| 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 ; |
| 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 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 { |
| 666 | u32 ; |
| 667 | u8 [] __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(syncp: &stats->syncp); |
| 718 | u64_stats_inc(p: &stats->cnt); |
| 719 | u64_stats_add(p: &stats->nsecs, val: duration); |
| 720 | u64_stats_update_end_irqrestore(syncp: &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, dfunc: 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 | #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN |
| 753 | |
| 754 | struct bpf_skb_data_end { |
| 755 | struct qdisc_skb_cb qdisc_cb; |
| 756 | void *data_meta; |
| 757 | void *data_end; |
| 758 | }; |
| 759 | |
| 760 | struct bpf_nh_params { |
| 761 | u32 nh_family; |
| 762 | union { |
| 763 | u32 ipv4_nh; |
| 764 | struct in6_addr ipv6_nh; |
| 765 | }; |
| 766 | }; |
| 767 | |
| 768 | /* flags for bpf_redirect_info kern_flags */ |
| 769 | #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */ |
| 770 | #define BPF_RI_F_RI_INIT BIT(1) |
| 771 | #define BPF_RI_F_CPU_MAP_INIT BIT(2) |
| 772 | #define BPF_RI_F_DEV_MAP_INIT BIT(3) |
| 773 | #define BPF_RI_F_XSK_MAP_INIT BIT(4) |
| 774 | |
| 775 | struct bpf_redirect_info { |
| 776 | u64 tgt_index; |
| 777 | void *tgt_value; |
| 778 | struct bpf_map *map; |
| 779 | u32 flags; |
| 780 | u32 map_id; |
| 781 | enum bpf_map_type map_type; |
| 782 | struct bpf_nh_params nh; |
| 783 | u32 kern_flags; |
| 784 | }; |
| 785 | |
| 786 | struct bpf_net_context { |
| 787 | struct bpf_redirect_info ri; |
| 788 | struct list_head cpu_map_flush_list; |
| 789 | struct list_head dev_map_flush_list; |
| 790 | struct list_head xskmap_map_flush_list; |
| 791 | }; |
| 792 | |
| 793 | static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx) |
| 794 | { |
| 795 | struct task_struct *tsk = current; |
| 796 | |
| 797 | if (tsk->bpf_net_context != NULL) |
| 798 | return NULL; |
| 799 | bpf_net_ctx->ri.kern_flags = 0; |
| 800 | |
| 801 | tsk->bpf_net_context = bpf_net_ctx; |
| 802 | return bpf_net_ctx; |
| 803 | } |
| 804 | |
| 805 | static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx) |
| 806 | { |
| 807 | if (bpf_net_ctx) |
| 808 | current->bpf_net_context = NULL; |
| 809 | } |
| 810 | |
| 811 | static inline struct bpf_net_context *bpf_net_ctx_get(void) |
| 812 | { |
| 813 | return current->bpf_net_context; |
| 814 | } |
| 815 | |
| 816 | static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void) |
| 817 | { |
| 818 | struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); |
| 819 | |
| 820 | if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) { |
| 821 | memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh)); |
| 822 | bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT; |
| 823 | } |
| 824 | |
| 825 | return &bpf_net_ctx->ri; |
| 826 | } |
| 827 | |
| 828 | static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void) |
| 829 | { |
| 830 | struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); |
| 831 | |
| 832 | if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) { |
| 833 | INIT_LIST_HEAD(list: &bpf_net_ctx->cpu_map_flush_list); |
| 834 | bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT; |
| 835 | } |
| 836 | |
| 837 | return &bpf_net_ctx->cpu_map_flush_list; |
| 838 | } |
| 839 | |
| 840 | static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void) |
| 841 | { |
| 842 | struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); |
| 843 | |
| 844 | if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) { |
| 845 | INIT_LIST_HEAD(list: &bpf_net_ctx->dev_map_flush_list); |
| 846 | bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT; |
| 847 | } |
| 848 | |
| 849 | return &bpf_net_ctx->dev_map_flush_list; |
| 850 | } |
| 851 | |
| 852 | static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void) |
| 853 | { |
| 854 | struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); |
| 855 | |
| 856 | if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) { |
| 857 | INIT_LIST_HEAD(list: &bpf_net_ctx->xskmap_map_flush_list); |
| 858 | bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT; |
| 859 | } |
| 860 | |
| 861 | return &bpf_net_ctx->xskmap_map_flush_list; |
| 862 | } |
| 863 | |
| 864 | static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map, |
| 865 | struct list_head **lh_dev, |
| 866 | struct list_head **lh_xsk) |
| 867 | { |
| 868 | struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get(); |
| 869 | u32 kern_flags = bpf_net_ctx->ri.kern_flags; |
| 870 | struct list_head *lh; |
| 871 | |
| 872 | *lh_map = *lh_dev = *lh_xsk = NULL; |
| 873 | |
| 874 | if (!IS_ENABLED(CONFIG_BPF_SYSCALL)) |
| 875 | return; |
| 876 | |
| 877 | lh = &bpf_net_ctx->dev_map_flush_list; |
| 878 | if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(head: lh)) |
| 879 | *lh_dev = lh; |
| 880 | |
| 881 | lh = &bpf_net_ctx->cpu_map_flush_list; |
| 882 | if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(head: lh)) |
| 883 | *lh_map = lh; |
| 884 | |
| 885 | lh = &bpf_net_ctx->xskmap_map_flush_list; |
| 886 | if (IS_ENABLED(CONFIG_XDP_SOCKETS) && |
| 887 | kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(head: lh)) |
| 888 | *lh_xsk = lh; |
| 889 | } |
| 890 | |
| 891 | /* Compute the linear packet data range [data, data_end) which |
| 892 | * will be accessed by various program types (cls_bpf, act_bpf, |
| 893 | * lwt, ...). Subsystems allowing direct data access must (!) |
| 894 | * ensure that cb[] area can be written to when BPF program is |
| 895 | * invoked (otherwise cb[] save/restore is necessary). |
| 896 | */ |
| 897 | static inline void bpf_compute_data_pointers(struct sk_buff *skb) |
| 898 | { |
| 899 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; |
| 900 | |
| 901 | BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb)); |
| 902 | cb->data_meta = skb->data - skb_metadata_len(skb); |
| 903 | cb->data_end = skb->data + skb_headlen(skb); |
| 904 | } |
| 905 | |
| 906 | static inline int bpf_prog_run_data_pointers( |
| 907 | const struct bpf_prog *prog, |
| 908 | struct sk_buff *skb) |
| 909 | { |
| 910 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; |
| 911 | void *save_data_meta, *save_data_end; |
| 912 | int res; |
| 913 | |
| 914 | save_data_meta = cb->data_meta; |
| 915 | save_data_end = cb->data_end; |
| 916 | |
| 917 | bpf_compute_data_pointers(skb); |
| 918 | res = bpf_prog_run(prog, ctx: skb); |
| 919 | |
| 920 | cb->data_meta = save_data_meta; |
| 921 | cb->data_end = save_data_end; |
| 922 | |
| 923 | return res; |
| 924 | } |
| 925 | |
| 926 | /* Similar to bpf_compute_data_pointers(), except that save orginal |
| 927 | * data in cb->data and cb->meta_data for restore. |
| 928 | */ |
| 929 | static inline void bpf_compute_and_save_data_end( |
| 930 | struct sk_buff *skb, void **saved_data_end) |
| 931 | { |
| 932 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; |
| 933 | |
| 934 | *saved_data_end = cb->data_end; |
| 935 | cb->data_end = skb->data + skb_headlen(skb); |
| 936 | } |
| 937 | |
| 938 | /* Restore data saved by bpf_compute_and_save_data_end(). */ |
| 939 | static inline void bpf_restore_data_end( |
| 940 | struct sk_buff *skb, void *saved_data_end) |
| 941 | { |
| 942 | struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb; |
| 943 | |
| 944 | cb->data_end = saved_data_end; |
| 945 | } |
| 946 | |
| 947 | static inline u8 *bpf_skb_cb(const struct sk_buff *skb) |
| 948 | { |
| 949 | /* eBPF programs may read/write skb->cb[] area to transfer meta |
| 950 | * data between tail calls. Since this also needs to work with |
| 951 | * tc, that scratch memory is mapped to qdisc_skb_cb's data area. |
| 952 | * |
| 953 | * In some socket filter cases, the cb unfortunately needs to be |
| 954 | * saved/restored so that protocol specific skb->cb[] data won't |
| 955 | * be lost. In any case, due to unpriviledged eBPF programs |
| 956 | * attached to sockets, we need to clear the bpf_skb_cb() area |
| 957 | * to not leak previous contents to user space. |
| 958 | */ |
| 959 | BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN); |
| 960 | BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != |
| 961 | sizeof_field(struct qdisc_skb_cb, data)); |
| 962 | |
| 963 | return qdisc_skb_cb(skb)->data; |
| 964 | } |
| 965 | |
| 966 | /* Must be invoked with migration disabled */ |
| 967 | static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog, |
| 968 | const void *ctx) |
| 969 | { |
| 970 | const struct sk_buff *skb = ctx; |
| 971 | u8 *cb_data = bpf_skb_cb(skb); |
| 972 | u8 cb_saved[BPF_SKB_CB_LEN]; |
| 973 | u32 res; |
| 974 | |
| 975 | if (unlikely(prog->cb_access)) { |
| 976 | memcpy(cb_saved, cb_data, sizeof(cb_saved)); |
| 977 | memset(cb_data, 0, sizeof(cb_saved)); |
| 978 | } |
| 979 | |
| 980 | res = bpf_prog_run(prog, ctx: skb); |
| 981 | |
| 982 | if (unlikely(prog->cb_access)) |
| 983 | memcpy(cb_data, cb_saved, sizeof(cb_saved)); |
| 984 | |
| 985 | return res; |
| 986 | } |
| 987 | |
| 988 | static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog, |
| 989 | struct sk_buff *skb) |
| 990 | { |
| 991 | u32 res; |
| 992 | |
| 993 | migrate_disable(); |
| 994 | res = __bpf_prog_run_save_cb(prog, ctx: skb); |
| 995 | migrate_enable(); |
| 996 | return res; |
| 997 | } |
| 998 | |
| 999 | static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog, |
| 1000 | struct sk_buff *skb) |
| 1001 | { |
| 1002 | u8 *cb_data = bpf_skb_cb(skb); |
| 1003 | u32 res; |
| 1004 | |
| 1005 | if (unlikely(prog->cb_access)) |
| 1006 | memset(cb_data, 0, BPF_SKB_CB_LEN); |
| 1007 | |
| 1008 | res = bpf_prog_run_pin_on_cpu(prog, ctx: skb); |
| 1009 | return res; |
| 1010 | } |
| 1011 | |
| 1012 | DECLARE_BPF_DISPATCHER(xdp) |
| 1013 | |
| 1014 | DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); |
| 1015 | |
| 1016 | u32 xdp_master_redirect(struct xdp_buff *xdp); |
| 1017 | |
| 1018 | void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog); |
| 1019 | |
| 1020 | static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog) |
| 1021 | { |
| 1022 | return prog->len * sizeof(struct bpf_insn); |
| 1023 | } |
| 1024 | |
| 1025 | static inline unsigned int bpf_prog_size(unsigned int proglen) |
| 1026 | { |
| 1027 | return max(sizeof(struct bpf_prog), |
| 1028 | offsetof(struct bpf_prog, insns[proglen])); |
| 1029 | } |
| 1030 | |
| 1031 | static inline bool bpf_prog_was_classic(const struct bpf_prog *prog) |
| 1032 | { |
| 1033 | /* When classic BPF programs have been loaded and the arch |
| 1034 | * does not have a classic BPF JIT (anymore), they have been |
| 1035 | * converted via bpf_migrate_filter() to eBPF and thus always |
| 1036 | * have an unspec program type. |
| 1037 | */ |
| 1038 | return prog->type == BPF_PROG_TYPE_UNSPEC; |
| 1039 | } |
| 1040 | |
| 1041 | static inline u32 bpf_ctx_off_adjust_machine(u32 size) |
| 1042 | { |
| 1043 | const u32 size_machine = sizeof(unsigned long); |
| 1044 | |
| 1045 | if (size > size_machine && size % size_machine == 0) |
| 1046 | size = size_machine; |
| 1047 | |
| 1048 | return size; |
| 1049 | } |
| 1050 | |
| 1051 | static inline bool |
| 1052 | bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default) |
| 1053 | { |
| 1054 | return size <= size_default && (size & (size - 1)) == 0; |
| 1055 | } |
| 1056 | |
| 1057 | static inline u8 |
| 1058 | bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default) |
| 1059 | { |
| 1060 | u8 access_off = off & (size_default - 1); |
| 1061 | |
| 1062 | #ifdef __LITTLE_ENDIAN |
| 1063 | return access_off; |
| 1064 | #else |
| 1065 | return size_default - (access_off + size); |
| 1066 | #endif |
| 1067 | } |
| 1068 | |
| 1069 | #define bpf_ctx_wide_access_ok(off, size, type, field) \ |
| 1070 | (size == sizeof(__u64) && \ |
| 1071 | off >= offsetof(type, field) && \ |
| 1072 | off + sizeof(__u64) <= offsetofend(type, field) && \ |
| 1073 | off % sizeof(__u64) == 0) |
| 1074 | |
| 1075 | #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0])) |
| 1076 | |
| 1077 | static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp) |
| 1078 | { |
| 1079 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
| 1080 | if (!fp->jited) { |
| 1081 | set_vm_flush_reset_perms(fp); |
| 1082 | return set_memory_ro((unsigned long)fp, fp->pages); |
| 1083 | } |
| 1084 | #endif |
| 1085 | return 0; |
| 1086 | } |
| 1087 | |
| 1088 | static inline int __must_check |
| 1089 | bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr) |
| 1090 | { |
| 1091 | set_vm_flush_reset_perms(hdr); |
| 1092 | return set_memory_rox(addr: (unsigned long)hdr, numpages: hdr->size >> PAGE_SHIFT); |
| 1093 | } |
| 1094 | |
| 1095 | int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap, |
| 1096 | enum skb_drop_reason *reason); |
| 1097 | |
| 1098 | static inline int sk_filter(struct sock *sk, struct sk_buff *skb) |
| 1099 | { |
| 1100 | enum skb_drop_reason ignore_reason; |
| 1101 | |
| 1102 | return sk_filter_trim_cap(sk, skb, cap: 1, reason: &ignore_reason); |
| 1103 | } |
| 1104 | |
| 1105 | static inline int sk_filter_reason(struct sock *sk, struct sk_buff *skb, |
| 1106 | enum skb_drop_reason *reason) |
| 1107 | { |
| 1108 | return sk_filter_trim_cap(sk, skb, cap: 1, reason); |
| 1109 | } |
| 1110 | |
| 1111 | struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err); |
| 1112 | void bpf_prog_free(struct bpf_prog *fp); |
| 1113 | |
| 1114 | bool bpf_opcode_in_insntable(u8 code); |
| 1115 | |
| 1116 | void bpf_prog_fill_jited_linfo(struct bpf_prog *prog, |
| 1117 | const u32 *insn_to_jit_off); |
| 1118 | int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog); |
| 1119 | void bpf_prog_jit_attempt_done(struct bpf_prog *prog); |
| 1120 | |
| 1121 | struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t ); |
| 1122 | struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t ); |
| 1123 | struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size, |
| 1124 | gfp_t ); |
| 1125 | void __bpf_prog_free(struct bpf_prog *fp); |
| 1126 | |
| 1127 | static inline void bpf_prog_unlock_free(struct bpf_prog *fp) |
| 1128 | { |
| 1129 | __bpf_prog_free(fp); |
| 1130 | } |
| 1131 | |
| 1132 | typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter, |
| 1133 | unsigned int flen); |
| 1134 | |
| 1135 | int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog); |
| 1136 | int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, |
| 1137 | bpf_aux_classic_check_t trans, bool save_orig); |
| 1138 | void bpf_prog_destroy(struct bpf_prog *fp); |
| 1139 | |
| 1140 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
| 1141 | int sk_attach_bpf(u32 ufd, struct sock *sk); |
| 1142 | int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk); |
| 1143 | int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk); |
| 1144 | void sk_reuseport_prog_free(struct bpf_prog *prog); |
| 1145 | int sk_detach_filter(struct sock *sk); |
| 1146 | int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len); |
| 1147 | |
| 1148 | bool sk_filter_charge(struct sock *sk, struct sk_filter *fp); |
| 1149 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp); |
| 1150 | |
| 1151 | u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
| 1152 | #define __bpf_call_base_args \ |
| 1153 | ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \ |
| 1154 | (void *)__bpf_call_base) |
| 1155 | |
| 1156 | struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog); |
| 1157 | void bpf_jit_compile(struct bpf_prog *prog); |
| 1158 | bool bpf_jit_needs_zext(void); |
| 1159 | bool bpf_jit_inlines_helper_call(s32 imm); |
| 1160 | bool bpf_jit_supports_subprog_tailcalls(void); |
| 1161 | bool bpf_jit_supports_percpu_insn(void); |
| 1162 | bool bpf_jit_supports_kfunc_call(void); |
| 1163 | bool bpf_jit_supports_far_kfunc_call(void); |
| 1164 | bool bpf_jit_supports_exceptions(void); |
| 1165 | bool bpf_jit_supports_ptr_xchg(void); |
| 1166 | bool bpf_jit_supports_arena(void); |
| 1167 | bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena); |
| 1168 | bool bpf_jit_supports_private_stack(void); |
| 1169 | bool bpf_jit_supports_timed_may_goto(void); |
| 1170 | u64 bpf_arch_uaddress_limit(void); |
| 1171 | void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie); |
| 1172 | u64 arch_bpf_timed_may_goto(void); |
| 1173 | u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *); |
| 1174 | bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id); |
| 1175 | |
| 1176 | static inline bool bpf_dump_raw_ok(const struct cred *cred) |
| 1177 | { |
| 1178 | /* Reconstruction of call-sites is dependent on kallsyms, |
| 1179 | * thus make dump the same restriction. |
| 1180 | */ |
| 1181 | return kallsyms_show_value(cred); |
| 1182 | } |
| 1183 | |
| 1184 | struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off, |
| 1185 | const struct bpf_insn *patch, u32 len); |
| 1186 | int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt); |
| 1187 | |
| 1188 | static inline bool xdp_return_frame_no_direct(void) |
| 1189 | { |
| 1190 | struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); |
| 1191 | |
| 1192 | return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT; |
| 1193 | } |
| 1194 | |
| 1195 | static inline void xdp_set_return_frame_no_direct(void) |
| 1196 | { |
| 1197 | struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); |
| 1198 | |
| 1199 | ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT; |
| 1200 | } |
| 1201 | |
| 1202 | static inline void xdp_clear_return_frame_no_direct(void) |
| 1203 | { |
| 1204 | struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); |
| 1205 | |
| 1206 | ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT; |
| 1207 | } |
| 1208 | |
| 1209 | static inline int xdp_ok_fwd_dev(const struct net_device *fwd, |
| 1210 | unsigned int pktlen) |
| 1211 | { |
| 1212 | unsigned int len; |
| 1213 | |
| 1214 | if (unlikely(!(fwd->flags & IFF_UP))) |
| 1215 | return -ENETDOWN; |
| 1216 | |
| 1217 | len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN; |
| 1218 | if (pktlen > len) |
| 1219 | return -EMSGSIZE; |
| 1220 | |
| 1221 | return 0; |
| 1222 | } |
| 1223 | |
| 1224 | /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the |
| 1225 | * same cpu context. Further for best results no more than a single map |
| 1226 | * for the do_redirect/do_flush pair should be used. This limitation is |
| 1227 | * because we only track one map and force a flush when the map changes. |
| 1228 | * This does not appear to be a real limitation for existing software. |
| 1229 | */ |
| 1230 | int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, |
| 1231 | struct xdp_buff *xdp, const struct bpf_prog *prog); |
| 1232 | int xdp_do_redirect(struct net_device *dev, |
| 1233 | struct xdp_buff *xdp, |
| 1234 | const struct bpf_prog *prog); |
| 1235 | int xdp_do_redirect_frame(struct net_device *dev, |
| 1236 | struct xdp_buff *xdp, |
| 1237 | struct xdp_frame *xdpf, |
| 1238 | const struct bpf_prog *prog); |
| 1239 | void xdp_do_flush(void); |
| 1240 | |
| 1241 | void bpf_warn_invalid_xdp_action(const struct net_device *dev, |
| 1242 | const struct bpf_prog *prog, u32 act); |
| 1243 | |
| 1244 | #ifdef CONFIG_INET |
| 1245 | struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, |
| 1246 | struct bpf_prog *prog, struct sk_buff *skb, |
| 1247 | struct sock *migrating_sk, |
| 1248 | u32 hash); |
| 1249 | #else |
| 1250 | static inline struct sock * |
| 1251 | bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, |
| 1252 | struct bpf_prog *prog, struct sk_buff *skb, |
| 1253 | struct sock *migrating_sk, |
| 1254 | u32 hash) |
| 1255 | { |
| 1256 | return NULL; |
| 1257 | } |
| 1258 | #endif |
| 1259 | |
| 1260 | #ifdef CONFIG_BPF_JIT |
| 1261 | extern int bpf_jit_enable; |
| 1262 | extern int bpf_jit_harden; |
| 1263 | extern int bpf_jit_kallsyms; |
| 1264 | extern long bpf_jit_limit; |
| 1265 | extern long bpf_jit_limit_max; |
| 1266 | |
| 1267 | typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size); |
| 1268 | |
| 1269 | void bpf_jit_fill_hole_with_zero(void *area, unsigned int size); |
| 1270 | |
| 1271 | struct bpf_binary_header * |
| 1272 | bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, |
| 1273 | unsigned int alignment, |
| 1274 | bpf_jit_fill_hole_t bpf_fill_ill_insns); |
| 1275 | void bpf_jit_binary_free(struct bpf_binary_header *hdr); |
| 1276 | u64 bpf_jit_alloc_exec_limit(void); |
| 1277 | void *bpf_jit_alloc_exec(unsigned long size); |
| 1278 | void bpf_jit_free_exec(void *addr); |
| 1279 | void bpf_jit_free(struct bpf_prog *fp); |
| 1280 | struct bpf_binary_header * |
| 1281 | bpf_jit_binary_pack_hdr(const struct bpf_prog *fp); |
| 1282 | |
| 1283 | void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns); |
| 1284 | void bpf_prog_pack_free(void *ptr, u32 size); |
| 1285 | |
| 1286 | static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp) |
| 1287 | { |
| 1288 | return list_empty(head: &fp->aux->ksym.lnode) || |
| 1289 | fp->aux->ksym.lnode.prev == LIST_POISON2; |
| 1290 | } |
| 1291 | |
| 1292 | struct bpf_binary_header * |
| 1293 | bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image, |
| 1294 | unsigned int alignment, |
| 1295 | struct bpf_binary_header **rw_hdr, |
| 1296 | u8 **rw_image, |
| 1297 | bpf_jit_fill_hole_t bpf_fill_ill_insns); |
| 1298 | int bpf_jit_binary_pack_finalize(struct bpf_binary_header *, |
| 1299 | struct bpf_binary_header *); |
| 1300 | void bpf_jit_binary_pack_free(struct bpf_binary_header *, |
| 1301 | struct bpf_binary_header *); |
| 1302 | |
| 1303 | int bpf_jit_add_poke_descriptor(struct bpf_prog *prog, |
| 1304 | struct bpf_jit_poke_descriptor *poke); |
| 1305 | |
| 1306 | int bpf_jit_get_func_addr(const struct bpf_prog *prog, |
| 1307 | const struct bpf_insn *insn, bool , |
| 1308 | u64 *func_addr, bool *func_addr_fixed); |
| 1309 | |
| 1310 | const char *bpf_jit_get_prog_name(struct bpf_prog *prog); |
| 1311 | |
| 1312 | struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp); |
| 1313 | void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other); |
| 1314 | |
| 1315 | static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen, |
| 1316 | u32 pass, void *image) |
| 1317 | { |
| 1318 | pr_err("flen=%u proglen=%u pass=%u image=%p from=%s pid=%d\n" , flen, |
| 1319 | proglen, pass, image, current->comm, task_pid_nr(current)); |
| 1320 | |
| 1321 | if (image) |
| 1322 | print_hex_dump(KERN_ERR, prefix_str: "JIT code: " , prefix_type: DUMP_PREFIX_OFFSET, |
| 1323 | rowsize: 16, groupsize: 1, buf: image, len: proglen, ascii: false); |
| 1324 | } |
| 1325 | |
| 1326 | static inline bool bpf_jit_is_ebpf(void) |
| 1327 | { |
| 1328 | # ifdef CONFIG_HAVE_EBPF_JIT |
| 1329 | return true; |
| 1330 | # else |
| 1331 | return false; |
| 1332 | # endif |
| 1333 | } |
| 1334 | |
| 1335 | static inline bool ebpf_jit_enabled(void) |
| 1336 | { |
| 1337 | return bpf_jit_enable && bpf_jit_is_ebpf(); |
| 1338 | } |
| 1339 | |
| 1340 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
| 1341 | { |
| 1342 | return fp->jited && bpf_jit_is_ebpf(); |
| 1343 | } |
| 1344 | |
| 1345 | static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) |
| 1346 | { |
| 1347 | /* These are the prerequisites, should someone ever have the |
| 1348 | * idea to call blinding outside of them, we make sure to |
| 1349 | * bail out. |
| 1350 | */ |
| 1351 | if (!bpf_jit_is_ebpf()) |
| 1352 | return false; |
| 1353 | if (!prog->jit_requested) |
| 1354 | return false; |
| 1355 | if (!bpf_jit_harden) |
| 1356 | return false; |
| 1357 | if (bpf_jit_harden == 1 && bpf_token_capable(token: prog->aux->token, CAP_BPF)) |
| 1358 | return false; |
| 1359 | |
| 1360 | return true; |
| 1361 | } |
| 1362 | |
| 1363 | static inline bool bpf_jit_kallsyms_enabled(void) |
| 1364 | { |
| 1365 | /* There are a couple of corner cases where kallsyms should |
| 1366 | * not be enabled f.e. on hardening. |
| 1367 | */ |
| 1368 | if (bpf_jit_harden) |
| 1369 | return false; |
| 1370 | if (!bpf_jit_kallsyms) |
| 1371 | return false; |
| 1372 | if (bpf_jit_kallsyms == 1) |
| 1373 | return true; |
| 1374 | |
| 1375 | return false; |
| 1376 | } |
| 1377 | |
| 1378 | int __bpf_address_lookup(unsigned long addr, unsigned long *size, |
| 1379 | unsigned long *off, char *sym); |
| 1380 | bool is_bpf_text_address(unsigned long addr); |
| 1381 | int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, |
| 1382 | char *sym); |
| 1383 | struct bpf_prog *bpf_prog_ksym_find(unsigned long addr); |
| 1384 | |
| 1385 | static inline int |
| 1386 | bpf_address_lookup(unsigned long addr, unsigned long *size, |
| 1387 | unsigned long *off, char **modname, char *sym) |
| 1388 | { |
| 1389 | int ret = __bpf_address_lookup(addr, size, off, sym); |
| 1390 | |
| 1391 | if (ret && modname) |
| 1392 | *modname = NULL; |
| 1393 | return ret; |
| 1394 | } |
| 1395 | |
| 1396 | void bpf_prog_kallsyms_add(struct bpf_prog *fp); |
| 1397 | void bpf_prog_kallsyms_del(struct bpf_prog *fp); |
| 1398 | |
| 1399 | #else /* CONFIG_BPF_JIT */ |
| 1400 | |
| 1401 | static inline bool ebpf_jit_enabled(void) |
| 1402 | { |
| 1403 | return false; |
| 1404 | } |
| 1405 | |
| 1406 | static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog) |
| 1407 | { |
| 1408 | return false; |
| 1409 | } |
| 1410 | |
| 1411 | static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp) |
| 1412 | { |
| 1413 | return false; |
| 1414 | } |
| 1415 | |
| 1416 | static inline int |
| 1417 | bpf_jit_add_poke_descriptor(struct bpf_prog *prog, |
| 1418 | struct bpf_jit_poke_descriptor *poke) |
| 1419 | { |
| 1420 | return -ENOTSUPP; |
| 1421 | } |
| 1422 | |
| 1423 | static inline void bpf_jit_free(struct bpf_prog *fp) |
| 1424 | { |
| 1425 | bpf_prog_unlock_free(fp); |
| 1426 | } |
| 1427 | |
| 1428 | static inline bool bpf_jit_kallsyms_enabled(void) |
| 1429 | { |
| 1430 | return false; |
| 1431 | } |
| 1432 | |
| 1433 | static inline int |
| 1434 | __bpf_address_lookup(unsigned long addr, unsigned long *size, |
| 1435 | unsigned long *off, char *sym) |
| 1436 | { |
| 1437 | return 0; |
| 1438 | } |
| 1439 | |
| 1440 | static inline bool is_bpf_text_address(unsigned long addr) |
| 1441 | { |
| 1442 | return false; |
| 1443 | } |
| 1444 | |
| 1445 | static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value, |
| 1446 | char *type, char *sym) |
| 1447 | { |
| 1448 | return -ERANGE; |
| 1449 | } |
| 1450 | |
| 1451 | static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr) |
| 1452 | { |
| 1453 | return NULL; |
| 1454 | } |
| 1455 | |
| 1456 | static inline int |
| 1457 | bpf_address_lookup(unsigned long addr, unsigned long *size, |
| 1458 | unsigned long *off, char **modname, char *sym) |
| 1459 | { |
| 1460 | return 0; |
| 1461 | } |
| 1462 | |
| 1463 | static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp) |
| 1464 | { |
| 1465 | } |
| 1466 | |
| 1467 | static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp) |
| 1468 | { |
| 1469 | } |
| 1470 | |
| 1471 | #endif /* CONFIG_BPF_JIT */ |
| 1472 | |
| 1473 | void bpf_prog_kallsyms_del_all(struct bpf_prog *fp); |
| 1474 | |
| 1475 | #define BPF_ANC BIT(15) |
| 1476 | |
| 1477 | static inline bool bpf_needs_clear_a(const struct sock_filter *first) |
| 1478 | { |
| 1479 | switch (first->code) { |
| 1480 | case BPF_RET | BPF_K: |
| 1481 | case BPF_LD | BPF_W | BPF_LEN: |
| 1482 | return false; |
| 1483 | |
| 1484 | case BPF_LD | BPF_W | BPF_ABS: |
| 1485 | case BPF_LD | BPF_H | BPF_ABS: |
| 1486 | case BPF_LD | BPF_B | BPF_ABS: |
| 1487 | if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X) |
| 1488 | return true; |
| 1489 | return false; |
| 1490 | |
| 1491 | default: |
| 1492 | return true; |
| 1493 | } |
| 1494 | } |
| 1495 | |
| 1496 | static inline u16 bpf_anc_helper(const struct sock_filter *ftest) |
| 1497 | { |
| 1498 | BUG_ON(ftest->code & BPF_ANC); |
| 1499 | |
| 1500 | switch (ftest->code) { |
| 1501 | case BPF_LD | BPF_W | BPF_ABS: |
| 1502 | case BPF_LD | BPF_H | BPF_ABS: |
| 1503 | case BPF_LD | BPF_B | BPF_ABS: |
| 1504 | #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ |
| 1505 | return BPF_ANC | SKF_AD_##CODE |
| 1506 | switch (ftest->k) { |
| 1507 | BPF_ANCILLARY(PROTOCOL); |
| 1508 | BPF_ANCILLARY(PKTTYPE); |
| 1509 | BPF_ANCILLARY(IFINDEX); |
| 1510 | BPF_ANCILLARY(NLATTR); |
| 1511 | BPF_ANCILLARY(NLATTR_NEST); |
| 1512 | BPF_ANCILLARY(MARK); |
| 1513 | BPF_ANCILLARY(QUEUE); |
| 1514 | BPF_ANCILLARY(HATYPE); |
| 1515 | BPF_ANCILLARY(RXHASH); |
| 1516 | BPF_ANCILLARY(CPU); |
| 1517 | BPF_ANCILLARY(ALU_XOR_X); |
| 1518 | BPF_ANCILLARY(VLAN_TAG); |
| 1519 | BPF_ANCILLARY(VLAN_TAG_PRESENT); |
| 1520 | BPF_ANCILLARY(PAY_OFFSET); |
| 1521 | BPF_ANCILLARY(RANDOM); |
| 1522 | BPF_ANCILLARY(VLAN_TPID); |
| 1523 | } |
| 1524 | fallthrough; |
| 1525 | default: |
| 1526 | return ftest->code; |
| 1527 | } |
| 1528 | } |
| 1529 | |
| 1530 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, |
| 1531 | int k, unsigned int size); |
| 1532 | |
| 1533 | static inline int bpf_tell_extensions(void) |
| 1534 | { |
| 1535 | return SKF_AD_MAX; |
| 1536 | } |
| 1537 | |
| 1538 | struct bpf_sock_addr_kern { |
| 1539 | struct sock *sk; |
| 1540 | struct sockaddr_unsized *uaddr; |
| 1541 | /* Temporary "register" to make indirect stores to nested structures |
| 1542 | * defined above. We need three registers to make such a store, but |
| 1543 | * only two (src and dst) are available at convert_ctx_access time |
| 1544 | */ |
| 1545 | u64 tmp_reg; |
| 1546 | void *t_ctx; /* Attach type specific context. */ |
| 1547 | u32 uaddrlen; |
| 1548 | }; |
| 1549 | |
| 1550 | struct bpf_sock_ops_kern { |
| 1551 | struct sock *sk; |
| 1552 | union { |
| 1553 | u32 args[4]; |
| 1554 | u32 reply; |
| 1555 | u32 replylong[4]; |
| 1556 | }; |
| 1557 | struct sk_buff *syn_skb; |
| 1558 | struct sk_buff *skb; |
| 1559 | void *skb_data_end; |
| 1560 | u8 op; |
| 1561 | u8 is_fullsock; |
| 1562 | u8 is_locked_tcp_sock; |
| 1563 | u8 remaining_opt_len; |
| 1564 | u64 temp; /* temp and everything after is not |
| 1565 | * initialized to 0 before calling |
| 1566 | * the BPF program. New fields that |
| 1567 | * should be initialized to 0 should |
| 1568 | * be inserted before temp. |
| 1569 | * temp is scratch storage used by |
| 1570 | * sock_ops_convert_ctx_access |
| 1571 | * as temporary storage of a register. |
| 1572 | */ |
| 1573 | }; |
| 1574 | |
| 1575 | struct bpf_sysctl_kern { |
| 1576 | struct ctl_table_header *head; |
| 1577 | const struct ctl_table *table; |
| 1578 | void *cur_val; |
| 1579 | size_t cur_len; |
| 1580 | void *new_val; |
| 1581 | size_t new_len; |
| 1582 | int new_updated; |
| 1583 | int write; |
| 1584 | loff_t *ppos; |
| 1585 | /* Temporary "register" for indirect stores to ppos. */ |
| 1586 | u64 tmp_reg; |
| 1587 | }; |
| 1588 | |
| 1589 | #define BPF_SOCKOPT_KERN_BUF_SIZE 32 |
| 1590 | struct bpf_sockopt_buf { |
| 1591 | u8 data[BPF_SOCKOPT_KERN_BUF_SIZE]; |
| 1592 | }; |
| 1593 | |
| 1594 | struct bpf_sockopt_kern { |
| 1595 | struct sock *sk; |
| 1596 | u8 *optval; |
| 1597 | u8 *optval_end; |
| 1598 | s32 level; |
| 1599 | s32 optname; |
| 1600 | s32 optlen; |
| 1601 | /* for retval in struct bpf_cg_run_ctx */ |
| 1602 | struct task_struct *current_task; |
| 1603 | /* Temporary "register" for indirect stores to ppos. */ |
| 1604 | u64 tmp_reg; |
| 1605 | }; |
| 1606 | |
| 1607 | int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len); |
| 1608 | |
| 1609 | struct bpf_sk_lookup_kern { |
| 1610 | u16 family; |
| 1611 | u16 protocol; |
| 1612 | __be16 sport; |
| 1613 | u16 dport; |
| 1614 | struct { |
| 1615 | __be32 saddr; |
| 1616 | __be32 daddr; |
| 1617 | } v4; |
| 1618 | struct { |
| 1619 | const struct in6_addr *saddr; |
| 1620 | const struct in6_addr *daddr; |
| 1621 | } v6; |
| 1622 | struct sock *selected_sk; |
| 1623 | u32 ingress_ifindex; |
| 1624 | bool no_reuseport; |
| 1625 | }; |
| 1626 | |
| 1627 | extern struct static_key_false bpf_sk_lookup_enabled; |
| 1628 | |
| 1629 | /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup. |
| 1630 | * |
| 1631 | * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and |
| 1632 | * SK_DROP. Their meaning is as follows: |
| 1633 | * |
| 1634 | * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result |
| 1635 | * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup |
| 1636 | * SK_DROP : terminate lookup with -ECONNREFUSED |
| 1637 | * |
| 1638 | * This macro aggregates return values and selected sockets from |
| 1639 | * multiple BPF programs according to following rules in order: |
| 1640 | * |
| 1641 | * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk, |
| 1642 | * macro result is SK_PASS and last ctx.selected_sk is used. |
| 1643 | * 2. If any program returned SK_DROP return value, |
| 1644 | * macro result is SK_DROP. |
| 1645 | * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL. |
| 1646 | * |
| 1647 | * Caller must ensure that the prog array is non-NULL, and that the |
| 1648 | * array as well as the programs it contains remain valid. |
| 1649 | */ |
| 1650 | #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \ |
| 1651 | ({ \ |
| 1652 | struct bpf_sk_lookup_kern *_ctx = &(ctx); \ |
| 1653 | struct bpf_prog_array_item *_item; \ |
| 1654 | struct sock *_selected_sk = NULL; \ |
| 1655 | bool _no_reuseport = false; \ |
| 1656 | struct bpf_prog *_prog; \ |
| 1657 | bool _all_pass = true; \ |
| 1658 | u32 _ret; \ |
| 1659 | \ |
| 1660 | migrate_disable(); \ |
| 1661 | _item = &(array)->items[0]; \ |
| 1662 | while ((_prog = READ_ONCE(_item->prog))) { \ |
| 1663 | /* restore most recent selection */ \ |
| 1664 | _ctx->selected_sk = _selected_sk; \ |
| 1665 | _ctx->no_reuseport = _no_reuseport; \ |
| 1666 | \ |
| 1667 | _ret = func(_prog, _ctx); \ |
| 1668 | if (_ret == SK_PASS && _ctx->selected_sk) { \ |
| 1669 | /* remember last non-NULL socket */ \ |
| 1670 | _selected_sk = _ctx->selected_sk; \ |
| 1671 | _no_reuseport = _ctx->no_reuseport; \ |
| 1672 | } else if (_ret == SK_DROP && _all_pass) { \ |
| 1673 | _all_pass = false; \ |
| 1674 | } \ |
| 1675 | _item++; \ |
| 1676 | } \ |
| 1677 | _ctx->selected_sk = _selected_sk; \ |
| 1678 | _ctx->no_reuseport = _no_reuseport; \ |
| 1679 | migrate_enable(); \ |
| 1680 | _all_pass || _selected_sk ? SK_PASS : SK_DROP; \ |
| 1681 | }) |
| 1682 | |
| 1683 | static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol, |
| 1684 | const __be32 saddr, const __be16 sport, |
| 1685 | const __be32 daddr, const u16 dport, |
| 1686 | const int ifindex, struct sock **psk) |
| 1687 | { |
| 1688 | struct bpf_prog_array *run_array; |
| 1689 | struct sock *selected_sk = NULL; |
| 1690 | bool no_reuseport = false; |
| 1691 | |
| 1692 | rcu_read_lock(); |
| 1693 | run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); |
| 1694 | if (run_array) { |
| 1695 | struct bpf_sk_lookup_kern ctx = { |
| 1696 | .family = AF_INET, |
| 1697 | .protocol = protocol, |
| 1698 | .v4.saddr = saddr, |
| 1699 | .v4.daddr = daddr, |
| 1700 | .sport = sport, |
| 1701 | .dport = dport, |
| 1702 | .ingress_ifindex = ifindex, |
| 1703 | }; |
| 1704 | u32 act; |
| 1705 | |
| 1706 | act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); |
| 1707 | if (act == SK_PASS) { |
| 1708 | selected_sk = ctx.selected_sk; |
| 1709 | no_reuseport = ctx.no_reuseport; |
| 1710 | } else { |
| 1711 | selected_sk = ERR_PTR(error: -ECONNREFUSED); |
| 1712 | } |
| 1713 | } |
| 1714 | rcu_read_unlock(); |
| 1715 | *psk = selected_sk; |
| 1716 | return no_reuseport; |
| 1717 | } |
| 1718 | |
| 1719 | #if IS_ENABLED(CONFIG_IPV6) |
| 1720 | static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol, |
| 1721 | const struct in6_addr *saddr, |
| 1722 | const __be16 sport, |
| 1723 | const struct in6_addr *daddr, |
| 1724 | const u16 dport, |
| 1725 | const int ifindex, struct sock **psk) |
| 1726 | { |
| 1727 | struct bpf_prog_array *run_array; |
| 1728 | struct sock *selected_sk = NULL; |
| 1729 | bool no_reuseport = false; |
| 1730 | |
| 1731 | rcu_read_lock(); |
| 1732 | run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]); |
| 1733 | if (run_array) { |
| 1734 | struct bpf_sk_lookup_kern ctx = { |
| 1735 | .family = AF_INET6, |
| 1736 | .protocol = protocol, |
| 1737 | .v6.saddr = saddr, |
| 1738 | .v6.daddr = daddr, |
| 1739 | .sport = sport, |
| 1740 | .dport = dport, |
| 1741 | .ingress_ifindex = ifindex, |
| 1742 | }; |
| 1743 | u32 act; |
| 1744 | |
| 1745 | act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run); |
| 1746 | if (act == SK_PASS) { |
| 1747 | selected_sk = ctx.selected_sk; |
| 1748 | no_reuseport = ctx.no_reuseport; |
| 1749 | } else { |
| 1750 | selected_sk = ERR_PTR(error: -ECONNREFUSED); |
| 1751 | } |
| 1752 | } |
| 1753 | rcu_read_unlock(); |
| 1754 | *psk = selected_sk; |
| 1755 | return no_reuseport; |
| 1756 | } |
| 1757 | #endif /* IS_ENABLED(CONFIG_IPV6) */ |
| 1758 | |
| 1759 | static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index, |
| 1760 | u64 flags, const u64 flag_mask, |
| 1761 | void *lookup_elem(struct bpf_map *map, u32 key)) |
| 1762 | { |
| 1763 | struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); |
| 1764 | const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX; |
| 1765 | |
| 1766 | /* Lower bits of the flags are used as return code on lookup failure */ |
| 1767 | if (unlikely(flags & ~(action_mask | flag_mask))) |
| 1768 | return XDP_ABORTED; |
| 1769 | |
| 1770 | ri->tgt_value = lookup_elem(map, index); |
| 1771 | if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) { |
| 1772 | /* If the lookup fails we want to clear out the state in the |
| 1773 | * redirect_info struct completely, so that if an eBPF program |
| 1774 | * performs multiple lookups, the last one always takes |
| 1775 | * precedence. |
| 1776 | */ |
| 1777 | ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */ |
| 1778 | ri->map_type = BPF_MAP_TYPE_UNSPEC; |
| 1779 | return flags & action_mask; |
| 1780 | } |
| 1781 | |
| 1782 | ri->tgt_index = index; |
| 1783 | ri->map_id = map->id; |
| 1784 | ri->map_type = map->map_type; |
| 1785 | |
| 1786 | if (flags & BPF_F_BROADCAST) { |
| 1787 | WRITE_ONCE(ri->map, map); |
| 1788 | ri->flags = flags; |
| 1789 | } else { |
| 1790 | WRITE_ONCE(ri->map, NULL); |
| 1791 | ri->flags = 0; |
| 1792 | } |
| 1793 | |
| 1794 | return XDP_REDIRECT; |
| 1795 | } |
| 1796 | |
| 1797 | #ifdef CONFIG_NET |
| 1798 | int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len); |
| 1799 | int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, |
| 1800 | u32 len, u64 flags); |
| 1801 | int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); |
| 1802 | int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len); |
| 1803 | void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len); |
| 1804 | void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, |
| 1805 | void *buf, unsigned long len, bool flush); |
| 1806 | int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset, |
| 1807 | const void *from, u32 len, u64 flags); |
| 1808 | void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset); |
| 1809 | #else /* CONFIG_NET */ |
| 1810 | static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, |
| 1811 | void *to, u32 len) |
| 1812 | { |
| 1813 | return -EOPNOTSUPP; |
| 1814 | } |
| 1815 | |
| 1816 | static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, |
| 1817 | const void *from, u32 len, u64 flags) |
| 1818 | { |
| 1819 | return -EOPNOTSUPP; |
| 1820 | } |
| 1821 | |
| 1822 | static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, |
| 1823 | void *buf, u32 len) |
| 1824 | { |
| 1825 | return -EOPNOTSUPP; |
| 1826 | } |
| 1827 | |
| 1828 | static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, |
| 1829 | void *buf, u32 len) |
| 1830 | { |
| 1831 | return -EOPNOTSUPP; |
| 1832 | } |
| 1833 | |
| 1834 | static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) |
| 1835 | { |
| 1836 | return NULL; |
| 1837 | } |
| 1838 | |
| 1839 | static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf, |
| 1840 | unsigned long len, bool flush) |
| 1841 | { |
| 1842 | } |
| 1843 | |
| 1844 | static inline int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset, |
| 1845 | const void *from, u32 len, |
| 1846 | u64 flags) |
| 1847 | { |
| 1848 | return -EOPNOTSUPP; |
| 1849 | } |
| 1850 | |
| 1851 | static inline void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset) |
| 1852 | { |
| 1853 | return ERR_PTR(-EOPNOTSUPP); |
| 1854 | } |
| 1855 | #endif /* CONFIG_NET */ |
| 1856 | |
| 1857 | #endif /* __LINUX_FILTER_H__ */ |
| 1858 | |