1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 2 * Copyright (c) 2016 Facebook 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of version 2 of the GNU General Public 6 * License as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, but 9 * WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 */ 13 #include <linux/kernel.h> 14 #include <linux/types.h> 15 #include <linux/slab.h> 16 #include <linux/bpf.h> 17 #include <linux/bpf_verifier.h> 18 #include <linux/filter.h> 19 #include <net/netlink.h> 20 #include <linux/file.h> 21 #include <linux/vmalloc.h> 22 #include <linux/stringify.h> 23 24 /* bpf_check() is a static code analyzer that walks eBPF program 25 * instruction by instruction and updates register/stack state. 26 * All paths of conditional branches are analyzed until 'bpf_exit' insn. 27 * 28 * The first pass is depth-first-search to check that the program is a DAG. 29 * It rejects the following programs: 30 * - larger than BPF_MAXINSNS insns 31 * - if loop is present (detected via back-edge) 32 * - unreachable insns exist (shouldn't be a forest. program = one function) 33 * - out of bounds or malformed jumps 34 * The second pass is all possible path descent from the 1st insn. 35 * Since it's analyzing all pathes through the program, the length of the 36 * analysis is limited to 64k insn, which may be hit even if total number of 37 * insn is less then 4K, but there are too many branches that change stack/regs. 38 * Number of 'branches to be analyzed' is limited to 1k 39 * 40 * On entry to each instruction, each register has a type, and the instruction 41 * changes the types of the registers depending on instruction semantics. 42 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is 43 * copied to R1. 44 * 45 * All registers are 64-bit. 46 * R0 - return register 47 * R1-R5 argument passing registers 48 * R6-R9 callee saved registers 49 * R10 - frame pointer read-only 50 * 51 * At the start of BPF program the register R1 contains a pointer to bpf_context 52 * and has type PTR_TO_CTX. 53 * 54 * Verifier tracks arithmetic operations on pointers in case: 55 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), 56 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), 57 * 1st insn copies R10 (which has FRAME_PTR) type into R1 58 * and 2nd arithmetic instruction is pattern matched to recognize 59 * that it wants to construct a pointer to some element within stack. 60 * So after 2nd insn, the register R1 has type PTR_TO_STACK 61 * (and -20 constant is saved for further stack bounds checking). 62 * Meaning that this reg is a pointer to stack plus known immediate constant. 63 * 64 * Most of the time the registers have UNKNOWN_VALUE type, which 65 * means the register has some value, but it's not a valid pointer. 66 * (like pointer plus pointer becomes UNKNOWN_VALUE type) 67 * 68 * When verifier sees load or store instructions the type of base register 69 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer 70 * types recognized by check_mem_access() function. 71 * 72 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' 73 * and the range of [ptr, ptr + map's value_size) is accessible. 74 * 75 * registers used to pass values to function calls are checked against 76 * function argument constraints. 77 * 78 * ARG_PTR_TO_MAP_KEY is one of such argument constraints. 79 * It means that the register type passed to this function must be 80 * PTR_TO_STACK and it will be used inside the function as 81 * 'pointer to map element key' 82 * 83 * For example the argument constraints for bpf_map_lookup_elem(): 84 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, 85 * .arg1_type = ARG_CONST_MAP_PTR, 86 * .arg2_type = ARG_PTR_TO_MAP_KEY, 87 * 88 * ret_type says that this function returns 'pointer to map elem value or null' 89 * function expects 1st argument to be a const pointer to 'struct bpf_map' and 90 * 2nd argument should be a pointer to stack, which will be used inside 91 * the helper function as a pointer to map element key. 92 * 93 * On the kernel side the helper function looks like: 94 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) 95 * { 96 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; 97 * void *key = (void *) (unsigned long) r2; 98 * void *value; 99 * 100 * here kernel can access 'key' and 'map' pointers safely, knowing that 101 * [key, key + map->key_size) bytes are valid and were initialized on 102 * the stack of eBPF program. 103 * } 104 * 105 * Corresponding eBPF program may look like: 106 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR 107 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK 108 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP 109 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), 110 * here verifier looks at prototype of map_lookup_elem() and sees: 111 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, 112 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes 113 * 114 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, 115 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits 116 * and were initialized prior to this call. 117 * If it's ok, then verifier allows this BPF_CALL insn and looks at 118 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets 119 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function 120 * returns ether pointer to map value or NULL. 121 * 122 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' 123 * insn, the register holding that pointer in the true branch changes state to 124 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false 125 * branch. See check_cond_jmp_op(). 126 * 127 * After the call R0 is set to return type of the function and registers R1-R5 128 * are set to NOT_INIT to indicate that they are no longer readable. 129 */ 130 131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */ 132 struct bpf_verifier_stack_elem { 133 /* verifer state is 'st' 134 * before processing instruction 'insn_idx' 135 * and after processing instruction 'prev_insn_idx' 136 */ 137 struct bpf_verifier_state st; 138 int insn_idx; 139 int prev_insn_idx; 140 struct bpf_verifier_stack_elem *next; 141 }; 142 143 #define BPF_COMPLEXITY_LIMIT_INSNS 98304 144 #define BPF_COMPLEXITY_LIMIT_STACK 1024 145 146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA) 147 148 struct bpf_call_arg_meta { 149 struct bpf_map *map_ptr; 150 bool raw_mode; 151 bool pkt_access; 152 int regno; 153 int access_size; 154 }; 155 156 /* verbose verifier prints what it's seeing 157 * bpf_check() is called under lock, so no race to access these global vars 158 */ 159 static u32 log_level, log_size, log_len; 160 static char *log_buf; 161 162 static DEFINE_MUTEX(bpf_verifier_lock); 163 164 /* log_level controls verbosity level of eBPF verifier. 165 * verbose() is used to dump the verification trace to the log, so the user 166 * can figure out what's wrong with the program 167 */ 168 static __printf(1, 2) void verbose(const char *fmt, ...) 169 { 170 va_list args; 171 172 if (log_level == 0 || log_len >= log_size - 1) 173 return; 174 175 va_start(args, fmt); 176 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); 177 va_end(args); 178 } 179 180 /* string representation of 'enum bpf_reg_type' */ 181 static const char * const reg_type_str[] = { 182 [NOT_INIT] = "?", 183 [UNKNOWN_VALUE] = "inv", 184 [PTR_TO_CTX] = "ctx", 185 [CONST_PTR_TO_MAP] = "map_ptr", 186 [PTR_TO_MAP_VALUE] = "map_value", 187 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", 188 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj", 189 [FRAME_PTR] = "fp", 190 [PTR_TO_STACK] = "fp", 191 [CONST_IMM] = "imm", 192 [PTR_TO_PACKET] = "pkt", 193 [PTR_TO_PACKET_END] = "pkt_end", 194 }; 195 196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x) 197 static const char * const func_id_str[] = { 198 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN) 199 }; 200 #undef __BPF_FUNC_STR_FN 201 202 static const char *func_id_name(int id) 203 { 204 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID); 205 206 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id]) 207 return func_id_str[id]; 208 else 209 return "unknown"; 210 } 211 212 static void print_verifier_state(struct bpf_verifier_state *state) 213 { 214 struct bpf_reg_state *reg; 215 enum bpf_reg_type t; 216 int i; 217 218 for (i = 0; i < MAX_BPF_REG; i++) { 219 reg = &state->regs[i]; 220 t = reg->type; 221 if (t == NOT_INIT) 222 continue; 223 verbose(" R%d=%s", i, reg_type_str[t]); 224 if (t == CONST_IMM || t == PTR_TO_STACK) 225 verbose("%lld", reg->imm); 226 else if (t == PTR_TO_PACKET) 227 verbose("(id=%d,off=%d,r=%d)", 228 reg->id, reg->off, reg->range); 229 else if (t == UNKNOWN_VALUE && reg->imm) 230 verbose("%lld", reg->imm); 231 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || 232 t == PTR_TO_MAP_VALUE_OR_NULL || 233 t == PTR_TO_MAP_VALUE_ADJ) 234 verbose("(ks=%d,vs=%d,id=%u)", 235 reg->map_ptr->key_size, 236 reg->map_ptr->value_size, 237 reg->id); 238 if (reg->min_value != BPF_REGISTER_MIN_RANGE) 239 verbose(",min_value=%lld", 240 (long long)reg->min_value); 241 if (reg->max_value != BPF_REGISTER_MAX_RANGE) 242 verbose(",max_value=%llu", 243 (unsigned long long)reg->max_value); 244 if (reg->min_align) 245 verbose(",min_align=%u", reg->min_align); 246 if (reg->aux_off) 247 verbose(",aux_off=%u", reg->aux_off); 248 if (reg->aux_off_align) 249 verbose(",aux_off_align=%u", reg->aux_off_align); 250 } 251 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 252 if (state->stack_slot_type[i] == STACK_SPILL) 253 verbose(" fp%d=%s", -MAX_BPF_STACK + i, 254 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); 255 } 256 verbose("\n"); 257 } 258 259 static const char *const bpf_class_string[] = { 260 [BPF_LD] = "ld", 261 [BPF_LDX] = "ldx", 262 [BPF_ST] = "st", 263 [BPF_STX] = "stx", 264 [BPF_ALU] = "alu", 265 [BPF_JMP] = "jmp", 266 [BPF_RET] = "BUG", 267 [BPF_ALU64] = "alu64", 268 }; 269 270 static const char *const bpf_alu_string[16] = { 271 [BPF_ADD >> 4] = "+=", 272 [BPF_SUB >> 4] = "-=", 273 [BPF_MUL >> 4] = "*=", 274 [BPF_DIV >> 4] = "/=", 275 [BPF_OR >> 4] = "|=", 276 [BPF_AND >> 4] = "&=", 277 [BPF_LSH >> 4] = "<<=", 278 [BPF_RSH >> 4] = ">>=", 279 [BPF_NEG >> 4] = "neg", 280 [BPF_MOD >> 4] = "%=", 281 [BPF_XOR >> 4] = "^=", 282 [BPF_MOV >> 4] = "=", 283 [BPF_ARSH >> 4] = "s>>=", 284 [BPF_END >> 4] = "endian", 285 }; 286 287 static const char *const bpf_ldst_string[] = { 288 [BPF_W >> 3] = "u32", 289 [BPF_H >> 3] = "u16", 290 [BPF_B >> 3] = "u8", 291 [BPF_DW >> 3] = "u64", 292 }; 293 294 static const char *const bpf_jmp_string[16] = { 295 [BPF_JA >> 4] = "jmp", 296 [BPF_JEQ >> 4] = "==", 297 [BPF_JGT >> 4] = ">", 298 [BPF_JGE >> 4] = ">=", 299 [BPF_JSET >> 4] = "&", 300 [BPF_JNE >> 4] = "!=", 301 [BPF_JSGT >> 4] = "s>", 302 [BPF_JSGE >> 4] = "s>=", 303 [BPF_CALL >> 4] = "call", 304 [BPF_EXIT >> 4] = "exit", 305 }; 306 307 static void print_bpf_insn(const struct bpf_verifier_env *env, 308 const struct bpf_insn *insn) 309 { 310 u8 class = BPF_CLASS(insn->code); 311 312 if (class == BPF_ALU || class == BPF_ALU64) { 313 if (BPF_SRC(insn->code) == BPF_X) 314 verbose("(%02x) %sr%d %s %sr%d\n", 315 insn->code, class == BPF_ALU ? "(u32) " : "", 316 insn->dst_reg, 317 bpf_alu_string[BPF_OP(insn->code) >> 4], 318 class == BPF_ALU ? "(u32) " : "", 319 insn->src_reg); 320 else 321 verbose("(%02x) %sr%d %s %s%d\n", 322 insn->code, class == BPF_ALU ? "(u32) " : "", 323 insn->dst_reg, 324 bpf_alu_string[BPF_OP(insn->code) >> 4], 325 class == BPF_ALU ? "(u32) " : "", 326 insn->imm); 327 } else if (class == BPF_STX) { 328 if (BPF_MODE(insn->code) == BPF_MEM) 329 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", 330 insn->code, 331 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 332 insn->dst_reg, 333 insn->off, insn->src_reg); 334 else if (BPF_MODE(insn->code) == BPF_XADD) 335 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", 336 insn->code, 337 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 338 insn->dst_reg, insn->off, 339 insn->src_reg); 340 else 341 verbose("BUG_%02x\n", insn->code); 342 } else if (class == BPF_ST) { 343 if (BPF_MODE(insn->code) != BPF_MEM) { 344 verbose("BUG_st_%02x\n", insn->code); 345 return; 346 } 347 verbose("(%02x) *(%s *)(r%d %+d) = %d\n", 348 insn->code, 349 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 350 insn->dst_reg, 351 insn->off, insn->imm); 352 } else if (class == BPF_LDX) { 353 if (BPF_MODE(insn->code) != BPF_MEM) { 354 verbose("BUG_ldx_%02x\n", insn->code); 355 return; 356 } 357 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", 358 insn->code, insn->dst_reg, 359 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 360 insn->src_reg, insn->off); 361 } else if (class == BPF_LD) { 362 if (BPF_MODE(insn->code) == BPF_ABS) { 363 verbose("(%02x) r0 = *(%s *)skb[%d]\n", 364 insn->code, 365 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 366 insn->imm); 367 } else if (BPF_MODE(insn->code) == BPF_IND) { 368 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", 369 insn->code, 370 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 371 insn->src_reg, insn->imm); 372 } else if (BPF_MODE(insn->code) == BPF_IMM && 373 BPF_SIZE(insn->code) == BPF_DW) { 374 /* At this point, we already made sure that the second 375 * part of the ldimm64 insn is accessible. 376 */ 377 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; 378 bool map_ptr = insn->src_reg == BPF_PSEUDO_MAP_FD; 379 380 if (map_ptr && !env->allow_ptr_leaks) 381 imm = 0; 382 383 verbose("(%02x) r%d = 0x%llx\n", insn->code, 384 insn->dst_reg, (unsigned long long)imm); 385 } else { 386 verbose("BUG_ld_%02x\n", insn->code); 387 return; 388 } 389 } else if (class == BPF_JMP) { 390 u8 opcode = BPF_OP(insn->code); 391 392 if (opcode == BPF_CALL) { 393 verbose("(%02x) call %s#%d\n", insn->code, 394 func_id_name(insn->imm), insn->imm); 395 } else if (insn->code == (BPF_JMP | BPF_JA)) { 396 verbose("(%02x) goto pc%+d\n", 397 insn->code, insn->off); 398 } else if (insn->code == (BPF_JMP | BPF_EXIT)) { 399 verbose("(%02x) exit\n", insn->code); 400 } else if (BPF_SRC(insn->code) == BPF_X) { 401 verbose("(%02x) if r%d %s r%d goto pc%+d\n", 402 insn->code, insn->dst_reg, 403 bpf_jmp_string[BPF_OP(insn->code) >> 4], 404 insn->src_reg, insn->off); 405 } else { 406 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", 407 insn->code, insn->dst_reg, 408 bpf_jmp_string[BPF_OP(insn->code) >> 4], 409 insn->imm, insn->off); 410 } 411 } else { 412 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); 413 } 414 } 415 416 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) 417 { 418 struct bpf_verifier_stack_elem *elem; 419 int insn_idx; 420 421 if (env->head == NULL) 422 return -1; 423 424 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); 425 insn_idx = env->head->insn_idx; 426 if (prev_insn_idx) 427 *prev_insn_idx = env->head->prev_insn_idx; 428 elem = env->head->next; 429 kfree(env->head); 430 env->head = elem; 431 env->stack_size--; 432 return insn_idx; 433 } 434 435 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, 436 int insn_idx, int prev_insn_idx) 437 { 438 struct bpf_verifier_stack_elem *elem; 439 440 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); 441 if (!elem) 442 goto err; 443 444 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); 445 elem->insn_idx = insn_idx; 446 elem->prev_insn_idx = prev_insn_idx; 447 elem->next = env->head; 448 env->head = elem; 449 env->stack_size++; 450 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { 451 verbose("BPF program is too complex\n"); 452 goto err; 453 } 454 return &elem->st; 455 err: 456 /* pop all elements and return */ 457 while (pop_stack(env, NULL) >= 0); 458 return NULL; 459 } 460 461 #define CALLER_SAVED_REGS 6 462 static const int caller_saved[CALLER_SAVED_REGS] = { 463 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 464 }; 465 466 static void mark_reg_not_init(struct bpf_reg_state *regs, u32 regno) 467 { 468 BUG_ON(regno >= MAX_BPF_REG); 469 470 memset(®s[regno], 0, sizeof(regs[regno])); 471 regs[regno].type = NOT_INIT; 472 regs[regno].min_value = BPF_REGISTER_MIN_RANGE; 473 regs[regno].max_value = BPF_REGISTER_MAX_RANGE; 474 } 475 476 static void init_reg_state(struct bpf_reg_state *regs) 477 { 478 int i; 479 480 for (i = 0; i < MAX_BPF_REG; i++) 481 mark_reg_not_init(regs, i); 482 483 /* frame pointer */ 484 regs[BPF_REG_FP].type = FRAME_PTR; 485 486 /* 1st arg to a function */ 487 regs[BPF_REG_1].type = PTR_TO_CTX; 488 } 489 490 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) 491 { 492 regs[regno].type = UNKNOWN_VALUE; 493 regs[regno].id = 0; 494 regs[regno].imm = 0; 495 } 496 497 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) 498 { 499 BUG_ON(regno >= MAX_BPF_REG); 500 __mark_reg_unknown_value(regs, regno); 501 } 502 503 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno) 504 { 505 regs[regno].min_value = BPF_REGISTER_MIN_RANGE; 506 regs[regno].max_value = BPF_REGISTER_MAX_RANGE; 507 regs[regno].value_from_signed = false; 508 regs[regno].min_align = 0; 509 } 510 511 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs, 512 u32 regno) 513 { 514 mark_reg_unknown_value(regs, regno); 515 reset_reg_range_values(regs, regno); 516 } 517 518 enum reg_arg_type { 519 SRC_OP, /* register is used as source operand */ 520 DST_OP, /* register is used as destination operand */ 521 DST_OP_NO_MARK /* same as above, check only, don't mark */ 522 }; 523 524 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, 525 enum reg_arg_type t) 526 { 527 if (regno >= MAX_BPF_REG) { 528 verbose("R%d is invalid\n", regno); 529 return -EINVAL; 530 } 531 532 if (t == SRC_OP) { 533 /* check whether register used as source operand can be read */ 534 if (regs[regno].type == NOT_INIT) { 535 verbose("R%d !read_ok\n", regno); 536 return -EACCES; 537 } 538 } else { 539 /* check whether register used as dest operand can be written to */ 540 if (regno == BPF_REG_FP) { 541 verbose("frame pointer is read only\n"); 542 return -EACCES; 543 } 544 if (t == DST_OP) 545 mark_reg_unknown_value(regs, regno); 546 } 547 return 0; 548 } 549 550 static bool is_spillable_regtype(enum bpf_reg_type type) 551 { 552 switch (type) { 553 case PTR_TO_MAP_VALUE: 554 case PTR_TO_MAP_VALUE_OR_NULL: 555 case PTR_TO_MAP_VALUE_ADJ: 556 case PTR_TO_STACK: 557 case PTR_TO_CTX: 558 case PTR_TO_PACKET: 559 case PTR_TO_PACKET_END: 560 case FRAME_PTR: 561 case CONST_PTR_TO_MAP: 562 return true; 563 default: 564 return false; 565 } 566 } 567 568 /* check_stack_read/write functions track spill/fill of registers, 569 * stack boundary and alignment are checked in check_mem_access() 570 */ 571 static int check_stack_write(struct bpf_verifier_state *state, int off, 572 int size, int value_regno) 573 { 574 int i; 575 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, 576 * so it's aligned access and [off, off + size) are within stack limits 577 */ 578 579 if (value_regno >= 0 && 580 is_spillable_regtype(state->regs[value_regno].type)) { 581 582 /* register containing pointer is being spilled into stack */ 583 if (size != BPF_REG_SIZE) { 584 verbose("invalid size of register spill\n"); 585 return -EACCES; 586 } 587 588 /* save register state */ 589 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 590 state->regs[value_regno]; 591 592 for (i = 0; i < BPF_REG_SIZE; i++) 593 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; 594 } else { 595 /* regular write of data into stack */ 596 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 597 (struct bpf_reg_state) {}; 598 599 for (i = 0; i < size; i++) 600 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; 601 } 602 return 0; 603 } 604 605 static int check_stack_read(struct bpf_verifier_state *state, int off, int size, 606 int value_regno) 607 { 608 u8 *slot_type; 609 int i; 610 611 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; 612 613 if (slot_type[0] == STACK_SPILL) { 614 if (size != BPF_REG_SIZE) { 615 verbose("invalid size of register spill\n"); 616 return -EACCES; 617 } 618 for (i = 1; i < BPF_REG_SIZE; i++) { 619 if (slot_type[i] != STACK_SPILL) { 620 verbose("corrupted spill memory\n"); 621 return -EACCES; 622 } 623 } 624 625 if (value_regno >= 0) 626 /* restore register state from stack */ 627 state->regs[value_regno] = 628 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; 629 return 0; 630 } else { 631 for (i = 0; i < size; i++) { 632 if (slot_type[i] != STACK_MISC) { 633 verbose("invalid read from stack off %d+%d size %d\n", 634 off, i, size); 635 return -EACCES; 636 } 637 } 638 if (value_regno >= 0) 639 /* have read misc data from the stack */ 640 mark_reg_unknown_value_and_range(state->regs, 641 value_regno); 642 return 0; 643 } 644 } 645 646 /* check read/write into map element returned by bpf_map_lookup_elem() */ 647 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off, 648 int size) 649 { 650 struct bpf_map *map = env->cur_state.regs[regno].map_ptr; 651 652 if (off < 0 || size <= 0 || off + size > map->value_size) { 653 verbose("invalid access to map value, value_size=%d off=%d size=%d\n", 654 map->value_size, off, size); 655 return -EACCES; 656 } 657 return 0; 658 } 659 660 /* check read/write into an adjusted map element */ 661 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno, 662 int off, int size) 663 { 664 struct bpf_verifier_state *state = &env->cur_state; 665 struct bpf_reg_state *reg = &state->regs[regno]; 666 int err; 667 668 /* We adjusted the register to this map value, so we 669 * need to change off and size to min_value and max_value 670 * respectively to make sure our theoretical access will be 671 * safe. 672 */ 673 if (log_level) 674 print_verifier_state(state); 675 env->varlen_map_value_access = true; 676 /* The minimum value is only important with signed 677 * comparisons where we can't assume the floor of a 678 * value is 0. If we are using signed variables for our 679 * index'es we need to make sure that whatever we use 680 * will have a set floor within our range. 681 */ 682 if (reg->min_value < 0) { 683 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", 684 regno); 685 return -EACCES; 686 } 687 err = check_map_access(env, regno, reg->min_value + off, size); 688 if (err) { 689 verbose("R%d min value is outside of the array range\n", 690 regno); 691 return err; 692 } 693 694 /* If we haven't set a max value then we need to bail 695 * since we can't be sure we won't do bad things. 696 */ 697 if (reg->max_value == BPF_REGISTER_MAX_RANGE) { 698 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n", 699 regno); 700 return -EACCES; 701 } 702 return check_map_access(env, regno, reg->max_value + off, size); 703 } 704 705 #define MAX_PACKET_OFF 0xffff 706 707 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, 708 const struct bpf_call_arg_meta *meta, 709 enum bpf_access_type t) 710 { 711 switch (env->prog->type) { 712 case BPF_PROG_TYPE_LWT_IN: 713 case BPF_PROG_TYPE_LWT_OUT: 714 /* dst_input() and dst_output() can't write for now */ 715 if (t == BPF_WRITE) 716 return false; 717 /* fallthrough */ 718 case BPF_PROG_TYPE_SCHED_CLS: 719 case BPF_PROG_TYPE_SCHED_ACT: 720 case BPF_PROG_TYPE_XDP: 721 case BPF_PROG_TYPE_LWT_XMIT: 722 if (meta) 723 return meta->pkt_access; 724 725 env->seen_direct_write = true; 726 return true; 727 default: 728 return false; 729 } 730 } 731 732 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, 733 int size) 734 { 735 struct bpf_reg_state *regs = env->cur_state.regs; 736 struct bpf_reg_state *reg = ®s[regno]; 737 738 off += reg->off; 739 if (off < 0 || size <= 0 || off + size > reg->range) { 740 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", 741 off, size, regno, reg->id, reg->off, reg->range); 742 return -EACCES; 743 } 744 return 0; 745 } 746 747 /* check access to 'struct bpf_context' fields */ 748 static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, 749 enum bpf_access_type t, enum bpf_reg_type *reg_type) 750 { 751 struct bpf_insn_access_aux info = { 752 .reg_type = *reg_type, 753 }; 754 755 /* for analyzer ctx accesses are already validated and converted */ 756 if (env->analyzer_ops) 757 return 0; 758 759 if (env->prog->aux->ops->is_valid_access && 760 env->prog->aux->ops->is_valid_access(off, size, t, &info)) { 761 /* A non zero info.ctx_field_size indicates that this field is a 762 * candidate for later verifier transformation to load the whole 763 * field and then apply a mask when accessed with a narrower 764 * access than actual ctx access size. A zero info.ctx_field_size 765 * will only allow for whole field access and rejects any other 766 * type of narrower access. 767 */ 768 env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; 769 *reg_type = info.reg_type; 770 771 /* remember the offset of last byte accessed in ctx */ 772 if (env->prog->aux->max_ctx_offset < off + size) 773 env->prog->aux->max_ctx_offset = off + size; 774 return 0; 775 } 776 777 verbose("invalid bpf_context access off=%d size=%d\n", off, size); 778 return -EACCES; 779 } 780 781 static bool __is_pointer_value(bool allow_ptr_leaks, 782 const struct bpf_reg_state *reg) 783 { 784 if (allow_ptr_leaks) 785 return false; 786 787 switch (reg->type) { 788 case UNKNOWN_VALUE: 789 case CONST_IMM: 790 return false; 791 default: 792 return true; 793 } 794 } 795 796 static bool is_pointer_value(struct bpf_verifier_env *env, int regno) 797 { 798 return __is_pointer_value(env->allow_ptr_leaks, &env->cur_state.regs[regno]); 799 } 800 801 static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg, 802 int off, int size, bool strict) 803 { 804 int ip_align; 805 int reg_off; 806 807 /* Byte size accesses are always allowed. */ 808 if (!strict || size == 1) 809 return 0; 810 811 reg_off = reg->off; 812 if (reg->id) { 813 if (reg->aux_off_align % size) { 814 verbose("Packet access is only %u byte aligned, %d byte access not allowed\n", 815 reg->aux_off_align, size); 816 return -EACCES; 817 } 818 reg_off += reg->aux_off; 819 } 820 821 /* For platforms that do not have a Kconfig enabling 822 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of 823 * NET_IP_ALIGN is universally set to '2'. And on platforms 824 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get 825 * to this code only in strict mode where we want to emulate 826 * the NET_IP_ALIGN==2 checking. Therefore use an 827 * unconditional IP align value of '2'. 828 */ 829 ip_align = 2; 830 if ((ip_align + reg_off + off) % size != 0) { 831 verbose("misaligned packet access off %d+%d+%d size %d\n", 832 ip_align, reg_off, off, size); 833 return -EACCES; 834 } 835 836 return 0; 837 } 838 839 static int check_val_ptr_alignment(const struct bpf_reg_state *reg, 840 int size, bool strict) 841 { 842 if (strict && size != 1) { 843 verbose("Unknown alignment. Only byte-sized access allowed in value access.\n"); 844 return -EACCES; 845 } 846 847 return 0; 848 } 849 850 static int check_ptr_alignment(struct bpf_verifier_env *env, 851 const struct bpf_reg_state *reg, 852 int off, int size) 853 { 854 bool strict = env->strict_alignment; 855 856 switch (reg->type) { 857 case PTR_TO_PACKET: 858 return check_pkt_ptr_alignment(reg, off, size, strict); 859 case PTR_TO_MAP_VALUE_ADJ: 860 return check_val_ptr_alignment(reg, size, strict); 861 default: 862 if (off % size != 0) { 863 verbose("misaligned access off %d size %d\n", 864 off, size); 865 return -EACCES; 866 } 867 868 return 0; 869 } 870 } 871 872 /* check whether memory at (regno + off) is accessible for t = (read | write) 873 * if t==write, value_regno is a register which value is stored into memory 874 * if t==read, value_regno is a register which will receive the value from memory 875 * if t==write && value_regno==-1, some unknown value is stored into memory 876 * if t==read && value_regno==-1, don't care what we read from memory 877 */ 878 static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off, 879 int bpf_size, enum bpf_access_type t, 880 int value_regno) 881 { 882 struct bpf_verifier_state *state = &env->cur_state; 883 struct bpf_reg_state *reg = &state->regs[regno]; 884 int size, err = 0; 885 886 if (reg->type == PTR_TO_STACK) 887 off += reg->imm; 888 889 size = bpf_size_to_bytes(bpf_size); 890 if (size < 0) 891 return size; 892 893 err = check_ptr_alignment(env, reg, off, size); 894 if (err) 895 return err; 896 897 if (reg->type == PTR_TO_MAP_VALUE || 898 reg->type == PTR_TO_MAP_VALUE_ADJ) { 899 if (t == BPF_WRITE && value_regno >= 0 && 900 is_pointer_value(env, value_regno)) { 901 verbose("R%d leaks addr into map\n", value_regno); 902 return -EACCES; 903 } 904 905 if (reg->type == PTR_TO_MAP_VALUE_ADJ) 906 err = check_map_access_adj(env, regno, off, size); 907 else 908 err = check_map_access(env, regno, off, size); 909 if (!err && t == BPF_READ && value_regno >= 0) 910 mark_reg_unknown_value_and_range(state->regs, 911 value_regno); 912 913 } else if (reg->type == PTR_TO_CTX) { 914 enum bpf_reg_type reg_type = UNKNOWN_VALUE; 915 916 if (t == BPF_WRITE && value_regno >= 0 && 917 is_pointer_value(env, value_regno)) { 918 verbose("R%d leaks addr into ctx\n", value_regno); 919 return -EACCES; 920 } 921 err = check_ctx_access(env, insn_idx, off, size, t, ®_type); 922 if (!err && t == BPF_READ && value_regno >= 0) { 923 mark_reg_unknown_value_and_range(state->regs, 924 value_regno); 925 /* note that reg.[id|off|range] == 0 */ 926 state->regs[value_regno].type = reg_type; 927 state->regs[value_regno].aux_off = 0; 928 state->regs[value_regno].aux_off_align = 0; 929 } 930 931 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { 932 if (off >= 0 || off < -MAX_BPF_STACK) { 933 verbose("invalid stack off=%d size=%d\n", off, size); 934 return -EACCES; 935 } 936 937 if (env->prog->aux->stack_depth < -off) 938 env->prog->aux->stack_depth = -off; 939 940 if (t == BPF_WRITE) { 941 if (!env->allow_ptr_leaks && 942 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && 943 size != BPF_REG_SIZE) { 944 verbose("attempt to corrupt spilled pointer on stack\n"); 945 return -EACCES; 946 } 947 err = check_stack_write(state, off, size, value_regno); 948 } else { 949 err = check_stack_read(state, off, size, value_regno); 950 } 951 } else if (state->regs[regno].type == PTR_TO_PACKET) { 952 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { 953 verbose("cannot write into packet\n"); 954 return -EACCES; 955 } 956 if (t == BPF_WRITE && value_regno >= 0 && 957 is_pointer_value(env, value_regno)) { 958 verbose("R%d leaks addr into packet\n", value_regno); 959 return -EACCES; 960 } 961 err = check_packet_access(env, regno, off, size); 962 if (!err && t == BPF_READ && value_regno >= 0) 963 mark_reg_unknown_value_and_range(state->regs, 964 value_regno); 965 } else { 966 verbose("R%d invalid mem access '%s'\n", 967 regno, reg_type_str[reg->type]); 968 return -EACCES; 969 } 970 971 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && 972 state->regs[value_regno].type == UNKNOWN_VALUE) { 973 /* 1 or 2 byte load zero-extends, determine the number of 974 * zero upper bits. Not doing it fo 4 byte load, since 975 * such values cannot be added to ptr_to_packet anyway. 976 */ 977 state->regs[value_regno].imm = 64 - size * 8; 978 } 979 return err; 980 } 981 982 static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) 983 { 984 struct bpf_reg_state *regs = env->cur_state.regs; 985 int err; 986 987 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || 988 insn->imm != 0) { 989 verbose("BPF_XADD uses reserved fields\n"); 990 return -EINVAL; 991 } 992 993 /* check src1 operand */ 994 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 995 if (err) 996 return err; 997 998 /* check src2 operand */ 999 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1000 if (err) 1001 return err; 1002 1003 if (is_pointer_value(env, insn->src_reg)) { 1004 verbose("R%d leaks addr into mem\n", insn->src_reg); 1005 return -EACCES; 1006 } 1007 1008 /* check whether atomic_add can read the memory */ 1009 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 1010 BPF_SIZE(insn->code), BPF_READ, -1); 1011 if (err) 1012 return err; 1013 1014 /* check whether atomic_add can write into the same memory */ 1015 return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 1016 BPF_SIZE(insn->code), BPF_WRITE, -1); 1017 } 1018 1019 /* when register 'regno' is passed into function that will read 'access_size' 1020 * bytes from that pointer, make sure that it's within stack boundary 1021 * and all elements of stack are initialized 1022 */ 1023 static int check_stack_boundary(struct bpf_verifier_env *env, int regno, 1024 int access_size, bool zero_size_allowed, 1025 struct bpf_call_arg_meta *meta) 1026 { 1027 struct bpf_verifier_state *state = &env->cur_state; 1028 struct bpf_reg_state *regs = state->regs; 1029 int off, i; 1030 1031 if (regs[regno].type != PTR_TO_STACK) { 1032 if (zero_size_allowed && access_size == 0 && 1033 regs[regno].type == CONST_IMM && 1034 regs[regno].imm == 0) 1035 return 0; 1036 1037 verbose("R%d type=%s expected=%s\n", regno, 1038 reg_type_str[regs[regno].type], 1039 reg_type_str[PTR_TO_STACK]); 1040 return -EACCES; 1041 } 1042 1043 off = regs[regno].imm; 1044 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || 1045 access_size <= 0) { 1046 verbose("invalid stack type R%d off=%d access_size=%d\n", 1047 regno, off, access_size); 1048 return -EACCES; 1049 } 1050 1051 if (env->prog->aux->stack_depth < -off) 1052 env->prog->aux->stack_depth = -off; 1053 1054 if (meta && meta->raw_mode) { 1055 meta->access_size = access_size; 1056 meta->regno = regno; 1057 return 0; 1058 } 1059 1060 for (i = 0; i < access_size; i++) { 1061 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { 1062 verbose("invalid indirect read from stack off %d+%d size %d\n", 1063 off, i, access_size); 1064 return -EACCES; 1065 } 1066 } 1067 return 0; 1068 } 1069 1070 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, 1071 int access_size, bool zero_size_allowed, 1072 struct bpf_call_arg_meta *meta) 1073 { 1074 struct bpf_reg_state *regs = env->cur_state.regs; 1075 1076 switch (regs[regno].type) { 1077 case PTR_TO_PACKET: 1078 return check_packet_access(env, regno, 0, access_size); 1079 case PTR_TO_MAP_VALUE: 1080 return check_map_access(env, regno, 0, access_size); 1081 case PTR_TO_MAP_VALUE_ADJ: 1082 return check_map_access_adj(env, regno, 0, access_size); 1083 default: /* const_imm|ptr_to_stack or invalid ptr */ 1084 return check_stack_boundary(env, regno, access_size, 1085 zero_size_allowed, meta); 1086 } 1087 } 1088 1089 static int check_func_arg(struct bpf_verifier_env *env, u32 regno, 1090 enum bpf_arg_type arg_type, 1091 struct bpf_call_arg_meta *meta) 1092 { 1093 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; 1094 enum bpf_reg_type expected_type, type = reg->type; 1095 int err = 0; 1096 1097 if (arg_type == ARG_DONTCARE) 1098 return 0; 1099 1100 if (type == NOT_INIT) { 1101 verbose("R%d !read_ok\n", regno); 1102 return -EACCES; 1103 } 1104 1105 if (arg_type == ARG_ANYTHING) { 1106 if (is_pointer_value(env, regno)) { 1107 verbose("R%d leaks addr into helper function\n", regno); 1108 return -EACCES; 1109 } 1110 return 0; 1111 } 1112 1113 if (type == PTR_TO_PACKET && 1114 !may_access_direct_pkt_data(env, meta, BPF_READ)) { 1115 verbose("helper access to the packet is not allowed\n"); 1116 return -EACCES; 1117 } 1118 1119 if (arg_type == ARG_PTR_TO_MAP_KEY || 1120 arg_type == ARG_PTR_TO_MAP_VALUE) { 1121 expected_type = PTR_TO_STACK; 1122 if (type != PTR_TO_PACKET && type != expected_type) 1123 goto err_type; 1124 } else if (arg_type == ARG_CONST_SIZE || 1125 arg_type == ARG_CONST_SIZE_OR_ZERO) { 1126 expected_type = CONST_IMM; 1127 /* One exception. Allow UNKNOWN_VALUE registers when the 1128 * boundaries are known and don't cause unsafe memory accesses 1129 */ 1130 if (type != UNKNOWN_VALUE && type != expected_type) 1131 goto err_type; 1132 } else if (arg_type == ARG_CONST_MAP_PTR) { 1133 expected_type = CONST_PTR_TO_MAP; 1134 if (type != expected_type) 1135 goto err_type; 1136 } else if (arg_type == ARG_PTR_TO_CTX) { 1137 expected_type = PTR_TO_CTX; 1138 if (type != expected_type) 1139 goto err_type; 1140 } else if (arg_type == ARG_PTR_TO_MEM || 1141 arg_type == ARG_PTR_TO_UNINIT_MEM) { 1142 expected_type = PTR_TO_STACK; 1143 /* One exception here. In case function allows for NULL to be 1144 * passed in as argument, it's a CONST_IMM type. Final test 1145 * happens during stack boundary checking. 1146 */ 1147 if (type == CONST_IMM && reg->imm == 0) 1148 /* final test in check_stack_boundary() */; 1149 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE && 1150 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type) 1151 goto err_type; 1152 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; 1153 } else { 1154 verbose("unsupported arg_type %d\n", arg_type); 1155 return -EFAULT; 1156 } 1157 1158 if (arg_type == ARG_CONST_MAP_PTR) { 1159 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ 1160 meta->map_ptr = reg->map_ptr; 1161 } else if (arg_type == ARG_PTR_TO_MAP_KEY) { 1162 /* bpf_map_xxx(..., map_ptr, ..., key) call: 1163 * check that [key, key + map->key_size) are within 1164 * stack limits and initialized 1165 */ 1166 if (!meta->map_ptr) { 1167 /* in function declaration map_ptr must come before 1168 * map_key, so that it's verified and known before 1169 * we have to check map_key here. Otherwise it means 1170 * that kernel subsystem misconfigured verifier 1171 */ 1172 verbose("invalid map_ptr to access map->key\n"); 1173 return -EACCES; 1174 } 1175 if (type == PTR_TO_PACKET) 1176 err = check_packet_access(env, regno, 0, 1177 meta->map_ptr->key_size); 1178 else 1179 err = check_stack_boundary(env, regno, 1180 meta->map_ptr->key_size, 1181 false, NULL); 1182 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { 1183 /* bpf_map_xxx(..., map_ptr, ..., value) call: 1184 * check [value, value + map->value_size) validity 1185 */ 1186 if (!meta->map_ptr) { 1187 /* kernel subsystem misconfigured verifier */ 1188 verbose("invalid map_ptr to access map->value\n"); 1189 return -EACCES; 1190 } 1191 if (type == PTR_TO_PACKET) 1192 err = check_packet_access(env, regno, 0, 1193 meta->map_ptr->value_size); 1194 else 1195 err = check_stack_boundary(env, regno, 1196 meta->map_ptr->value_size, 1197 false, NULL); 1198 } else if (arg_type == ARG_CONST_SIZE || 1199 arg_type == ARG_CONST_SIZE_OR_ZERO) { 1200 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); 1201 1202 /* bpf_xxx(..., buf, len) call will access 'len' bytes 1203 * from stack pointer 'buf'. Check it 1204 * note: regno == len, regno - 1 == buf 1205 */ 1206 if (regno == 0) { 1207 /* kernel subsystem misconfigured verifier */ 1208 verbose("ARG_CONST_SIZE cannot be first argument\n"); 1209 return -EACCES; 1210 } 1211 1212 /* If the register is UNKNOWN_VALUE, the access check happens 1213 * using its boundaries. Otherwise, just use its imm 1214 */ 1215 if (type == UNKNOWN_VALUE) { 1216 /* For unprivileged variable accesses, disable raw 1217 * mode so that the program is required to 1218 * initialize all the memory that the helper could 1219 * just partially fill up. 1220 */ 1221 meta = NULL; 1222 1223 if (reg->min_value < 0) { 1224 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n", 1225 regno); 1226 return -EACCES; 1227 } 1228 1229 if (reg->min_value == 0) { 1230 err = check_helper_mem_access(env, regno - 1, 0, 1231 zero_size_allowed, 1232 meta); 1233 if (err) 1234 return err; 1235 } 1236 1237 if (reg->max_value == BPF_REGISTER_MAX_RANGE) { 1238 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", 1239 regno); 1240 return -EACCES; 1241 } 1242 err = check_helper_mem_access(env, regno - 1, 1243 reg->max_value, 1244 zero_size_allowed, meta); 1245 if (err) 1246 return err; 1247 } else { 1248 /* register is CONST_IMM */ 1249 err = check_helper_mem_access(env, regno - 1, reg->imm, 1250 zero_size_allowed, meta); 1251 } 1252 } 1253 1254 return err; 1255 err_type: 1256 verbose("R%d type=%s expected=%s\n", regno, 1257 reg_type_str[type], reg_type_str[expected_type]); 1258 return -EACCES; 1259 } 1260 1261 static int check_map_func_compatibility(struct bpf_map *map, int func_id) 1262 { 1263 if (!map) 1264 return 0; 1265 1266 /* We need a two way check, first is from map perspective ... */ 1267 switch (map->map_type) { 1268 case BPF_MAP_TYPE_PROG_ARRAY: 1269 if (func_id != BPF_FUNC_tail_call) 1270 goto error; 1271 break; 1272 case BPF_MAP_TYPE_PERF_EVENT_ARRAY: 1273 if (func_id != BPF_FUNC_perf_event_read && 1274 func_id != BPF_FUNC_perf_event_output) 1275 goto error; 1276 break; 1277 case BPF_MAP_TYPE_STACK_TRACE: 1278 if (func_id != BPF_FUNC_get_stackid) 1279 goto error; 1280 break; 1281 case BPF_MAP_TYPE_CGROUP_ARRAY: 1282 if (func_id != BPF_FUNC_skb_under_cgroup && 1283 func_id != BPF_FUNC_current_task_under_cgroup) 1284 goto error; 1285 break; 1286 /* devmap returns a pointer to a live net_device ifindex that we cannot 1287 * allow to be modified from bpf side. So do not allow lookup elements 1288 * for now. 1289 */ 1290 case BPF_MAP_TYPE_DEVMAP: 1291 if (func_id != BPF_FUNC_redirect_map) 1292 goto error; 1293 break; 1294 case BPF_MAP_TYPE_ARRAY_OF_MAPS: 1295 case BPF_MAP_TYPE_HASH_OF_MAPS: 1296 if (func_id != BPF_FUNC_map_lookup_elem) 1297 goto error; 1298 default: 1299 break; 1300 } 1301 1302 /* ... and second from the function itself. */ 1303 switch (func_id) { 1304 case BPF_FUNC_tail_call: 1305 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 1306 goto error; 1307 break; 1308 case BPF_FUNC_perf_event_read: 1309 case BPF_FUNC_perf_event_output: 1310 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) 1311 goto error; 1312 break; 1313 case BPF_FUNC_get_stackid: 1314 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) 1315 goto error; 1316 break; 1317 case BPF_FUNC_current_task_under_cgroup: 1318 case BPF_FUNC_skb_under_cgroup: 1319 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) 1320 goto error; 1321 break; 1322 case BPF_FUNC_redirect_map: 1323 if (map->map_type != BPF_MAP_TYPE_DEVMAP) 1324 goto error; 1325 break; 1326 default: 1327 break; 1328 } 1329 1330 return 0; 1331 error: 1332 verbose("cannot pass map_type %d into func %s#%d\n", 1333 map->map_type, func_id_name(func_id), func_id); 1334 return -EINVAL; 1335 } 1336 1337 static int check_raw_mode(const struct bpf_func_proto *fn) 1338 { 1339 int count = 0; 1340 1341 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) 1342 count++; 1343 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) 1344 count++; 1345 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) 1346 count++; 1347 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) 1348 count++; 1349 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) 1350 count++; 1351 1352 return count > 1 ? -EINVAL : 0; 1353 } 1354 1355 static void clear_all_pkt_pointers(struct bpf_verifier_env *env) 1356 { 1357 struct bpf_verifier_state *state = &env->cur_state; 1358 struct bpf_reg_state *regs = state->regs, *reg; 1359 int i; 1360 1361 for (i = 0; i < MAX_BPF_REG; i++) 1362 if (regs[i].type == PTR_TO_PACKET || 1363 regs[i].type == PTR_TO_PACKET_END) 1364 mark_reg_unknown_value(regs, i); 1365 1366 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1367 if (state->stack_slot_type[i] != STACK_SPILL) 1368 continue; 1369 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1370 if (reg->type != PTR_TO_PACKET && 1371 reg->type != PTR_TO_PACKET_END) 1372 continue; 1373 __mark_reg_unknown_value(state->spilled_regs, 1374 i / BPF_REG_SIZE); 1375 } 1376 } 1377 1378 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) 1379 { 1380 struct bpf_verifier_state *state = &env->cur_state; 1381 const struct bpf_func_proto *fn = NULL; 1382 struct bpf_reg_state *regs = state->regs; 1383 struct bpf_call_arg_meta meta; 1384 bool changes_data; 1385 int i, err; 1386 1387 /* find function prototype */ 1388 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { 1389 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id); 1390 return -EINVAL; 1391 } 1392 1393 if (env->prog->aux->ops->get_func_proto) 1394 fn = env->prog->aux->ops->get_func_proto(func_id); 1395 1396 if (!fn) { 1397 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id); 1398 return -EINVAL; 1399 } 1400 1401 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1402 if (!env->prog->gpl_compatible && fn->gpl_only) { 1403 verbose("cannot call GPL only function from proprietary program\n"); 1404 return -EINVAL; 1405 } 1406 1407 changes_data = bpf_helper_changes_pkt_data(fn->func); 1408 1409 memset(&meta, 0, sizeof(meta)); 1410 meta.pkt_access = fn->pkt_access; 1411 1412 /* We only support one arg being in raw mode at the moment, which 1413 * is sufficient for the helper functions we have right now. 1414 */ 1415 err = check_raw_mode(fn); 1416 if (err) { 1417 verbose("kernel subsystem misconfigured func %s#%d\n", 1418 func_id_name(func_id), func_id); 1419 return err; 1420 } 1421 1422 /* check args */ 1423 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); 1424 if (err) 1425 return err; 1426 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); 1427 if (err) 1428 return err; 1429 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); 1430 if (err) 1431 return err; 1432 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); 1433 if (err) 1434 return err; 1435 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); 1436 if (err) 1437 return err; 1438 1439 /* Mark slots with STACK_MISC in case of raw mode, stack offset 1440 * is inferred from register state. 1441 */ 1442 for (i = 0; i < meta.access_size; i++) { 1443 err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1); 1444 if (err) 1445 return err; 1446 } 1447 1448 /* reset caller saved regs */ 1449 for (i = 0; i < CALLER_SAVED_REGS; i++) 1450 mark_reg_not_init(regs, caller_saved[i]); 1451 1452 /* update return register */ 1453 if (fn->ret_type == RET_INTEGER) { 1454 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1455 } else if (fn->ret_type == RET_VOID) { 1456 regs[BPF_REG_0].type = NOT_INIT; 1457 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { 1458 struct bpf_insn_aux_data *insn_aux; 1459 1460 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; 1461 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0; 1462 /* remember map_ptr, so that check_map_access() 1463 * can check 'value_size' boundary of memory access 1464 * to map element returned from bpf_map_lookup_elem() 1465 */ 1466 if (meta.map_ptr == NULL) { 1467 verbose("kernel subsystem misconfigured verifier\n"); 1468 return -EINVAL; 1469 } 1470 regs[BPF_REG_0].map_ptr = meta.map_ptr; 1471 regs[BPF_REG_0].id = ++env->id_gen; 1472 insn_aux = &env->insn_aux_data[insn_idx]; 1473 if (!insn_aux->map_ptr) 1474 insn_aux->map_ptr = meta.map_ptr; 1475 else if (insn_aux->map_ptr != meta.map_ptr) 1476 insn_aux->map_ptr = BPF_MAP_PTR_POISON; 1477 } else { 1478 verbose("unknown return type %d of func %s#%d\n", 1479 fn->ret_type, func_id_name(func_id), func_id); 1480 return -EINVAL; 1481 } 1482 1483 err = check_map_func_compatibility(meta.map_ptr, func_id); 1484 if (err) 1485 return err; 1486 1487 if (changes_data) 1488 clear_all_pkt_pointers(env); 1489 return 0; 1490 } 1491 1492 static int check_packet_ptr_add(struct bpf_verifier_env *env, 1493 struct bpf_insn *insn) 1494 { 1495 struct bpf_reg_state *regs = env->cur_state.regs; 1496 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1497 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1498 struct bpf_reg_state tmp_reg; 1499 s32 imm; 1500 1501 if (BPF_SRC(insn->code) == BPF_K) { 1502 /* pkt_ptr += imm */ 1503 imm = insn->imm; 1504 1505 add_imm: 1506 if (imm < 0) { 1507 verbose("addition of negative constant to packet pointer is not allowed\n"); 1508 return -EACCES; 1509 } 1510 if (imm >= MAX_PACKET_OFF || 1511 imm + dst_reg->off >= MAX_PACKET_OFF) { 1512 verbose("constant %d is too large to add to packet pointer\n", 1513 imm); 1514 return -EACCES; 1515 } 1516 /* a constant was added to pkt_ptr. 1517 * Remember it while keeping the same 'id' 1518 */ 1519 dst_reg->off += imm; 1520 } else { 1521 bool had_id; 1522 1523 if (src_reg->type == PTR_TO_PACKET) { 1524 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ 1525 tmp_reg = *dst_reg; /* save r7 state */ 1526 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ 1527 src_reg = &tmp_reg; /* pretend it's src_reg state */ 1528 /* if the checks below reject it, the copy won't matter, 1529 * since we're rejecting the whole program. If all ok, 1530 * then imm22 state will be added to r7 1531 * and r7 will be pkt(id=0,off=22,r=62) while 1532 * r6 will stay as pkt(id=0,off=0,r=62) 1533 */ 1534 } 1535 1536 if (src_reg->type == CONST_IMM) { 1537 /* pkt_ptr += reg where reg is known constant */ 1538 imm = src_reg->imm; 1539 goto add_imm; 1540 } 1541 /* disallow pkt_ptr += reg 1542 * if reg is not uknown_value with guaranteed zero upper bits 1543 * otherwise pkt_ptr may overflow and addition will become 1544 * subtraction which is not allowed 1545 */ 1546 if (src_reg->type != UNKNOWN_VALUE) { 1547 verbose("cannot add '%s' to ptr_to_packet\n", 1548 reg_type_str[src_reg->type]); 1549 return -EACCES; 1550 } 1551 if (src_reg->imm < 48) { 1552 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", 1553 src_reg->imm); 1554 return -EACCES; 1555 } 1556 1557 had_id = (dst_reg->id != 0); 1558 1559 /* dst_reg stays as pkt_ptr type and since some positive 1560 * integer value was added to the pointer, increment its 'id' 1561 */ 1562 dst_reg->id = ++env->id_gen; 1563 1564 /* something was added to pkt_ptr, set range to zero */ 1565 dst_reg->aux_off += dst_reg->off; 1566 dst_reg->off = 0; 1567 dst_reg->range = 0; 1568 if (had_id) 1569 dst_reg->aux_off_align = min(dst_reg->aux_off_align, 1570 src_reg->min_align); 1571 else 1572 dst_reg->aux_off_align = src_reg->min_align; 1573 } 1574 return 0; 1575 } 1576 1577 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) 1578 { 1579 struct bpf_reg_state *regs = env->cur_state.regs; 1580 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1581 u8 opcode = BPF_OP(insn->code); 1582 s64 imm_log2; 1583 1584 /* for type == UNKNOWN_VALUE: 1585 * imm > 0 -> number of zero upper bits 1586 * imm == 0 -> don't track which is the same as all bits can be non-zero 1587 */ 1588 1589 if (BPF_SRC(insn->code) == BPF_X) { 1590 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1591 1592 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && 1593 dst_reg->imm && opcode == BPF_ADD) { 1594 /* dreg += sreg 1595 * where both have zero upper bits. Adding them 1596 * can only result making one more bit non-zero 1597 * in the larger value. 1598 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1599 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1600 */ 1601 dst_reg->imm = min(dst_reg->imm, src_reg->imm); 1602 dst_reg->imm--; 1603 return 0; 1604 } 1605 if (src_reg->type == CONST_IMM && src_reg->imm > 0 && 1606 dst_reg->imm && opcode == BPF_ADD) { 1607 /* dreg += sreg 1608 * where dreg has zero upper bits and sreg is const. 1609 * Adding them can only result making one more bit 1610 * non-zero in the larger value. 1611 */ 1612 imm_log2 = __ilog2_u64((long long)src_reg->imm); 1613 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1614 dst_reg->imm--; 1615 return 0; 1616 } 1617 /* all other cases non supported yet, just mark dst_reg */ 1618 dst_reg->imm = 0; 1619 return 0; 1620 } 1621 1622 /* sign extend 32-bit imm into 64-bit to make sure that 1623 * negative values occupy bit 63. Note ilog2() would have 1624 * been incorrect, since sizeof(insn->imm) == 4 1625 */ 1626 imm_log2 = __ilog2_u64((long long)insn->imm); 1627 1628 if (dst_reg->imm && opcode == BPF_LSH) { 1629 /* reg <<= imm 1630 * if reg was a result of 2 byte load, then its imm == 48 1631 * which means that upper 48 bits are zero and shifting this reg 1632 * left by 4 would mean that upper 44 bits are still zero 1633 */ 1634 dst_reg->imm -= insn->imm; 1635 } else if (dst_reg->imm && opcode == BPF_MUL) { 1636 /* reg *= imm 1637 * if multiplying by 14 subtract 4 1638 * This is conservative calculation of upper zero bits. 1639 * It's not trying to special case insn->imm == 1 or 0 cases 1640 */ 1641 dst_reg->imm -= imm_log2 + 1; 1642 } else if (opcode == BPF_AND) { 1643 /* reg &= imm */ 1644 dst_reg->imm = 63 - imm_log2; 1645 } else if (dst_reg->imm && opcode == BPF_ADD) { 1646 /* reg += imm */ 1647 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1648 dst_reg->imm--; 1649 } else if (opcode == BPF_RSH) { 1650 /* reg >>= imm 1651 * which means that after right shift, upper bits will be zero 1652 * note that verifier already checked that 1653 * 0 <= imm < 64 for shift insn 1654 */ 1655 dst_reg->imm += insn->imm; 1656 if (unlikely(dst_reg->imm > 64)) 1657 /* some dumb code did: 1658 * r2 = *(u32 *)mem; 1659 * r2 >>= 32; 1660 * and all bits are zero now */ 1661 dst_reg->imm = 64; 1662 } else { 1663 /* all other alu ops, means that we don't know what will 1664 * happen to the value, mark it with unknown number of zero bits 1665 */ 1666 dst_reg->imm = 0; 1667 } 1668 1669 if (dst_reg->imm < 0) { 1670 /* all 64 bits of the register can contain non-zero bits 1671 * and such value cannot be added to ptr_to_packet, since it 1672 * may overflow, mark it as unknown to avoid further eval 1673 */ 1674 dst_reg->imm = 0; 1675 } 1676 return 0; 1677 } 1678 1679 static int evaluate_reg_imm_alu_unknown(struct bpf_verifier_env *env, 1680 struct bpf_insn *insn) 1681 { 1682 struct bpf_reg_state *regs = env->cur_state.regs; 1683 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1684 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1685 u8 opcode = BPF_OP(insn->code); 1686 s64 imm_log2 = __ilog2_u64((long long)dst_reg->imm); 1687 1688 /* BPF_X code with src_reg->type UNKNOWN_VALUE here. */ 1689 if (src_reg->imm > 0 && dst_reg->imm) { 1690 switch (opcode) { 1691 case BPF_ADD: 1692 /* dreg += sreg 1693 * where both have zero upper bits. Adding them 1694 * can only result making one more bit non-zero 1695 * in the larger value. 1696 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1697 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1698 */ 1699 dst_reg->imm = min(src_reg->imm, 63 - imm_log2); 1700 dst_reg->imm--; 1701 break; 1702 case BPF_AND: 1703 /* dreg &= sreg 1704 * AND can not extend zero bits only shrink 1705 * Ex. 0x00..00ffffff 1706 * & 0x0f..ffffffff 1707 * ---------------- 1708 * 0x00..00ffffff 1709 */ 1710 dst_reg->imm = max(src_reg->imm, 63 - imm_log2); 1711 break; 1712 case BPF_OR: 1713 /* dreg |= sreg 1714 * OR can only extend zero bits 1715 * Ex. 0x00..00ffffff 1716 * | 0x0f..ffffffff 1717 * ---------------- 1718 * 0x0f..00ffffff 1719 */ 1720 dst_reg->imm = min(src_reg->imm, 63 - imm_log2); 1721 break; 1722 case BPF_SUB: 1723 case BPF_MUL: 1724 case BPF_RSH: 1725 case BPF_LSH: 1726 /* These may be flushed out later */ 1727 default: 1728 mark_reg_unknown_value(regs, insn->dst_reg); 1729 } 1730 } else { 1731 mark_reg_unknown_value(regs, insn->dst_reg); 1732 } 1733 1734 dst_reg->type = UNKNOWN_VALUE; 1735 return 0; 1736 } 1737 1738 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, 1739 struct bpf_insn *insn) 1740 { 1741 struct bpf_reg_state *regs = env->cur_state.regs; 1742 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1743 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1744 u8 opcode = BPF_OP(insn->code); 1745 u64 dst_imm = dst_reg->imm; 1746 1747 if (BPF_SRC(insn->code) == BPF_X && src_reg->type == UNKNOWN_VALUE) 1748 return evaluate_reg_imm_alu_unknown(env, insn); 1749 1750 /* dst_reg->type == CONST_IMM here. Simulate execution of insns 1751 * containing ALU ops. Don't care about overflow or negative 1752 * values, just add/sub/... them; registers are in u64. 1753 */ 1754 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) { 1755 dst_imm += insn->imm; 1756 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && 1757 src_reg->type == CONST_IMM) { 1758 dst_imm += src_reg->imm; 1759 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) { 1760 dst_imm -= insn->imm; 1761 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X && 1762 src_reg->type == CONST_IMM) { 1763 dst_imm -= src_reg->imm; 1764 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) { 1765 dst_imm *= insn->imm; 1766 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X && 1767 src_reg->type == CONST_IMM) { 1768 dst_imm *= src_reg->imm; 1769 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) { 1770 dst_imm |= insn->imm; 1771 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X && 1772 src_reg->type == CONST_IMM) { 1773 dst_imm |= src_reg->imm; 1774 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) { 1775 dst_imm &= insn->imm; 1776 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X && 1777 src_reg->type == CONST_IMM) { 1778 dst_imm &= src_reg->imm; 1779 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) { 1780 dst_imm >>= insn->imm; 1781 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X && 1782 src_reg->type == CONST_IMM) { 1783 dst_imm >>= src_reg->imm; 1784 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) { 1785 dst_imm <<= insn->imm; 1786 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X && 1787 src_reg->type == CONST_IMM) { 1788 dst_imm <<= src_reg->imm; 1789 } else { 1790 mark_reg_unknown_value(regs, insn->dst_reg); 1791 goto out; 1792 } 1793 1794 dst_reg->imm = dst_imm; 1795 out: 1796 return 0; 1797 } 1798 1799 static void check_reg_overflow(struct bpf_reg_state *reg) 1800 { 1801 if (reg->max_value > BPF_REGISTER_MAX_RANGE) 1802 reg->max_value = BPF_REGISTER_MAX_RANGE; 1803 if (reg->min_value < BPF_REGISTER_MIN_RANGE || 1804 reg->min_value > BPF_REGISTER_MAX_RANGE) 1805 reg->min_value = BPF_REGISTER_MIN_RANGE; 1806 } 1807 1808 static u32 calc_align(u32 imm) 1809 { 1810 if (!imm) 1811 return 1U << 31; 1812 return imm - ((imm - 1) & imm); 1813 } 1814 1815 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env, 1816 struct bpf_insn *insn) 1817 { 1818 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1819 s64 min_val = BPF_REGISTER_MIN_RANGE; 1820 u64 max_val = BPF_REGISTER_MAX_RANGE; 1821 u8 opcode = BPF_OP(insn->code); 1822 u32 dst_align, src_align; 1823 1824 dst_reg = ®s[insn->dst_reg]; 1825 src_align = 0; 1826 if (BPF_SRC(insn->code) == BPF_X) { 1827 check_reg_overflow(®s[insn->src_reg]); 1828 min_val = regs[insn->src_reg].min_value; 1829 max_val = regs[insn->src_reg].max_value; 1830 1831 /* If the source register is a random pointer then the 1832 * min_value/max_value values represent the range of the known 1833 * accesses into that value, not the actual min/max value of the 1834 * register itself. In this case we have to reset the reg range 1835 * values so we know it is not safe to look at. 1836 */ 1837 if (regs[insn->src_reg].type != CONST_IMM && 1838 regs[insn->src_reg].type != UNKNOWN_VALUE) { 1839 min_val = BPF_REGISTER_MIN_RANGE; 1840 max_val = BPF_REGISTER_MAX_RANGE; 1841 src_align = 0; 1842 } else { 1843 src_align = regs[insn->src_reg].min_align; 1844 } 1845 } else if (insn->imm < BPF_REGISTER_MAX_RANGE && 1846 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) { 1847 min_val = max_val = insn->imm; 1848 src_align = calc_align(insn->imm); 1849 } 1850 1851 dst_align = dst_reg->min_align; 1852 1853 /* We don't know anything about what was done to this register, mark it 1854 * as unknown. Also, if both derived bounds came from signed/unsigned 1855 * mixed compares and one side is unbounded, we cannot really do anything 1856 * with them as boundaries cannot be trusted. Thus, arithmetic of two 1857 * regs of such kind will get invalidated bounds on the dst side. 1858 */ 1859 if ((min_val == BPF_REGISTER_MIN_RANGE && 1860 max_val == BPF_REGISTER_MAX_RANGE) || 1861 (BPF_SRC(insn->code) == BPF_X && 1862 ((min_val != BPF_REGISTER_MIN_RANGE && 1863 max_val == BPF_REGISTER_MAX_RANGE) || 1864 (min_val == BPF_REGISTER_MIN_RANGE && 1865 max_val != BPF_REGISTER_MAX_RANGE) || 1866 (dst_reg->min_value != BPF_REGISTER_MIN_RANGE && 1867 dst_reg->max_value == BPF_REGISTER_MAX_RANGE) || 1868 (dst_reg->min_value == BPF_REGISTER_MIN_RANGE && 1869 dst_reg->max_value != BPF_REGISTER_MAX_RANGE)) && 1870 regs[insn->dst_reg].value_from_signed != 1871 regs[insn->src_reg].value_from_signed)) { 1872 reset_reg_range_values(regs, insn->dst_reg); 1873 return; 1874 } 1875 1876 /* If one of our values was at the end of our ranges then we can't just 1877 * do our normal operations to the register, we need to set the values 1878 * to the min/max since they are undefined. 1879 */ 1880 if (opcode != BPF_SUB) { 1881 if (min_val == BPF_REGISTER_MIN_RANGE) 1882 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1883 if (max_val == BPF_REGISTER_MAX_RANGE) 1884 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1885 } 1886 1887 switch (opcode) { 1888 case BPF_ADD: 1889 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1890 dst_reg->min_value += min_val; 1891 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1892 dst_reg->max_value += max_val; 1893 dst_reg->min_align = min(src_align, dst_align); 1894 break; 1895 case BPF_SUB: 1896 /* If one of our values was at the end of our ranges, then the 1897 * _opposite_ value in the dst_reg goes to the end of our range. 1898 */ 1899 if (min_val == BPF_REGISTER_MIN_RANGE) 1900 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1901 if (max_val == BPF_REGISTER_MAX_RANGE) 1902 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1903 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1904 dst_reg->min_value -= max_val; 1905 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1906 dst_reg->max_value -= min_val; 1907 dst_reg->min_align = min(src_align, dst_align); 1908 break; 1909 case BPF_MUL: 1910 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1911 dst_reg->min_value *= min_val; 1912 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1913 dst_reg->max_value *= max_val; 1914 dst_reg->min_align = max(src_align, dst_align); 1915 break; 1916 case BPF_AND: 1917 /* Disallow AND'ing of negative numbers, ain't nobody got time 1918 * for that. Otherwise the minimum is 0 and the max is the max 1919 * value we could AND against. 1920 */ 1921 if (min_val < 0) 1922 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1923 else 1924 dst_reg->min_value = 0; 1925 dst_reg->max_value = max_val; 1926 dst_reg->min_align = max(src_align, dst_align); 1927 break; 1928 case BPF_LSH: 1929 /* Gotta have special overflow logic here, if we're shifting 1930 * more than MAX_RANGE then just assume we have an invalid 1931 * range. 1932 */ 1933 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) { 1934 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1935 dst_reg->min_align = 1; 1936 } else { 1937 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1938 dst_reg->min_value <<= min_val; 1939 if (!dst_reg->min_align) 1940 dst_reg->min_align = 1; 1941 dst_reg->min_align <<= min_val; 1942 } 1943 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) 1944 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1945 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1946 dst_reg->max_value <<= max_val; 1947 break; 1948 case BPF_RSH: 1949 /* RSH by a negative number is undefined, and the BPF_RSH is an 1950 * unsigned shift, so make the appropriate casts. 1951 */ 1952 if (min_val < 0 || dst_reg->min_value < 0) { 1953 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1954 } else { 1955 dst_reg->min_value = 1956 (u64)(dst_reg->min_value) >> min_val; 1957 } 1958 if (min_val < 0) { 1959 dst_reg->min_align = 1; 1960 } else { 1961 dst_reg->min_align >>= (u64) min_val; 1962 if (!dst_reg->min_align) 1963 dst_reg->min_align = 1; 1964 } 1965 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1966 dst_reg->max_value >>= max_val; 1967 break; 1968 default: 1969 reset_reg_range_values(regs, insn->dst_reg); 1970 break; 1971 } 1972 1973 check_reg_overflow(dst_reg); 1974 } 1975 1976 /* check validity of 32-bit and 64-bit arithmetic operations */ 1977 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) 1978 { 1979 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1980 u8 opcode = BPF_OP(insn->code); 1981 int err; 1982 1983 if (opcode == BPF_END || opcode == BPF_NEG) { 1984 if (opcode == BPF_NEG) { 1985 if (BPF_SRC(insn->code) != 0 || 1986 insn->src_reg != BPF_REG_0 || 1987 insn->off != 0 || insn->imm != 0) { 1988 verbose("BPF_NEG uses reserved fields\n"); 1989 return -EINVAL; 1990 } 1991 } else { 1992 if (insn->src_reg != BPF_REG_0 || insn->off != 0 || 1993 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { 1994 verbose("BPF_END uses reserved fields\n"); 1995 return -EINVAL; 1996 } 1997 } 1998 1999 /* check src operand */ 2000 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2001 if (err) 2002 return err; 2003 2004 if (is_pointer_value(env, insn->dst_reg)) { 2005 verbose("R%d pointer arithmetic prohibited\n", 2006 insn->dst_reg); 2007 return -EACCES; 2008 } 2009 2010 /* check dest operand */ 2011 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2012 if (err) 2013 return err; 2014 2015 } else if (opcode == BPF_MOV) { 2016 2017 if (BPF_SRC(insn->code) == BPF_X) { 2018 if (insn->imm != 0 || insn->off != 0) { 2019 verbose("BPF_MOV uses reserved fields\n"); 2020 return -EINVAL; 2021 } 2022 2023 /* check src operand */ 2024 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2025 if (err) 2026 return err; 2027 } else { 2028 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 2029 verbose("BPF_MOV uses reserved fields\n"); 2030 return -EINVAL; 2031 } 2032 } 2033 2034 /* check dest operand */ 2035 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2036 if (err) 2037 return err; 2038 2039 /* we are setting our register to something new, we need to 2040 * reset its range values. 2041 */ 2042 reset_reg_range_values(regs, insn->dst_reg); 2043 2044 if (BPF_SRC(insn->code) == BPF_X) { 2045 if (BPF_CLASS(insn->code) == BPF_ALU64) { 2046 /* case: R1 = R2 2047 * copy register state to dest reg 2048 */ 2049 regs[insn->dst_reg] = regs[insn->src_reg]; 2050 } else { 2051 if (is_pointer_value(env, insn->src_reg)) { 2052 verbose("R%d partial copy of pointer\n", 2053 insn->src_reg); 2054 return -EACCES; 2055 } 2056 mark_reg_unknown_value(regs, insn->dst_reg); 2057 } 2058 } else { 2059 /* case: R = imm 2060 * remember the value we stored into this reg 2061 */ 2062 regs[insn->dst_reg].type = CONST_IMM; 2063 regs[insn->dst_reg].imm = insn->imm; 2064 regs[insn->dst_reg].id = 0; 2065 regs[insn->dst_reg].max_value = insn->imm; 2066 regs[insn->dst_reg].min_value = insn->imm; 2067 regs[insn->dst_reg].min_align = calc_align(insn->imm); 2068 regs[insn->dst_reg].value_from_signed = false; 2069 } 2070 2071 } else if (opcode > BPF_END) { 2072 verbose("invalid BPF_ALU opcode %x\n", opcode); 2073 return -EINVAL; 2074 2075 } else { /* all other ALU ops: and, sub, xor, add, ... */ 2076 2077 if (BPF_SRC(insn->code) == BPF_X) { 2078 if (insn->imm != 0 || insn->off != 0) { 2079 verbose("BPF_ALU uses reserved fields\n"); 2080 return -EINVAL; 2081 } 2082 /* check src1 operand */ 2083 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2084 if (err) 2085 return err; 2086 } else { 2087 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 2088 verbose("BPF_ALU uses reserved fields\n"); 2089 return -EINVAL; 2090 } 2091 } 2092 2093 /* check src2 operand */ 2094 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2095 if (err) 2096 return err; 2097 2098 if ((opcode == BPF_MOD || opcode == BPF_DIV) && 2099 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { 2100 verbose("div by zero\n"); 2101 return -EINVAL; 2102 } 2103 2104 if ((opcode == BPF_LSH || opcode == BPF_RSH || 2105 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { 2106 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; 2107 2108 if (insn->imm < 0 || insn->imm >= size) { 2109 verbose("invalid shift %d\n", insn->imm); 2110 return -EINVAL; 2111 } 2112 } 2113 2114 /* check dest operand */ 2115 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 2116 if (err) 2117 return err; 2118 2119 dst_reg = ®s[insn->dst_reg]; 2120 2121 /* first we want to adjust our ranges. */ 2122 adjust_reg_min_max_vals(env, insn); 2123 2124 /* pattern match 'bpf_add Rx, imm' instruction */ 2125 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && 2126 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { 2127 dst_reg->type = PTR_TO_STACK; 2128 dst_reg->imm = insn->imm; 2129 return 0; 2130 } else if (opcode == BPF_ADD && 2131 BPF_CLASS(insn->code) == BPF_ALU64 && 2132 dst_reg->type == PTR_TO_STACK && 2133 ((BPF_SRC(insn->code) == BPF_X && 2134 regs[insn->src_reg].type == CONST_IMM) || 2135 BPF_SRC(insn->code) == BPF_K)) { 2136 if (BPF_SRC(insn->code) == BPF_X) 2137 dst_reg->imm += regs[insn->src_reg].imm; 2138 else 2139 dst_reg->imm += insn->imm; 2140 return 0; 2141 } else if (opcode == BPF_ADD && 2142 BPF_CLASS(insn->code) == BPF_ALU64 && 2143 (dst_reg->type == PTR_TO_PACKET || 2144 (BPF_SRC(insn->code) == BPF_X && 2145 regs[insn->src_reg].type == PTR_TO_PACKET))) { 2146 /* ptr_to_packet += K|X */ 2147 return check_packet_ptr_add(env, insn); 2148 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 2149 dst_reg->type == UNKNOWN_VALUE && 2150 env->allow_ptr_leaks) { 2151 /* unknown += K|X */ 2152 return evaluate_reg_alu(env, insn); 2153 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 2154 dst_reg->type == CONST_IMM && 2155 env->allow_ptr_leaks) { 2156 /* reg_imm += K|X */ 2157 return evaluate_reg_imm_alu(env, insn); 2158 } else if (is_pointer_value(env, insn->dst_reg)) { 2159 verbose("R%d pointer arithmetic prohibited\n", 2160 insn->dst_reg); 2161 return -EACCES; 2162 } else if (BPF_SRC(insn->code) == BPF_X && 2163 is_pointer_value(env, insn->src_reg)) { 2164 verbose("R%d pointer arithmetic prohibited\n", 2165 insn->src_reg); 2166 return -EACCES; 2167 } 2168 2169 /* If we did pointer math on a map value then just set it to our 2170 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or 2171 * loads to this register appropriately, otherwise just mark the 2172 * register as unknown. 2173 */ 2174 if (env->allow_ptr_leaks && 2175 BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD && 2176 (dst_reg->type == PTR_TO_MAP_VALUE || 2177 dst_reg->type == PTR_TO_MAP_VALUE_ADJ)) 2178 dst_reg->type = PTR_TO_MAP_VALUE_ADJ; 2179 else 2180 mark_reg_unknown_value(regs, insn->dst_reg); 2181 } 2182 2183 return 0; 2184 } 2185 2186 static void find_good_pkt_pointers(struct bpf_verifier_state *state, 2187 struct bpf_reg_state *dst_reg) 2188 { 2189 struct bpf_reg_state *regs = state->regs, *reg; 2190 int i; 2191 2192 /* LLVM can generate two kind of checks: 2193 * 2194 * Type 1: 2195 * 2196 * r2 = r3; 2197 * r2 += 8; 2198 * if (r2 > pkt_end) goto <handle exception> 2199 * <access okay> 2200 * 2201 * Where: 2202 * r2 == dst_reg, pkt_end == src_reg 2203 * r2=pkt(id=n,off=8,r=0) 2204 * r3=pkt(id=n,off=0,r=0) 2205 * 2206 * Type 2: 2207 * 2208 * r2 = r3; 2209 * r2 += 8; 2210 * if (pkt_end >= r2) goto <access okay> 2211 * <handle exception> 2212 * 2213 * Where: 2214 * pkt_end == dst_reg, r2 == src_reg 2215 * r2=pkt(id=n,off=8,r=0) 2216 * r3=pkt(id=n,off=0,r=0) 2217 * 2218 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) 2219 * so that range of bytes [r3, r3 + 8) is safe to access. 2220 */ 2221 2222 for (i = 0; i < MAX_BPF_REG; i++) 2223 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) 2224 /* keep the maximum range already checked */ 2225 regs[i].range = max(regs[i].range, dst_reg->off); 2226 2227 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 2228 if (state->stack_slot_type[i] != STACK_SPILL) 2229 continue; 2230 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 2231 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) 2232 reg->range = max(reg->range, dst_reg->off); 2233 } 2234 } 2235 2236 /* Adjusts the register min/max values in the case that the dst_reg is the 2237 * variable register that we are working on, and src_reg is a constant or we're 2238 * simply doing a BPF_K check. 2239 */ 2240 static void reg_set_min_max(struct bpf_reg_state *true_reg, 2241 struct bpf_reg_state *false_reg, u64 val, 2242 u8 opcode) 2243 { 2244 bool value_from_signed = true; 2245 bool is_range = true; 2246 2247 switch (opcode) { 2248 case BPF_JEQ: 2249 /* If this is false then we know nothing Jon Snow, but if it is 2250 * true then we know for sure. 2251 */ 2252 true_reg->max_value = true_reg->min_value = val; 2253 is_range = false; 2254 break; 2255 case BPF_JNE: 2256 /* If this is true we know nothing Jon Snow, but if it is false 2257 * we know the value for sure; 2258 */ 2259 false_reg->max_value = false_reg->min_value = val; 2260 is_range = false; 2261 break; 2262 case BPF_JGT: 2263 value_from_signed = false; 2264 /* fallthrough */ 2265 case BPF_JSGT: 2266 if (true_reg->value_from_signed != value_from_signed) 2267 reset_reg_range_values(true_reg, 0); 2268 if (false_reg->value_from_signed != value_from_signed) 2269 reset_reg_range_values(false_reg, 0); 2270 if (opcode == BPF_JGT) { 2271 /* Unsigned comparison, the minimum value is 0. */ 2272 false_reg->min_value = 0; 2273 } 2274 /* If this is false then we know the maximum val is val, 2275 * otherwise we know the min val is val+1. 2276 */ 2277 false_reg->max_value = val; 2278 false_reg->value_from_signed = value_from_signed; 2279 true_reg->min_value = val + 1; 2280 true_reg->value_from_signed = value_from_signed; 2281 break; 2282 case BPF_JGE: 2283 value_from_signed = false; 2284 /* fallthrough */ 2285 case BPF_JSGE: 2286 if (true_reg->value_from_signed != value_from_signed) 2287 reset_reg_range_values(true_reg, 0); 2288 if (false_reg->value_from_signed != value_from_signed) 2289 reset_reg_range_values(false_reg, 0); 2290 if (opcode == BPF_JGE) { 2291 /* Unsigned comparison, the minimum value is 0. */ 2292 false_reg->min_value = 0; 2293 } 2294 /* If this is false then we know the maximum value is val - 1, 2295 * otherwise we know the mimimum value is val. 2296 */ 2297 false_reg->max_value = val - 1; 2298 false_reg->value_from_signed = value_from_signed; 2299 true_reg->min_value = val; 2300 true_reg->value_from_signed = value_from_signed; 2301 break; 2302 default: 2303 break; 2304 } 2305 2306 check_reg_overflow(false_reg); 2307 check_reg_overflow(true_reg); 2308 if (is_range) { 2309 if (__is_pointer_value(false, false_reg)) 2310 reset_reg_range_values(false_reg, 0); 2311 if (__is_pointer_value(false, true_reg)) 2312 reset_reg_range_values(true_reg, 0); 2313 } 2314 } 2315 2316 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg 2317 * is the variable reg. 2318 */ 2319 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, 2320 struct bpf_reg_state *false_reg, u64 val, 2321 u8 opcode) 2322 { 2323 bool value_from_signed = true; 2324 bool is_range = true; 2325 2326 switch (opcode) { 2327 case BPF_JEQ: 2328 /* If this is false then we know nothing Jon Snow, but if it is 2329 * true then we know for sure. 2330 */ 2331 true_reg->max_value = true_reg->min_value = val; 2332 is_range = false; 2333 break; 2334 case BPF_JNE: 2335 /* If this is true we know nothing Jon Snow, but if it is false 2336 * we know the value for sure; 2337 */ 2338 false_reg->max_value = false_reg->min_value = val; 2339 is_range = false; 2340 break; 2341 case BPF_JGT: 2342 value_from_signed = false; 2343 /* fallthrough */ 2344 case BPF_JSGT: 2345 if (true_reg->value_from_signed != value_from_signed) 2346 reset_reg_range_values(true_reg, 0); 2347 if (false_reg->value_from_signed != value_from_signed) 2348 reset_reg_range_values(false_reg, 0); 2349 if (opcode == BPF_JGT) { 2350 /* Unsigned comparison, the minimum value is 0. */ 2351 true_reg->min_value = 0; 2352 } 2353 /* 2354 * If this is false, then the val is <= the register, if it is 2355 * true the register <= to the val. 2356 */ 2357 false_reg->min_value = val; 2358 false_reg->value_from_signed = value_from_signed; 2359 true_reg->max_value = val - 1; 2360 true_reg->value_from_signed = value_from_signed; 2361 break; 2362 case BPF_JGE: 2363 value_from_signed = false; 2364 /* fallthrough */ 2365 case BPF_JSGE: 2366 if (true_reg->value_from_signed != value_from_signed) 2367 reset_reg_range_values(true_reg, 0); 2368 if (false_reg->value_from_signed != value_from_signed) 2369 reset_reg_range_values(false_reg, 0); 2370 if (opcode == BPF_JGE) { 2371 /* Unsigned comparison, the minimum value is 0. */ 2372 true_reg->min_value = 0; 2373 } 2374 /* If this is false then constant < register, if it is true then 2375 * the register < constant. 2376 */ 2377 false_reg->min_value = val + 1; 2378 false_reg->value_from_signed = value_from_signed; 2379 true_reg->max_value = val; 2380 true_reg->value_from_signed = value_from_signed; 2381 break; 2382 default: 2383 break; 2384 } 2385 2386 check_reg_overflow(false_reg); 2387 check_reg_overflow(true_reg); 2388 if (is_range) { 2389 if (__is_pointer_value(false, false_reg)) 2390 reset_reg_range_values(false_reg, 0); 2391 if (__is_pointer_value(false, true_reg)) 2392 reset_reg_range_values(true_reg, 0); 2393 } 2394 } 2395 2396 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, 2397 enum bpf_reg_type type) 2398 { 2399 struct bpf_reg_state *reg = ®s[regno]; 2400 2401 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { 2402 if (type == UNKNOWN_VALUE) { 2403 __mark_reg_unknown_value(regs, regno); 2404 } else if (reg->map_ptr->inner_map_meta) { 2405 reg->type = CONST_PTR_TO_MAP; 2406 reg->map_ptr = reg->map_ptr->inner_map_meta; 2407 } else { 2408 reg->type = type; 2409 } 2410 /* We don't need id from this point onwards anymore, thus we 2411 * should better reset it, so that state pruning has chances 2412 * to take effect. 2413 */ 2414 reg->id = 0; 2415 } 2416 } 2417 2418 /* The logic is similar to find_good_pkt_pointers(), both could eventually 2419 * be folded together at some point. 2420 */ 2421 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, 2422 enum bpf_reg_type type) 2423 { 2424 struct bpf_reg_state *regs = state->regs; 2425 u32 id = regs[regno].id; 2426 int i; 2427 2428 for (i = 0; i < MAX_BPF_REG; i++) 2429 mark_map_reg(regs, i, id, type); 2430 2431 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 2432 if (state->stack_slot_type[i] != STACK_SPILL) 2433 continue; 2434 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type); 2435 } 2436 } 2437 2438 static int check_cond_jmp_op(struct bpf_verifier_env *env, 2439 struct bpf_insn *insn, int *insn_idx) 2440 { 2441 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; 2442 struct bpf_reg_state *regs = this_branch->regs, *dst_reg; 2443 u8 opcode = BPF_OP(insn->code); 2444 int err; 2445 2446 if (opcode > BPF_EXIT) { 2447 verbose("invalid BPF_JMP opcode %x\n", opcode); 2448 return -EINVAL; 2449 } 2450 2451 if (BPF_SRC(insn->code) == BPF_X) { 2452 if (insn->imm != 0) { 2453 verbose("BPF_JMP uses reserved fields\n"); 2454 return -EINVAL; 2455 } 2456 2457 /* check src1 operand */ 2458 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2459 if (err) 2460 return err; 2461 2462 if (is_pointer_value(env, insn->src_reg)) { 2463 verbose("R%d pointer comparison prohibited\n", 2464 insn->src_reg); 2465 return -EACCES; 2466 } 2467 } else { 2468 if (insn->src_reg != BPF_REG_0) { 2469 verbose("BPF_JMP uses reserved fields\n"); 2470 return -EINVAL; 2471 } 2472 } 2473 2474 /* check src2 operand */ 2475 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2476 if (err) 2477 return err; 2478 2479 dst_reg = ®s[insn->dst_reg]; 2480 2481 /* detect if R == 0 where R was initialized to zero earlier */ 2482 if (BPF_SRC(insn->code) == BPF_K && 2483 (opcode == BPF_JEQ || opcode == BPF_JNE) && 2484 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { 2485 if (opcode == BPF_JEQ) { 2486 /* if (imm == imm) goto pc+off; 2487 * only follow the goto, ignore fall-through 2488 */ 2489 *insn_idx += insn->off; 2490 return 0; 2491 } else { 2492 /* if (imm != imm) goto pc+off; 2493 * only follow fall-through branch, since 2494 * that's where the program will go 2495 */ 2496 return 0; 2497 } 2498 } 2499 2500 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); 2501 if (!other_branch) 2502 return -EFAULT; 2503 2504 /* detect if we are comparing against a constant value so we can adjust 2505 * our min/max values for our dst register. 2506 */ 2507 if (BPF_SRC(insn->code) == BPF_X) { 2508 if (regs[insn->src_reg].type == CONST_IMM) 2509 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2510 dst_reg, regs[insn->src_reg].imm, 2511 opcode); 2512 else if (dst_reg->type == CONST_IMM) 2513 reg_set_min_max_inv(&other_branch->regs[insn->src_reg], 2514 ®s[insn->src_reg], dst_reg->imm, 2515 opcode); 2516 } else { 2517 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2518 dst_reg, insn->imm, opcode); 2519 } 2520 2521 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ 2522 if (BPF_SRC(insn->code) == BPF_K && 2523 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && 2524 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { 2525 /* Mark all identical map registers in each branch as either 2526 * safe or unknown depending R == 0 or R != 0 conditional. 2527 */ 2528 mark_map_regs(this_branch, insn->dst_reg, 2529 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE); 2530 mark_map_regs(other_branch, insn->dst_reg, 2531 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE); 2532 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && 2533 dst_reg->type == PTR_TO_PACKET && 2534 regs[insn->src_reg].type == PTR_TO_PACKET_END) { 2535 find_good_pkt_pointers(this_branch, dst_reg); 2536 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && 2537 dst_reg->type == PTR_TO_PACKET_END && 2538 regs[insn->src_reg].type == PTR_TO_PACKET) { 2539 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); 2540 } else if (is_pointer_value(env, insn->dst_reg)) { 2541 verbose("R%d pointer comparison prohibited\n", insn->dst_reg); 2542 return -EACCES; 2543 } 2544 if (log_level) 2545 print_verifier_state(this_branch); 2546 return 0; 2547 } 2548 2549 /* return the map pointer stored inside BPF_LD_IMM64 instruction */ 2550 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) 2551 { 2552 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; 2553 2554 return (struct bpf_map *) (unsigned long) imm64; 2555 } 2556 2557 /* verify BPF_LD_IMM64 instruction */ 2558 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) 2559 { 2560 struct bpf_reg_state *regs = env->cur_state.regs; 2561 int err; 2562 2563 if (BPF_SIZE(insn->code) != BPF_DW) { 2564 verbose("invalid BPF_LD_IMM insn\n"); 2565 return -EINVAL; 2566 } 2567 if (insn->off != 0) { 2568 verbose("BPF_LD_IMM64 uses reserved fields\n"); 2569 return -EINVAL; 2570 } 2571 2572 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2573 if (err) 2574 return err; 2575 2576 if (insn->src_reg == 0) { 2577 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; 2578 2579 regs[insn->dst_reg].type = CONST_IMM; 2580 regs[insn->dst_reg].imm = imm; 2581 regs[insn->dst_reg].id = 0; 2582 return 0; 2583 } 2584 2585 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ 2586 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); 2587 2588 regs[insn->dst_reg].type = CONST_PTR_TO_MAP; 2589 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); 2590 return 0; 2591 } 2592 2593 static bool may_access_skb(enum bpf_prog_type type) 2594 { 2595 switch (type) { 2596 case BPF_PROG_TYPE_SOCKET_FILTER: 2597 case BPF_PROG_TYPE_SCHED_CLS: 2598 case BPF_PROG_TYPE_SCHED_ACT: 2599 return true; 2600 default: 2601 return false; 2602 } 2603 } 2604 2605 /* verify safety of LD_ABS|LD_IND instructions: 2606 * - they can only appear in the programs where ctx == skb 2607 * - since they are wrappers of function calls, they scratch R1-R5 registers, 2608 * preserve R6-R9, and store return value into R0 2609 * 2610 * Implicit input: 2611 * ctx == skb == R6 == CTX 2612 * 2613 * Explicit input: 2614 * SRC == any register 2615 * IMM == 32-bit immediate 2616 * 2617 * Output: 2618 * R0 - 8/16/32-bit skb data converted to cpu endianness 2619 */ 2620 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) 2621 { 2622 struct bpf_reg_state *regs = env->cur_state.regs; 2623 u8 mode = BPF_MODE(insn->code); 2624 int i, err; 2625 2626 if (!may_access_skb(env->prog->type)) { 2627 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); 2628 return -EINVAL; 2629 } 2630 2631 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || 2632 BPF_SIZE(insn->code) == BPF_DW || 2633 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { 2634 verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); 2635 return -EINVAL; 2636 } 2637 2638 /* check whether implicit source operand (register R6) is readable */ 2639 err = check_reg_arg(regs, BPF_REG_6, SRC_OP); 2640 if (err) 2641 return err; 2642 2643 if (regs[BPF_REG_6].type != PTR_TO_CTX) { 2644 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); 2645 return -EINVAL; 2646 } 2647 2648 if (mode == BPF_IND) { 2649 /* check explicit source operand */ 2650 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2651 if (err) 2652 return err; 2653 } 2654 2655 /* reset caller saved regs to unreadable */ 2656 for (i = 0; i < CALLER_SAVED_REGS; i++) 2657 mark_reg_not_init(regs, caller_saved[i]); 2658 2659 /* mark destination R0 register as readable, since it contains 2660 * the value fetched from the packet 2661 */ 2662 regs[BPF_REG_0].type = UNKNOWN_VALUE; 2663 return 0; 2664 } 2665 2666 /* non-recursive DFS pseudo code 2667 * 1 procedure DFS-iterative(G,v): 2668 * 2 label v as discovered 2669 * 3 let S be a stack 2670 * 4 S.push(v) 2671 * 5 while S is not empty 2672 * 6 t <- S.pop() 2673 * 7 if t is what we're looking for: 2674 * 8 return t 2675 * 9 for all edges e in G.adjacentEdges(t) do 2676 * 10 if edge e is already labelled 2677 * 11 continue with the next edge 2678 * 12 w <- G.adjacentVertex(t,e) 2679 * 13 if vertex w is not discovered and not explored 2680 * 14 label e as tree-edge 2681 * 15 label w as discovered 2682 * 16 S.push(w) 2683 * 17 continue at 5 2684 * 18 else if vertex w is discovered 2685 * 19 label e as back-edge 2686 * 20 else 2687 * 21 // vertex w is explored 2688 * 22 label e as forward- or cross-edge 2689 * 23 label t as explored 2690 * 24 S.pop() 2691 * 2692 * convention: 2693 * 0x10 - discovered 2694 * 0x11 - discovered and fall-through edge labelled 2695 * 0x12 - discovered and fall-through and branch edges labelled 2696 * 0x20 - explored 2697 */ 2698 2699 enum { 2700 DISCOVERED = 0x10, 2701 EXPLORED = 0x20, 2702 FALLTHROUGH = 1, 2703 BRANCH = 2, 2704 }; 2705 2706 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) 2707 2708 static int *insn_stack; /* stack of insns to process */ 2709 static int cur_stack; /* current stack index */ 2710 static int *insn_state; 2711 2712 /* t, w, e - match pseudo-code above: 2713 * t - index of current instruction 2714 * w - next instruction 2715 * e - edge 2716 */ 2717 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) 2718 { 2719 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) 2720 return 0; 2721 2722 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) 2723 return 0; 2724 2725 if (w < 0 || w >= env->prog->len) { 2726 verbose("jump out of range from insn %d to %d\n", t, w); 2727 return -EINVAL; 2728 } 2729 2730 if (e == BRANCH) 2731 /* mark branch target for state pruning */ 2732 env->explored_states[w] = STATE_LIST_MARK; 2733 2734 if (insn_state[w] == 0) { 2735 /* tree-edge */ 2736 insn_state[t] = DISCOVERED | e; 2737 insn_state[w] = DISCOVERED; 2738 if (cur_stack >= env->prog->len) 2739 return -E2BIG; 2740 insn_stack[cur_stack++] = w; 2741 return 1; 2742 } else if ((insn_state[w] & 0xF0) == DISCOVERED) { 2743 verbose("back-edge from insn %d to %d\n", t, w); 2744 return -EINVAL; 2745 } else if (insn_state[w] == EXPLORED) { 2746 /* forward- or cross-edge */ 2747 insn_state[t] = DISCOVERED | e; 2748 } else { 2749 verbose("insn state internal bug\n"); 2750 return -EFAULT; 2751 } 2752 return 0; 2753 } 2754 2755 /* non-recursive depth-first-search to detect loops in BPF program 2756 * loop == back-edge in directed graph 2757 */ 2758 static int check_cfg(struct bpf_verifier_env *env) 2759 { 2760 struct bpf_insn *insns = env->prog->insnsi; 2761 int insn_cnt = env->prog->len; 2762 int ret = 0; 2763 int i, t; 2764 2765 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2766 if (!insn_state) 2767 return -ENOMEM; 2768 2769 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2770 if (!insn_stack) { 2771 kfree(insn_state); 2772 return -ENOMEM; 2773 } 2774 2775 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ 2776 insn_stack[0] = 0; /* 0 is the first instruction */ 2777 cur_stack = 1; 2778 2779 peek_stack: 2780 if (cur_stack == 0) 2781 goto check_state; 2782 t = insn_stack[cur_stack - 1]; 2783 2784 if (BPF_CLASS(insns[t].code) == BPF_JMP) { 2785 u8 opcode = BPF_OP(insns[t].code); 2786 2787 if (opcode == BPF_EXIT) { 2788 goto mark_explored; 2789 } else if (opcode == BPF_CALL) { 2790 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2791 if (ret == 1) 2792 goto peek_stack; 2793 else if (ret < 0) 2794 goto err_free; 2795 if (t + 1 < insn_cnt) 2796 env->explored_states[t + 1] = STATE_LIST_MARK; 2797 } else if (opcode == BPF_JA) { 2798 if (BPF_SRC(insns[t].code) != BPF_K) { 2799 ret = -EINVAL; 2800 goto err_free; 2801 } 2802 /* unconditional jump with single edge */ 2803 ret = push_insn(t, t + insns[t].off + 1, 2804 FALLTHROUGH, env); 2805 if (ret == 1) 2806 goto peek_stack; 2807 else if (ret < 0) 2808 goto err_free; 2809 /* tell verifier to check for equivalent states 2810 * after every call and jump 2811 */ 2812 if (t + 1 < insn_cnt) 2813 env->explored_states[t + 1] = STATE_LIST_MARK; 2814 } else { 2815 /* conditional jump with two edges */ 2816 env->explored_states[t] = STATE_LIST_MARK; 2817 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2818 if (ret == 1) 2819 goto peek_stack; 2820 else if (ret < 0) 2821 goto err_free; 2822 2823 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); 2824 if (ret == 1) 2825 goto peek_stack; 2826 else if (ret < 0) 2827 goto err_free; 2828 } 2829 } else { 2830 /* all other non-branch instructions with single 2831 * fall-through edge 2832 */ 2833 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2834 if (ret == 1) 2835 goto peek_stack; 2836 else if (ret < 0) 2837 goto err_free; 2838 } 2839 2840 mark_explored: 2841 insn_state[t] = EXPLORED; 2842 if (cur_stack-- <= 0) { 2843 verbose("pop stack internal bug\n"); 2844 ret = -EFAULT; 2845 goto err_free; 2846 } 2847 goto peek_stack; 2848 2849 check_state: 2850 for (i = 0; i < insn_cnt; i++) { 2851 if (insn_state[i] != EXPLORED) { 2852 verbose("unreachable insn %d\n", i); 2853 ret = -EINVAL; 2854 goto err_free; 2855 } 2856 } 2857 ret = 0; /* cfg looks good */ 2858 2859 err_free: 2860 kfree(insn_state); 2861 kfree(insn_stack); 2862 return ret; 2863 } 2864 2865 /* the following conditions reduce the number of explored insns 2866 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet 2867 */ 2868 static bool compare_ptrs_to_packet(struct bpf_verifier_env *env, 2869 struct bpf_reg_state *old, 2870 struct bpf_reg_state *cur) 2871 { 2872 if (old->id != cur->id) 2873 return false; 2874 2875 /* old ptr_to_packet is more conservative, since it allows smaller 2876 * range. Ex: 2877 * old(off=0,r=10) is equal to cur(off=0,r=20), because 2878 * old(off=0,r=10) means that with range=10 the verifier proceeded 2879 * further and found no issues with the program. Now we're in the same 2880 * spot with cur(off=0,r=20), so we're safe too, since anything further 2881 * will only be looking at most 10 bytes after this pointer. 2882 */ 2883 if (old->off == cur->off && old->range < cur->range) 2884 return true; 2885 2886 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) 2887 * since both cannot be used for packet access and safe(old) 2888 * pointer has smaller off that could be used for further 2889 * 'if (ptr > data_end)' check 2890 * Ex: 2891 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean 2892 * that we cannot access the packet. 2893 * The safe range is: 2894 * [ptr, ptr + range - off) 2895 * so whenever off >=range, it means no safe bytes from this pointer. 2896 * When comparing old->off <= cur->off, it means that older code 2897 * went with smaller offset and that offset was later 2898 * used to figure out the safe range after 'if (ptr > data_end)' check 2899 * Say, 'old' state was explored like: 2900 * ... R3(off=0, r=0) 2901 * R4 = R3 + 20 2902 * ... now R4(off=20,r=0) <-- here 2903 * if (R4 > data_end) 2904 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. 2905 * ... the code further went all the way to bpf_exit. 2906 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). 2907 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier 2908 * goes further, such cur_R4 will give larger safe packet range after 2909 * 'if (R4 > data_end)' and all further insn were already good with r=20, 2910 * so they will be good with r=30 and we can prune the search. 2911 */ 2912 if (!env->strict_alignment && old->off <= cur->off && 2913 old->off >= old->range && cur->off >= cur->range) 2914 return true; 2915 2916 return false; 2917 } 2918 2919 /* compare two verifier states 2920 * 2921 * all states stored in state_list are known to be valid, since 2922 * verifier reached 'bpf_exit' instruction through them 2923 * 2924 * this function is called when verifier exploring different branches of 2925 * execution popped from the state stack. If it sees an old state that has 2926 * more strict register state and more strict stack state then this execution 2927 * branch doesn't need to be explored further, since verifier already 2928 * concluded that more strict state leads to valid finish. 2929 * 2930 * Therefore two states are equivalent if register state is more conservative 2931 * and explored stack state is more conservative than the current one. 2932 * Example: 2933 * explored current 2934 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) 2935 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) 2936 * 2937 * In other words if current stack state (one being explored) has more 2938 * valid slots than old one that already passed validation, it means 2939 * the verifier can stop exploring and conclude that current state is valid too 2940 * 2941 * Similarly with registers. If explored state has register type as invalid 2942 * whereas register type in current state is meaningful, it means that 2943 * the current state will reach 'bpf_exit' instruction safely 2944 */ 2945 static bool states_equal(struct bpf_verifier_env *env, 2946 struct bpf_verifier_state *old, 2947 struct bpf_verifier_state *cur) 2948 { 2949 bool varlen_map_access = env->varlen_map_value_access; 2950 struct bpf_reg_state *rold, *rcur; 2951 int i; 2952 2953 for (i = 0; i < MAX_BPF_REG; i++) { 2954 rold = &old->regs[i]; 2955 rcur = &cur->regs[i]; 2956 2957 if (memcmp(rold, rcur, sizeof(*rold)) == 0) 2958 continue; 2959 2960 /* If the ranges were not the same, but everything else was and 2961 * we didn't do a variable access into a map then we are a-ok. 2962 */ 2963 if (!varlen_map_access && 2964 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0) 2965 continue; 2966 2967 /* If we didn't map access then again we don't care about the 2968 * mismatched range values and it's ok if our old type was 2969 * UNKNOWN and we didn't go to a NOT_INIT'ed reg. 2970 */ 2971 if (rold->type == NOT_INIT || 2972 (!varlen_map_access && rold->type == UNKNOWN_VALUE && 2973 rcur->type != NOT_INIT)) 2974 continue; 2975 2976 /* Don't care about the reg->id in this case. */ 2977 if (rold->type == PTR_TO_MAP_VALUE_OR_NULL && 2978 rcur->type == PTR_TO_MAP_VALUE_OR_NULL && 2979 rold->map_ptr == rcur->map_ptr) 2980 continue; 2981 2982 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && 2983 compare_ptrs_to_packet(env, rold, rcur)) 2984 continue; 2985 2986 return false; 2987 } 2988 2989 for (i = 0; i < MAX_BPF_STACK; i++) { 2990 if (old->stack_slot_type[i] == STACK_INVALID) 2991 continue; 2992 if (old->stack_slot_type[i] != cur->stack_slot_type[i]) 2993 /* Ex: old explored (safe) state has STACK_SPILL in 2994 * this stack slot, but current has has STACK_MISC -> 2995 * this verifier states are not equivalent, 2996 * return false to continue verification of this path 2997 */ 2998 return false; 2999 if (i % BPF_REG_SIZE) 3000 continue; 3001 if (old->stack_slot_type[i] != STACK_SPILL) 3002 continue; 3003 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], 3004 &cur->spilled_regs[i / BPF_REG_SIZE], 3005 sizeof(old->spilled_regs[0]))) 3006 /* when explored and current stack slot types are 3007 * the same, check that stored pointers types 3008 * are the same as well. 3009 * Ex: explored safe path could have stored 3010 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} 3011 * but current path has stored: 3012 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} 3013 * such verifier states are not equivalent. 3014 * return false to continue verification of this path 3015 */ 3016 return false; 3017 else 3018 continue; 3019 } 3020 return true; 3021 } 3022 3023 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) 3024 { 3025 struct bpf_verifier_state_list *new_sl; 3026 struct bpf_verifier_state_list *sl; 3027 3028 sl = env->explored_states[insn_idx]; 3029 if (!sl) 3030 /* this 'insn_idx' instruction wasn't marked, so we will not 3031 * be doing state search here 3032 */ 3033 return 0; 3034 3035 while (sl != STATE_LIST_MARK) { 3036 if (states_equal(env, &sl->state, &env->cur_state)) 3037 /* reached equivalent register/stack state, 3038 * prune the search 3039 */ 3040 return 1; 3041 sl = sl->next; 3042 } 3043 3044 /* there were no equivalent states, remember current one. 3045 * technically the current state is not proven to be safe yet, 3046 * but it will either reach bpf_exit (which means it's safe) or 3047 * it will be rejected. Since there are no loops, we won't be 3048 * seeing this 'insn_idx' instruction again on the way to bpf_exit 3049 */ 3050 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); 3051 if (!new_sl) 3052 return -ENOMEM; 3053 3054 /* add new state to the head of linked list */ 3055 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); 3056 new_sl->next = env->explored_states[insn_idx]; 3057 env->explored_states[insn_idx] = new_sl; 3058 return 0; 3059 } 3060 3061 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, 3062 int insn_idx, int prev_insn_idx) 3063 { 3064 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) 3065 return 0; 3066 3067 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); 3068 } 3069 3070 static int do_check(struct bpf_verifier_env *env) 3071 { 3072 struct bpf_verifier_state *state = &env->cur_state; 3073 struct bpf_insn *insns = env->prog->insnsi; 3074 struct bpf_reg_state *regs = state->regs; 3075 int insn_cnt = env->prog->len; 3076 int insn_idx, prev_insn_idx = 0; 3077 int insn_processed = 0; 3078 bool do_print_state = false; 3079 3080 init_reg_state(regs); 3081 insn_idx = 0; 3082 env->varlen_map_value_access = false; 3083 for (;;) { 3084 struct bpf_insn *insn; 3085 u8 class; 3086 int err; 3087 3088 if (insn_idx >= insn_cnt) { 3089 verbose("invalid insn idx %d insn_cnt %d\n", 3090 insn_idx, insn_cnt); 3091 return -EFAULT; 3092 } 3093 3094 insn = &insns[insn_idx]; 3095 class = BPF_CLASS(insn->code); 3096 3097 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { 3098 verbose("BPF program is too large. Processed %d insn\n", 3099 insn_processed); 3100 return -E2BIG; 3101 } 3102 3103 err = is_state_visited(env, insn_idx); 3104 if (err < 0) 3105 return err; 3106 if (err == 1) { 3107 /* found equivalent state, can prune the search */ 3108 if (log_level) { 3109 if (do_print_state) 3110 verbose("\nfrom %d to %d: safe\n", 3111 prev_insn_idx, insn_idx); 3112 else 3113 verbose("%d: safe\n", insn_idx); 3114 } 3115 goto process_bpf_exit; 3116 } 3117 3118 if (need_resched()) 3119 cond_resched(); 3120 3121 if (log_level > 1 || (log_level && do_print_state)) { 3122 if (log_level > 1) 3123 verbose("%d:", insn_idx); 3124 else 3125 verbose("\nfrom %d to %d:", 3126 prev_insn_idx, insn_idx); 3127 print_verifier_state(&env->cur_state); 3128 do_print_state = false; 3129 } 3130 3131 if (log_level) { 3132 verbose("%d: ", insn_idx); 3133 print_bpf_insn(env, insn); 3134 } 3135 3136 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); 3137 if (err) 3138 return err; 3139 3140 if (class == BPF_ALU || class == BPF_ALU64) { 3141 err = check_alu_op(env, insn); 3142 if (err) 3143 return err; 3144 3145 } else if (class == BPF_LDX) { 3146 enum bpf_reg_type *prev_src_type, src_reg_type; 3147 3148 /* check for reserved fields is already done */ 3149 3150 /* check src operand */ 3151 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 3152 if (err) 3153 return err; 3154 3155 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 3156 if (err) 3157 return err; 3158 3159 src_reg_type = regs[insn->src_reg].type; 3160 3161 /* check that memory (src_reg + off) is readable, 3162 * the state of dst_reg will be updated by this func 3163 */ 3164 err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, 3165 BPF_SIZE(insn->code), BPF_READ, 3166 insn->dst_reg); 3167 if (err) 3168 return err; 3169 3170 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; 3171 3172 if (*prev_src_type == NOT_INIT) { 3173 /* saw a valid insn 3174 * dst_reg = *(u32 *)(src_reg + off) 3175 * save type to validate intersecting paths 3176 */ 3177 *prev_src_type = src_reg_type; 3178 3179 } else if (src_reg_type != *prev_src_type && 3180 (src_reg_type == PTR_TO_CTX || 3181 *prev_src_type == PTR_TO_CTX)) { 3182 /* ABuser program is trying to use the same insn 3183 * dst_reg = *(u32*) (src_reg + off) 3184 * with different pointer types: 3185 * src_reg == ctx in one branch and 3186 * src_reg == stack|map in some other branch. 3187 * Reject it. 3188 */ 3189 verbose("same insn cannot be used with different pointers\n"); 3190 return -EINVAL; 3191 } 3192 3193 } else if (class == BPF_STX) { 3194 enum bpf_reg_type *prev_dst_type, dst_reg_type; 3195 3196 if (BPF_MODE(insn->code) == BPF_XADD) { 3197 err = check_xadd(env, insn_idx, insn); 3198 if (err) 3199 return err; 3200 insn_idx++; 3201 continue; 3202 } 3203 3204 /* check src1 operand */ 3205 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 3206 if (err) 3207 return err; 3208 /* check src2 operand */ 3209 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 3210 if (err) 3211 return err; 3212 3213 dst_reg_type = regs[insn->dst_reg].type; 3214 3215 /* check that memory (dst_reg + off) is writeable */ 3216 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 3217 BPF_SIZE(insn->code), BPF_WRITE, 3218 insn->src_reg); 3219 if (err) 3220 return err; 3221 3222 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; 3223 3224 if (*prev_dst_type == NOT_INIT) { 3225 *prev_dst_type = dst_reg_type; 3226 } else if (dst_reg_type != *prev_dst_type && 3227 (dst_reg_type == PTR_TO_CTX || 3228 *prev_dst_type == PTR_TO_CTX)) { 3229 verbose("same insn cannot be used with different pointers\n"); 3230 return -EINVAL; 3231 } 3232 3233 } else if (class == BPF_ST) { 3234 if (BPF_MODE(insn->code) != BPF_MEM || 3235 insn->src_reg != BPF_REG_0) { 3236 verbose("BPF_ST uses reserved fields\n"); 3237 return -EINVAL; 3238 } 3239 /* check src operand */ 3240 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 3241 if (err) 3242 return err; 3243 3244 /* check that memory (dst_reg + off) is writeable */ 3245 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 3246 BPF_SIZE(insn->code), BPF_WRITE, 3247 -1); 3248 if (err) 3249 return err; 3250 3251 } else if (class == BPF_JMP) { 3252 u8 opcode = BPF_OP(insn->code); 3253 3254 if (opcode == BPF_CALL) { 3255 if (BPF_SRC(insn->code) != BPF_K || 3256 insn->off != 0 || 3257 insn->src_reg != BPF_REG_0 || 3258 insn->dst_reg != BPF_REG_0) { 3259 verbose("BPF_CALL uses reserved fields\n"); 3260 return -EINVAL; 3261 } 3262 3263 err = check_call(env, insn->imm, insn_idx); 3264 if (err) 3265 return err; 3266 3267 } else if (opcode == BPF_JA) { 3268 if (BPF_SRC(insn->code) != BPF_K || 3269 insn->imm != 0 || 3270 insn->src_reg != BPF_REG_0 || 3271 insn->dst_reg != BPF_REG_0) { 3272 verbose("BPF_JA uses reserved fields\n"); 3273 return -EINVAL; 3274 } 3275 3276 insn_idx += insn->off + 1; 3277 continue; 3278 3279 } else if (opcode == BPF_EXIT) { 3280 if (BPF_SRC(insn->code) != BPF_K || 3281 insn->imm != 0 || 3282 insn->src_reg != BPF_REG_0 || 3283 insn->dst_reg != BPF_REG_0) { 3284 verbose("BPF_EXIT uses reserved fields\n"); 3285 return -EINVAL; 3286 } 3287 3288 /* eBPF calling convetion is such that R0 is used 3289 * to return the value from eBPF program. 3290 * Make sure that it's readable at this time 3291 * of bpf_exit, which means that program wrote 3292 * something into it earlier 3293 */ 3294 err = check_reg_arg(regs, BPF_REG_0, SRC_OP); 3295 if (err) 3296 return err; 3297 3298 if (is_pointer_value(env, BPF_REG_0)) { 3299 verbose("R0 leaks addr as return value\n"); 3300 return -EACCES; 3301 } 3302 3303 process_bpf_exit: 3304 insn_idx = pop_stack(env, &prev_insn_idx); 3305 if (insn_idx < 0) { 3306 break; 3307 } else { 3308 do_print_state = true; 3309 continue; 3310 } 3311 } else { 3312 err = check_cond_jmp_op(env, insn, &insn_idx); 3313 if (err) 3314 return err; 3315 } 3316 } else if (class == BPF_LD) { 3317 u8 mode = BPF_MODE(insn->code); 3318 3319 if (mode == BPF_ABS || mode == BPF_IND) { 3320 err = check_ld_abs(env, insn); 3321 if (err) 3322 return err; 3323 3324 } else if (mode == BPF_IMM) { 3325 err = check_ld_imm(env, insn); 3326 if (err) 3327 return err; 3328 3329 insn_idx++; 3330 } else { 3331 verbose("invalid BPF_LD mode\n"); 3332 return -EINVAL; 3333 } 3334 reset_reg_range_values(regs, insn->dst_reg); 3335 } else { 3336 verbose("unknown insn class %d\n", class); 3337 return -EINVAL; 3338 } 3339 3340 insn_idx++; 3341 } 3342 3343 verbose("processed %d insns, stack depth %d\n", 3344 insn_processed, env->prog->aux->stack_depth); 3345 return 0; 3346 } 3347 3348 static int check_map_prealloc(struct bpf_map *map) 3349 { 3350 return (map->map_type != BPF_MAP_TYPE_HASH && 3351 map->map_type != BPF_MAP_TYPE_PERCPU_HASH && 3352 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || 3353 !(map->map_flags & BPF_F_NO_PREALLOC); 3354 } 3355 3356 static int check_map_prog_compatibility(struct bpf_map *map, 3357 struct bpf_prog *prog) 3358 3359 { 3360 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use 3361 * preallocated hash maps, since doing memory allocation 3362 * in overflow_handler can crash depending on where nmi got 3363 * triggered. 3364 */ 3365 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { 3366 if (!check_map_prealloc(map)) { 3367 verbose("perf_event programs can only use preallocated hash map\n"); 3368 return -EINVAL; 3369 } 3370 if (map->inner_map_meta && 3371 !check_map_prealloc(map->inner_map_meta)) { 3372 verbose("perf_event programs can only use preallocated inner hash map\n"); 3373 return -EINVAL; 3374 } 3375 } 3376 return 0; 3377 } 3378 3379 /* look for pseudo eBPF instructions that access map FDs and 3380 * replace them with actual map pointers 3381 */ 3382 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) 3383 { 3384 struct bpf_insn *insn = env->prog->insnsi; 3385 int insn_cnt = env->prog->len; 3386 int i, j, err; 3387 3388 err = bpf_prog_calc_tag(env->prog); 3389 if (err) 3390 return err; 3391 3392 for (i = 0; i < insn_cnt; i++, insn++) { 3393 if (BPF_CLASS(insn->code) == BPF_LDX && 3394 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { 3395 verbose("BPF_LDX uses reserved fields\n"); 3396 return -EINVAL; 3397 } 3398 3399 if (BPF_CLASS(insn->code) == BPF_STX && 3400 ((BPF_MODE(insn->code) != BPF_MEM && 3401 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { 3402 verbose("BPF_STX uses reserved fields\n"); 3403 return -EINVAL; 3404 } 3405 3406 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { 3407 struct bpf_map *map; 3408 struct fd f; 3409 3410 if (i == insn_cnt - 1 || insn[1].code != 0 || 3411 insn[1].dst_reg != 0 || insn[1].src_reg != 0 || 3412 insn[1].off != 0) { 3413 verbose("invalid bpf_ld_imm64 insn\n"); 3414 return -EINVAL; 3415 } 3416 3417 if (insn->src_reg == 0) 3418 /* valid generic load 64-bit imm */ 3419 goto next_insn; 3420 3421 if (insn->src_reg != BPF_PSEUDO_MAP_FD) { 3422 verbose("unrecognized bpf_ld_imm64 insn\n"); 3423 return -EINVAL; 3424 } 3425 3426 f = fdget(insn->imm); 3427 map = __bpf_map_get(f); 3428 if (IS_ERR(map)) { 3429 verbose("fd %d is not pointing to valid bpf_map\n", 3430 insn->imm); 3431 return PTR_ERR(map); 3432 } 3433 3434 err = check_map_prog_compatibility(map, env->prog); 3435 if (err) { 3436 fdput(f); 3437 return err; 3438 } 3439 3440 /* store map pointer inside BPF_LD_IMM64 instruction */ 3441 insn[0].imm = (u32) (unsigned long) map; 3442 insn[1].imm = ((u64) (unsigned long) map) >> 32; 3443 3444 /* check whether we recorded this map already */ 3445 for (j = 0; j < env->used_map_cnt; j++) 3446 if (env->used_maps[j] == map) { 3447 fdput(f); 3448 goto next_insn; 3449 } 3450 3451 if (env->used_map_cnt >= MAX_USED_MAPS) { 3452 fdput(f); 3453 return -E2BIG; 3454 } 3455 3456 /* hold the map. If the program is rejected by verifier, 3457 * the map will be released by release_maps() or it 3458 * will be used by the valid program until it's unloaded 3459 * and all maps are released in free_bpf_prog_info() 3460 */ 3461 map = bpf_map_inc(map, false); 3462 if (IS_ERR(map)) { 3463 fdput(f); 3464 return PTR_ERR(map); 3465 } 3466 env->used_maps[env->used_map_cnt++] = map; 3467 3468 fdput(f); 3469 next_insn: 3470 insn++; 3471 i++; 3472 } 3473 } 3474 3475 /* now all pseudo BPF_LD_IMM64 instructions load valid 3476 * 'struct bpf_map *' into a register instead of user map_fd. 3477 * These pointers will be used later by verifier to validate map access. 3478 */ 3479 return 0; 3480 } 3481 3482 /* drop refcnt of maps used by the rejected program */ 3483 static void release_maps(struct bpf_verifier_env *env) 3484 { 3485 int i; 3486 3487 for (i = 0; i < env->used_map_cnt; i++) 3488 bpf_map_put(env->used_maps[i]); 3489 } 3490 3491 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ 3492 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) 3493 { 3494 struct bpf_insn *insn = env->prog->insnsi; 3495 int insn_cnt = env->prog->len; 3496 int i; 3497 3498 for (i = 0; i < insn_cnt; i++, insn++) 3499 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) 3500 insn->src_reg = 0; 3501 } 3502 3503 /* single env->prog->insni[off] instruction was replaced with the range 3504 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying 3505 * [0, off) and [off, end) to new locations, so the patched range stays zero 3506 */ 3507 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, 3508 u32 off, u32 cnt) 3509 { 3510 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; 3511 3512 if (cnt == 1) 3513 return 0; 3514 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); 3515 if (!new_data) 3516 return -ENOMEM; 3517 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); 3518 memcpy(new_data + off + cnt - 1, old_data + off, 3519 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); 3520 env->insn_aux_data = new_data; 3521 vfree(old_data); 3522 return 0; 3523 } 3524 3525 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, 3526 const struct bpf_insn *patch, u32 len) 3527 { 3528 struct bpf_prog *new_prog; 3529 3530 new_prog = bpf_patch_insn_single(env->prog, off, patch, len); 3531 if (!new_prog) 3532 return NULL; 3533 if (adjust_insn_aux_data(env, new_prog->len, off, len)) 3534 return NULL; 3535 return new_prog; 3536 } 3537 3538 /* convert load instructions that access fields of 'struct __sk_buff' 3539 * into sequence of instructions that access fields of 'struct sk_buff' 3540 */ 3541 static int convert_ctx_accesses(struct bpf_verifier_env *env) 3542 { 3543 const struct bpf_verifier_ops *ops = env->prog->aux->ops; 3544 int i, cnt, size, ctx_field_size, delta = 0; 3545 const int insn_cnt = env->prog->len; 3546 struct bpf_insn insn_buf[16], *insn; 3547 struct bpf_prog *new_prog; 3548 enum bpf_access_type type; 3549 bool is_narrower_load; 3550 u32 target_size; 3551 3552 if (ops->gen_prologue) { 3553 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, 3554 env->prog); 3555 if (cnt >= ARRAY_SIZE(insn_buf)) { 3556 verbose("bpf verifier is misconfigured\n"); 3557 return -EINVAL; 3558 } else if (cnt) { 3559 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); 3560 if (!new_prog) 3561 return -ENOMEM; 3562 3563 env->prog = new_prog; 3564 delta += cnt - 1; 3565 } 3566 } 3567 3568 if (!ops->convert_ctx_access) 3569 return 0; 3570 3571 insn = env->prog->insnsi + delta; 3572 3573 for (i = 0; i < insn_cnt; i++, insn++) { 3574 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || 3575 insn->code == (BPF_LDX | BPF_MEM | BPF_H) || 3576 insn->code == (BPF_LDX | BPF_MEM | BPF_W) || 3577 insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) 3578 type = BPF_READ; 3579 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || 3580 insn->code == (BPF_STX | BPF_MEM | BPF_H) || 3581 insn->code == (BPF_STX | BPF_MEM | BPF_W) || 3582 insn->code == (BPF_STX | BPF_MEM | BPF_DW)) 3583 type = BPF_WRITE; 3584 else 3585 continue; 3586 3587 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) 3588 continue; 3589 3590 ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; 3591 size = BPF_LDST_BYTES(insn); 3592 3593 /* If the read access is a narrower load of the field, 3594 * convert to a 4/8-byte load, to minimum program type specific 3595 * convert_ctx_access changes. If conversion is successful, 3596 * we will apply proper mask to the result. 3597 */ 3598 is_narrower_load = size < ctx_field_size; 3599 if (is_narrower_load) { 3600 u32 off = insn->off; 3601 u8 size_code; 3602 3603 if (type == BPF_WRITE) { 3604 verbose("bpf verifier narrow ctx access misconfigured\n"); 3605 return -EINVAL; 3606 } 3607 3608 size_code = BPF_H; 3609 if (ctx_field_size == 4) 3610 size_code = BPF_W; 3611 else if (ctx_field_size == 8) 3612 size_code = BPF_DW; 3613 3614 insn->off = off & ~(ctx_field_size - 1); 3615 insn->code = BPF_LDX | BPF_MEM | size_code; 3616 } 3617 3618 target_size = 0; 3619 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, 3620 &target_size); 3621 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || 3622 (ctx_field_size && !target_size)) { 3623 verbose("bpf verifier is misconfigured\n"); 3624 return -EINVAL; 3625 } 3626 3627 if (is_narrower_load && size < target_size) { 3628 if (ctx_field_size <= 4) 3629 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, 3630 (1 << size * 8) - 1); 3631 else 3632 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, 3633 (1 << size * 8) - 1); 3634 } 3635 3636 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); 3637 if (!new_prog) 3638 return -ENOMEM; 3639 3640 delta += cnt - 1; 3641 3642 /* keep walking new program and skip insns we just inserted */ 3643 env->prog = new_prog; 3644 insn = new_prog->insnsi + i + delta; 3645 } 3646 3647 return 0; 3648 } 3649 3650 /* fixup insn->imm field of bpf_call instructions 3651 * and inline eligible helpers as explicit sequence of BPF instructions 3652 * 3653 * this function is called after eBPF program passed verification 3654 */ 3655 static int fixup_bpf_calls(struct bpf_verifier_env *env) 3656 { 3657 struct bpf_prog *prog = env->prog; 3658 struct bpf_insn *insn = prog->insnsi; 3659 const struct bpf_func_proto *fn; 3660 const int insn_cnt = prog->len; 3661 struct bpf_insn insn_buf[16]; 3662 struct bpf_prog *new_prog; 3663 struct bpf_map *map_ptr; 3664 int i, cnt, delta = 0; 3665 3666 for (i = 0; i < insn_cnt; i++, insn++) { 3667 if (insn->code != (BPF_JMP | BPF_CALL)) 3668 continue; 3669 3670 if (insn->imm == BPF_FUNC_get_route_realm) 3671 prog->dst_needed = 1; 3672 if (insn->imm == BPF_FUNC_get_prandom_u32) 3673 bpf_user_rnd_init_once(); 3674 if (insn->imm == BPF_FUNC_tail_call) { 3675 /* If we tail call into other programs, we 3676 * cannot make any assumptions since they can 3677 * be replaced dynamically during runtime in 3678 * the program array. 3679 */ 3680 prog->cb_access = 1; 3681 env->prog->aux->stack_depth = MAX_BPF_STACK; 3682 3683 /* mark bpf_tail_call as different opcode to avoid 3684 * conditional branch in the interpeter for every normal 3685 * call and to prevent accidental JITing by JIT compiler 3686 * that doesn't support bpf_tail_call yet 3687 */ 3688 insn->imm = 0; 3689 insn->code = BPF_JMP | BPF_TAIL_CALL; 3690 continue; 3691 } 3692 3693 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) { 3694 map_ptr = env->insn_aux_data[i + delta].map_ptr; 3695 if (map_ptr == BPF_MAP_PTR_POISON || 3696 !map_ptr->ops->map_gen_lookup) 3697 goto patch_call_imm; 3698 3699 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); 3700 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { 3701 verbose("bpf verifier is misconfigured\n"); 3702 return -EINVAL; 3703 } 3704 3705 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 3706 cnt); 3707 if (!new_prog) 3708 return -ENOMEM; 3709 3710 delta += cnt - 1; 3711 3712 /* keep walking new program and skip insns we just inserted */ 3713 env->prog = prog = new_prog; 3714 insn = new_prog->insnsi + i + delta; 3715 continue; 3716 } 3717 3718 patch_call_imm: 3719 fn = prog->aux->ops->get_func_proto(insn->imm); 3720 /* all functions that have prototype and verifier allowed 3721 * programs to call them, must be real in-kernel functions 3722 */ 3723 if (!fn->func) { 3724 verbose("kernel subsystem misconfigured func %s#%d\n", 3725 func_id_name(insn->imm), insn->imm); 3726 return -EFAULT; 3727 } 3728 insn->imm = fn->func - __bpf_call_base; 3729 } 3730 3731 return 0; 3732 } 3733 3734 static void free_states(struct bpf_verifier_env *env) 3735 { 3736 struct bpf_verifier_state_list *sl, *sln; 3737 int i; 3738 3739 if (!env->explored_states) 3740 return; 3741 3742 for (i = 0; i < env->prog->len; i++) { 3743 sl = env->explored_states[i]; 3744 3745 if (sl) 3746 while (sl != STATE_LIST_MARK) { 3747 sln = sl->next; 3748 kfree(sl); 3749 sl = sln; 3750 } 3751 } 3752 3753 kfree(env->explored_states); 3754 } 3755 3756 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) 3757 { 3758 char __user *log_ubuf = NULL; 3759 struct bpf_verifier_env *env; 3760 int ret = -EINVAL; 3761 3762 /* 'struct bpf_verifier_env' can be global, but since it's not small, 3763 * allocate/free it every time bpf_check() is called 3764 */ 3765 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3766 if (!env) 3767 return -ENOMEM; 3768 3769 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3770 (*prog)->len); 3771 ret = -ENOMEM; 3772 if (!env->insn_aux_data) 3773 goto err_free_env; 3774 env->prog = *prog; 3775 3776 /* grab the mutex to protect few globals used by verifier */ 3777 mutex_lock(&bpf_verifier_lock); 3778 3779 if (attr->log_level || attr->log_buf || attr->log_size) { 3780 /* user requested verbose verifier output 3781 * and supplied buffer to store the verification trace 3782 */ 3783 log_level = attr->log_level; 3784 log_ubuf = (char __user *) (unsigned long) attr->log_buf; 3785 log_size = attr->log_size; 3786 log_len = 0; 3787 3788 ret = -EINVAL; 3789 /* log_* values have to be sane */ 3790 if (log_size < 128 || log_size > UINT_MAX >> 8 || 3791 log_level == 0 || log_ubuf == NULL) 3792 goto err_unlock; 3793 3794 ret = -ENOMEM; 3795 log_buf = vmalloc(log_size); 3796 if (!log_buf) 3797 goto err_unlock; 3798 } else { 3799 log_level = 0; 3800 } 3801 3802 env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); 3803 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 3804 env->strict_alignment = true; 3805 3806 ret = replace_map_fd_with_map_ptr(env); 3807 if (ret < 0) 3808 goto skip_full_check; 3809 3810 env->explored_states = kcalloc(env->prog->len, 3811 sizeof(struct bpf_verifier_state_list *), 3812 GFP_USER); 3813 ret = -ENOMEM; 3814 if (!env->explored_states) 3815 goto skip_full_check; 3816 3817 ret = check_cfg(env); 3818 if (ret < 0) 3819 goto skip_full_check; 3820 3821 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3822 3823 ret = do_check(env); 3824 3825 skip_full_check: 3826 while (pop_stack(env, NULL) >= 0); 3827 free_states(env); 3828 3829 if (ret == 0) 3830 /* program is valid, convert *(u32*)(ctx + off) accesses */ 3831 ret = convert_ctx_accesses(env); 3832 3833 if (ret == 0) 3834 ret = fixup_bpf_calls(env); 3835 3836 if (log_level && log_len >= log_size - 1) { 3837 BUG_ON(log_len >= log_size); 3838 /* verifier log exceeded user supplied buffer */ 3839 ret = -ENOSPC; 3840 /* fall through to return what was recorded */ 3841 } 3842 3843 /* copy verifier log back to user space including trailing zero */ 3844 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { 3845 ret = -EFAULT; 3846 goto free_log_buf; 3847 } 3848 3849 if (ret == 0 && env->used_map_cnt) { 3850 /* if program passed verifier, update used_maps in bpf_prog_info */ 3851 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, 3852 sizeof(env->used_maps[0]), 3853 GFP_KERNEL); 3854 3855 if (!env->prog->aux->used_maps) { 3856 ret = -ENOMEM; 3857 goto free_log_buf; 3858 } 3859 3860 memcpy(env->prog->aux->used_maps, env->used_maps, 3861 sizeof(env->used_maps[0]) * env->used_map_cnt); 3862 env->prog->aux->used_map_cnt = env->used_map_cnt; 3863 3864 /* program is valid. Convert pseudo bpf_ld_imm64 into generic 3865 * bpf_ld_imm64 instructions 3866 */ 3867 convert_pseudo_ld_imm64(env); 3868 } 3869 3870 free_log_buf: 3871 if (log_level) 3872 vfree(log_buf); 3873 if (!env->prog->aux->used_maps) 3874 /* if we didn't copy map pointers into bpf_prog_info, release 3875 * them now. Otherwise free_bpf_prog_info() will release them. 3876 */ 3877 release_maps(env); 3878 *prog = env->prog; 3879 err_unlock: 3880 mutex_unlock(&bpf_verifier_lock); 3881 vfree(env->insn_aux_data); 3882 err_free_env: 3883 kfree(env); 3884 return ret; 3885 } 3886 3887 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, 3888 void *priv) 3889 { 3890 struct bpf_verifier_env *env; 3891 int ret; 3892 3893 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3894 if (!env) 3895 return -ENOMEM; 3896 3897 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3898 prog->len); 3899 ret = -ENOMEM; 3900 if (!env->insn_aux_data) 3901 goto err_free_env; 3902 env->prog = prog; 3903 env->analyzer_ops = ops; 3904 env->analyzer_priv = priv; 3905 3906 /* grab the mutex to protect few globals used by verifier */ 3907 mutex_lock(&bpf_verifier_lock); 3908 3909 log_level = 0; 3910 3911 env->strict_alignment = false; 3912 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 3913 env->strict_alignment = true; 3914 3915 env->explored_states = kcalloc(env->prog->len, 3916 sizeof(struct bpf_verifier_state_list *), 3917 GFP_KERNEL); 3918 ret = -ENOMEM; 3919 if (!env->explored_states) 3920 goto skip_full_check; 3921 3922 ret = check_cfg(env); 3923 if (ret < 0) 3924 goto skip_full_check; 3925 3926 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3927 3928 ret = do_check(env); 3929 3930 skip_full_check: 3931 while (pop_stack(env, NULL) >= 0); 3932 free_states(env); 3933 3934 mutex_unlock(&bpf_verifier_lock); 3935 vfree(env->insn_aux_data); 3936 err_free_env: 3937 kfree(env); 3938 return ret; 3939 } 3940 EXPORT_SYMBOL_GPL(bpf_analyzer); 3941