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 case BPF_MAP_TYPE_ARRAY_OF_MAPS: 1287 case BPF_MAP_TYPE_HASH_OF_MAPS: 1288 if (func_id != BPF_FUNC_map_lookup_elem) 1289 goto error; 1290 default: 1291 break; 1292 } 1293 1294 /* ... and second from the function itself. */ 1295 switch (func_id) { 1296 case BPF_FUNC_tail_call: 1297 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 1298 goto error; 1299 break; 1300 case BPF_FUNC_perf_event_read: 1301 case BPF_FUNC_perf_event_output: 1302 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) 1303 goto error; 1304 break; 1305 case BPF_FUNC_get_stackid: 1306 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) 1307 goto error; 1308 break; 1309 case BPF_FUNC_current_task_under_cgroup: 1310 case BPF_FUNC_skb_under_cgroup: 1311 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) 1312 goto error; 1313 break; 1314 default: 1315 break; 1316 } 1317 1318 return 0; 1319 error: 1320 verbose("cannot pass map_type %d into func %s#%d\n", 1321 map->map_type, func_id_name(func_id), func_id); 1322 return -EINVAL; 1323 } 1324 1325 static int check_raw_mode(const struct bpf_func_proto *fn) 1326 { 1327 int count = 0; 1328 1329 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) 1330 count++; 1331 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) 1332 count++; 1333 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) 1334 count++; 1335 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) 1336 count++; 1337 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) 1338 count++; 1339 1340 return count > 1 ? -EINVAL : 0; 1341 } 1342 1343 static void clear_all_pkt_pointers(struct bpf_verifier_env *env) 1344 { 1345 struct bpf_verifier_state *state = &env->cur_state; 1346 struct bpf_reg_state *regs = state->regs, *reg; 1347 int i; 1348 1349 for (i = 0; i < MAX_BPF_REG; i++) 1350 if (regs[i].type == PTR_TO_PACKET || 1351 regs[i].type == PTR_TO_PACKET_END) 1352 mark_reg_unknown_value(regs, i); 1353 1354 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1355 if (state->stack_slot_type[i] != STACK_SPILL) 1356 continue; 1357 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1358 if (reg->type != PTR_TO_PACKET && 1359 reg->type != PTR_TO_PACKET_END) 1360 continue; 1361 __mark_reg_unknown_value(state->spilled_regs, 1362 i / BPF_REG_SIZE); 1363 } 1364 } 1365 1366 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) 1367 { 1368 struct bpf_verifier_state *state = &env->cur_state; 1369 const struct bpf_func_proto *fn = NULL; 1370 struct bpf_reg_state *regs = state->regs; 1371 struct bpf_call_arg_meta meta; 1372 bool changes_data; 1373 int i, err; 1374 1375 /* find function prototype */ 1376 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { 1377 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id); 1378 return -EINVAL; 1379 } 1380 1381 if (env->prog->aux->ops->get_func_proto) 1382 fn = env->prog->aux->ops->get_func_proto(func_id); 1383 1384 if (!fn) { 1385 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id); 1386 return -EINVAL; 1387 } 1388 1389 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1390 if (!env->prog->gpl_compatible && fn->gpl_only) { 1391 verbose("cannot call GPL only function from proprietary program\n"); 1392 return -EINVAL; 1393 } 1394 1395 changes_data = bpf_helper_changes_pkt_data(fn->func); 1396 1397 memset(&meta, 0, sizeof(meta)); 1398 meta.pkt_access = fn->pkt_access; 1399 1400 /* We only support one arg being in raw mode at the moment, which 1401 * is sufficient for the helper functions we have right now. 1402 */ 1403 err = check_raw_mode(fn); 1404 if (err) { 1405 verbose("kernel subsystem misconfigured func %s#%d\n", 1406 func_id_name(func_id), func_id); 1407 return err; 1408 } 1409 1410 /* check args */ 1411 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); 1412 if (err) 1413 return err; 1414 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); 1415 if (err) 1416 return err; 1417 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); 1418 if (err) 1419 return err; 1420 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); 1421 if (err) 1422 return err; 1423 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); 1424 if (err) 1425 return err; 1426 1427 /* Mark slots with STACK_MISC in case of raw mode, stack offset 1428 * is inferred from register state. 1429 */ 1430 for (i = 0; i < meta.access_size; i++) { 1431 err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1); 1432 if (err) 1433 return err; 1434 } 1435 1436 /* reset caller saved regs */ 1437 for (i = 0; i < CALLER_SAVED_REGS; i++) 1438 mark_reg_not_init(regs, caller_saved[i]); 1439 1440 /* update return register */ 1441 if (fn->ret_type == RET_INTEGER) { 1442 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1443 } else if (fn->ret_type == RET_VOID) { 1444 regs[BPF_REG_0].type = NOT_INIT; 1445 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { 1446 struct bpf_insn_aux_data *insn_aux; 1447 1448 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; 1449 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0; 1450 /* remember map_ptr, so that check_map_access() 1451 * can check 'value_size' boundary of memory access 1452 * to map element returned from bpf_map_lookup_elem() 1453 */ 1454 if (meta.map_ptr == NULL) { 1455 verbose("kernel subsystem misconfigured verifier\n"); 1456 return -EINVAL; 1457 } 1458 regs[BPF_REG_0].map_ptr = meta.map_ptr; 1459 regs[BPF_REG_0].id = ++env->id_gen; 1460 insn_aux = &env->insn_aux_data[insn_idx]; 1461 if (!insn_aux->map_ptr) 1462 insn_aux->map_ptr = meta.map_ptr; 1463 else if (insn_aux->map_ptr != meta.map_ptr) 1464 insn_aux->map_ptr = BPF_MAP_PTR_POISON; 1465 } else { 1466 verbose("unknown return type %d of func %s#%d\n", 1467 fn->ret_type, func_id_name(func_id), func_id); 1468 return -EINVAL; 1469 } 1470 1471 err = check_map_func_compatibility(meta.map_ptr, func_id); 1472 if (err) 1473 return err; 1474 1475 if (changes_data) 1476 clear_all_pkt_pointers(env); 1477 return 0; 1478 } 1479 1480 static int check_packet_ptr_add(struct bpf_verifier_env *env, 1481 struct bpf_insn *insn) 1482 { 1483 struct bpf_reg_state *regs = env->cur_state.regs; 1484 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1485 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1486 struct bpf_reg_state tmp_reg; 1487 s32 imm; 1488 1489 if (BPF_SRC(insn->code) == BPF_K) { 1490 /* pkt_ptr += imm */ 1491 imm = insn->imm; 1492 1493 add_imm: 1494 if (imm < 0) { 1495 verbose("addition of negative constant to packet pointer is not allowed\n"); 1496 return -EACCES; 1497 } 1498 if (imm >= MAX_PACKET_OFF || 1499 imm + dst_reg->off >= MAX_PACKET_OFF) { 1500 verbose("constant %d is too large to add to packet pointer\n", 1501 imm); 1502 return -EACCES; 1503 } 1504 /* a constant was added to pkt_ptr. 1505 * Remember it while keeping the same 'id' 1506 */ 1507 dst_reg->off += imm; 1508 } else { 1509 bool had_id; 1510 1511 if (src_reg->type == PTR_TO_PACKET) { 1512 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ 1513 tmp_reg = *dst_reg; /* save r7 state */ 1514 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ 1515 src_reg = &tmp_reg; /* pretend it's src_reg state */ 1516 /* if the checks below reject it, the copy won't matter, 1517 * since we're rejecting the whole program. If all ok, 1518 * then imm22 state will be added to r7 1519 * and r7 will be pkt(id=0,off=22,r=62) while 1520 * r6 will stay as pkt(id=0,off=0,r=62) 1521 */ 1522 } 1523 1524 if (src_reg->type == CONST_IMM) { 1525 /* pkt_ptr += reg where reg is known constant */ 1526 imm = src_reg->imm; 1527 goto add_imm; 1528 } 1529 /* disallow pkt_ptr += reg 1530 * if reg is not uknown_value with guaranteed zero upper bits 1531 * otherwise pkt_ptr may overflow and addition will become 1532 * subtraction which is not allowed 1533 */ 1534 if (src_reg->type != UNKNOWN_VALUE) { 1535 verbose("cannot add '%s' to ptr_to_packet\n", 1536 reg_type_str[src_reg->type]); 1537 return -EACCES; 1538 } 1539 if (src_reg->imm < 48) { 1540 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", 1541 src_reg->imm); 1542 return -EACCES; 1543 } 1544 1545 had_id = (dst_reg->id != 0); 1546 1547 /* dst_reg stays as pkt_ptr type and since some positive 1548 * integer value was added to the pointer, increment its 'id' 1549 */ 1550 dst_reg->id = ++env->id_gen; 1551 1552 /* something was added to pkt_ptr, set range to zero */ 1553 dst_reg->aux_off += dst_reg->off; 1554 dst_reg->off = 0; 1555 dst_reg->range = 0; 1556 if (had_id) 1557 dst_reg->aux_off_align = min(dst_reg->aux_off_align, 1558 src_reg->min_align); 1559 else 1560 dst_reg->aux_off_align = src_reg->min_align; 1561 } 1562 return 0; 1563 } 1564 1565 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) 1566 { 1567 struct bpf_reg_state *regs = env->cur_state.regs; 1568 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1569 u8 opcode = BPF_OP(insn->code); 1570 s64 imm_log2; 1571 1572 /* for type == UNKNOWN_VALUE: 1573 * imm > 0 -> number of zero upper bits 1574 * imm == 0 -> don't track which is the same as all bits can be non-zero 1575 */ 1576 1577 if (BPF_SRC(insn->code) == BPF_X) { 1578 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1579 1580 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && 1581 dst_reg->imm && opcode == BPF_ADD) { 1582 /* dreg += sreg 1583 * where both have zero upper bits. Adding them 1584 * can only result making one more bit non-zero 1585 * in the larger value. 1586 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1587 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1588 */ 1589 dst_reg->imm = min(dst_reg->imm, src_reg->imm); 1590 dst_reg->imm--; 1591 return 0; 1592 } 1593 if (src_reg->type == CONST_IMM && src_reg->imm > 0 && 1594 dst_reg->imm && opcode == BPF_ADD) { 1595 /* dreg += sreg 1596 * where dreg has zero upper bits and sreg is const. 1597 * Adding them can only result making one more bit 1598 * non-zero in the larger value. 1599 */ 1600 imm_log2 = __ilog2_u64((long long)src_reg->imm); 1601 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1602 dst_reg->imm--; 1603 return 0; 1604 } 1605 /* all other cases non supported yet, just mark dst_reg */ 1606 dst_reg->imm = 0; 1607 return 0; 1608 } 1609 1610 /* sign extend 32-bit imm into 64-bit to make sure that 1611 * negative values occupy bit 63. Note ilog2() would have 1612 * been incorrect, since sizeof(insn->imm) == 4 1613 */ 1614 imm_log2 = __ilog2_u64((long long)insn->imm); 1615 1616 if (dst_reg->imm && opcode == BPF_LSH) { 1617 /* reg <<= imm 1618 * if reg was a result of 2 byte load, then its imm == 48 1619 * which means that upper 48 bits are zero and shifting this reg 1620 * left by 4 would mean that upper 44 bits are still zero 1621 */ 1622 dst_reg->imm -= insn->imm; 1623 } else if (dst_reg->imm && opcode == BPF_MUL) { 1624 /* reg *= imm 1625 * if multiplying by 14 subtract 4 1626 * This is conservative calculation of upper zero bits. 1627 * It's not trying to special case insn->imm == 1 or 0 cases 1628 */ 1629 dst_reg->imm -= imm_log2 + 1; 1630 } else if (opcode == BPF_AND) { 1631 /* reg &= imm */ 1632 dst_reg->imm = 63 - imm_log2; 1633 } else if (dst_reg->imm && opcode == BPF_ADD) { 1634 /* reg += imm */ 1635 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1636 dst_reg->imm--; 1637 } else if (opcode == BPF_RSH) { 1638 /* reg >>= imm 1639 * which means that after right shift, upper bits will be zero 1640 * note that verifier already checked that 1641 * 0 <= imm < 64 for shift insn 1642 */ 1643 dst_reg->imm += insn->imm; 1644 if (unlikely(dst_reg->imm > 64)) 1645 /* some dumb code did: 1646 * r2 = *(u32 *)mem; 1647 * r2 >>= 32; 1648 * and all bits are zero now */ 1649 dst_reg->imm = 64; 1650 } else { 1651 /* all other alu ops, means that we don't know what will 1652 * happen to the value, mark it with unknown number of zero bits 1653 */ 1654 dst_reg->imm = 0; 1655 } 1656 1657 if (dst_reg->imm < 0) { 1658 /* all 64 bits of the register can contain non-zero bits 1659 * and such value cannot be added to ptr_to_packet, since it 1660 * may overflow, mark it as unknown to avoid further eval 1661 */ 1662 dst_reg->imm = 0; 1663 } 1664 return 0; 1665 } 1666 1667 static int evaluate_reg_imm_alu_unknown(struct bpf_verifier_env *env, 1668 struct bpf_insn *insn) 1669 { 1670 struct bpf_reg_state *regs = env->cur_state.regs; 1671 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1672 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1673 u8 opcode = BPF_OP(insn->code); 1674 s64 imm_log2 = __ilog2_u64((long long)dst_reg->imm); 1675 1676 /* BPF_X code with src_reg->type UNKNOWN_VALUE here. */ 1677 if (src_reg->imm > 0 && dst_reg->imm) { 1678 switch (opcode) { 1679 case BPF_ADD: 1680 /* dreg += sreg 1681 * where both have zero upper bits. Adding them 1682 * can only result making one more bit non-zero 1683 * in the larger value. 1684 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1685 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1686 */ 1687 dst_reg->imm = min(src_reg->imm, 63 - imm_log2); 1688 dst_reg->imm--; 1689 break; 1690 case BPF_AND: 1691 /* dreg &= sreg 1692 * AND can not extend zero bits only shrink 1693 * Ex. 0x00..00ffffff 1694 * & 0x0f..ffffffff 1695 * ---------------- 1696 * 0x00..00ffffff 1697 */ 1698 dst_reg->imm = max(src_reg->imm, 63 - imm_log2); 1699 break; 1700 case BPF_OR: 1701 /* dreg |= sreg 1702 * OR can only extend zero bits 1703 * Ex. 0x00..00ffffff 1704 * | 0x0f..ffffffff 1705 * ---------------- 1706 * 0x0f..00ffffff 1707 */ 1708 dst_reg->imm = min(src_reg->imm, 63 - imm_log2); 1709 break; 1710 case BPF_SUB: 1711 case BPF_MUL: 1712 case BPF_RSH: 1713 case BPF_LSH: 1714 /* These may be flushed out later */ 1715 default: 1716 mark_reg_unknown_value(regs, insn->dst_reg); 1717 } 1718 } else { 1719 mark_reg_unknown_value(regs, insn->dst_reg); 1720 } 1721 1722 dst_reg->type = UNKNOWN_VALUE; 1723 return 0; 1724 } 1725 1726 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, 1727 struct bpf_insn *insn) 1728 { 1729 struct bpf_reg_state *regs = env->cur_state.regs; 1730 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1731 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1732 u8 opcode = BPF_OP(insn->code); 1733 u64 dst_imm = dst_reg->imm; 1734 1735 if (BPF_SRC(insn->code) == BPF_X && src_reg->type == UNKNOWN_VALUE) 1736 return evaluate_reg_imm_alu_unknown(env, insn); 1737 1738 /* dst_reg->type == CONST_IMM here. Simulate execution of insns 1739 * containing ALU ops. Don't care about overflow or negative 1740 * values, just add/sub/... them; registers are in u64. 1741 */ 1742 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) { 1743 dst_imm += insn->imm; 1744 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && 1745 src_reg->type == CONST_IMM) { 1746 dst_imm += src_reg->imm; 1747 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) { 1748 dst_imm -= insn->imm; 1749 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X && 1750 src_reg->type == CONST_IMM) { 1751 dst_imm -= src_reg->imm; 1752 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) { 1753 dst_imm *= insn->imm; 1754 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X && 1755 src_reg->type == CONST_IMM) { 1756 dst_imm *= src_reg->imm; 1757 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) { 1758 dst_imm |= insn->imm; 1759 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X && 1760 src_reg->type == CONST_IMM) { 1761 dst_imm |= src_reg->imm; 1762 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) { 1763 dst_imm &= insn->imm; 1764 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X && 1765 src_reg->type == CONST_IMM) { 1766 dst_imm &= src_reg->imm; 1767 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) { 1768 dst_imm >>= insn->imm; 1769 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X && 1770 src_reg->type == CONST_IMM) { 1771 dst_imm >>= src_reg->imm; 1772 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) { 1773 dst_imm <<= insn->imm; 1774 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X && 1775 src_reg->type == CONST_IMM) { 1776 dst_imm <<= src_reg->imm; 1777 } else { 1778 mark_reg_unknown_value(regs, insn->dst_reg); 1779 goto out; 1780 } 1781 1782 dst_reg->imm = dst_imm; 1783 out: 1784 return 0; 1785 } 1786 1787 static void check_reg_overflow(struct bpf_reg_state *reg) 1788 { 1789 if (reg->max_value > BPF_REGISTER_MAX_RANGE) 1790 reg->max_value = BPF_REGISTER_MAX_RANGE; 1791 if (reg->min_value < BPF_REGISTER_MIN_RANGE || 1792 reg->min_value > BPF_REGISTER_MAX_RANGE) 1793 reg->min_value = BPF_REGISTER_MIN_RANGE; 1794 } 1795 1796 static u32 calc_align(u32 imm) 1797 { 1798 if (!imm) 1799 return 1U << 31; 1800 return imm - ((imm - 1) & imm); 1801 } 1802 1803 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env, 1804 struct bpf_insn *insn) 1805 { 1806 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1807 s64 min_val = BPF_REGISTER_MIN_RANGE; 1808 u64 max_val = BPF_REGISTER_MAX_RANGE; 1809 u8 opcode = BPF_OP(insn->code); 1810 u32 dst_align, src_align; 1811 1812 dst_reg = ®s[insn->dst_reg]; 1813 src_align = 0; 1814 if (BPF_SRC(insn->code) == BPF_X) { 1815 check_reg_overflow(®s[insn->src_reg]); 1816 min_val = regs[insn->src_reg].min_value; 1817 max_val = regs[insn->src_reg].max_value; 1818 1819 /* If the source register is a random pointer then the 1820 * min_value/max_value values represent the range of the known 1821 * accesses into that value, not the actual min/max value of the 1822 * register itself. In this case we have to reset the reg range 1823 * values so we know it is not safe to look at. 1824 */ 1825 if (regs[insn->src_reg].type != CONST_IMM && 1826 regs[insn->src_reg].type != UNKNOWN_VALUE) { 1827 min_val = BPF_REGISTER_MIN_RANGE; 1828 max_val = BPF_REGISTER_MAX_RANGE; 1829 src_align = 0; 1830 } else { 1831 src_align = regs[insn->src_reg].min_align; 1832 } 1833 } else if (insn->imm < BPF_REGISTER_MAX_RANGE && 1834 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) { 1835 min_val = max_val = insn->imm; 1836 src_align = calc_align(insn->imm); 1837 } 1838 1839 dst_align = dst_reg->min_align; 1840 1841 /* We don't know anything about what was done to this register, mark it 1842 * as unknown. Also, if both derived bounds came from signed/unsigned 1843 * mixed compares and one side is unbounded, we cannot really do anything 1844 * with them as boundaries cannot be trusted. Thus, arithmetic of two 1845 * regs of such kind will get invalidated bounds on the dst side. 1846 */ 1847 if ((min_val == BPF_REGISTER_MIN_RANGE && 1848 max_val == BPF_REGISTER_MAX_RANGE) || 1849 (BPF_SRC(insn->code) == BPF_X && 1850 ((min_val != BPF_REGISTER_MIN_RANGE && 1851 max_val == BPF_REGISTER_MAX_RANGE) || 1852 (min_val == BPF_REGISTER_MIN_RANGE && 1853 max_val != BPF_REGISTER_MAX_RANGE) || 1854 (dst_reg->min_value != BPF_REGISTER_MIN_RANGE && 1855 dst_reg->max_value == BPF_REGISTER_MAX_RANGE) || 1856 (dst_reg->min_value == BPF_REGISTER_MIN_RANGE && 1857 dst_reg->max_value != BPF_REGISTER_MAX_RANGE)) && 1858 regs[insn->dst_reg].value_from_signed != 1859 regs[insn->src_reg].value_from_signed)) { 1860 reset_reg_range_values(regs, insn->dst_reg); 1861 return; 1862 } 1863 1864 /* If one of our values was at the end of our ranges then we can't just 1865 * do our normal operations to the register, we need to set the values 1866 * to the min/max since they are undefined. 1867 */ 1868 if (opcode != BPF_SUB) { 1869 if (min_val == BPF_REGISTER_MIN_RANGE) 1870 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1871 if (max_val == BPF_REGISTER_MAX_RANGE) 1872 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1873 } 1874 1875 switch (opcode) { 1876 case BPF_ADD: 1877 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1878 dst_reg->min_value += min_val; 1879 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1880 dst_reg->max_value += max_val; 1881 dst_reg->min_align = min(src_align, dst_align); 1882 break; 1883 case BPF_SUB: 1884 /* If one of our values was at the end of our ranges, then the 1885 * _opposite_ value in the dst_reg goes to the end of our range. 1886 */ 1887 if (min_val == BPF_REGISTER_MIN_RANGE) 1888 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1889 if (max_val == BPF_REGISTER_MAX_RANGE) 1890 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1891 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1892 dst_reg->min_value -= max_val; 1893 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1894 dst_reg->max_value -= min_val; 1895 dst_reg->min_align = min(src_align, dst_align); 1896 break; 1897 case BPF_MUL: 1898 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1899 dst_reg->min_value *= min_val; 1900 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1901 dst_reg->max_value *= max_val; 1902 dst_reg->min_align = max(src_align, dst_align); 1903 break; 1904 case BPF_AND: 1905 /* Disallow AND'ing of negative numbers, ain't nobody got time 1906 * for that. Otherwise the minimum is 0 and the max is the max 1907 * value we could AND against. 1908 */ 1909 if (min_val < 0) 1910 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1911 else 1912 dst_reg->min_value = 0; 1913 dst_reg->max_value = max_val; 1914 dst_reg->min_align = max(src_align, dst_align); 1915 break; 1916 case BPF_LSH: 1917 /* Gotta have special overflow logic here, if we're shifting 1918 * more than MAX_RANGE then just assume we have an invalid 1919 * range. 1920 */ 1921 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) { 1922 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1923 dst_reg->min_align = 1; 1924 } else { 1925 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1926 dst_reg->min_value <<= min_val; 1927 if (!dst_reg->min_align) 1928 dst_reg->min_align = 1; 1929 dst_reg->min_align <<= min_val; 1930 } 1931 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) 1932 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1933 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1934 dst_reg->max_value <<= max_val; 1935 break; 1936 case BPF_RSH: 1937 /* RSH by a negative number is undefined, and the BPF_RSH is an 1938 * unsigned shift, so make the appropriate casts. 1939 */ 1940 if (min_val < 0 || dst_reg->min_value < 0) { 1941 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1942 } else { 1943 dst_reg->min_value = 1944 (u64)(dst_reg->min_value) >> min_val; 1945 } 1946 if (min_val < 0) { 1947 dst_reg->min_align = 1; 1948 } else { 1949 dst_reg->min_align >>= (u64) min_val; 1950 if (!dst_reg->min_align) 1951 dst_reg->min_align = 1; 1952 } 1953 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1954 dst_reg->max_value >>= max_val; 1955 break; 1956 default: 1957 reset_reg_range_values(regs, insn->dst_reg); 1958 break; 1959 } 1960 1961 check_reg_overflow(dst_reg); 1962 } 1963 1964 /* check validity of 32-bit and 64-bit arithmetic operations */ 1965 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) 1966 { 1967 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1968 u8 opcode = BPF_OP(insn->code); 1969 int err; 1970 1971 if (opcode == BPF_END || opcode == BPF_NEG) { 1972 if (opcode == BPF_NEG) { 1973 if (BPF_SRC(insn->code) != 0 || 1974 insn->src_reg != BPF_REG_0 || 1975 insn->off != 0 || insn->imm != 0) { 1976 verbose("BPF_NEG uses reserved fields\n"); 1977 return -EINVAL; 1978 } 1979 } else { 1980 if (insn->src_reg != BPF_REG_0 || insn->off != 0 || 1981 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { 1982 verbose("BPF_END uses reserved fields\n"); 1983 return -EINVAL; 1984 } 1985 } 1986 1987 /* check src operand */ 1988 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1989 if (err) 1990 return err; 1991 1992 if (is_pointer_value(env, insn->dst_reg)) { 1993 verbose("R%d pointer arithmetic prohibited\n", 1994 insn->dst_reg); 1995 return -EACCES; 1996 } 1997 1998 /* check dest operand */ 1999 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2000 if (err) 2001 return err; 2002 2003 } else if (opcode == BPF_MOV) { 2004 2005 if (BPF_SRC(insn->code) == BPF_X) { 2006 if (insn->imm != 0 || insn->off != 0) { 2007 verbose("BPF_MOV uses reserved fields\n"); 2008 return -EINVAL; 2009 } 2010 2011 /* check src operand */ 2012 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2013 if (err) 2014 return err; 2015 } else { 2016 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 2017 verbose("BPF_MOV uses reserved fields\n"); 2018 return -EINVAL; 2019 } 2020 } 2021 2022 /* check dest operand */ 2023 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2024 if (err) 2025 return err; 2026 2027 /* we are setting our register to something new, we need to 2028 * reset its range values. 2029 */ 2030 reset_reg_range_values(regs, insn->dst_reg); 2031 2032 if (BPF_SRC(insn->code) == BPF_X) { 2033 if (BPF_CLASS(insn->code) == BPF_ALU64) { 2034 /* case: R1 = R2 2035 * copy register state to dest reg 2036 */ 2037 regs[insn->dst_reg] = regs[insn->src_reg]; 2038 } else { 2039 if (is_pointer_value(env, insn->src_reg)) { 2040 verbose("R%d partial copy of pointer\n", 2041 insn->src_reg); 2042 return -EACCES; 2043 } 2044 mark_reg_unknown_value(regs, insn->dst_reg); 2045 } 2046 } else { 2047 /* case: R = imm 2048 * remember the value we stored into this reg 2049 */ 2050 regs[insn->dst_reg].type = CONST_IMM; 2051 regs[insn->dst_reg].imm = insn->imm; 2052 regs[insn->dst_reg].id = 0; 2053 regs[insn->dst_reg].max_value = insn->imm; 2054 regs[insn->dst_reg].min_value = insn->imm; 2055 regs[insn->dst_reg].min_align = calc_align(insn->imm); 2056 regs[insn->dst_reg].value_from_signed = false; 2057 } 2058 2059 } else if (opcode > BPF_END) { 2060 verbose("invalid BPF_ALU opcode %x\n", opcode); 2061 return -EINVAL; 2062 2063 } else { /* all other ALU ops: and, sub, xor, add, ... */ 2064 2065 if (BPF_SRC(insn->code) == BPF_X) { 2066 if (insn->imm != 0 || insn->off != 0) { 2067 verbose("BPF_ALU uses reserved fields\n"); 2068 return -EINVAL; 2069 } 2070 /* check src1 operand */ 2071 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2072 if (err) 2073 return err; 2074 } else { 2075 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 2076 verbose("BPF_ALU uses reserved fields\n"); 2077 return -EINVAL; 2078 } 2079 } 2080 2081 /* check src2 operand */ 2082 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2083 if (err) 2084 return err; 2085 2086 if ((opcode == BPF_MOD || opcode == BPF_DIV) && 2087 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { 2088 verbose("div by zero\n"); 2089 return -EINVAL; 2090 } 2091 2092 if ((opcode == BPF_LSH || opcode == BPF_RSH || 2093 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { 2094 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; 2095 2096 if (insn->imm < 0 || insn->imm >= size) { 2097 verbose("invalid shift %d\n", insn->imm); 2098 return -EINVAL; 2099 } 2100 } 2101 2102 /* check dest operand */ 2103 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 2104 if (err) 2105 return err; 2106 2107 dst_reg = ®s[insn->dst_reg]; 2108 2109 /* first we want to adjust our ranges. */ 2110 adjust_reg_min_max_vals(env, insn); 2111 2112 /* pattern match 'bpf_add Rx, imm' instruction */ 2113 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && 2114 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { 2115 dst_reg->type = PTR_TO_STACK; 2116 dst_reg->imm = insn->imm; 2117 return 0; 2118 } else if (opcode == BPF_ADD && 2119 BPF_CLASS(insn->code) == BPF_ALU64 && 2120 dst_reg->type == PTR_TO_STACK && 2121 ((BPF_SRC(insn->code) == BPF_X && 2122 regs[insn->src_reg].type == CONST_IMM) || 2123 BPF_SRC(insn->code) == BPF_K)) { 2124 if (BPF_SRC(insn->code) == BPF_X) 2125 dst_reg->imm += regs[insn->src_reg].imm; 2126 else 2127 dst_reg->imm += insn->imm; 2128 return 0; 2129 } else if (opcode == BPF_ADD && 2130 BPF_CLASS(insn->code) == BPF_ALU64 && 2131 (dst_reg->type == PTR_TO_PACKET || 2132 (BPF_SRC(insn->code) == BPF_X && 2133 regs[insn->src_reg].type == PTR_TO_PACKET))) { 2134 /* ptr_to_packet += K|X */ 2135 return check_packet_ptr_add(env, insn); 2136 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 2137 dst_reg->type == UNKNOWN_VALUE && 2138 env->allow_ptr_leaks) { 2139 /* unknown += K|X */ 2140 return evaluate_reg_alu(env, insn); 2141 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 2142 dst_reg->type == CONST_IMM && 2143 env->allow_ptr_leaks) { 2144 /* reg_imm += K|X */ 2145 return evaluate_reg_imm_alu(env, insn); 2146 } else if (is_pointer_value(env, insn->dst_reg)) { 2147 verbose("R%d pointer arithmetic prohibited\n", 2148 insn->dst_reg); 2149 return -EACCES; 2150 } else if (BPF_SRC(insn->code) == BPF_X && 2151 is_pointer_value(env, insn->src_reg)) { 2152 verbose("R%d pointer arithmetic prohibited\n", 2153 insn->src_reg); 2154 return -EACCES; 2155 } 2156 2157 /* If we did pointer math on a map value then just set it to our 2158 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or 2159 * loads to this register appropriately, otherwise just mark the 2160 * register as unknown. 2161 */ 2162 if (env->allow_ptr_leaks && 2163 BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD && 2164 (dst_reg->type == PTR_TO_MAP_VALUE || 2165 dst_reg->type == PTR_TO_MAP_VALUE_ADJ)) 2166 dst_reg->type = PTR_TO_MAP_VALUE_ADJ; 2167 else 2168 mark_reg_unknown_value(regs, insn->dst_reg); 2169 } 2170 2171 return 0; 2172 } 2173 2174 static void find_good_pkt_pointers(struct bpf_verifier_state *state, 2175 struct bpf_reg_state *dst_reg) 2176 { 2177 struct bpf_reg_state *regs = state->regs, *reg; 2178 int i; 2179 2180 /* LLVM can generate two kind of checks: 2181 * 2182 * Type 1: 2183 * 2184 * r2 = r3; 2185 * r2 += 8; 2186 * if (r2 > pkt_end) goto <handle exception> 2187 * <access okay> 2188 * 2189 * Where: 2190 * r2 == dst_reg, pkt_end == src_reg 2191 * r2=pkt(id=n,off=8,r=0) 2192 * r3=pkt(id=n,off=0,r=0) 2193 * 2194 * Type 2: 2195 * 2196 * r2 = r3; 2197 * r2 += 8; 2198 * if (pkt_end >= r2) goto <access okay> 2199 * <handle exception> 2200 * 2201 * Where: 2202 * pkt_end == dst_reg, r2 == src_reg 2203 * r2=pkt(id=n,off=8,r=0) 2204 * r3=pkt(id=n,off=0,r=0) 2205 * 2206 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) 2207 * so that range of bytes [r3, r3 + 8) is safe to access. 2208 */ 2209 2210 for (i = 0; i < MAX_BPF_REG; i++) 2211 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) 2212 /* keep the maximum range already checked */ 2213 regs[i].range = max(regs[i].range, dst_reg->off); 2214 2215 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 2216 if (state->stack_slot_type[i] != STACK_SPILL) 2217 continue; 2218 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 2219 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) 2220 reg->range = max(reg->range, dst_reg->off); 2221 } 2222 } 2223 2224 /* Adjusts the register min/max values in the case that the dst_reg is the 2225 * variable register that we are working on, and src_reg is a constant or we're 2226 * simply doing a BPF_K check. 2227 */ 2228 static void reg_set_min_max(struct bpf_reg_state *true_reg, 2229 struct bpf_reg_state *false_reg, u64 val, 2230 u8 opcode) 2231 { 2232 bool value_from_signed = true; 2233 bool is_range = true; 2234 2235 switch (opcode) { 2236 case BPF_JEQ: 2237 /* If this is false then we know nothing Jon Snow, but if it is 2238 * true then we know for sure. 2239 */ 2240 true_reg->max_value = true_reg->min_value = val; 2241 is_range = false; 2242 break; 2243 case BPF_JNE: 2244 /* If this is true we know nothing Jon Snow, but if it is false 2245 * we know the value for sure; 2246 */ 2247 false_reg->max_value = false_reg->min_value = val; 2248 is_range = false; 2249 break; 2250 case BPF_JGT: 2251 value_from_signed = false; 2252 /* fallthrough */ 2253 case BPF_JSGT: 2254 if (true_reg->value_from_signed != value_from_signed) 2255 reset_reg_range_values(true_reg, 0); 2256 if (false_reg->value_from_signed != value_from_signed) 2257 reset_reg_range_values(false_reg, 0); 2258 if (opcode == BPF_JGT) { 2259 /* Unsigned comparison, the minimum value is 0. */ 2260 false_reg->min_value = 0; 2261 } 2262 /* If this is false then we know the maximum val is val, 2263 * otherwise we know the min val is val+1. 2264 */ 2265 false_reg->max_value = val; 2266 false_reg->value_from_signed = value_from_signed; 2267 true_reg->min_value = val + 1; 2268 true_reg->value_from_signed = value_from_signed; 2269 break; 2270 case BPF_JGE: 2271 value_from_signed = false; 2272 /* fallthrough */ 2273 case BPF_JSGE: 2274 if (true_reg->value_from_signed != value_from_signed) 2275 reset_reg_range_values(true_reg, 0); 2276 if (false_reg->value_from_signed != value_from_signed) 2277 reset_reg_range_values(false_reg, 0); 2278 if (opcode == BPF_JGE) { 2279 /* Unsigned comparison, the minimum value is 0. */ 2280 false_reg->min_value = 0; 2281 } 2282 /* If this is false then we know the maximum value is val - 1, 2283 * otherwise we know the mimimum value is val. 2284 */ 2285 false_reg->max_value = val - 1; 2286 false_reg->value_from_signed = value_from_signed; 2287 true_reg->min_value = val; 2288 true_reg->value_from_signed = value_from_signed; 2289 break; 2290 default: 2291 break; 2292 } 2293 2294 check_reg_overflow(false_reg); 2295 check_reg_overflow(true_reg); 2296 if (is_range) { 2297 if (__is_pointer_value(false, false_reg)) 2298 reset_reg_range_values(false_reg, 0); 2299 if (__is_pointer_value(false, true_reg)) 2300 reset_reg_range_values(true_reg, 0); 2301 } 2302 } 2303 2304 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg 2305 * is the variable reg. 2306 */ 2307 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, 2308 struct bpf_reg_state *false_reg, u64 val, 2309 u8 opcode) 2310 { 2311 bool value_from_signed = true; 2312 bool is_range = true; 2313 2314 switch (opcode) { 2315 case BPF_JEQ: 2316 /* If this is false then we know nothing Jon Snow, but if it is 2317 * true then we know for sure. 2318 */ 2319 true_reg->max_value = true_reg->min_value = val; 2320 is_range = false; 2321 break; 2322 case BPF_JNE: 2323 /* If this is true we know nothing Jon Snow, but if it is false 2324 * we know the value for sure; 2325 */ 2326 false_reg->max_value = false_reg->min_value = val; 2327 is_range = false; 2328 break; 2329 case BPF_JGT: 2330 value_from_signed = false; 2331 /* fallthrough */ 2332 case BPF_JSGT: 2333 if (true_reg->value_from_signed != value_from_signed) 2334 reset_reg_range_values(true_reg, 0); 2335 if (false_reg->value_from_signed != value_from_signed) 2336 reset_reg_range_values(false_reg, 0); 2337 if (opcode == BPF_JGT) { 2338 /* Unsigned comparison, the minimum value is 0. */ 2339 true_reg->min_value = 0; 2340 } 2341 /* 2342 * If this is false, then the val is <= the register, if it is 2343 * true the register <= to the val. 2344 */ 2345 false_reg->min_value = val; 2346 false_reg->value_from_signed = value_from_signed; 2347 true_reg->max_value = val - 1; 2348 true_reg->value_from_signed = value_from_signed; 2349 break; 2350 case BPF_JGE: 2351 value_from_signed = false; 2352 /* fallthrough */ 2353 case BPF_JSGE: 2354 if (true_reg->value_from_signed != value_from_signed) 2355 reset_reg_range_values(true_reg, 0); 2356 if (false_reg->value_from_signed != value_from_signed) 2357 reset_reg_range_values(false_reg, 0); 2358 if (opcode == BPF_JGE) { 2359 /* Unsigned comparison, the minimum value is 0. */ 2360 true_reg->min_value = 0; 2361 } 2362 /* If this is false then constant < register, if it is true then 2363 * the register < constant. 2364 */ 2365 false_reg->min_value = val + 1; 2366 false_reg->value_from_signed = value_from_signed; 2367 true_reg->max_value = val; 2368 true_reg->value_from_signed = value_from_signed; 2369 break; 2370 default: 2371 break; 2372 } 2373 2374 check_reg_overflow(false_reg); 2375 check_reg_overflow(true_reg); 2376 if (is_range) { 2377 if (__is_pointer_value(false, false_reg)) 2378 reset_reg_range_values(false_reg, 0); 2379 if (__is_pointer_value(false, true_reg)) 2380 reset_reg_range_values(true_reg, 0); 2381 } 2382 } 2383 2384 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, 2385 enum bpf_reg_type type) 2386 { 2387 struct bpf_reg_state *reg = ®s[regno]; 2388 2389 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { 2390 if (type == UNKNOWN_VALUE) { 2391 __mark_reg_unknown_value(regs, regno); 2392 } else if (reg->map_ptr->inner_map_meta) { 2393 reg->type = CONST_PTR_TO_MAP; 2394 reg->map_ptr = reg->map_ptr->inner_map_meta; 2395 } else { 2396 reg->type = type; 2397 } 2398 /* We don't need id from this point onwards anymore, thus we 2399 * should better reset it, so that state pruning has chances 2400 * to take effect. 2401 */ 2402 reg->id = 0; 2403 } 2404 } 2405 2406 /* The logic is similar to find_good_pkt_pointers(), both could eventually 2407 * be folded together at some point. 2408 */ 2409 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, 2410 enum bpf_reg_type type) 2411 { 2412 struct bpf_reg_state *regs = state->regs; 2413 u32 id = regs[regno].id; 2414 int i; 2415 2416 for (i = 0; i < MAX_BPF_REG; i++) 2417 mark_map_reg(regs, i, id, type); 2418 2419 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 2420 if (state->stack_slot_type[i] != STACK_SPILL) 2421 continue; 2422 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type); 2423 } 2424 } 2425 2426 static int check_cond_jmp_op(struct bpf_verifier_env *env, 2427 struct bpf_insn *insn, int *insn_idx) 2428 { 2429 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; 2430 struct bpf_reg_state *regs = this_branch->regs, *dst_reg; 2431 u8 opcode = BPF_OP(insn->code); 2432 int err; 2433 2434 if (opcode > BPF_EXIT) { 2435 verbose("invalid BPF_JMP opcode %x\n", opcode); 2436 return -EINVAL; 2437 } 2438 2439 if (BPF_SRC(insn->code) == BPF_X) { 2440 if (insn->imm != 0) { 2441 verbose("BPF_JMP uses reserved fields\n"); 2442 return -EINVAL; 2443 } 2444 2445 /* check src1 operand */ 2446 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2447 if (err) 2448 return err; 2449 2450 if (is_pointer_value(env, insn->src_reg)) { 2451 verbose("R%d pointer comparison prohibited\n", 2452 insn->src_reg); 2453 return -EACCES; 2454 } 2455 } else { 2456 if (insn->src_reg != BPF_REG_0) { 2457 verbose("BPF_JMP uses reserved fields\n"); 2458 return -EINVAL; 2459 } 2460 } 2461 2462 /* check src2 operand */ 2463 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2464 if (err) 2465 return err; 2466 2467 dst_reg = ®s[insn->dst_reg]; 2468 2469 /* detect if R == 0 where R was initialized to zero earlier */ 2470 if (BPF_SRC(insn->code) == BPF_K && 2471 (opcode == BPF_JEQ || opcode == BPF_JNE) && 2472 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { 2473 if (opcode == BPF_JEQ) { 2474 /* if (imm == imm) goto pc+off; 2475 * only follow the goto, ignore fall-through 2476 */ 2477 *insn_idx += insn->off; 2478 return 0; 2479 } else { 2480 /* if (imm != imm) goto pc+off; 2481 * only follow fall-through branch, since 2482 * that's where the program will go 2483 */ 2484 return 0; 2485 } 2486 } 2487 2488 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); 2489 if (!other_branch) 2490 return -EFAULT; 2491 2492 /* detect if we are comparing against a constant value so we can adjust 2493 * our min/max values for our dst register. 2494 */ 2495 if (BPF_SRC(insn->code) == BPF_X) { 2496 if (regs[insn->src_reg].type == CONST_IMM) 2497 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2498 dst_reg, regs[insn->src_reg].imm, 2499 opcode); 2500 else if (dst_reg->type == CONST_IMM) 2501 reg_set_min_max_inv(&other_branch->regs[insn->src_reg], 2502 ®s[insn->src_reg], dst_reg->imm, 2503 opcode); 2504 } else { 2505 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2506 dst_reg, insn->imm, opcode); 2507 } 2508 2509 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ 2510 if (BPF_SRC(insn->code) == BPF_K && 2511 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && 2512 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { 2513 /* Mark all identical map registers in each branch as either 2514 * safe or unknown depending R == 0 or R != 0 conditional. 2515 */ 2516 mark_map_regs(this_branch, insn->dst_reg, 2517 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE); 2518 mark_map_regs(other_branch, insn->dst_reg, 2519 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE); 2520 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && 2521 dst_reg->type == PTR_TO_PACKET && 2522 regs[insn->src_reg].type == PTR_TO_PACKET_END) { 2523 find_good_pkt_pointers(this_branch, dst_reg); 2524 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && 2525 dst_reg->type == PTR_TO_PACKET_END && 2526 regs[insn->src_reg].type == PTR_TO_PACKET) { 2527 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); 2528 } else if (is_pointer_value(env, insn->dst_reg)) { 2529 verbose("R%d pointer comparison prohibited\n", insn->dst_reg); 2530 return -EACCES; 2531 } 2532 if (log_level) 2533 print_verifier_state(this_branch); 2534 return 0; 2535 } 2536 2537 /* return the map pointer stored inside BPF_LD_IMM64 instruction */ 2538 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) 2539 { 2540 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; 2541 2542 return (struct bpf_map *) (unsigned long) imm64; 2543 } 2544 2545 /* verify BPF_LD_IMM64 instruction */ 2546 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) 2547 { 2548 struct bpf_reg_state *regs = env->cur_state.regs; 2549 int err; 2550 2551 if (BPF_SIZE(insn->code) != BPF_DW) { 2552 verbose("invalid BPF_LD_IMM insn\n"); 2553 return -EINVAL; 2554 } 2555 if (insn->off != 0) { 2556 verbose("BPF_LD_IMM64 uses reserved fields\n"); 2557 return -EINVAL; 2558 } 2559 2560 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2561 if (err) 2562 return err; 2563 2564 if (insn->src_reg == 0) { 2565 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; 2566 2567 regs[insn->dst_reg].type = CONST_IMM; 2568 regs[insn->dst_reg].imm = imm; 2569 regs[insn->dst_reg].id = 0; 2570 return 0; 2571 } 2572 2573 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ 2574 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); 2575 2576 regs[insn->dst_reg].type = CONST_PTR_TO_MAP; 2577 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); 2578 return 0; 2579 } 2580 2581 static bool may_access_skb(enum bpf_prog_type type) 2582 { 2583 switch (type) { 2584 case BPF_PROG_TYPE_SOCKET_FILTER: 2585 case BPF_PROG_TYPE_SCHED_CLS: 2586 case BPF_PROG_TYPE_SCHED_ACT: 2587 return true; 2588 default: 2589 return false; 2590 } 2591 } 2592 2593 /* verify safety of LD_ABS|LD_IND instructions: 2594 * - they can only appear in the programs where ctx == skb 2595 * - since they are wrappers of function calls, they scratch R1-R5 registers, 2596 * preserve R6-R9, and store return value into R0 2597 * 2598 * Implicit input: 2599 * ctx == skb == R6 == CTX 2600 * 2601 * Explicit input: 2602 * SRC == any register 2603 * IMM == 32-bit immediate 2604 * 2605 * Output: 2606 * R0 - 8/16/32-bit skb data converted to cpu endianness 2607 */ 2608 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) 2609 { 2610 struct bpf_reg_state *regs = env->cur_state.regs; 2611 u8 mode = BPF_MODE(insn->code); 2612 int i, err; 2613 2614 if (!may_access_skb(env->prog->type)) { 2615 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); 2616 return -EINVAL; 2617 } 2618 2619 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || 2620 BPF_SIZE(insn->code) == BPF_DW || 2621 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { 2622 verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); 2623 return -EINVAL; 2624 } 2625 2626 /* check whether implicit source operand (register R6) is readable */ 2627 err = check_reg_arg(regs, BPF_REG_6, SRC_OP); 2628 if (err) 2629 return err; 2630 2631 if (regs[BPF_REG_6].type != PTR_TO_CTX) { 2632 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); 2633 return -EINVAL; 2634 } 2635 2636 if (mode == BPF_IND) { 2637 /* check explicit source operand */ 2638 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2639 if (err) 2640 return err; 2641 } 2642 2643 /* reset caller saved regs to unreadable */ 2644 for (i = 0; i < CALLER_SAVED_REGS; i++) 2645 mark_reg_not_init(regs, caller_saved[i]); 2646 2647 /* mark destination R0 register as readable, since it contains 2648 * the value fetched from the packet 2649 */ 2650 regs[BPF_REG_0].type = UNKNOWN_VALUE; 2651 return 0; 2652 } 2653 2654 /* non-recursive DFS pseudo code 2655 * 1 procedure DFS-iterative(G,v): 2656 * 2 label v as discovered 2657 * 3 let S be a stack 2658 * 4 S.push(v) 2659 * 5 while S is not empty 2660 * 6 t <- S.pop() 2661 * 7 if t is what we're looking for: 2662 * 8 return t 2663 * 9 for all edges e in G.adjacentEdges(t) do 2664 * 10 if edge e is already labelled 2665 * 11 continue with the next edge 2666 * 12 w <- G.adjacentVertex(t,e) 2667 * 13 if vertex w is not discovered and not explored 2668 * 14 label e as tree-edge 2669 * 15 label w as discovered 2670 * 16 S.push(w) 2671 * 17 continue at 5 2672 * 18 else if vertex w is discovered 2673 * 19 label e as back-edge 2674 * 20 else 2675 * 21 // vertex w is explored 2676 * 22 label e as forward- or cross-edge 2677 * 23 label t as explored 2678 * 24 S.pop() 2679 * 2680 * convention: 2681 * 0x10 - discovered 2682 * 0x11 - discovered and fall-through edge labelled 2683 * 0x12 - discovered and fall-through and branch edges labelled 2684 * 0x20 - explored 2685 */ 2686 2687 enum { 2688 DISCOVERED = 0x10, 2689 EXPLORED = 0x20, 2690 FALLTHROUGH = 1, 2691 BRANCH = 2, 2692 }; 2693 2694 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) 2695 2696 static int *insn_stack; /* stack of insns to process */ 2697 static int cur_stack; /* current stack index */ 2698 static int *insn_state; 2699 2700 /* t, w, e - match pseudo-code above: 2701 * t - index of current instruction 2702 * w - next instruction 2703 * e - edge 2704 */ 2705 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) 2706 { 2707 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) 2708 return 0; 2709 2710 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) 2711 return 0; 2712 2713 if (w < 0 || w >= env->prog->len) { 2714 verbose("jump out of range from insn %d to %d\n", t, w); 2715 return -EINVAL; 2716 } 2717 2718 if (e == BRANCH) 2719 /* mark branch target for state pruning */ 2720 env->explored_states[w] = STATE_LIST_MARK; 2721 2722 if (insn_state[w] == 0) { 2723 /* tree-edge */ 2724 insn_state[t] = DISCOVERED | e; 2725 insn_state[w] = DISCOVERED; 2726 if (cur_stack >= env->prog->len) 2727 return -E2BIG; 2728 insn_stack[cur_stack++] = w; 2729 return 1; 2730 } else if ((insn_state[w] & 0xF0) == DISCOVERED) { 2731 verbose("back-edge from insn %d to %d\n", t, w); 2732 return -EINVAL; 2733 } else if (insn_state[w] == EXPLORED) { 2734 /* forward- or cross-edge */ 2735 insn_state[t] = DISCOVERED | e; 2736 } else { 2737 verbose("insn state internal bug\n"); 2738 return -EFAULT; 2739 } 2740 return 0; 2741 } 2742 2743 /* non-recursive depth-first-search to detect loops in BPF program 2744 * loop == back-edge in directed graph 2745 */ 2746 static int check_cfg(struct bpf_verifier_env *env) 2747 { 2748 struct bpf_insn *insns = env->prog->insnsi; 2749 int insn_cnt = env->prog->len; 2750 int ret = 0; 2751 int i, t; 2752 2753 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2754 if (!insn_state) 2755 return -ENOMEM; 2756 2757 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2758 if (!insn_stack) { 2759 kfree(insn_state); 2760 return -ENOMEM; 2761 } 2762 2763 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ 2764 insn_stack[0] = 0; /* 0 is the first instruction */ 2765 cur_stack = 1; 2766 2767 peek_stack: 2768 if (cur_stack == 0) 2769 goto check_state; 2770 t = insn_stack[cur_stack - 1]; 2771 2772 if (BPF_CLASS(insns[t].code) == BPF_JMP) { 2773 u8 opcode = BPF_OP(insns[t].code); 2774 2775 if (opcode == BPF_EXIT) { 2776 goto mark_explored; 2777 } else if (opcode == BPF_CALL) { 2778 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2779 if (ret == 1) 2780 goto peek_stack; 2781 else if (ret < 0) 2782 goto err_free; 2783 if (t + 1 < insn_cnt) 2784 env->explored_states[t + 1] = STATE_LIST_MARK; 2785 } else if (opcode == BPF_JA) { 2786 if (BPF_SRC(insns[t].code) != BPF_K) { 2787 ret = -EINVAL; 2788 goto err_free; 2789 } 2790 /* unconditional jump with single edge */ 2791 ret = push_insn(t, t + insns[t].off + 1, 2792 FALLTHROUGH, env); 2793 if (ret == 1) 2794 goto peek_stack; 2795 else if (ret < 0) 2796 goto err_free; 2797 /* tell verifier to check for equivalent states 2798 * after every call and jump 2799 */ 2800 if (t + 1 < insn_cnt) 2801 env->explored_states[t + 1] = STATE_LIST_MARK; 2802 } else { 2803 /* conditional jump with two edges */ 2804 env->explored_states[t] = STATE_LIST_MARK; 2805 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2806 if (ret == 1) 2807 goto peek_stack; 2808 else if (ret < 0) 2809 goto err_free; 2810 2811 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); 2812 if (ret == 1) 2813 goto peek_stack; 2814 else if (ret < 0) 2815 goto err_free; 2816 } 2817 } else { 2818 /* all other non-branch instructions with single 2819 * fall-through edge 2820 */ 2821 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2822 if (ret == 1) 2823 goto peek_stack; 2824 else if (ret < 0) 2825 goto err_free; 2826 } 2827 2828 mark_explored: 2829 insn_state[t] = EXPLORED; 2830 if (cur_stack-- <= 0) { 2831 verbose("pop stack internal bug\n"); 2832 ret = -EFAULT; 2833 goto err_free; 2834 } 2835 goto peek_stack; 2836 2837 check_state: 2838 for (i = 0; i < insn_cnt; i++) { 2839 if (insn_state[i] != EXPLORED) { 2840 verbose("unreachable insn %d\n", i); 2841 ret = -EINVAL; 2842 goto err_free; 2843 } 2844 } 2845 ret = 0; /* cfg looks good */ 2846 2847 err_free: 2848 kfree(insn_state); 2849 kfree(insn_stack); 2850 return ret; 2851 } 2852 2853 /* the following conditions reduce the number of explored insns 2854 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet 2855 */ 2856 static bool compare_ptrs_to_packet(struct bpf_verifier_env *env, 2857 struct bpf_reg_state *old, 2858 struct bpf_reg_state *cur) 2859 { 2860 if (old->id != cur->id) 2861 return false; 2862 2863 /* old ptr_to_packet is more conservative, since it allows smaller 2864 * range. Ex: 2865 * old(off=0,r=10) is equal to cur(off=0,r=20), because 2866 * old(off=0,r=10) means that with range=10 the verifier proceeded 2867 * further and found no issues with the program. Now we're in the same 2868 * spot with cur(off=0,r=20), so we're safe too, since anything further 2869 * will only be looking at most 10 bytes after this pointer. 2870 */ 2871 if (old->off == cur->off && old->range < cur->range) 2872 return true; 2873 2874 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) 2875 * since both cannot be used for packet access and safe(old) 2876 * pointer has smaller off that could be used for further 2877 * 'if (ptr > data_end)' check 2878 * Ex: 2879 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean 2880 * that we cannot access the packet. 2881 * The safe range is: 2882 * [ptr, ptr + range - off) 2883 * so whenever off >=range, it means no safe bytes from this pointer. 2884 * When comparing old->off <= cur->off, it means that older code 2885 * went with smaller offset and that offset was later 2886 * used to figure out the safe range after 'if (ptr > data_end)' check 2887 * Say, 'old' state was explored like: 2888 * ... R3(off=0, r=0) 2889 * R4 = R3 + 20 2890 * ... now R4(off=20,r=0) <-- here 2891 * if (R4 > data_end) 2892 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. 2893 * ... the code further went all the way to bpf_exit. 2894 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). 2895 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier 2896 * goes further, such cur_R4 will give larger safe packet range after 2897 * 'if (R4 > data_end)' and all further insn were already good with r=20, 2898 * so they will be good with r=30 and we can prune the search. 2899 */ 2900 if (!env->strict_alignment && old->off <= cur->off && 2901 old->off >= old->range && cur->off >= cur->range) 2902 return true; 2903 2904 return false; 2905 } 2906 2907 /* compare two verifier states 2908 * 2909 * all states stored in state_list are known to be valid, since 2910 * verifier reached 'bpf_exit' instruction through them 2911 * 2912 * this function is called when verifier exploring different branches of 2913 * execution popped from the state stack. If it sees an old state that has 2914 * more strict register state and more strict stack state then this execution 2915 * branch doesn't need to be explored further, since verifier already 2916 * concluded that more strict state leads to valid finish. 2917 * 2918 * Therefore two states are equivalent if register state is more conservative 2919 * and explored stack state is more conservative than the current one. 2920 * Example: 2921 * explored current 2922 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) 2923 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) 2924 * 2925 * In other words if current stack state (one being explored) has more 2926 * valid slots than old one that already passed validation, it means 2927 * the verifier can stop exploring and conclude that current state is valid too 2928 * 2929 * Similarly with registers. If explored state has register type as invalid 2930 * whereas register type in current state is meaningful, it means that 2931 * the current state will reach 'bpf_exit' instruction safely 2932 */ 2933 static bool states_equal(struct bpf_verifier_env *env, 2934 struct bpf_verifier_state *old, 2935 struct bpf_verifier_state *cur) 2936 { 2937 bool varlen_map_access = env->varlen_map_value_access; 2938 struct bpf_reg_state *rold, *rcur; 2939 int i; 2940 2941 for (i = 0; i < MAX_BPF_REG; i++) { 2942 rold = &old->regs[i]; 2943 rcur = &cur->regs[i]; 2944 2945 if (memcmp(rold, rcur, sizeof(*rold)) == 0) 2946 continue; 2947 2948 /* If the ranges were not the same, but everything else was and 2949 * we didn't do a variable access into a map then we are a-ok. 2950 */ 2951 if (!varlen_map_access && 2952 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0) 2953 continue; 2954 2955 /* If we didn't map access then again we don't care about the 2956 * mismatched range values and it's ok if our old type was 2957 * UNKNOWN and we didn't go to a NOT_INIT'ed reg. 2958 */ 2959 if (rold->type == NOT_INIT || 2960 (!varlen_map_access && rold->type == UNKNOWN_VALUE && 2961 rcur->type != NOT_INIT)) 2962 continue; 2963 2964 /* Don't care about the reg->id in this case. */ 2965 if (rold->type == PTR_TO_MAP_VALUE_OR_NULL && 2966 rcur->type == PTR_TO_MAP_VALUE_OR_NULL && 2967 rold->map_ptr == rcur->map_ptr) 2968 continue; 2969 2970 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && 2971 compare_ptrs_to_packet(env, rold, rcur)) 2972 continue; 2973 2974 return false; 2975 } 2976 2977 for (i = 0; i < MAX_BPF_STACK; i++) { 2978 if (old->stack_slot_type[i] == STACK_INVALID) 2979 continue; 2980 if (old->stack_slot_type[i] != cur->stack_slot_type[i]) 2981 /* Ex: old explored (safe) state has STACK_SPILL in 2982 * this stack slot, but current has has STACK_MISC -> 2983 * this verifier states are not equivalent, 2984 * return false to continue verification of this path 2985 */ 2986 return false; 2987 if (i % BPF_REG_SIZE) 2988 continue; 2989 if (old->stack_slot_type[i] != STACK_SPILL) 2990 continue; 2991 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], 2992 &cur->spilled_regs[i / BPF_REG_SIZE], 2993 sizeof(old->spilled_regs[0]))) 2994 /* when explored and current stack slot types are 2995 * the same, check that stored pointers types 2996 * are the same as well. 2997 * Ex: explored safe path could have stored 2998 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} 2999 * but current path has stored: 3000 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} 3001 * such verifier states are not equivalent. 3002 * return false to continue verification of this path 3003 */ 3004 return false; 3005 else 3006 continue; 3007 } 3008 return true; 3009 } 3010 3011 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) 3012 { 3013 struct bpf_verifier_state_list *new_sl; 3014 struct bpf_verifier_state_list *sl; 3015 3016 sl = env->explored_states[insn_idx]; 3017 if (!sl) 3018 /* this 'insn_idx' instruction wasn't marked, so we will not 3019 * be doing state search here 3020 */ 3021 return 0; 3022 3023 while (sl != STATE_LIST_MARK) { 3024 if (states_equal(env, &sl->state, &env->cur_state)) 3025 /* reached equivalent register/stack state, 3026 * prune the search 3027 */ 3028 return 1; 3029 sl = sl->next; 3030 } 3031 3032 /* there were no equivalent states, remember current one. 3033 * technically the current state is not proven to be safe yet, 3034 * but it will either reach bpf_exit (which means it's safe) or 3035 * it will be rejected. Since there are no loops, we won't be 3036 * seeing this 'insn_idx' instruction again on the way to bpf_exit 3037 */ 3038 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); 3039 if (!new_sl) 3040 return -ENOMEM; 3041 3042 /* add new state to the head of linked list */ 3043 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); 3044 new_sl->next = env->explored_states[insn_idx]; 3045 env->explored_states[insn_idx] = new_sl; 3046 return 0; 3047 } 3048 3049 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, 3050 int insn_idx, int prev_insn_idx) 3051 { 3052 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) 3053 return 0; 3054 3055 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); 3056 } 3057 3058 static int do_check(struct bpf_verifier_env *env) 3059 { 3060 struct bpf_verifier_state *state = &env->cur_state; 3061 struct bpf_insn *insns = env->prog->insnsi; 3062 struct bpf_reg_state *regs = state->regs; 3063 int insn_cnt = env->prog->len; 3064 int insn_idx, prev_insn_idx = 0; 3065 int insn_processed = 0; 3066 bool do_print_state = false; 3067 3068 init_reg_state(regs); 3069 insn_idx = 0; 3070 env->varlen_map_value_access = false; 3071 for (;;) { 3072 struct bpf_insn *insn; 3073 u8 class; 3074 int err; 3075 3076 if (insn_idx >= insn_cnt) { 3077 verbose("invalid insn idx %d insn_cnt %d\n", 3078 insn_idx, insn_cnt); 3079 return -EFAULT; 3080 } 3081 3082 insn = &insns[insn_idx]; 3083 class = BPF_CLASS(insn->code); 3084 3085 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { 3086 verbose("BPF program is too large. Processed %d insn\n", 3087 insn_processed); 3088 return -E2BIG; 3089 } 3090 3091 err = is_state_visited(env, insn_idx); 3092 if (err < 0) 3093 return err; 3094 if (err == 1) { 3095 /* found equivalent state, can prune the search */ 3096 if (log_level) { 3097 if (do_print_state) 3098 verbose("\nfrom %d to %d: safe\n", 3099 prev_insn_idx, insn_idx); 3100 else 3101 verbose("%d: safe\n", insn_idx); 3102 } 3103 goto process_bpf_exit; 3104 } 3105 3106 if (need_resched()) 3107 cond_resched(); 3108 3109 if (log_level > 1 || (log_level && do_print_state)) { 3110 if (log_level > 1) 3111 verbose("%d:", insn_idx); 3112 else 3113 verbose("\nfrom %d to %d:", 3114 prev_insn_idx, insn_idx); 3115 print_verifier_state(&env->cur_state); 3116 do_print_state = false; 3117 } 3118 3119 if (log_level) { 3120 verbose("%d: ", insn_idx); 3121 print_bpf_insn(env, insn); 3122 } 3123 3124 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); 3125 if (err) 3126 return err; 3127 3128 if (class == BPF_ALU || class == BPF_ALU64) { 3129 err = check_alu_op(env, insn); 3130 if (err) 3131 return err; 3132 3133 } else if (class == BPF_LDX) { 3134 enum bpf_reg_type *prev_src_type, src_reg_type; 3135 3136 /* check for reserved fields is already done */ 3137 3138 /* check src operand */ 3139 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 3140 if (err) 3141 return err; 3142 3143 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 3144 if (err) 3145 return err; 3146 3147 src_reg_type = regs[insn->src_reg].type; 3148 3149 /* check that memory (src_reg + off) is readable, 3150 * the state of dst_reg will be updated by this func 3151 */ 3152 err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, 3153 BPF_SIZE(insn->code), BPF_READ, 3154 insn->dst_reg); 3155 if (err) 3156 return err; 3157 3158 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; 3159 3160 if (*prev_src_type == NOT_INIT) { 3161 /* saw a valid insn 3162 * dst_reg = *(u32 *)(src_reg + off) 3163 * save type to validate intersecting paths 3164 */ 3165 *prev_src_type = src_reg_type; 3166 3167 } else if (src_reg_type != *prev_src_type && 3168 (src_reg_type == PTR_TO_CTX || 3169 *prev_src_type == PTR_TO_CTX)) { 3170 /* ABuser program is trying to use the same insn 3171 * dst_reg = *(u32*) (src_reg + off) 3172 * with different pointer types: 3173 * src_reg == ctx in one branch and 3174 * src_reg == stack|map in some other branch. 3175 * Reject it. 3176 */ 3177 verbose("same insn cannot be used with different pointers\n"); 3178 return -EINVAL; 3179 } 3180 3181 } else if (class == BPF_STX) { 3182 enum bpf_reg_type *prev_dst_type, dst_reg_type; 3183 3184 if (BPF_MODE(insn->code) == BPF_XADD) { 3185 err = check_xadd(env, insn_idx, insn); 3186 if (err) 3187 return err; 3188 insn_idx++; 3189 continue; 3190 } 3191 3192 /* check src1 operand */ 3193 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 3194 if (err) 3195 return err; 3196 /* check src2 operand */ 3197 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 3198 if (err) 3199 return err; 3200 3201 dst_reg_type = regs[insn->dst_reg].type; 3202 3203 /* check that memory (dst_reg + off) is writeable */ 3204 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 3205 BPF_SIZE(insn->code), BPF_WRITE, 3206 insn->src_reg); 3207 if (err) 3208 return err; 3209 3210 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; 3211 3212 if (*prev_dst_type == NOT_INIT) { 3213 *prev_dst_type = dst_reg_type; 3214 } else if (dst_reg_type != *prev_dst_type && 3215 (dst_reg_type == PTR_TO_CTX || 3216 *prev_dst_type == PTR_TO_CTX)) { 3217 verbose("same insn cannot be used with different pointers\n"); 3218 return -EINVAL; 3219 } 3220 3221 } else if (class == BPF_ST) { 3222 if (BPF_MODE(insn->code) != BPF_MEM || 3223 insn->src_reg != BPF_REG_0) { 3224 verbose("BPF_ST uses reserved fields\n"); 3225 return -EINVAL; 3226 } 3227 /* check src operand */ 3228 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 3229 if (err) 3230 return err; 3231 3232 /* check that memory (dst_reg + off) is writeable */ 3233 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, 3234 BPF_SIZE(insn->code), BPF_WRITE, 3235 -1); 3236 if (err) 3237 return err; 3238 3239 } else if (class == BPF_JMP) { 3240 u8 opcode = BPF_OP(insn->code); 3241 3242 if (opcode == BPF_CALL) { 3243 if (BPF_SRC(insn->code) != BPF_K || 3244 insn->off != 0 || 3245 insn->src_reg != BPF_REG_0 || 3246 insn->dst_reg != BPF_REG_0) { 3247 verbose("BPF_CALL uses reserved fields\n"); 3248 return -EINVAL; 3249 } 3250 3251 err = check_call(env, insn->imm, insn_idx); 3252 if (err) 3253 return err; 3254 3255 } else if (opcode == BPF_JA) { 3256 if (BPF_SRC(insn->code) != BPF_K || 3257 insn->imm != 0 || 3258 insn->src_reg != BPF_REG_0 || 3259 insn->dst_reg != BPF_REG_0) { 3260 verbose("BPF_JA uses reserved fields\n"); 3261 return -EINVAL; 3262 } 3263 3264 insn_idx += insn->off + 1; 3265 continue; 3266 3267 } else if (opcode == BPF_EXIT) { 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_EXIT uses reserved fields\n"); 3273 return -EINVAL; 3274 } 3275 3276 /* eBPF calling convetion is such that R0 is used 3277 * to return the value from eBPF program. 3278 * Make sure that it's readable at this time 3279 * of bpf_exit, which means that program wrote 3280 * something into it earlier 3281 */ 3282 err = check_reg_arg(regs, BPF_REG_0, SRC_OP); 3283 if (err) 3284 return err; 3285 3286 if (is_pointer_value(env, BPF_REG_0)) { 3287 verbose("R0 leaks addr as return value\n"); 3288 return -EACCES; 3289 } 3290 3291 process_bpf_exit: 3292 insn_idx = pop_stack(env, &prev_insn_idx); 3293 if (insn_idx < 0) { 3294 break; 3295 } else { 3296 do_print_state = true; 3297 continue; 3298 } 3299 } else { 3300 err = check_cond_jmp_op(env, insn, &insn_idx); 3301 if (err) 3302 return err; 3303 } 3304 } else if (class == BPF_LD) { 3305 u8 mode = BPF_MODE(insn->code); 3306 3307 if (mode == BPF_ABS || mode == BPF_IND) { 3308 err = check_ld_abs(env, insn); 3309 if (err) 3310 return err; 3311 3312 } else if (mode == BPF_IMM) { 3313 err = check_ld_imm(env, insn); 3314 if (err) 3315 return err; 3316 3317 insn_idx++; 3318 } else { 3319 verbose("invalid BPF_LD mode\n"); 3320 return -EINVAL; 3321 } 3322 reset_reg_range_values(regs, insn->dst_reg); 3323 } else { 3324 verbose("unknown insn class %d\n", class); 3325 return -EINVAL; 3326 } 3327 3328 insn_idx++; 3329 } 3330 3331 verbose("processed %d insns, stack depth %d\n", 3332 insn_processed, env->prog->aux->stack_depth); 3333 return 0; 3334 } 3335 3336 static int check_map_prealloc(struct bpf_map *map) 3337 { 3338 return (map->map_type != BPF_MAP_TYPE_HASH && 3339 map->map_type != BPF_MAP_TYPE_PERCPU_HASH && 3340 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || 3341 !(map->map_flags & BPF_F_NO_PREALLOC); 3342 } 3343 3344 static int check_map_prog_compatibility(struct bpf_map *map, 3345 struct bpf_prog *prog) 3346 3347 { 3348 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use 3349 * preallocated hash maps, since doing memory allocation 3350 * in overflow_handler can crash depending on where nmi got 3351 * triggered. 3352 */ 3353 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { 3354 if (!check_map_prealloc(map)) { 3355 verbose("perf_event programs can only use preallocated hash map\n"); 3356 return -EINVAL; 3357 } 3358 if (map->inner_map_meta && 3359 !check_map_prealloc(map->inner_map_meta)) { 3360 verbose("perf_event programs can only use preallocated inner hash map\n"); 3361 return -EINVAL; 3362 } 3363 } 3364 return 0; 3365 } 3366 3367 /* look for pseudo eBPF instructions that access map FDs and 3368 * replace them with actual map pointers 3369 */ 3370 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) 3371 { 3372 struct bpf_insn *insn = env->prog->insnsi; 3373 int insn_cnt = env->prog->len; 3374 int i, j, err; 3375 3376 err = bpf_prog_calc_tag(env->prog); 3377 if (err) 3378 return err; 3379 3380 for (i = 0; i < insn_cnt; i++, insn++) { 3381 if (BPF_CLASS(insn->code) == BPF_LDX && 3382 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { 3383 verbose("BPF_LDX uses reserved fields\n"); 3384 return -EINVAL; 3385 } 3386 3387 if (BPF_CLASS(insn->code) == BPF_STX && 3388 ((BPF_MODE(insn->code) != BPF_MEM && 3389 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { 3390 verbose("BPF_STX uses reserved fields\n"); 3391 return -EINVAL; 3392 } 3393 3394 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { 3395 struct bpf_map *map; 3396 struct fd f; 3397 3398 if (i == insn_cnt - 1 || insn[1].code != 0 || 3399 insn[1].dst_reg != 0 || insn[1].src_reg != 0 || 3400 insn[1].off != 0) { 3401 verbose("invalid bpf_ld_imm64 insn\n"); 3402 return -EINVAL; 3403 } 3404 3405 if (insn->src_reg == 0) 3406 /* valid generic load 64-bit imm */ 3407 goto next_insn; 3408 3409 if (insn->src_reg != BPF_PSEUDO_MAP_FD) { 3410 verbose("unrecognized bpf_ld_imm64 insn\n"); 3411 return -EINVAL; 3412 } 3413 3414 f = fdget(insn->imm); 3415 map = __bpf_map_get(f); 3416 if (IS_ERR(map)) { 3417 verbose("fd %d is not pointing to valid bpf_map\n", 3418 insn->imm); 3419 return PTR_ERR(map); 3420 } 3421 3422 err = check_map_prog_compatibility(map, env->prog); 3423 if (err) { 3424 fdput(f); 3425 return err; 3426 } 3427 3428 /* store map pointer inside BPF_LD_IMM64 instruction */ 3429 insn[0].imm = (u32) (unsigned long) map; 3430 insn[1].imm = ((u64) (unsigned long) map) >> 32; 3431 3432 /* check whether we recorded this map already */ 3433 for (j = 0; j < env->used_map_cnt; j++) 3434 if (env->used_maps[j] == map) { 3435 fdput(f); 3436 goto next_insn; 3437 } 3438 3439 if (env->used_map_cnt >= MAX_USED_MAPS) { 3440 fdput(f); 3441 return -E2BIG; 3442 } 3443 3444 /* hold the map. If the program is rejected by verifier, 3445 * the map will be released by release_maps() or it 3446 * will be used by the valid program until it's unloaded 3447 * and all maps are released in free_bpf_prog_info() 3448 */ 3449 map = bpf_map_inc(map, false); 3450 if (IS_ERR(map)) { 3451 fdput(f); 3452 return PTR_ERR(map); 3453 } 3454 env->used_maps[env->used_map_cnt++] = map; 3455 3456 fdput(f); 3457 next_insn: 3458 insn++; 3459 i++; 3460 } 3461 } 3462 3463 /* now all pseudo BPF_LD_IMM64 instructions load valid 3464 * 'struct bpf_map *' into a register instead of user map_fd. 3465 * These pointers will be used later by verifier to validate map access. 3466 */ 3467 return 0; 3468 } 3469 3470 /* drop refcnt of maps used by the rejected program */ 3471 static void release_maps(struct bpf_verifier_env *env) 3472 { 3473 int i; 3474 3475 for (i = 0; i < env->used_map_cnt; i++) 3476 bpf_map_put(env->used_maps[i]); 3477 } 3478 3479 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ 3480 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) 3481 { 3482 struct bpf_insn *insn = env->prog->insnsi; 3483 int insn_cnt = env->prog->len; 3484 int i; 3485 3486 for (i = 0; i < insn_cnt; i++, insn++) 3487 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) 3488 insn->src_reg = 0; 3489 } 3490 3491 /* single env->prog->insni[off] instruction was replaced with the range 3492 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying 3493 * [0, off) and [off, end) to new locations, so the patched range stays zero 3494 */ 3495 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, 3496 u32 off, u32 cnt) 3497 { 3498 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; 3499 3500 if (cnt == 1) 3501 return 0; 3502 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); 3503 if (!new_data) 3504 return -ENOMEM; 3505 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); 3506 memcpy(new_data + off + cnt - 1, old_data + off, 3507 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); 3508 env->insn_aux_data = new_data; 3509 vfree(old_data); 3510 return 0; 3511 } 3512 3513 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, 3514 const struct bpf_insn *patch, u32 len) 3515 { 3516 struct bpf_prog *new_prog; 3517 3518 new_prog = bpf_patch_insn_single(env->prog, off, patch, len); 3519 if (!new_prog) 3520 return NULL; 3521 if (adjust_insn_aux_data(env, new_prog->len, off, len)) 3522 return NULL; 3523 return new_prog; 3524 } 3525 3526 /* convert load instructions that access fields of 'struct __sk_buff' 3527 * into sequence of instructions that access fields of 'struct sk_buff' 3528 */ 3529 static int convert_ctx_accesses(struct bpf_verifier_env *env) 3530 { 3531 const struct bpf_verifier_ops *ops = env->prog->aux->ops; 3532 int i, cnt, size, ctx_field_size, delta = 0; 3533 const int insn_cnt = env->prog->len; 3534 struct bpf_insn insn_buf[16], *insn; 3535 struct bpf_prog *new_prog; 3536 enum bpf_access_type type; 3537 bool is_narrower_load; 3538 u32 target_size; 3539 3540 if (ops->gen_prologue) { 3541 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, 3542 env->prog); 3543 if (cnt >= ARRAY_SIZE(insn_buf)) { 3544 verbose("bpf verifier is misconfigured\n"); 3545 return -EINVAL; 3546 } else if (cnt) { 3547 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); 3548 if (!new_prog) 3549 return -ENOMEM; 3550 3551 env->prog = new_prog; 3552 delta += cnt - 1; 3553 } 3554 } 3555 3556 if (!ops->convert_ctx_access) 3557 return 0; 3558 3559 insn = env->prog->insnsi + delta; 3560 3561 for (i = 0; i < insn_cnt; i++, insn++) { 3562 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || 3563 insn->code == (BPF_LDX | BPF_MEM | BPF_H) || 3564 insn->code == (BPF_LDX | BPF_MEM | BPF_W) || 3565 insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) 3566 type = BPF_READ; 3567 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || 3568 insn->code == (BPF_STX | BPF_MEM | BPF_H) || 3569 insn->code == (BPF_STX | BPF_MEM | BPF_W) || 3570 insn->code == (BPF_STX | BPF_MEM | BPF_DW)) 3571 type = BPF_WRITE; 3572 else 3573 continue; 3574 3575 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) 3576 continue; 3577 3578 ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; 3579 size = BPF_LDST_BYTES(insn); 3580 3581 /* If the read access is a narrower load of the field, 3582 * convert to a 4/8-byte load, to minimum program type specific 3583 * convert_ctx_access changes. If conversion is successful, 3584 * we will apply proper mask to the result. 3585 */ 3586 is_narrower_load = size < ctx_field_size; 3587 if (is_narrower_load) { 3588 u32 off = insn->off; 3589 u8 size_code; 3590 3591 if (type == BPF_WRITE) { 3592 verbose("bpf verifier narrow ctx access misconfigured\n"); 3593 return -EINVAL; 3594 } 3595 3596 size_code = BPF_H; 3597 if (ctx_field_size == 4) 3598 size_code = BPF_W; 3599 else if (ctx_field_size == 8) 3600 size_code = BPF_DW; 3601 3602 insn->off = off & ~(ctx_field_size - 1); 3603 insn->code = BPF_LDX | BPF_MEM | size_code; 3604 } 3605 3606 target_size = 0; 3607 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, 3608 &target_size); 3609 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || 3610 (ctx_field_size && !target_size)) { 3611 verbose("bpf verifier is misconfigured\n"); 3612 return -EINVAL; 3613 } 3614 3615 if (is_narrower_load && size < target_size) { 3616 if (ctx_field_size <= 4) 3617 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, 3618 (1 << size * 8) - 1); 3619 else 3620 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, 3621 (1 << size * 8) - 1); 3622 } 3623 3624 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); 3625 if (!new_prog) 3626 return -ENOMEM; 3627 3628 delta += cnt - 1; 3629 3630 /* keep walking new program and skip insns we just inserted */ 3631 env->prog = new_prog; 3632 insn = new_prog->insnsi + i + delta; 3633 } 3634 3635 return 0; 3636 } 3637 3638 /* fixup insn->imm field of bpf_call instructions 3639 * and inline eligible helpers as explicit sequence of BPF instructions 3640 * 3641 * this function is called after eBPF program passed verification 3642 */ 3643 static int fixup_bpf_calls(struct bpf_verifier_env *env) 3644 { 3645 struct bpf_prog *prog = env->prog; 3646 struct bpf_insn *insn = prog->insnsi; 3647 const struct bpf_func_proto *fn; 3648 const int insn_cnt = prog->len; 3649 struct bpf_insn insn_buf[16]; 3650 struct bpf_prog *new_prog; 3651 struct bpf_map *map_ptr; 3652 int i, cnt, delta = 0; 3653 3654 for (i = 0; i < insn_cnt; i++, insn++) { 3655 if (insn->code != (BPF_JMP | BPF_CALL)) 3656 continue; 3657 3658 if (insn->imm == BPF_FUNC_get_route_realm) 3659 prog->dst_needed = 1; 3660 if (insn->imm == BPF_FUNC_get_prandom_u32) 3661 bpf_user_rnd_init_once(); 3662 if (insn->imm == BPF_FUNC_tail_call) { 3663 /* If we tail call into other programs, we 3664 * cannot make any assumptions since they can 3665 * be replaced dynamically during runtime in 3666 * the program array. 3667 */ 3668 prog->cb_access = 1; 3669 env->prog->aux->stack_depth = MAX_BPF_STACK; 3670 3671 /* mark bpf_tail_call as different opcode to avoid 3672 * conditional branch in the interpeter for every normal 3673 * call and to prevent accidental JITing by JIT compiler 3674 * that doesn't support bpf_tail_call yet 3675 */ 3676 insn->imm = 0; 3677 insn->code = BPF_JMP | BPF_TAIL_CALL; 3678 continue; 3679 } 3680 3681 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) { 3682 map_ptr = env->insn_aux_data[i + delta].map_ptr; 3683 if (map_ptr == BPF_MAP_PTR_POISON || 3684 !map_ptr->ops->map_gen_lookup) 3685 goto patch_call_imm; 3686 3687 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); 3688 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { 3689 verbose("bpf verifier is misconfigured\n"); 3690 return -EINVAL; 3691 } 3692 3693 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 3694 cnt); 3695 if (!new_prog) 3696 return -ENOMEM; 3697 3698 delta += cnt - 1; 3699 3700 /* keep walking new program and skip insns we just inserted */ 3701 env->prog = prog = new_prog; 3702 insn = new_prog->insnsi + i + delta; 3703 continue; 3704 } 3705 3706 patch_call_imm: 3707 fn = prog->aux->ops->get_func_proto(insn->imm); 3708 /* all functions that have prototype and verifier allowed 3709 * programs to call them, must be real in-kernel functions 3710 */ 3711 if (!fn->func) { 3712 verbose("kernel subsystem misconfigured func %s#%d\n", 3713 func_id_name(insn->imm), insn->imm); 3714 return -EFAULT; 3715 } 3716 insn->imm = fn->func - __bpf_call_base; 3717 } 3718 3719 return 0; 3720 } 3721 3722 static void free_states(struct bpf_verifier_env *env) 3723 { 3724 struct bpf_verifier_state_list *sl, *sln; 3725 int i; 3726 3727 if (!env->explored_states) 3728 return; 3729 3730 for (i = 0; i < env->prog->len; i++) { 3731 sl = env->explored_states[i]; 3732 3733 if (sl) 3734 while (sl != STATE_LIST_MARK) { 3735 sln = sl->next; 3736 kfree(sl); 3737 sl = sln; 3738 } 3739 } 3740 3741 kfree(env->explored_states); 3742 } 3743 3744 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) 3745 { 3746 char __user *log_ubuf = NULL; 3747 struct bpf_verifier_env *env; 3748 int ret = -EINVAL; 3749 3750 /* 'struct bpf_verifier_env' can be global, but since it's not small, 3751 * allocate/free it every time bpf_check() is called 3752 */ 3753 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3754 if (!env) 3755 return -ENOMEM; 3756 3757 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3758 (*prog)->len); 3759 ret = -ENOMEM; 3760 if (!env->insn_aux_data) 3761 goto err_free_env; 3762 env->prog = *prog; 3763 3764 /* grab the mutex to protect few globals used by verifier */ 3765 mutex_lock(&bpf_verifier_lock); 3766 3767 if (attr->log_level || attr->log_buf || attr->log_size) { 3768 /* user requested verbose verifier output 3769 * and supplied buffer to store the verification trace 3770 */ 3771 log_level = attr->log_level; 3772 log_ubuf = (char __user *) (unsigned long) attr->log_buf; 3773 log_size = attr->log_size; 3774 log_len = 0; 3775 3776 ret = -EINVAL; 3777 /* log_* values have to be sane */ 3778 if (log_size < 128 || log_size > UINT_MAX >> 8 || 3779 log_level == 0 || log_ubuf == NULL) 3780 goto err_unlock; 3781 3782 ret = -ENOMEM; 3783 log_buf = vmalloc(log_size); 3784 if (!log_buf) 3785 goto err_unlock; 3786 } else { 3787 log_level = 0; 3788 } 3789 3790 env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); 3791 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 3792 env->strict_alignment = true; 3793 3794 ret = replace_map_fd_with_map_ptr(env); 3795 if (ret < 0) 3796 goto skip_full_check; 3797 3798 env->explored_states = kcalloc(env->prog->len, 3799 sizeof(struct bpf_verifier_state_list *), 3800 GFP_USER); 3801 ret = -ENOMEM; 3802 if (!env->explored_states) 3803 goto skip_full_check; 3804 3805 ret = check_cfg(env); 3806 if (ret < 0) 3807 goto skip_full_check; 3808 3809 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3810 3811 ret = do_check(env); 3812 3813 skip_full_check: 3814 while (pop_stack(env, NULL) >= 0); 3815 free_states(env); 3816 3817 if (ret == 0) 3818 /* program is valid, convert *(u32*)(ctx + off) accesses */ 3819 ret = convert_ctx_accesses(env); 3820 3821 if (ret == 0) 3822 ret = fixup_bpf_calls(env); 3823 3824 if (log_level && log_len >= log_size - 1) { 3825 BUG_ON(log_len >= log_size); 3826 /* verifier log exceeded user supplied buffer */ 3827 ret = -ENOSPC; 3828 /* fall through to return what was recorded */ 3829 } 3830 3831 /* copy verifier log back to user space including trailing zero */ 3832 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { 3833 ret = -EFAULT; 3834 goto free_log_buf; 3835 } 3836 3837 if (ret == 0 && env->used_map_cnt) { 3838 /* if program passed verifier, update used_maps in bpf_prog_info */ 3839 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, 3840 sizeof(env->used_maps[0]), 3841 GFP_KERNEL); 3842 3843 if (!env->prog->aux->used_maps) { 3844 ret = -ENOMEM; 3845 goto free_log_buf; 3846 } 3847 3848 memcpy(env->prog->aux->used_maps, env->used_maps, 3849 sizeof(env->used_maps[0]) * env->used_map_cnt); 3850 env->prog->aux->used_map_cnt = env->used_map_cnt; 3851 3852 /* program is valid. Convert pseudo bpf_ld_imm64 into generic 3853 * bpf_ld_imm64 instructions 3854 */ 3855 convert_pseudo_ld_imm64(env); 3856 } 3857 3858 free_log_buf: 3859 if (log_level) 3860 vfree(log_buf); 3861 if (!env->prog->aux->used_maps) 3862 /* if we didn't copy map pointers into bpf_prog_info, release 3863 * them now. Otherwise free_bpf_prog_info() will release them. 3864 */ 3865 release_maps(env); 3866 *prog = env->prog; 3867 err_unlock: 3868 mutex_unlock(&bpf_verifier_lock); 3869 vfree(env->insn_aux_data); 3870 err_free_env: 3871 kfree(env); 3872 return ret; 3873 } 3874 3875 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, 3876 void *priv) 3877 { 3878 struct bpf_verifier_env *env; 3879 int ret; 3880 3881 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3882 if (!env) 3883 return -ENOMEM; 3884 3885 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3886 prog->len); 3887 ret = -ENOMEM; 3888 if (!env->insn_aux_data) 3889 goto err_free_env; 3890 env->prog = prog; 3891 env->analyzer_ops = ops; 3892 env->analyzer_priv = priv; 3893 3894 /* grab the mutex to protect few globals used by verifier */ 3895 mutex_lock(&bpf_verifier_lock); 3896 3897 log_level = 0; 3898 3899 env->strict_alignment = false; 3900 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 3901 env->strict_alignment = true; 3902 3903 env->explored_states = kcalloc(env->prog->len, 3904 sizeof(struct bpf_verifier_state_list *), 3905 GFP_KERNEL); 3906 ret = -ENOMEM; 3907 if (!env->explored_states) 3908 goto skip_full_check; 3909 3910 ret = check_cfg(env); 3911 if (ret < 0) 3912 goto skip_full_check; 3913 3914 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3915 3916 ret = do_check(env); 3917 3918 skip_full_check: 3919 while (pop_stack(env, NULL) >= 0); 3920 free_states(env); 3921 3922 mutex_unlock(&bpf_verifier_lock); 3923 vfree(env->insn_aux_data); 3924 err_free_env: 3925 kfree(env); 3926 return ret; 3927 } 3928 EXPORT_SYMBOL_GPL(bpf_analyzer); 3929