG_TYPE_UNSPEC; 1003 } 1004 bpf_ctx_off_adjust_machine(u32 size)1005 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 1006 { 1007 const u32 size_machine = sizeof(unsigned long); 1008 1009 if (size > size_machine && size % size_machine == 0) 1010 size = size_machine; 1011 1012 return size; 1013 } 1014 1015 static inline bool bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)1016 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 1017 { 1018 return size <= size_default && (size & (size - 1)) == 0; 1019 } 1020 1021 static inline u8 bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)1022 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 1023 { 1024 u8 access_off = off & (size_default - 1); 1025 1026 #ifdef __LITTLE_ENDIAN 1027 return access_off; 1028 #else 1029 return size_default - (access_off + size); 1030 #endif 1031 } 1032 1033 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 1034 (size == sizeof(__u64) && \ 1035 off >= offsetof(type, field) && \ 1036 off + sizeof(__u64) <= offsetofend(type, field) && \ 1037 off % sizeof(__u64) == 0) 1038 1039 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 1040 bpf_prog_lock_ro(struct bpf_prog * fp)1041 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) 1042 { 1043 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 1044 if (!fp->jited) { 1045 set_vm_flush_reset_perms(fp); 1046 return set_memory_ro((unsigned long)fp, fp->pages); 1047 } 1048 #endif 1049 return 0; 1050 } 1051 1052 static inline int __must_check bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)1053 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 1054 { 1055 set_vm_flush_reset_perms(hdr); 1056 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 1057 } 1058 1059 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); sk_filter(struct sock * sk,struct sk_buff * skb)1060 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 1061 { 1062 return sk_filter_trim_cap(sk, skb, 1); 1063 } 1064 1065 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 1066 void bpf_prog_free(struct bpf_prog *fp); 1067 1068 bool bpf_opcode_in_insntable(u8 code); 1069 1070 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 1071 const u32 *insn_to_jit_off); 1072 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 1073 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 1074 1075 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 1076 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 1077 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 1078 gfp_t gfp_extra_flags); 1079 void __bpf_prog_free(struct bpf_prog *fp); 1080 bpf_prog_unlock_free(struct bpf_prog * fp)1081 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 1082 { 1083 __bpf_prog_free(fp); 1084 } 1085 1086 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 1087 unsigned int flen); 1088 1089 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 1090 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1091 bpf_aux_classic_check_t trans, bool save_orig); 1092 void bpf_prog_destroy(struct bpf_prog *fp); 1093 1094 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1095 int sk_attach_bpf(u32 ufd, struct sock *sk); 1096 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1097 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 1098 void sk_reuseport_prog_free(struct bpf_prog *prog); 1099 int sk_detach_filter(struct sock *sk); 1100 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 1101 1102 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 1103 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 1104 1105 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1106 #define __bpf_call_base_args \ 1107 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ 1108 (void *)__bpf_call_base) 1109 1110 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); 1111 void bpf_jit_compile(struct bpf_prog *prog); 1112 bool bpf_jit_needs_zext(void); 1113 bool bpf_jit_inlines_helper_call(s32 imm); 1114 bool bpf_jit_supports_subprog_tailcalls(void); 1115 bool bpf_jit_supports_percpu_insn(void); 1116 bool bpf_jit_supports_kfunc_call(void); 1117 bool bpf_jit_supports_far_kfunc_call(void); 1118 bool bpf_jit_supports_exceptions(void); 1119 bool bpf_jit_supports_ptr_xchg(void); 1120 bool bpf_jit_supports_arena(void); 1121 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena); 1122 u64 bpf_arch_uaddress_limit(void); 1123 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 1124 bool bpf_helper_changes_pkt_data(void *func); 1125 bpf_dump_raw_ok(const struct cred * cred)1126 static inline bool bpf_dump_raw_ok(const struct cred *cred) 1127 { 1128 /* Reconstruction of call-sites is dependent on kallsyms, 1129 * thus make dump the same restriction. 1130 */ 1131 return kallsyms_show_value(cred); 1132 } 1133 1134 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 1135 const struct bpf_insn *patch, u32 len); 1136 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 1137 xdp_return_frame_no_direct(void)1138 static inline bool xdp_return_frame_no_direct(void) 1139 { 1140 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1141 1142 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 1143 } 1144 xdp_set_return_frame_no_direct(void)1145 static inline void xdp_set_return_frame_no_direct(void) 1146 { 1147 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1148 1149 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 1150 } 1151 xdp_clear_return_frame_no_direct(void)1152 static inline void xdp_clear_return_frame_no_direct(void) 1153 { 1154 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1155 1156 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1157 } 1158 xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)1159 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1160 unsigned int pktlen) 1161 { 1162 unsigned int len; 1163 1164 if (unlikely(!(fwd->flags & IFF_UP))) 1165 return -ENETDOWN; 1166 1167 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1168 if (pktlen > len) 1169 return -EMSGSIZE; 1170 1171 return 0; 1172 } 1173 1174 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1175 * same cpu context. Further for best results no more than a single map 1176 * for the do_redirect/do_flush pair should be used. This limitation is 1177 * because we only track one map and force a flush when the map changes. 1178 * This does not appear to be a real limitation for existing software. 1179 */ 1180 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1181 struct xdp_buff *xdp, struct bpf_prog *prog); 1182 int xdp_do_redirect(struct net_device *dev, 1183 struct xdp_buff *xdp, 1184 struct bpf_prog *prog); 1185 int xdp_do_redirect_frame(struct net_device *dev, 1186 struct xdp_buff *xdp, 1187 struct xdp_frame *xdpf, 1188 struct bpf_prog *prog); 1189 void xdp_do_flush(void); 1190 1191 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act); 1192 1193 #ifdef CONFIG_INET 1194 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1195 struct bpf_prog *prog, struct sk_buff *skb, 1196 struct sock *migrating_sk, 1197 u32 hash); 1198 #else 1199 static inline struct sock * bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)1200 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1201 struct bpf_prog *prog, struct sk_buff *skb, 1202 struct sock *migrating_sk, 1203 u32 hash) 1204 { 1205 return NULL; 1206 } 1207 #endif 1208 1209 #ifdef CONFIG_BPF_JIT 1210 extern int bpf_jit_enable; 1211 extern int bpf_jit_harden; 1212 extern int bpf_jit_kallsyms; 1213 extern long bpf_jit_limit; 1214 extern long bpf_jit_limit_max; 1215 1216 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1217 1218 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1219 1220 struct bpf_binary_header * 1221 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1222 unsigned int alignment, 1223 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1224 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1225 u64 bpf_jit_alloc_exec_limit(void); 1226 void *bpf_jit_alloc_exec(unsigned long size); 1227 void bpf_jit_free_exec(void *addr); 1228 void bpf_jit_free(struct bpf_prog *fp); 1229 struct bpf_binary_header * 1230 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1231 1232 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1233 void bpf_prog_pack_free(void *ptr, u32 size); 1234 bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1235 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1236 { 1237 return list_empty(&fp->aux->ksym.lnode) || 1238 fp->aux->ksym.lnode.prev == LIST_POISON2; 1239 } 1240 1241 struct bpf_binary_header * 1242 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1243 unsigned int alignment, 1244 struct bpf_binary_header **rw_hdr, 1245 u8 **rw_image, 1246 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1247 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header, 1248 struct bpf_binary_header *rw_header); 1249 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1250 struct bpf_binary_header *rw_header); 1251 1252 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1253 struct bpf_jit_poke_descriptor *poke); 1254 1255 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1256 const struct bpf_insn *insn, bool extra_pass, 1257 u64 *func_addr, bool *func_addr_fixed); 1258 1259 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); 1260 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1261 bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1262 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1263 u32 pass, void *image) 1264 { 1265 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen, 1266 proglen, pass, image, current->comm, task_pid_nr(current)); 1267 1268 if (image) 1269 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1270 16, 1, image, proglen, false); 1271 } 1272 bpf_jit_is_ebpf(void)1273 static inline bool bpf_jit_is_ebpf(void) 1274 { 1275 # ifdef CONFIG_HAVE_EBPF_JIT 1276 return true; 1277 # else 1278 return false; 1279 # endif 1280 } 1281 ebpf_jit_enabled(void)1282 static inline bool ebpf_jit_enabled(void) 1283 { 1284 return bpf_jit_enable && bpf_jit_is_ebpf(); 1285 } 1286 bpf_prog_ebpf_jited(const struct bpf_prog * fp)1287 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1288 { 1289 return fp->jited && bpf_jit_is_ebpf(); 1290 } 1291 bpf_jit_blinding_enabled(struct bpf_prog * prog)1292 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1293 { 1294 /* These are the prerequisites, should someone ever have the 1295 * idea to call blinding outside of them, we make sure to 1296 * bail out. 1297 */ 1298 if (!bpf_jit_is_ebpf()) 1299 return false; 1300 if (!prog->jit_requested) 1301 return false; 1302 if (!bpf_jit_harden) 1303 return false; 1304 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1305 return false; 1306 1307 return true; 1308 } 1309 bpf_jit_kallsyms_enabled(void)1310 static inline bool bpf_jit_kallsyms_enabled(void) 1311 { 1312 /* There are a couple of corner cases where kallsyms should 1313 * not be enabled f.e. on hardening. 1314 */ 1315 if (bpf_jit_harden) 1316 return false; 1317 if (!bpf_jit_kallsyms) 1318 return false; 1319 if (bpf_jit_kallsyms == 1) 1320 return true; 1321 1322 return false; 1323 } 1324 1325 int __bpf_address_lookup(unsigned long addr, unsigned long *size, 1326 unsigned long *off, char *sym); 1327 bool is_bpf_text_address(unsigned long addr); 1328 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1329 char *sym); 1330 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1331 1332 static inline int bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1333 bpf_address_lookup(unsigned long addr, unsigned long *size, 1334 unsigned long *off, char **modname, char *sym) 1335 { 1336 int ret = __bpf_address_lookup(addr, size, off, sym); 1337 1338 if (ret && modname) 1339 *modname = NULL; 1340 return ret; 1341 } 1342 1343 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1344 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1345 1346 #else /* CONFIG_BPF_JIT */ 1347 ebpf_jit_enabled(void)1348 static inline bool ebpf_jit_enabled(void) 1349 { 1350 return false; 1351 } 1352 bpf_jit_blinding_enabled(struct bpf_prog * prog)1353 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1354 { 1355 return false; 1356 } 1357 bpf_prog_ebpf_jited(const struct bpf_prog * fp)1358 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1359 { 1360 return false; 1361 } 1362 1363 static inline int bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1364 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1365 struct bpf_jit_poke_descriptor *poke) 1366 { 1367 return -ENOTSUPP; 1368 } 1369 bpf_jit_free(struct bpf_prog * fp)1370 static inline void bpf_jit_free(struct bpf_prog *fp) 1371 { 1372 bpf_prog_unlock_free(fp); 1373 } 1374 bpf_jit_kallsyms_enabled(void)1375 static inline bool bpf_jit_kallsyms_enabled(void) 1376 { 1377 return false; 1378 } 1379 1380 static inline int __bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1381 __bpf_address_lookup(unsigned long addr, unsigned long *size, 1382 unsigned long *off, char *sym) 1383 { 1384 return 0; 1385 } 1386 is_bpf_text_address(unsigned long addr)1387 static inline bool is_bpf_text_address(unsigned long addr) 1388 { 1389 return false; 1390 } 1391 bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1392 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1393 char *type, char *sym) 1394 { 1395 return -ERANGE; 1396 } 1397 bpf_prog_ksym_find(unsigned long addr)1398 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1399 { 1400 return NULL; 1401 } 1402 1403 static inline int bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1404 bpf_address_lookup(unsigned long addr, unsigned long *size, 1405 unsigned long *off, char **modname, char *sym) 1406 { 1407 return 0; 1408 } 1409 bpf_prog_kallsyms_add(struct bpf_prog * fp)1410 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1411 { 1412 } 1413 bpf_prog_kallsyms_del(struct bpf_prog * fp)1414 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1415 { 1416 } 1417 1418 #endif /* CONFIG_BPF_JIT */ 1419 1420 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1421 1422 #define BPF_ANC BIT(15) 1423 bpf_needs_clear_a(const struct sock_filter * first)1424 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1425 { 1426 switch (first->code) { 1427 case BPF_RET | BPF_K: 1428 case BPF_LD | BPF_W | BPF_LEN: 1429 return false; 1430 1431 case BPF_LD | BPF_W | BPF_ABS: 1432 case BPF_LD | BPF_H | BPF_ABS: 1433 case BPF_LD | BPF_B | BPF_ABS: 1434 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1435 return true; 1436 return false; 1437 1438 default: 1439 return true; 1440 } 1441 } 1442 bpf_anc_helper(const struct sock_filter * ftest)1443 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1444 { 1445 BUG_ON(ftest->code & BPF_ANC); 1446 1447 switch (ftest->code) { 1448 case BPF_LD | BPF_W | BPF_ABS: 1449 case BPF_LD | BPF_H | BPF_ABS: 1450 case BPF_LD | BPF_B | BPF_ABS: 1451 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1452 return BPF_ANC | SKF_AD_##CODE 1453 switch (ftest->k) { 1454 BPF_ANCILLARY(PROTOCOL); 1455 BPF_ANCILLARY(PKTTYPE); 1456 BPF_ANCILLARY(IFINDEX); 1457 BPF_ANCILLARY(NLATTR); 1458 BPF_ANCILLARY(NLATTR_NEST); 1459 BPF_ANCILLARY(MARK); 1460 BPF_ANCILLARY(QUEUE); 1461 BPF_ANCILLARY(HATYPE); 1462 BPF_ANCILLARY(RXHASH); 1463 BPF_ANCILLARY(CPU); 1464 BPF_ANCILLARY(ALU_XOR_X); 1465 BPF_ANCILLARY(VLAN_TAG); 1466 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1467 BPF_ANCILLARY(PAY_OFFSET); 1468 BPF_ANCILLARY(RANDOM); 1469 BPF_ANCILLARY(VLAN_TPID); 1470 } 1471 fallthrough; 1472 default: 1473 return ftest->code; 1474 } 1475 } 1476 1477 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1478 int k, unsigned int size); 1479 bpf_tell_extensions(void)1480 static inline int bpf_tell_extensions(void) 1481 { 1482 return SKF_AD_MAX; 1483 } 1484 1485 struct bpf_sock_addr_kern { 1486 struct sock *sk; 1487 struct sockaddr *uaddr; 1488 /* Temporary "register" to make indirect stores to nested structures 1489 * defined above. We need three registers to make such a store, but 1490 * only two (src and dst) are available at convert_ctx_access time 1491 */ 1492 u64 tmp_reg; 1493 void *t_ctx; /* Attach type specific context. */ 1494 u32 uaddrlen; 1495 }; 1496 1497 struct bpf_sock_ops_kern { 1498 struct sock *sk; 1499 union { 1500 u32 args[4]; 1501 u32 reply; 1502 u32 replylong[4]; 1503 }; 1504 struct sk_buff *syn_skb; 1505 struct sk_buff *skb; 1506 void *skb_data_end; 1507 u8 op; 1508 u8 is_fullsock; 1509 u8 remaining_opt_len; 1510 u64 temp; /* temp and everything after is not 1511 * initialized to 0 before calling 1512 * the BPF program. New fields that 1513 * should be initialized to 0 should 1514 * be inserted before temp. 1515 * temp is scratch storage used by 1516 * sock_ops_convert_ctx_access 1517 * as temporary storage of a register. 1518 */ 1519 }; 1520 1521 struct bpf_sysctl_kern { 1522 struct ctl_table_header *head; 1523 const struct ctl_table *table; 1524 void *cur_val; 1525 size_t cur_len; 1526 void *new_val; 1527 size_t new_len; 1528 int new_updated; 1529 int write; 1530 loff_t *ppos; 1531 /* Temporary "register" for indirect stores to ppos. */ 1532 u64 tmp_reg; 1533 }; 1534 1535 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1536 struct bpf_sockopt_buf { 1537 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1538 }; 1539 1540 struct bpf_sockopt_kern { 1541 struct sock *sk; 1542 u8 *optval; 1543 u8 *optval_end; 1544 s32 level; 1545 s32 optname; 1546 s32 optlen; 1547 /* for retval in struct bpf_cg_run_ctx */ 1548 struct task_struct *current_task; 1549 /* Temporary "register" for indirect stores to ppos. */ 1550 u64 tmp_reg; 1551 }; 1552 1553 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1554 1555 struct bpf_sk_lookup_kern { 1556 u16 family; 1557 u16 protocol; 1558 __be16 sport; 1559 u16 dport; 1560 struct { 1561 __be32 saddr; 1562 __be32 daddr; 1563 } v4; 1564 struct { 1565 const struct in6_addr *saddr; 1566 const struct in6_addr *daddr; 1567 } v6; 1568 struct sock *selected_sk; 1569 u32 ingress_ifindex; 1570 bool no_reuseport; 1571 }; 1572 1573 extern struct static_key_false bpf_sk_lookup_enabled; 1574 1575 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1576 * 1577 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1578 * SK_DROP. Their meaning is as follows: 1579 * 1580 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1581 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1582 * SK_DROP : terminate lookup with -ECONNREFUSED 1583 * 1584 * This macro aggregates return values and selected sockets from 1585 * multiple BPF programs according to following rules in order: 1586 * 1587 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1588 * macro result is SK_PASS and last ctx.selected_sk is used. 1589 * 2. If any program returned SK_DROP return value, 1590 * macro result is SK_DROP. 1591 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1592 * 1593 * Caller must ensure that the prog array is non-NULL, and that the 1594 * array as well as the programs it contains remain valid. 1595 */ 1596 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1597 ({ \ 1598 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1599 struct bpf_prog_array_item *_item; \ 1600 struct sock *_selected_sk = NULL; \ 1601 bool _no_reuseport = false; \ 1602 struct bpf_prog *_prog; \ 1603 bool _all_pass = true; \ 1604 u32 _ret; \ 1605 \ 1606 migrate_disable(); \ 1607 _item = &(array)->items[0]; \ 1608 while ((_prog = READ_ONCE(_item->prog))) { \ 1609 /* restore most recent selection */ \ 1610 _ctx->selected_sk = _selected_sk; \ 1611 _ctx->no_reuseport = _no_reuseport; \ 1612 \ 1613 _ret = func(_prog, _ctx); \ 1614 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1615 /* remember last non-NULL socket */ \ 1616 _selected_sk = _ctx->selected_sk; \ 1617 _no_reuseport = _ctx->no_reuseport; \ 1618 } else if (_ret == SK_DROP && _all_pass) { \ 1619 _all_pass = false; \ 1620 } \ 1621 _item++; \ 1622 } \ 1623 _ctx->selected_sk = _selected_sk; \ 1624 _ctx->no_reuseport = _no_reuseport; \ 1625 migrate_enable(); \ 1626 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1627 }) 1628 bpf_sk_lookup_run_v4(const struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1629 static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol, 1630 const __be32 saddr, const __be16 sport, 1631 const __be32 daddr, const u16 dport, 1632 const int ifindex, struct sock **psk) 1633 { 1634 struct bpf_prog_array *run_array; 1635 struct sock *selected_sk = NULL; 1636 bool no_reuseport = false; 1637 1638 rcu_read_lock(); 1639 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1640 if (run_array) { 1641 struct bpf_sk_lookup_kern ctx = { 1642 .family = AF_INET, 1643 .protocol = protocol, 1644 .v4.saddr = saddr, 1645 .v4.daddr = daddr, 1646 .sport = sport, 1647 .dport = dport, 1648 .ingress_ifindex = ifindex, 1649 }; 1650 u32 act; 1651 1652 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1653 if (act == SK_PASS) { 1654 selected_sk = ctx.selected_sk; 1655 no_reuseport = ctx.no_reuseport; 1656 } else { 1657 selected_sk = ERR_PTR(-ECONNREFUSED); 1658 } 1659 } 1660 rcu_read_unlock(); 1661 *psk = selected_sk; 1662 return no_reuseport; 1663 } 1664 1665 #if IS_ENABLED(CONFIG_IPV6) bpf_sk_lookup_run_v6(const struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,const int ifindex,struct sock ** psk)1666 static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol, 1667 const struct in6_addr *saddr, 1668 const __be16 sport, 1669 const struct in6_addr *daddr, 1670 const u16 dport, 1671 const int ifindex, struct sock **psk) 1672 { 1673 struct bpf_prog_array *run_array; 1674 struct sock *selected_sk = NULL; 1675 bool no_reuseport = false; 1676 1677 rcu_read_lock(); 1678 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1679 if (run_array) { 1680 struct bpf_sk_lookup_kern ctx = { 1681 .family = AF_INET6, 1682 .protocol = protocol, 1683 .v6.saddr = saddr, 1684 .v6.daddr = daddr, 1685 .sport = sport, 1686 .dport = dport, 1687 .ingress_ifindex = ifindex, 1688 }; 1689 u32 act; 1690 1691 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1692 if (act == SK_PASS) { 1693 selected_sk = ctx.selected_sk; 1694 no_reuseport = ctx.no_reuseport; 1695 } else { 1696 selected_sk = ERR_PTR(-ECONNREFUSED); 1697 } 1698 } 1699 rcu_read_unlock(); 1700 *psk = selected_sk; 1701 return no_reuseport; 1702 } 1703 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1704 __bpf_xdp_redirect_map(struct bpf_map * map,u64 index,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1705 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1706 u64 flags, const u64 flag_mask, 1707 void *lookup_elem(struct bpf_map *map, u32 key)) 1708 { 1709 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1710 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1711 1712 /* Lower bits of the flags are used as return code on lookup failure */ 1713 if (unlikely(flags & ~(action_mask | flag_mask))) 1714 return XDP_ABORTED; 1715 1716 ri->tgt_value = lookup_elem(map, index); 1717 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1718 /* If the lookup fails we want to clear out the state in the 1719 * redirect_info struct completely, so that if an eBPF program 1720 * performs multiple lookups, the last one always takes 1721 * precedence. 1722 */ 1723 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1724 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1725 return flags & action_mask; 1726 } 1727 1728 ri->tgt_index = index; 1729 ri->map_id = map->id; 1730 ri->map_type = map->map_type; 1731 1732 if (flags & BPF_F_BROADCAST) { 1733 WRITE_ONCE(ri->map, map); 1734 ri->flags = flags; 1735 } else { 1736 WRITE_ONCE(ri->map, NULL); 1737 ri->flags = 0; 1738 } 1739 1740 return XDP_REDIRECT; 1741 } 1742 1743 #ifdef CONFIG_NET 1744 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1745 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1746 u32 len, u64 flags); 1747 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1748 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1749 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1750 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1751 void *buf, unsigned long len, bool flush); 1752 #else /* CONFIG_NET */ __bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1753 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1754 void *to, u32 len) 1755 { 1756 return -EOPNOTSUPP; 1757 } 1758 __bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1759 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1760 const void *from, u32 len, u64 flags) 1761 { 1762 return -EOPNOTSUPP; 1763 } 1764 __bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1765 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1766 void *buf, u32 len) 1767 { 1768 return -EOPNOTSUPP; 1769 } 1770 __bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1771 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1772 void *buf, u32 len) 1773 { 1774 return -EOPNOTSUPP; 1775 } 1776 bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1777 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1778 { 1779 return NULL; 1780 } 1781 bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1782 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1783 unsigned long len, bool flush) 1784 { 1785 } 1786 #endif /* CONFIG_NET */ 1787 1788 #endif /* __LINUX_FILTER_H__ */ 1789