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 <linux/u64_stats_sync.h> 25 26 #include <net/sch_generic.h> 27 28 #include <asm/byteorder.h> 29 #include <uapi/linux/filter.h> 30 31 struct sk_buff; 32 struct sock; 33 struct seccomp_data; 34 struct bpf_prog_aux; 35 struct xdp_rxq_info; 36 struct xdp_buff; 37 struct sock_reuseport; 38 struct ctl_table; 39 struct ctl_table_header; 40 41 /* ArgX, context and stack frame pointer register positions. Note, 42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function 43 * calls in BPF_CALL instruction. 44 */ 45 #define BPF_REG_ARG1 BPF_REG_1 46 #define BPF_REG_ARG2 BPF_REG_2 47 #define BPF_REG_ARG3 BPF_REG_3 48 #define BPF_REG_ARG4 BPF_REG_4 49 #define BPF_REG_ARG5 BPF_REG_5 50 #define BPF_REG_CTX BPF_REG_6 51 #define BPF_REG_FP BPF_REG_10 52 53 /* Additional register mappings for converted user programs. */ 54 #define BPF_REG_A BPF_REG_0 55 #define BPF_REG_X BPF_REG_7 56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */ 57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */ 58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */ 59 60 /* Kernel hidden auxiliary/helper register. */ 61 #define BPF_REG_PARAMS MAX_BPF_REG 62 #define BPF_REG_AX (MAX_BPF_REG + 1) 63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 2) 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 special ldsx instruction. Same as BPF_NOSPEC */ 82 #define BPF_PROBE_MEM32SX 0xc0 83 84 /* unused opcode to mark call to interpreter with arguments */ 85 #define BPF_CALL_ARGS 0xe0 86 87 /* unused opcode to mark speculation barrier for mitigating 88 * Spectre v1 and v4 89 */ 90 #define BPF_NOSPEC 0xc0 91 92 /* As per nm, we expose JITed images as text (code) section for 93 * kallsyms. That way, tools like perf can find it to match 94 * addresses. 95 */ 96 #define BPF_SYM_ELF_TYPE 't' 97 98 /* BPF program can access up to 512 bytes of stack space. */ 99 #define MAX_BPF_STACK 512 100 101 /* Helper macros for filter block array initializers. */ 102 103 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */ 104 105 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \ 106 ((struct bpf_insn) { \ 107 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \ 108 .dst_reg = DST, \ 109 .src_reg = SRC, \ 110 .off = OFF, \ 111 .imm = 0 }) 112 113 #define BPF_ALU64_REG(OP, DST, SRC) \ 114 BPF_ALU64_REG_OFF(OP, DST, SRC, 0) 115 116 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \ 117 ((struct bpf_insn) { \ 118 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \ 119 .dst_reg = DST, \ 120 .src_reg = SRC, \ 121 .off = OFF, \ 122 .imm = 0 }) 123 124 #define BPF_ALU32_REG(OP, DST, SRC) \ 125 BPF_ALU32_REG_OFF(OP, DST, SRC, 0) 126 127 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */ 128 129 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \ 130 ((struct bpf_insn) { \ 131 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \ 132 .dst_reg = DST, \ 133 .src_reg = 0, \ 134 .off = OFF, \ 135 .imm = IMM }) 136 #define BPF_ALU64_IMM(OP, DST, IMM) \ 137 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0) 138 139 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \ 140 ((struct bpf_insn) { \ 141 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \ 142 .dst_reg = DST, \ 143 .src_reg = 0, \ 144 .off = OFF, \ 145 .imm = IMM }) 146 #define BPF_ALU32_IMM(OP, DST, IMM) \ 147 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0) 148 149 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */ 150 151 #define BPF_ENDIAN(TYPE, DST, LEN) \ 152 ((struct bpf_insn) { \ 153 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \ 154 .dst_reg = DST, \ 155 .src_reg = 0, \ 156 .off = 0, \ 157 .imm = LEN }) 158 159 /* Byte Swap, bswap16/32/64 */ 160 161 #define BPF_BSWAP(DST, LEN) \ 162 ((struct bpf_insn) { \ 163 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \ 164 .dst_reg = DST, \ 165 .src_reg = 0, \ 166 .off = 0, \ 167 .imm = LEN }) 168 169 /* Short form of mov, dst_reg = src_reg */ 170 171 #define BPF_MOV64_REG(DST, SRC) \ 172 ((struct bpf_insn) { \ 173 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 174 .dst_reg = DST, \ 175 .src_reg = SRC, \ 176 .off = 0, \ 177 .imm = 0 }) 178 179 #define BPF_MOV32_REG(DST, SRC) \ 180 ((struct bpf_insn) { \ 181 .code = BPF_ALU | BPF_MOV | BPF_X, \ 182 .dst_reg = DST, \ 183 .src_reg = SRC, \ 184 .off = 0, \ 185 .imm = 0 }) 186 187 /* Special (internal-only) form of mov, used to resolve per-CPU addrs: 188 * dst_reg = src_reg + <percpu_base_off> 189 * BPF_ADDR_PERCPU is used as a special insn->off value. 190 */ 191 #define BPF_ADDR_PERCPU (-1) 192 193 #define BPF_MOV64_PERCPU_REG(DST, SRC) \ 194 ((struct bpf_insn) { \ 195 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 196 .dst_reg = DST, \ 197 .src_reg = SRC, \ 198 .off = BPF_ADDR_PERCPU, \ 199 .imm = 0 }) 200 201 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn) 202 { 203 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU; 204 } 205 206 /* Short form of mov, dst_reg = imm32 */ 207 208 #define BPF_MOV64_IMM(DST, IMM) \ 209 ((struct bpf_insn) { \ 210 .code = BPF_ALU64 | BPF_MOV | BPF_K, \ 211 .dst_reg = DST, \ 212 .src_reg = 0, \ 213 .off = 0, \ 214 .imm = IMM }) 215 216 #define BPF_MOV32_IMM(DST, IMM) \ 217 ((struct bpf_insn) { \ 218 .code = BPF_ALU | BPF_MOV | BPF_K, \ 219 .dst_reg = DST, \ 220 .src_reg = 0, \ 221 .off = 0, \ 222 .imm = IMM }) 223 224 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */ 225 226 #define BPF_MOVSX64_REG(DST, SRC, OFF) \ 227 ((struct bpf_insn) { \ 228 .code = BPF_ALU64 | BPF_MOV | BPF_X, \ 229 .dst_reg = DST, \ 230 .src_reg = SRC, \ 231 .off = OFF, \ 232 .imm = 0 }) 233 234 #define BPF_MOVSX32_REG(DST, SRC, OFF) \ 235 ((struct bpf_insn) { \ 236 .code = BPF_ALU | BPF_MOV | BPF_X, \ 237 .dst_reg = DST, \ 238 .src_reg = SRC, \ 239 .off = OFF, \ 240 .imm = 0 }) 241 242 /* Special form of mov32, used for doing explicit zero extension on dst. */ 243 #define BPF_ZEXT_REG(DST) \ 244 ((struct bpf_insn) { \ 245 .code = BPF_ALU | BPF_MOV | BPF_X, \ 246 .dst_reg = DST, \ 247 .src_reg = DST, \ 248 .off = 0, \ 249 .imm = 1 }) 250 251 static inline bool insn_is_zext(const struct bpf_insn *insn) 252 { 253 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1; 254 } 255 256 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers 257 * to pointers in user vma. 258 */ 259 static inline bool insn_is_cast_user(const struct bpf_insn *insn) 260 { 261 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && 262 insn->off == BPF_ADDR_SPACE_CAST && 263 insn->imm == 1U << 16; 264 } 265 266 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */ 267 #define BPF_LD_IMM64(DST, IMM) \ 268 BPF_LD_IMM64_RAW(DST, 0, IMM) 269 270 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \ 271 ((struct bpf_insn) { \ 272 .code = BPF_LD | BPF_DW | BPF_IMM, \ 273 .dst_reg = DST, \ 274 .src_reg = SRC, \ 275 .off = 0, \ 276 .imm = (__u32) (IMM) }), \ 277 ((struct bpf_insn) { \ 278 .code = 0, /* zero is reserved opcode */ \ 279 .dst_reg = 0, \ 280 .src_reg = 0, \ 281 .off = 0, \ 282 .imm = ((__u64) (IMM)) >> 32 }) 283 284 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */ 285 #define BPF_LD_MAP_FD(DST, MAP_FD) \ 286 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD) 287 288 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */ 289 290 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \ 291 ((struct bpf_insn) { \ 292 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \ 293 .dst_reg = DST, \ 294 .src_reg = SRC, \ 295 .off = 0, \ 296 .imm = IMM }) 297 298 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \ 299 ((struct bpf_insn) { \ 300 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \ 301 .dst_reg = DST, \ 302 .src_reg = SRC, \ 303 .off = 0, \ 304 .imm = IMM }) 305 306 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */ 307 308 #define BPF_LD_ABS(SIZE, IMM) \ 309 ((struct bpf_insn) { \ 310 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \ 311 .dst_reg = 0, \ 312 .src_reg = 0, \ 313 .off = 0, \ 314 .imm = IMM }) 315 316 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */ 317 318 #define BPF_LD_IND(SIZE, SRC, IMM) \ 319 ((struct bpf_insn) { \ 320 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \ 321 .dst_reg = 0, \ 322 .src_reg = SRC, \ 323 .off = 0, \ 324 .imm = IMM }) 325 326 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */ 327 328 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \ 329 ((struct bpf_insn) { \ 330 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \ 331 .dst_reg = DST, \ 332 .src_reg = SRC, \ 333 .off = OFF, \ 334 .imm = 0 }) 335 336 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */ 337 338 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \ 339 ((struct bpf_insn) { \ 340 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \ 341 .dst_reg = DST, \ 342 .src_reg = SRC, \ 343 .off = OFF, \ 344 .imm = 0 }) 345 346 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */ 347 348 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \ 349 ((struct bpf_insn) { \ 350 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \ 351 .dst_reg = DST, \ 352 .src_reg = SRC, \ 353 .off = OFF, \ 354 .imm = 0 }) 355 356 357 /* 358 * Atomic operations: 359 * 360 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg 361 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg 362 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg 363 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg 364 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg); 365 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg); 366 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg); 367 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg); 368 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg) 369 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg) 370 * BPF_LOAD_ACQ dst_reg = smp_load_acquire(src_reg + off16) 371 * BPF_STORE_REL smp_store_release(dst_reg + off16, src_reg) 372 */ 373 374 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \ 375 ((struct bpf_insn) { \ 376 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \ 377 .dst_reg = DST, \ 378 .src_reg = SRC, \ 379 .off = OFF, \ 380 .imm = OP }) 381 382 /* Legacy alias */ 383 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF) 384 385 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */ 386 387 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \ 388 ((struct bpf_insn) { \ 389 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \ 390 .dst_reg = DST, \ 391 .src_reg = 0, \ 392 .off = OFF, \ 393 .imm = IMM }) 394 395 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */ 396 397 #define BPF_JMP_REG(OP, DST, SRC, OFF) \ 398 ((struct bpf_insn) { \ 399 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \ 400 .dst_reg = DST, \ 401 .src_reg = SRC, \ 402 .off = OFF, \ 403 .imm = 0 }) 404 405 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */ 406 407 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \ 408 ((struct bpf_insn) { \ 409 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \ 410 .dst_reg = DST, \ 411 .src_reg = 0, \ 412 .off = OFF, \ 413 .imm = IMM }) 414 415 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */ 416 417 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \ 418 ((struct bpf_insn) { \ 419 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \ 420 .dst_reg = DST, \ 421 .src_reg = SRC, \ 422 .off = OFF, \ 423 .imm = 0 }) 424 425 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */ 426 427 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \ 428 ((struct bpf_insn) { \ 429 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \ 430 .dst_reg = DST, \ 431 .src_reg = 0, \ 432 .off = OFF, \ 433 .imm = IMM }) 434 435 /* Unconditional jumps, goto pc + off16 */ 436 437 #define BPF_JMP_A(OFF) \ 438 ((struct bpf_insn) { \ 439 .code = BPF_JMP | BPF_JA, \ 440 .dst_reg = 0, \ 441 .src_reg = 0, \ 442 .off = OFF, \ 443 .imm = 0 }) 444 445 /* Unconditional jumps, gotol pc + imm32 */ 446 447 #define BPF_JMP32_A(IMM) \ 448 ((struct bpf_insn) { \ 449 .code = BPF_JMP32 | BPF_JA, \ 450 .dst_reg = 0, \ 451 .src_reg = 0, \ 452 .off = 0, \ 453 .imm = IMM }) 454 455 /* Relative call */ 456 457 #define BPF_CALL_REL(TGT) \ 458 ((struct bpf_insn) { \ 459 .code = BPF_JMP | BPF_CALL, \ 460 .dst_reg = 0, \ 461 .src_reg = BPF_PSEUDO_CALL, \ 462 .off = 0, \ 463 .imm = TGT }) 464 465 /* Convert function address to BPF immediate */ 466 467 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base) 468 469 #define BPF_EMIT_CALL(FUNC) \ 470 ((struct bpf_insn) { \ 471 .code = BPF_JMP | BPF_CALL, \ 472 .dst_reg = 0, \ 473 .src_reg = 0, \ 474 .off = 0, \ 475 .imm = BPF_CALL_IMM(FUNC) }) 476 477 /* Kfunc call */ 478 479 #define BPF_CALL_KFUNC(OFF, IMM) \ 480 ((struct bpf_insn) { \ 481 .code = BPF_JMP | BPF_CALL, \ 482 .dst_reg = 0, \ 483 .src_reg = BPF_PSEUDO_KFUNC_CALL, \ 484 .off = OFF, \ 485 .imm = IMM }) 486 487 /* Raw code statement block */ 488 489 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \ 490 ((struct bpf_insn) { \ 491 .code = CODE, \ 492 .dst_reg = DST, \ 493 .src_reg = SRC, \ 494 .off = OFF, \ 495 .imm = IMM }) 496 497 /* Program exit */ 498 499 #define BPF_EXIT_INSN() \ 500 ((struct bpf_insn) { \ 501 .code = BPF_JMP | BPF_EXIT, \ 502 .dst_reg = 0, \ 503 .src_reg = 0, \ 504 .off = 0, \ 505 .imm = 0 }) 506 507 /* Speculation barrier */ 508 509 #define BPF_ST_NOSPEC() \ 510 ((struct bpf_insn) { \ 511 .code = BPF_ST | BPF_NOSPEC, \ 512 .dst_reg = 0, \ 513 .src_reg = 0, \ 514 .off = 0, \ 515 .imm = 0 }) 516 517 /* Internal classic blocks for direct assignment */ 518 519 #define __BPF_STMT(CODE, K) \ 520 ((struct sock_filter) BPF_STMT(CODE, K)) 521 522 #define __BPF_JUMP(CODE, K, JT, JF) \ 523 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF)) 524 525 #define bytes_to_bpf_size(bytes) \ 526 ({ \ 527 int bpf_size = -EINVAL; \ 528 \ 529 if (bytes == sizeof(u8)) \ 530 bpf_size = BPF_B; \ 531 else if (bytes == sizeof(u16)) \ 532 bpf_size = BPF_H; \ 533 else if (bytes == sizeof(u32)) \ 534 bpf_size = BPF_W; \ 535 else if (bytes == sizeof(u64)) \ 536 bpf_size = BPF_DW; \ 537 \ 538 bpf_size; \ 539 }) 540 541 #define bpf_size_to_bytes(bpf_size) \ 542 ({ \ 543 int bytes = -EINVAL; \ 544 \ 545 if (bpf_size == BPF_B) \ 546 bytes = sizeof(u8); \ 547 else if (bpf_size == BPF_H) \ 548 bytes = sizeof(u16); \ 549 else if (bpf_size == BPF_W) \ 550 bytes = sizeof(u32); \ 551 else if (bpf_size == BPF_DW) \ 552 bytes = sizeof(u64); \ 553 \ 554 bytes; \ 555 }) 556 557 #define BPF_SIZEOF(type) \ 558 ({ \ 559 const int __size = bytes_to_bpf_size(sizeof(type)); \ 560 BUILD_BUG_ON(__size < 0); \ 561 __size; \ 562 }) 563 564 #define BPF_FIELD_SIZEOF(type, field) \ 565 ({ \ 566 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \ 567 BUILD_BUG_ON(__size < 0); \ 568 __size; \ 569 }) 570 571 #define BPF_LDST_BYTES(insn) \ 572 ({ \ 573 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \ 574 WARN_ON(__size < 0); \ 575 __size; \ 576 }) 577 578 #define __BPF_MAP_0(m, v, ...) v 579 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a) 580 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__) 581 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__) 582 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__) 583 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__) 584 585 #define __BPF_REG_0(...) __BPF_PAD(5) 586 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4) 587 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3) 588 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2) 589 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1) 590 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__) 591 592 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__) 593 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__) 594 595 #define __BPF_CAST(t, a) \ 596 (__force t) \ 597 (__force \ 598 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \ 599 (unsigned long)0, (t)0))) a 600 #define __BPF_V void 601 #define __BPF_N 602 603 #define __BPF_DECL_ARGS(t, a) t a 604 #define __BPF_DECL_REGS(t, a) u64 a 605 606 #define __BPF_PAD(n) \ 607 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \ 608 u64, __ur_3, u64, __ur_4, u64, __ur_5) 609 610 #define BPF_CALL_x(x, attr, name, ...) \ 611 static __always_inline \ 612 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 613 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \ 614 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \ 615 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \ 616 { \ 617 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\ 618 } \ 619 static __always_inline \ 620 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)) 621 622 #define __NOATTR 623 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__) 624 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__) 625 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__) 626 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__) 627 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__) 628 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__) 629 630 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__) 631 632 #define bpf_ctx_range(TYPE, MEMBER) \ 633 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 634 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \ 635 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1 636 #if BITS_PER_LONG == 64 637 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 638 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1 639 #else 640 # define bpf_ctx_range_ptr(TYPE, MEMBER) \ 641 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1 642 #endif /* BITS_PER_LONG == 64 */ 643 644 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \ 645 ({ \ 646 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \ 647 *(PTR_SIZE) = (SIZE); \ 648 offsetof(TYPE, MEMBER); \ 649 }) 650 651 /* A struct sock_filter is architecture independent. */ 652 struct compat_sock_fprog { 653 u16 len; 654 compat_uptr_t filter; /* struct sock_filter * */ 655 }; 656 657 struct sock_fprog_kern { 658 u16 len; 659 struct sock_filter *filter; 660 }; 661 662 /* Some arches need doubleword alignment for their instructions and/or data */ 663 #define BPF_IMAGE_ALIGNMENT 8 664 665 struct bpf_binary_header { 666 u32 size; 667 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT); 668 }; 669 670 struct bpf_prog_stats { 671 u64_stats_t cnt; 672 u64_stats_t nsecs; 673 u64_stats_t misses; 674 struct u64_stats_sync syncp; 675 } __aligned(2 * sizeof(u64)); 676 677 struct bpf_timed_may_goto { 678 u64 count; 679 u64 timestamp; 680 }; 681 682 struct sk_filter { 683 refcount_t refcnt; 684 struct rcu_head rcu; 685 struct bpf_prog *prog; 686 }; 687 688 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key); 689 690 extern struct mutex nf_conn_btf_access_lock; 691 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 692 const struct bpf_reg_state *reg, 693 int off, int size); 694 695 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx, 696 const struct bpf_insn *insnsi, 697 unsigned int (*bpf_func)(const void *, 698 const struct bpf_insn *)); 699 700 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, 701 const void *ctx, 702 bpf_dispatcher_fn dfunc) 703 { 704 u32 ret; 705 706 cant_migrate(); 707 if (static_branch_unlikely(&bpf_stats_enabled_key)) { 708 struct bpf_prog_stats *stats; 709 u64 duration, start = sched_clock(); 710 unsigned long flags; 711 712 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 713 714 duration = sched_clock() - start; 715 if (likely(prog->stats)) { 716 stats = this_cpu_ptr(prog->stats); 717 flags = u64_stats_update_begin_irqsave(&stats->syncp); 718 u64_stats_inc(&stats->cnt); 719 u64_stats_add(&stats->nsecs, duration); 720 u64_stats_update_end_irqrestore(&stats->syncp, flags); 721 } 722 } else { 723 ret = dfunc(ctx, prog->insnsi, prog->bpf_func); 724 } 725 return ret; 726 } 727 728 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx) 729 { 730 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func); 731 } 732 733 /* 734 * Use in preemptible and therefore migratable context to make sure that 735 * the execution of the BPF program runs on one CPU. 736 * 737 * This uses migrate_disable/enable() explicitly to document that the 738 * invocation of a BPF program does not require reentrancy protection 739 * against a BPF program which is invoked from a preempting task. 740 */ 741 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog, 742 const void *ctx) 743 { 744 u32 ret; 745 746 migrate_disable(); 747 ret = bpf_prog_run(prog, ctx); 748 migrate_enable(); 749 return ret; 750 } 751 752 static inline bool is_stack_arg_ldx(const struct bpf_insn *insn) 753 { 754 return insn->code == (BPF_LDX | BPF_MEM | BPF_DW) && 755 insn->src_reg == BPF_REG_PARAMS && 756 insn->off > 0 && insn->off % 8 == 0; 757 } 758 759 static inline bool is_stack_arg_st(const struct bpf_insn *insn) 760 { 761 return insn->code == (BPF_ST | BPF_MEM | BPF_DW) && 762 insn->dst_reg == BPF_REG_PARAMS && 763 insn->off < 0 && insn->off % 8 == 0; 764 } 765 766 static inline bool is_stack_arg_stx(const struct bpf_insn *insn) 767 { 768 return insn->code == (BPF_STX | BPF_MEM | BPF_DW) && 769 insn->dst_reg == BPF_REG_PARAMS && 770 insn->off < 0 && insn->off % 8 == 0; 771 } 772 773 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN 774 775 struct bpf_skb_data_end { 776 struct qdisc_skb_cb qdisc_cb; 777 void *data_meta; 778 void *data_end; 779 }; 780 781 struct bpf_nh_params { 782 u32 nh_family; 783 union { 784 u32 ipv4_nh; 785 struct in6_addr ipv6_nh; 786 }; 787 }; 788 789 /* flags for bpf_redirect_info kern_flags */ 790 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ 791 #define BPF_RI_F_RI_INIT BIT(1) 792 #define BPF_RI_F_CPU_MAP_INIT BIT(2) 793 #define BPF_RI_F_DEV_MAP_INIT BIT(3) 794 #define BPF_RI_F_XSK_MAP_INIT BIT(4) 795 796 struct bpf_redirect_info { 797 u64 tgt_index; 798 void *tgt_value; 799 struct bpf_map *map; 800 u32 flags; 801 u32 map_id; 802 enum bpf_map_type map_type; 803 struct bpf_nh_params nh; 804 u32 kern_flags; 805 }; 806 807 struct bpf_net_context { 808 struct bpf_redirect_info ri; 809 struct list_head cpu_map_flush_list; 810 struct list_head dev_map_flush_list; 811 struct list_head xskmap_map_flush_list; 812 }; 813 814 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx) 815 { 816 struct task_struct *tsk = current; 817 818 if (tsk->bpf_net_context != NULL) 819 return NULL; 820 bpf_net_ctx->ri.kern_flags = 0; 821 822 tsk->bpf_net_context = bpf_net_ctx; 823 return bpf_net_ctx; 824 } 825 826 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx) 827 { 828 if (bpf_net_ctx) 829 current->bpf_net_context = NULL; 830 } 831 832 static inline struct bpf_net_context *bpf_net_ctx_get(void) 833 { 834 return current->bpf_net_context; 835 } 836 837 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void) 838 { 839 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 840 841 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) { 842 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh)); 843 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT; 844 } 845 846 return &bpf_net_ctx->ri; 847 } 848 849 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void) 850 { 851 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 852 853 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) { 854 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list); 855 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT; 856 } 857 858 return &bpf_net_ctx->cpu_map_flush_list; 859 } 860 861 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void) 862 { 863 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 864 865 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) { 866 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list); 867 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT; 868 } 869 870 return &bpf_net_ctx->dev_map_flush_list; 871 } 872 873 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void) 874 { 875 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 876 877 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) { 878 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list); 879 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT; 880 } 881 882 return &bpf_net_ctx->xskmap_map_flush_list; 883 } 884 885 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map, 886 struct list_head **lh_dev, 887 struct list_head **lh_xsk) 888 { 889 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); 890 u32 kern_flags = bpf_net_ctx->ri.kern_flags; 891 struct list_head *lh; 892 893 *lh_map = *lh_dev = *lh_xsk = NULL; 894 895 if (!IS_ENABLED(CONFIG_BPF_SYSCALL)) 896 return; 897 898 lh = &bpf_net_ctx->dev_map_flush_list; 899 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh)) 900 *lh_dev = lh; 901 902 lh = &bpf_net_ctx->cpu_map_flush_list; 903 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh)) 904 *lh_map = lh; 905 906 lh = &bpf_net_ctx->xskmap_map_flush_list; 907 if (IS_ENABLED(CONFIG_XDP_SOCKETS) && 908 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh)) 909 *lh_xsk = lh; 910 } 911 912 /* Compute the linear packet data range [data, data_end) which 913 * will be accessed by various program types (cls_bpf, act_bpf, 914 * lwt, ...). Subsystems allowing direct data access must (!) 915 * ensure that cb[] area can be written to when BPF program is 916 * invoked (otherwise cb[] save/restore is necessary). 917 */ 918 static inline void bpf_compute_data_pointers(struct sk_buff *skb) 919 { 920 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 921 922 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); 923 cb->data_meta = skb->data - skb_metadata_len(skb); 924 cb->data_end = skb->data + skb_headlen(skb); 925 } 926 927 static inline int bpf_prog_run_data_pointers( 928 const struct bpf_prog *prog, 929 struct sk_buff *skb) 930 { 931 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 932 void *save_data_meta, *save_data_end; 933 int res; 934 935 save_data_meta = cb->data_meta; 936 save_data_end = cb->data_end; 937 938 bpf_compute_data_pointers(skb); 939 res = bpf_prog_run(prog, skb); 940 941 cb->data_meta = save_data_meta; 942 cb->data_end = save_data_end; 943 944 return res; 945 } 946 947 /* Similar to bpf_compute_data_pointers(), except that save orginal 948 * data in cb->data and cb->meta_data for restore. 949 */ 950 static inline void bpf_compute_and_save_data_end( 951 struct sk_buff *skb, void **saved_data_end) 952 { 953 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 954 955 *saved_data_end = cb->data_end; 956 cb->data_end = skb->data + skb_headlen(skb); 957 } 958 959 /* Restore data saved by bpf_compute_and_save_data_end(). */ 960 static inline void bpf_restore_data_end( 961 struct sk_buff *skb, void *saved_data_end) 962 { 963 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; 964 965 cb->data_end = saved_data_end; 966 } 967 968 static inline u8 *bpf_skb_cb(const struct sk_buff *skb) 969 { 970 /* eBPF programs may read/write skb->cb[] area to transfer meta 971 * data between tail calls. Since this also needs to work with 972 * tc, that scratch memory is mapped to qdisc_skb_cb's data area. 973 * 974 * In some socket filter cases, the cb unfortunately needs to be 975 * saved/restored so that protocol specific skb->cb[] data won't 976 * be lost. In any case, due to unpriviledged eBPF programs 977 * attached to sockets, we need to clear the bpf_skb_cb() area 978 * to not leak previous contents to user space. 979 */ 980 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); 981 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != 982 sizeof_field(struct qdisc_skb_cb, data)); 983 984 return qdisc_skb_cb(skb)->data; 985 } 986 987 /* Must be invoked with migration disabled */ 988 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, 989 const void *ctx) 990 { 991 const struct sk_buff *skb = ctx; 992 u8 *cb_data = bpf_skb_cb(skb); 993 u8 cb_saved[BPF_SKB_CB_LEN]; 994 u32 res; 995 996 if (unlikely(prog->cb_access)) { 997 memcpy(cb_saved, cb_data, sizeof(cb_saved)); 998 memset(cb_data, 0, sizeof(cb_saved)); 999 } 1000 1001 res = bpf_prog_run(prog, skb); 1002 1003 if (unlikely(prog->cb_access)) 1004 memcpy(cb_data, cb_saved, sizeof(cb_saved)); 1005 1006 return res; 1007 } 1008 1009 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, 1010 struct sk_buff *skb) 1011 { 1012 u32 res; 1013 1014 migrate_disable(); 1015 res = __bpf_prog_run_save_cb(prog, skb); 1016 migrate_enable(); 1017 return res; 1018 } 1019 1020 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, 1021 struct sk_buff *skb) 1022 { 1023 u8 *cb_data = bpf_skb_cb(skb); 1024 u32 res; 1025 1026 if (unlikely(prog->cb_access)) 1027 memset(cb_data, 0, BPF_SKB_CB_LEN); 1028 1029 res = bpf_prog_run_pin_on_cpu(prog, skb); 1030 return res; 1031 } 1032 1033 DECLARE_BPF_DISPATCHER(xdp) 1034 1035 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 1036 1037 u32 xdp_master_redirect(struct xdp_buff *xdp); 1038 1039 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); 1040 1041 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) 1042 { 1043 return prog->len * sizeof(struct bpf_insn); 1044 } 1045 1046 static inline unsigned int bpf_prog_size(unsigned int proglen) 1047 { 1048 return max(sizeof(struct bpf_prog), 1049 offsetof(struct bpf_prog, insns[proglen])); 1050 } 1051 1052 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) 1053 { 1054 /* When classic BPF programs have been loaded and the arch 1055 * does not have a classic BPF JIT (anymore), they have been 1056 * converted via bpf_migrate_filter() to eBPF and thus always 1057 * have an unspec program type. 1058 */ 1059 return prog->type == BPF_PROG_TYPE_UNSPEC; 1060 } 1061 1062 static inline u32 bpf_ctx_off_adjust_machine(u32 size) 1063 { 1064 const u32 size_machine = sizeof(unsigned long); 1065 1066 if (size > size_machine && size % size_machine == 0) 1067 size = size_machine; 1068 1069 return size; 1070 } 1071 1072 static inline bool 1073 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) 1074 { 1075 return size <= size_default && (size & (size - 1)) == 0; 1076 } 1077 1078 static inline u8 1079 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) 1080 { 1081 u8 access_off = off & (size_default - 1); 1082 1083 #ifdef __LITTLE_ENDIAN 1084 return access_off; 1085 #else 1086 return size_default - (access_off + size); 1087 #endif 1088 } 1089 1090 #define bpf_ctx_wide_access_ok(off, size, type, field) \ 1091 (size == sizeof(__u64) && \ 1092 off >= offsetof(type, field) && \ 1093 off + sizeof(__u64) <= offsetofend(type, field) && \ 1094 off % sizeof(__u64) == 0) 1095 1096 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) 1097 1098 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) 1099 { 1100 #ifndef CONFIG_BPF_JIT_ALWAYS_ON 1101 if (!fp->jited) { 1102 set_vm_flush_reset_perms(fp); 1103 return set_memory_ro((unsigned long)fp, fp->pages); 1104 } 1105 #endif 1106 return 0; 1107 } 1108 1109 static inline int __must_check 1110 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) 1111 { 1112 set_vm_flush_reset_perms(hdr); 1113 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT); 1114 } 1115 1116 enum skb_drop_reason 1117 sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap); 1118 1119 static inline int sk_filter(struct sock *sk, struct sk_buff *skb) 1120 { 1121 enum skb_drop_reason drop_reason; 1122 1123 drop_reason = sk_filter_trim_cap(sk, skb, 1); 1124 return drop_reason ? -EPERM : 0; 1125 } 1126 1127 static inline enum skb_drop_reason 1128 sk_filter_reason(struct sock *sk, struct sk_buff *skb) 1129 { 1130 return sk_filter_trim_cap(sk, skb, 1); 1131 } 1132 1133 struct bpf_prog *__bpf_prog_select_runtime(struct bpf_verifier_env *env, struct bpf_prog *fp, 1134 int *err); 1135 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); 1136 void bpf_prog_free(struct bpf_prog *fp); 1137 1138 bool bpf_opcode_in_insntable(u8 code); 1139 1140 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, 1141 const u32 *insn_to_jit_off); 1142 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); 1143 void bpf_prog_jit_attempt_done(struct bpf_prog *prog); 1144 1145 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags); 1146 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags); 1147 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, 1148 gfp_t gfp_extra_flags); 1149 void __bpf_prog_free(struct bpf_prog *fp); 1150 1151 static inline void bpf_prog_unlock_free(struct bpf_prog *fp) 1152 { 1153 __bpf_prog_free(fp); 1154 } 1155 1156 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, 1157 unsigned int flen); 1158 1159 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); 1160 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1161 bpf_aux_classic_check_t trans, bool save_orig); 1162 void bpf_prog_destroy(struct bpf_prog *fp); 1163 1164 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1165 int sk_attach_bpf(u32 ufd, struct sock *sk); 1166 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); 1167 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); 1168 void sk_reuseport_prog_free(struct bpf_prog *prog); 1169 int sk_detach_filter(struct sock *sk); 1170 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); 1171 1172 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); 1173 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); 1174 1175 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1176 1177 struct bpf_prog *bpf_int_jit_compile(struct bpf_verifier_env *env, struct bpf_prog *prog); 1178 void bpf_jit_compile(struct bpf_prog *prog); 1179 bool bpf_jit_needs_zext(void); 1180 bool bpf_jit_inlines_helper_call(s32 imm); 1181 bool bpf_jit_supports_subprog_tailcalls(void); 1182 bool bpf_jit_supports_percpu_insn(void); 1183 bool bpf_jit_supports_kfunc_call(void); 1184 bool bpf_jit_supports_stack_args(void); 1185 bool bpf_jit_supports_far_kfunc_call(void); 1186 bool bpf_jit_supports_exceptions(void); 1187 bool bpf_jit_supports_ptr_xchg(void); 1188 bool bpf_jit_supports_arena(void); 1189 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena); 1190 bool bpf_jit_supports_private_stack(void); 1191 bool bpf_jit_supports_timed_may_goto(void); 1192 bool bpf_jit_supports_fsession(void); 1193 u64 bpf_arch_uaddress_limit(void); 1194 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); 1195 u64 arch_bpf_timed_may_goto(void); 1196 u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *); 1197 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id); 1198 1199 static inline bool bpf_dump_raw_ok(const struct cred *cred) 1200 { 1201 /* Reconstruction of call-sites is dependent on kallsyms, 1202 * thus make dump the same restriction. 1203 */ 1204 return kallsyms_show_value(cred); 1205 } 1206 1207 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, 1208 const struct bpf_insn *patch, u32 len); 1209 1210 #ifdef CONFIG_BPF_SYSCALL 1211 struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, 1212 const struct bpf_insn *patch, u32 len); 1213 struct bpf_insn_aux_data *bpf_dup_insn_aux_data(struct bpf_verifier_env *env); 1214 void bpf_restore_insn_aux_data(struct bpf_verifier_env *env, 1215 struct bpf_insn_aux_data *orig_insn_aux); 1216 #else 1217 static inline struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, 1218 const struct bpf_insn *patch, u32 len) 1219 { 1220 return ERR_PTR(-ENOTSUPP); 1221 } 1222 1223 static inline struct bpf_insn_aux_data *bpf_dup_insn_aux_data(struct bpf_verifier_env *env) 1224 { 1225 return NULL; 1226 } 1227 1228 static inline void bpf_restore_insn_aux_data(struct bpf_verifier_env *env, 1229 struct bpf_insn_aux_data *orig_insn_aux) 1230 { 1231 } 1232 #endif /* CONFIG_BPF_SYSCALL */ 1233 1234 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); 1235 1236 static inline bool xdp_return_frame_no_direct(void) 1237 { 1238 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1239 1240 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; 1241 } 1242 1243 static inline void xdp_set_return_frame_no_direct(void) 1244 { 1245 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1246 1247 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; 1248 } 1249 1250 static inline void xdp_clear_return_frame_no_direct(void) 1251 { 1252 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1253 1254 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; 1255 } 1256 1257 static inline int xdp_ok_fwd_dev(const struct net_device *fwd, 1258 unsigned int pktlen) 1259 { 1260 unsigned int len; 1261 1262 if (unlikely(!(fwd->flags & IFF_UP))) 1263 return -ENETDOWN; 1264 1265 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; 1266 if (pktlen > len) 1267 return -EMSGSIZE; 1268 1269 return 0; 1270 } 1271 1272 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the 1273 * same cpu context. Further for best results no more than a single map 1274 * for the do_redirect/do_flush pair should be used. This limitation is 1275 * because we only track one map and force a flush when the map changes. 1276 * This does not appear to be a real limitation for existing software. 1277 */ 1278 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 1279 struct xdp_buff *xdp, const struct bpf_prog *prog); 1280 int xdp_do_redirect(struct net_device *dev, 1281 struct xdp_buff *xdp, 1282 const struct bpf_prog *prog); 1283 int xdp_do_redirect_frame(struct net_device *dev, 1284 struct xdp_buff *xdp, 1285 struct xdp_frame *xdpf, 1286 const struct bpf_prog *prog); 1287 void xdp_do_flush(void); 1288 1289 void bpf_warn_invalid_xdp_action(const struct net_device *dev, 1290 const struct bpf_prog *prog, u32 act); 1291 1292 #ifdef CONFIG_INET 1293 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1294 struct bpf_prog *prog, struct sk_buff *skb, 1295 struct sock *migrating_sk, 1296 u32 hash); 1297 #else 1298 static inline struct sock * 1299 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 1300 struct bpf_prog *prog, struct sk_buff *skb, 1301 struct sock *migrating_sk, 1302 u32 hash) 1303 { 1304 return NULL; 1305 } 1306 #endif 1307 1308 #ifdef CONFIG_BPF_JIT 1309 extern int bpf_jit_enable; 1310 extern int bpf_jit_harden; 1311 extern int bpf_jit_kallsyms; 1312 extern long bpf_jit_limit; 1313 extern long bpf_jit_limit_max; 1314 1315 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); 1316 1317 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); 1318 1319 struct bpf_binary_header * 1320 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, 1321 unsigned int alignment, 1322 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1323 void bpf_jit_binary_free(struct bpf_binary_header *hdr); 1324 u64 bpf_jit_alloc_exec_limit(void); 1325 void *bpf_jit_alloc_exec(unsigned long size); 1326 void bpf_jit_free_exec(void *addr); 1327 void bpf_jit_free(struct bpf_prog *fp); 1328 struct bpf_binary_header * 1329 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); 1330 1331 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); 1332 void bpf_prog_pack_free(void *ptr, u32 size); 1333 1334 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) 1335 { 1336 return list_empty(&fp->aux->ksym.lnode) || 1337 fp->aux->ksym.lnode.prev == LIST_POISON2; 1338 } 1339 1340 struct bpf_binary_header * 1341 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, 1342 unsigned int alignment, 1343 struct bpf_binary_header **rw_hdr, 1344 u8 **rw_image, 1345 bpf_jit_fill_hole_t bpf_fill_ill_insns); 1346 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header, 1347 struct bpf_binary_header *rw_header); 1348 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, 1349 struct bpf_binary_header *rw_header); 1350 1351 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1352 struct bpf_jit_poke_descriptor *poke); 1353 1354 int bpf_jit_get_func_addr(const struct bpf_prog *prog, 1355 const struct bpf_insn *insn, bool extra_pass, 1356 u64 *func_addr, bool *func_addr_fixed); 1357 1358 const char *bpf_jit_get_prog_name(struct bpf_prog *prog); 1359 1360 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog); 1361 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); 1362 1363 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog) 1364 { 1365 return prog->blinding_requested && !prog->blinded; 1366 } 1367 1368 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, 1369 u32 pass, void *image) 1370 { 1371 pr_err("flen=%u proglen=%u pass=%u image=%p from=%s pid=%d\n", flen, 1372 proglen, pass, image, current->comm, task_pid_nr(current)); 1373 1374 if (image) 1375 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET, 1376 16, 1, image, proglen, false); 1377 } 1378 1379 static inline bool bpf_jit_is_ebpf(void) 1380 { 1381 # ifdef CONFIG_HAVE_EBPF_JIT 1382 return true; 1383 # else 1384 return false; 1385 # endif 1386 } 1387 1388 static inline bool ebpf_jit_enabled(void) 1389 { 1390 return bpf_jit_enable && bpf_jit_is_ebpf(); 1391 } 1392 1393 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1394 { 1395 return fp->jited && bpf_jit_is_ebpf(); 1396 } 1397 1398 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1399 { 1400 /* These are the prerequisites, should someone ever have the 1401 * idea to call blinding outside of them, we make sure to 1402 * bail out. 1403 */ 1404 if (!bpf_jit_is_ebpf()) 1405 return false; 1406 if (!prog->jit_requested) 1407 return false; 1408 if (!bpf_jit_harden) 1409 return false; 1410 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF)) 1411 return false; 1412 1413 return true; 1414 } 1415 1416 static inline bool bpf_jit_kallsyms_enabled(void) 1417 { 1418 /* There are a couple of corner cases where kallsyms should 1419 * not be enabled f.e. on hardening. 1420 */ 1421 if (bpf_jit_harden) 1422 return false; 1423 if (!bpf_jit_kallsyms) 1424 return false; 1425 if (bpf_jit_kallsyms == 1) 1426 return true; 1427 1428 return false; 1429 } 1430 1431 int bpf_address_lookup(unsigned long addr, unsigned long *size, 1432 unsigned long *off, char *sym); 1433 bool is_bpf_text_address(unsigned long addr); 1434 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, 1435 char *sym); 1436 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); 1437 1438 void bpf_prog_kallsyms_add(struct bpf_prog *fp); 1439 void bpf_prog_kallsyms_del(struct bpf_prog *fp); 1440 1441 #else /* CONFIG_BPF_JIT */ 1442 1443 static inline bool ebpf_jit_enabled(void) 1444 { 1445 return false; 1446 } 1447 1448 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) 1449 { 1450 return false; 1451 } 1452 1453 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) 1454 { 1455 return false; 1456 } 1457 1458 static inline int 1459 bpf_jit_add_poke_descriptor(struct bpf_prog *prog, 1460 struct bpf_jit_poke_descriptor *poke) 1461 { 1462 return -ENOTSUPP; 1463 } 1464 1465 static inline void bpf_jit_free(struct bpf_prog *fp) 1466 { 1467 bpf_prog_unlock_free(fp); 1468 } 1469 1470 static inline bool bpf_jit_kallsyms_enabled(void) 1471 { 1472 return false; 1473 } 1474 1475 static inline int 1476 bpf_address_lookup(unsigned long addr, unsigned long *size, 1477 unsigned long *off, char *sym) 1478 { 1479 return 0; 1480 } 1481 1482 static inline bool is_bpf_text_address(unsigned long addr) 1483 { 1484 return false; 1485 } 1486 1487 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, 1488 char *type, char *sym) 1489 { 1490 return -ERANGE; 1491 } 1492 1493 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) 1494 { 1495 return NULL; 1496 } 1497 1498 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) 1499 { 1500 } 1501 1502 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) 1503 { 1504 } 1505 1506 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog) 1507 { 1508 return false; 1509 } 1510 1511 static inline 1512 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog) 1513 { 1514 return prog; 1515 } 1516 1517 static inline void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other) 1518 { 1519 } 1520 #endif /* CONFIG_BPF_JIT */ 1521 1522 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); 1523 1524 #define BPF_ANC BIT(15) 1525 1526 static inline bool bpf_needs_clear_a(const struct sock_filter *first) 1527 { 1528 switch (first->code) { 1529 case BPF_RET | BPF_K: 1530 case BPF_LD | BPF_W | BPF_LEN: 1531 return false; 1532 1533 case BPF_LD | BPF_W | BPF_ABS: 1534 case BPF_LD | BPF_H | BPF_ABS: 1535 case BPF_LD | BPF_B | BPF_ABS: 1536 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) 1537 return true; 1538 return false; 1539 1540 default: 1541 return true; 1542 } 1543 } 1544 1545 static inline u16 bpf_anc_helper(const struct sock_filter *ftest) 1546 { 1547 BUG_ON(ftest->code & BPF_ANC); 1548 1549 switch (ftest->code) { 1550 case BPF_LD | BPF_W | BPF_ABS: 1551 case BPF_LD | BPF_H | BPF_ABS: 1552 case BPF_LD | BPF_B | BPF_ABS: 1553 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ 1554 return BPF_ANC | SKF_AD_##CODE 1555 switch (ftest->k) { 1556 BPF_ANCILLARY(PROTOCOL); 1557 BPF_ANCILLARY(PKTTYPE); 1558 BPF_ANCILLARY(IFINDEX); 1559 BPF_ANCILLARY(NLATTR); 1560 BPF_ANCILLARY(NLATTR_NEST); 1561 BPF_ANCILLARY(MARK); 1562 BPF_ANCILLARY(QUEUE); 1563 BPF_ANCILLARY(HATYPE); 1564 BPF_ANCILLARY(RXHASH); 1565 BPF_ANCILLARY(CPU); 1566 BPF_ANCILLARY(ALU_XOR_X); 1567 BPF_ANCILLARY(VLAN_TAG); 1568 BPF_ANCILLARY(VLAN_TAG_PRESENT); 1569 BPF_ANCILLARY(PAY_OFFSET); 1570 BPF_ANCILLARY(RANDOM); 1571 BPF_ANCILLARY(VLAN_TPID); 1572 } 1573 fallthrough; 1574 default: 1575 return ftest->code; 1576 } 1577 } 1578 1579 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, 1580 int k, unsigned int size); 1581 1582 static inline int bpf_tell_extensions(void) 1583 { 1584 return SKF_AD_MAX; 1585 } 1586 1587 struct bpf_sock_addr_kern { 1588 struct sock *sk; 1589 struct sockaddr_unsized *uaddr; 1590 /* Temporary "register" to make indirect stores to nested structures 1591 * defined above. We need three registers to make such a store, but 1592 * only two (src and dst) are available at convert_ctx_access time 1593 */ 1594 u64 tmp_reg; 1595 void *t_ctx; /* Attach type specific context. */ 1596 u32 uaddrlen; 1597 }; 1598 1599 struct bpf_sock_ops_kern { 1600 struct sock *sk; 1601 union { 1602 u32 args[4]; 1603 u32 reply; 1604 u32 replylong[4]; 1605 }; 1606 struct sk_buff *syn_skb; 1607 struct sk_buff *skb; 1608 void *skb_data_end; 1609 u8 op; 1610 u8 is_fullsock; 1611 u8 is_locked_tcp_sock; 1612 u8 remaining_opt_len; 1613 u64 temp; /* temp and everything after is not 1614 * initialized to 0 before calling 1615 * the BPF program. New fields that 1616 * should be initialized to 0 should 1617 * be inserted before temp. 1618 * temp is scratch storage used by 1619 * sock_ops_convert_ctx_access 1620 * as temporary storage of a register. 1621 */ 1622 }; 1623 1624 struct bpf_sysctl_kern { 1625 struct ctl_table_header *head; 1626 const struct ctl_table *table; 1627 void *cur_val; 1628 size_t cur_len; 1629 void *new_val; 1630 size_t new_len; 1631 int new_updated; 1632 int write; 1633 loff_t *ppos; 1634 /* Temporary "register" for indirect stores to ppos. */ 1635 u64 tmp_reg; 1636 }; 1637 1638 #define BPF_SOCKOPT_KERN_BUF_SIZE 32 1639 struct bpf_sockopt_buf { 1640 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; 1641 }; 1642 1643 struct bpf_sockopt_kern { 1644 struct sock *sk; 1645 u8 *optval; 1646 u8 *optval_end; 1647 s32 level; 1648 s32 optname; 1649 s32 optlen; 1650 /* for retval in struct bpf_cg_run_ctx */ 1651 struct task_struct *current_task; 1652 /* Temporary "register" for indirect stores to ppos. */ 1653 u64 tmp_reg; 1654 }; 1655 1656 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); 1657 1658 struct bpf_sk_lookup_kern { 1659 u16 family; 1660 u16 protocol; 1661 __be16 sport; 1662 u16 dport; 1663 struct { 1664 __be32 saddr; 1665 __be32 daddr; 1666 } v4; 1667 struct { 1668 const struct in6_addr *saddr; 1669 const struct in6_addr *daddr; 1670 } v6; 1671 struct sock *selected_sk; 1672 u32 ingress_ifindex; 1673 bool no_reuseport; 1674 }; 1675 1676 extern struct static_key_false bpf_sk_lookup_enabled; 1677 1678 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. 1679 * 1680 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and 1681 * SK_DROP. Their meaning is as follows: 1682 * 1683 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result 1684 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup 1685 * SK_DROP : terminate lookup with -ECONNREFUSED 1686 * 1687 * This macro aggregates return values and selected sockets from 1688 * multiple BPF programs according to following rules in order: 1689 * 1690 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, 1691 * macro result is SK_PASS and last ctx.selected_sk is used. 1692 * 2. If any program returned SK_DROP return value, 1693 * macro result is SK_DROP. 1694 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. 1695 * 1696 * Caller must ensure that the prog array is non-NULL, and that the 1697 * array as well as the programs it contains remain valid. 1698 */ 1699 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ 1700 ({ \ 1701 struct bpf_sk_lookup_kern *_ctx = &(ctx); \ 1702 struct bpf_prog_array_item *_item; \ 1703 struct sock *_selected_sk = NULL; \ 1704 bool _no_reuseport = false; \ 1705 struct bpf_prog *_prog; \ 1706 bool _all_pass = true; \ 1707 u32 _ret; \ 1708 \ 1709 migrate_disable(); \ 1710 _item = &(array)->items[0]; \ 1711 while ((_prog = READ_ONCE(_item->prog))) { \ 1712 /* restore most recent selection */ \ 1713 _ctx->selected_sk = _selected_sk; \ 1714 _ctx->no_reuseport = _no_reuseport; \ 1715 \ 1716 _ret = func(_prog, _ctx); \ 1717 if (_ret == SK_PASS && _ctx->selected_sk) { \ 1718 /* remember last non-NULL socket */ \ 1719 _selected_sk = _ctx->selected_sk; \ 1720 _no_reuseport = _ctx->no_reuseport; \ 1721 } else if (_ret == SK_DROP && _all_pass) { \ 1722 _all_pass = false; \ 1723 } \ 1724 _item++; \ 1725 } \ 1726 _ctx->selected_sk = _selected_sk; \ 1727 _ctx->no_reuseport = _no_reuseport; \ 1728 migrate_enable(); \ 1729 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ 1730 }) 1731 1732 static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol, 1733 const __be32 saddr, const __be16 sport, 1734 const __be32 daddr, const u16 dport, 1735 const int ifindex, struct sock **psk) 1736 { 1737 struct bpf_prog_array *run_array; 1738 struct sock *selected_sk = NULL; 1739 bool no_reuseport = false; 1740 1741 rcu_read_lock(); 1742 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1743 if (run_array) { 1744 struct bpf_sk_lookup_kern ctx = { 1745 .family = AF_INET, 1746 .protocol = protocol, 1747 .v4.saddr = saddr, 1748 .v4.daddr = daddr, 1749 .sport = sport, 1750 .dport = dport, 1751 .ingress_ifindex = ifindex, 1752 }; 1753 u32 act; 1754 1755 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1756 if (act == SK_PASS) { 1757 selected_sk = ctx.selected_sk; 1758 no_reuseport = ctx.no_reuseport; 1759 } else { 1760 selected_sk = ERR_PTR(-ECONNREFUSED); 1761 } 1762 } 1763 rcu_read_unlock(); 1764 *psk = selected_sk; 1765 return no_reuseport; 1766 } 1767 1768 #if IS_ENABLED(CONFIG_IPV6) 1769 static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol, 1770 const struct in6_addr *saddr, 1771 const __be16 sport, 1772 const struct in6_addr *daddr, 1773 const u16 dport, 1774 const int ifindex, struct sock **psk) 1775 { 1776 struct bpf_prog_array *run_array; 1777 struct sock *selected_sk = NULL; 1778 bool no_reuseport = false; 1779 1780 rcu_read_lock(); 1781 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); 1782 if (run_array) { 1783 struct bpf_sk_lookup_kern ctx = { 1784 .family = AF_INET6, 1785 .protocol = protocol, 1786 .v6.saddr = saddr, 1787 .v6.daddr = daddr, 1788 .sport = sport, 1789 .dport = dport, 1790 .ingress_ifindex = ifindex, 1791 }; 1792 u32 act; 1793 1794 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); 1795 if (act == SK_PASS) { 1796 selected_sk = ctx.selected_sk; 1797 no_reuseport = ctx.no_reuseport; 1798 } else { 1799 selected_sk = ERR_PTR(-ECONNREFUSED); 1800 } 1801 } 1802 rcu_read_unlock(); 1803 *psk = selected_sk; 1804 return no_reuseport; 1805 } 1806 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1807 1808 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, 1809 u64 flags, const u64 flag_mask, 1810 void *lookup_elem(struct bpf_map *map, u32 key)) 1811 { 1812 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 1813 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; 1814 1815 /* Lower bits of the flags are used as return code on lookup failure */ 1816 if (unlikely(flags & ~(action_mask | flag_mask))) 1817 return XDP_ABORTED; 1818 1819 ri->tgt_value = lookup_elem(map, index); 1820 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { 1821 /* If the lookup fails we want to clear out the state in the 1822 * redirect_info struct completely, so that if an eBPF program 1823 * performs multiple lookups, the last one always takes 1824 * precedence. 1825 */ 1826 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ 1827 ri->map_type = BPF_MAP_TYPE_UNSPEC; 1828 return flags & action_mask; 1829 } 1830 1831 ri->tgt_index = index; 1832 ri->map_id = map->id; 1833 ri->map_type = map->map_type; 1834 1835 if (flags & BPF_F_BROADCAST) { 1836 WRITE_ONCE(ri->map, map); 1837 ri->flags = flags; 1838 } else { 1839 WRITE_ONCE(ri->map, NULL); 1840 ri->flags = 0; 1841 } 1842 1843 return XDP_REDIRECT; 1844 } 1845 1846 #ifdef CONFIG_NET 1847 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); 1848 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1849 u32 len, u64 flags); 1850 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1851 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); 1852 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); 1853 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 1854 void *buf, unsigned long len, bool flush); 1855 int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset, 1856 const void *from, u32 len, u64 flags); 1857 void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset); 1858 #else /* CONFIG_NET */ 1859 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, 1860 void *to, u32 len) 1861 { 1862 return -EOPNOTSUPP; 1863 } 1864 1865 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, 1866 const void *from, u32 len, u64 flags) 1867 { 1868 return -EOPNOTSUPP; 1869 } 1870 1871 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, 1872 void *buf, u32 len) 1873 { 1874 return -EOPNOTSUPP; 1875 } 1876 1877 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, 1878 void *buf, u32 len) 1879 { 1880 return -EOPNOTSUPP; 1881 } 1882 1883 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 1884 { 1885 return NULL; 1886 } 1887 1888 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, 1889 unsigned long len, bool flush) 1890 { 1891 } 1892 1893 static inline int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset, 1894 const void *from, u32 len, 1895 u64 flags) 1896 { 1897 return -EOPNOTSUPP; 1898 } 1899 1900 static inline void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset) 1901 { 1902 return ERR_PTR(-EOPNOTSUPP); 1903 } 1904 #endif /* CONFIG_NET */ 1905 1906 #endif /* __LINUX_FILTER_H__ */ 1907