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 65536 144 #define BPF_COMPLEXITY_LIMIT_STACK 1024 145 146 struct bpf_call_arg_meta { 147 struct bpf_map *map_ptr; 148 bool raw_mode; 149 bool pkt_access; 150 int regno; 151 int access_size; 152 }; 153 154 /* verbose verifier prints what it's seeing 155 * bpf_check() is called under lock, so no race to access these global vars 156 */ 157 static u32 log_level, log_size, log_len; 158 static char *log_buf; 159 160 static DEFINE_MUTEX(bpf_verifier_lock); 161 162 /* log_level controls verbosity level of eBPF verifier. 163 * verbose() is used to dump the verification trace to the log, so the user 164 * can figure out what's wrong with the program 165 */ 166 static __printf(1, 2) void verbose(const char *fmt, ...) 167 { 168 va_list args; 169 170 if (log_level == 0 || log_len >= log_size - 1) 171 return; 172 173 va_start(args, fmt); 174 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); 175 va_end(args); 176 } 177 178 /* string representation of 'enum bpf_reg_type' */ 179 static const char * const reg_type_str[] = { 180 [NOT_INIT] = "?", 181 [UNKNOWN_VALUE] = "inv", 182 [PTR_TO_CTX] = "ctx", 183 [CONST_PTR_TO_MAP] = "map_ptr", 184 [PTR_TO_MAP_VALUE] = "map_value", 185 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", 186 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj", 187 [FRAME_PTR] = "fp", 188 [PTR_TO_STACK] = "fp", 189 [CONST_IMM] = "imm", 190 [PTR_TO_PACKET] = "pkt", 191 [PTR_TO_PACKET_END] = "pkt_end", 192 }; 193 194 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x) 195 static const char * const func_id_str[] = { 196 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN) 197 }; 198 #undef __BPF_FUNC_STR_FN 199 200 static const char *func_id_name(int id) 201 { 202 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID); 203 204 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id]) 205 return func_id_str[id]; 206 else 207 return "unknown"; 208 } 209 210 static void print_verifier_state(struct bpf_verifier_state *state) 211 { 212 struct bpf_reg_state *reg; 213 enum bpf_reg_type t; 214 int i; 215 216 for (i = 0; i < MAX_BPF_REG; i++) { 217 reg = &state->regs[i]; 218 t = reg->type; 219 if (t == NOT_INIT) 220 continue; 221 verbose(" R%d=%s", i, reg_type_str[t]); 222 if (t == CONST_IMM || t == PTR_TO_STACK) 223 verbose("%lld", reg->imm); 224 else if (t == PTR_TO_PACKET) 225 verbose("(id=%d,off=%d,r=%d)", 226 reg->id, reg->off, reg->range); 227 else if (t == UNKNOWN_VALUE && reg->imm) 228 verbose("%lld", reg->imm); 229 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || 230 t == PTR_TO_MAP_VALUE_OR_NULL || 231 t == PTR_TO_MAP_VALUE_ADJ) 232 verbose("(ks=%d,vs=%d,id=%u)", 233 reg->map_ptr->key_size, 234 reg->map_ptr->value_size, 235 reg->id); 236 if (reg->min_value != BPF_REGISTER_MIN_RANGE) 237 verbose(",min_value=%lld", 238 (long long)reg->min_value); 239 if (reg->max_value != BPF_REGISTER_MAX_RANGE) 240 verbose(",max_value=%llu", 241 (unsigned long long)reg->max_value); 242 } 243 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 244 if (state->stack_slot_type[i] == STACK_SPILL) 245 verbose(" fp%d=%s", -MAX_BPF_STACK + i, 246 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); 247 } 248 verbose("\n"); 249 } 250 251 static const char *const bpf_class_string[] = { 252 [BPF_LD] = "ld", 253 [BPF_LDX] = "ldx", 254 [BPF_ST] = "st", 255 [BPF_STX] = "stx", 256 [BPF_ALU] = "alu", 257 [BPF_JMP] = "jmp", 258 [BPF_RET] = "BUG", 259 [BPF_ALU64] = "alu64", 260 }; 261 262 static const char *const bpf_alu_string[16] = { 263 [BPF_ADD >> 4] = "+=", 264 [BPF_SUB >> 4] = "-=", 265 [BPF_MUL >> 4] = "*=", 266 [BPF_DIV >> 4] = "/=", 267 [BPF_OR >> 4] = "|=", 268 [BPF_AND >> 4] = "&=", 269 [BPF_LSH >> 4] = "<<=", 270 [BPF_RSH >> 4] = ">>=", 271 [BPF_NEG >> 4] = "neg", 272 [BPF_MOD >> 4] = "%=", 273 [BPF_XOR >> 4] = "^=", 274 [BPF_MOV >> 4] = "=", 275 [BPF_ARSH >> 4] = "s>>=", 276 [BPF_END >> 4] = "endian", 277 }; 278 279 static const char *const bpf_ldst_string[] = { 280 [BPF_W >> 3] = "u32", 281 [BPF_H >> 3] = "u16", 282 [BPF_B >> 3] = "u8", 283 [BPF_DW >> 3] = "u64", 284 }; 285 286 static const char *const bpf_jmp_string[16] = { 287 [BPF_JA >> 4] = "jmp", 288 [BPF_JEQ >> 4] = "==", 289 [BPF_JGT >> 4] = ">", 290 [BPF_JGE >> 4] = ">=", 291 [BPF_JSET >> 4] = "&", 292 [BPF_JNE >> 4] = "!=", 293 [BPF_JSGT >> 4] = "s>", 294 [BPF_JSGE >> 4] = "s>=", 295 [BPF_CALL >> 4] = "call", 296 [BPF_EXIT >> 4] = "exit", 297 }; 298 299 static void print_bpf_insn(struct bpf_insn *insn) 300 { 301 u8 class = BPF_CLASS(insn->code); 302 303 if (class == BPF_ALU || class == BPF_ALU64) { 304 if (BPF_SRC(insn->code) == BPF_X) 305 verbose("(%02x) %sr%d %s %sr%d\n", 306 insn->code, class == BPF_ALU ? "(u32) " : "", 307 insn->dst_reg, 308 bpf_alu_string[BPF_OP(insn->code) >> 4], 309 class == BPF_ALU ? "(u32) " : "", 310 insn->src_reg); 311 else 312 verbose("(%02x) %sr%d %s %s%d\n", 313 insn->code, class == BPF_ALU ? "(u32) " : "", 314 insn->dst_reg, 315 bpf_alu_string[BPF_OP(insn->code) >> 4], 316 class == BPF_ALU ? "(u32) " : "", 317 insn->imm); 318 } else if (class == BPF_STX) { 319 if (BPF_MODE(insn->code) == BPF_MEM) 320 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", 321 insn->code, 322 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 323 insn->dst_reg, 324 insn->off, insn->src_reg); 325 else if (BPF_MODE(insn->code) == BPF_XADD) 326 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", 327 insn->code, 328 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 329 insn->dst_reg, insn->off, 330 insn->src_reg); 331 else 332 verbose("BUG_%02x\n", insn->code); 333 } else if (class == BPF_ST) { 334 if (BPF_MODE(insn->code) != BPF_MEM) { 335 verbose("BUG_st_%02x\n", insn->code); 336 return; 337 } 338 verbose("(%02x) *(%s *)(r%d %+d) = %d\n", 339 insn->code, 340 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 341 insn->dst_reg, 342 insn->off, insn->imm); 343 } else if (class == BPF_LDX) { 344 if (BPF_MODE(insn->code) != BPF_MEM) { 345 verbose("BUG_ldx_%02x\n", insn->code); 346 return; 347 } 348 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", 349 insn->code, insn->dst_reg, 350 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 351 insn->src_reg, insn->off); 352 } else if (class == BPF_LD) { 353 if (BPF_MODE(insn->code) == BPF_ABS) { 354 verbose("(%02x) r0 = *(%s *)skb[%d]\n", 355 insn->code, 356 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 357 insn->imm); 358 } else if (BPF_MODE(insn->code) == BPF_IND) { 359 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", 360 insn->code, 361 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 362 insn->src_reg, insn->imm); 363 } else if (BPF_MODE(insn->code) == BPF_IMM) { 364 verbose("(%02x) r%d = 0x%x\n", 365 insn->code, insn->dst_reg, insn->imm); 366 } else { 367 verbose("BUG_ld_%02x\n", insn->code); 368 return; 369 } 370 } else if (class == BPF_JMP) { 371 u8 opcode = BPF_OP(insn->code); 372 373 if (opcode == BPF_CALL) { 374 verbose("(%02x) call %s#%d\n", insn->code, 375 func_id_name(insn->imm), insn->imm); 376 } else if (insn->code == (BPF_JMP | BPF_JA)) { 377 verbose("(%02x) goto pc%+d\n", 378 insn->code, insn->off); 379 } else if (insn->code == (BPF_JMP | BPF_EXIT)) { 380 verbose("(%02x) exit\n", insn->code); 381 } else if (BPF_SRC(insn->code) == BPF_X) { 382 verbose("(%02x) if r%d %s r%d goto pc%+d\n", 383 insn->code, insn->dst_reg, 384 bpf_jmp_string[BPF_OP(insn->code) >> 4], 385 insn->src_reg, insn->off); 386 } else { 387 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", 388 insn->code, insn->dst_reg, 389 bpf_jmp_string[BPF_OP(insn->code) >> 4], 390 insn->imm, insn->off); 391 } 392 } else { 393 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); 394 } 395 } 396 397 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) 398 { 399 struct bpf_verifier_stack_elem *elem; 400 int insn_idx; 401 402 if (env->head == NULL) 403 return -1; 404 405 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); 406 insn_idx = env->head->insn_idx; 407 if (prev_insn_idx) 408 *prev_insn_idx = env->head->prev_insn_idx; 409 elem = env->head->next; 410 kfree(env->head); 411 env->head = elem; 412 env->stack_size--; 413 return insn_idx; 414 } 415 416 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, 417 int insn_idx, int prev_insn_idx) 418 { 419 struct bpf_verifier_stack_elem *elem; 420 421 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); 422 if (!elem) 423 goto err; 424 425 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); 426 elem->insn_idx = insn_idx; 427 elem->prev_insn_idx = prev_insn_idx; 428 elem->next = env->head; 429 env->head = elem; 430 env->stack_size++; 431 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { 432 verbose("BPF program is too complex\n"); 433 goto err; 434 } 435 return &elem->st; 436 err: 437 /* pop all elements and return */ 438 while (pop_stack(env, NULL) >= 0); 439 return NULL; 440 } 441 442 #define CALLER_SAVED_REGS 6 443 static const int caller_saved[CALLER_SAVED_REGS] = { 444 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 445 }; 446 447 static void init_reg_state(struct bpf_reg_state *regs) 448 { 449 int i; 450 451 for (i = 0; i < MAX_BPF_REG; i++) { 452 regs[i].type = NOT_INIT; 453 regs[i].imm = 0; 454 regs[i].min_value = BPF_REGISTER_MIN_RANGE; 455 regs[i].max_value = BPF_REGISTER_MAX_RANGE; 456 } 457 458 /* frame pointer */ 459 regs[BPF_REG_FP].type = FRAME_PTR; 460 461 /* 1st arg to a function */ 462 regs[BPF_REG_1].type = PTR_TO_CTX; 463 } 464 465 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) 466 { 467 regs[regno].type = UNKNOWN_VALUE; 468 regs[regno].id = 0; 469 regs[regno].imm = 0; 470 } 471 472 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) 473 { 474 BUG_ON(regno >= MAX_BPF_REG); 475 __mark_reg_unknown_value(regs, regno); 476 } 477 478 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno) 479 { 480 regs[regno].min_value = BPF_REGISTER_MIN_RANGE; 481 regs[regno].max_value = BPF_REGISTER_MAX_RANGE; 482 } 483 484 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs, 485 u32 regno) 486 { 487 mark_reg_unknown_value(regs, regno); 488 reset_reg_range_values(regs, regno); 489 } 490 491 enum reg_arg_type { 492 SRC_OP, /* register is used as source operand */ 493 DST_OP, /* register is used as destination operand */ 494 DST_OP_NO_MARK /* same as above, check only, don't mark */ 495 }; 496 497 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, 498 enum reg_arg_type t) 499 { 500 if (regno >= MAX_BPF_REG) { 501 verbose("R%d is invalid\n", regno); 502 return -EINVAL; 503 } 504 505 if (t == SRC_OP) { 506 /* check whether register used as source operand can be read */ 507 if (regs[regno].type == NOT_INIT) { 508 verbose("R%d !read_ok\n", regno); 509 return -EACCES; 510 } 511 } else { 512 /* check whether register used as dest operand can be written to */ 513 if (regno == BPF_REG_FP) { 514 verbose("frame pointer is read only\n"); 515 return -EACCES; 516 } 517 if (t == DST_OP) 518 mark_reg_unknown_value(regs, regno); 519 } 520 return 0; 521 } 522 523 static int bpf_size_to_bytes(int bpf_size) 524 { 525 if (bpf_size == BPF_W) 526 return 4; 527 else if (bpf_size == BPF_H) 528 return 2; 529 else if (bpf_size == BPF_B) 530 return 1; 531 else if (bpf_size == BPF_DW) 532 return 8; 533 else 534 return -EINVAL; 535 } 536 537 static bool is_spillable_regtype(enum bpf_reg_type type) 538 { 539 switch (type) { 540 case PTR_TO_MAP_VALUE: 541 case PTR_TO_MAP_VALUE_OR_NULL: 542 case PTR_TO_MAP_VALUE_ADJ: 543 case PTR_TO_STACK: 544 case PTR_TO_CTX: 545 case PTR_TO_PACKET: 546 case PTR_TO_PACKET_END: 547 case FRAME_PTR: 548 case CONST_PTR_TO_MAP: 549 return true; 550 default: 551 return false; 552 } 553 } 554 555 /* check_stack_read/write functions track spill/fill of registers, 556 * stack boundary and alignment are checked in check_mem_access() 557 */ 558 static int check_stack_write(struct bpf_verifier_state *state, int off, 559 int size, int value_regno) 560 { 561 int i; 562 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, 563 * so it's aligned access and [off, off + size) are within stack limits 564 */ 565 566 if (value_regno >= 0 && 567 is_spillable_regtype(state->regs[value_regno].type)) { 568 569 /* register containing pointer is being spilled into stack */ 570 if (size != BPF_REG_SIZE) { 571 verbose("invalid size of register spill\n"); 572 return -EACCES; 573 } 574 575 /* save register state */ 576 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 577 state->regs[value_regno]; 578 579 for (i = 0; i < BPF_REG_SIZE; i++) 580 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; 581 } else { 582 /* regular write of data into stack */ 583 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 584 (struct bpf_reg_state) {}; 585 586 for (i = 0; i < size; i++) 587 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; 588 } 589 return 0; 590 } 591 592 static int check_stack_read(struct bpf_verifier_state *state, int off, int size, 593 int value_regno) 594 { 595 u8 *slot_type; 596 int i; 597 598 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; 599 600 if (slot_type[0] == STACK_SPILL) { 601 if (size != BPF_REG_SIZE) { 602 verbose("invalid size of register spill\n"); 603 return -EACCES; 604 } 605 for (i = 1; i < BPF_REG_SIZE; i++) { 606 if (slot_type[i] != STACK_SPILL) { 607 verbose("corrupted spill memory\n"); 608 return -EACCES; 609 } 610 } 611 612 if (value_regno >= 0) 613 /* restore register state from stack */ 614 state->regs[value_regno] = 615 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; 616 return 0; 617 } else { 618 for (i = 0; i < size; i++) { 619 if (slot_type[i] != STACK_MISC) { 620 verbose("invalid read from stack off %d+%d size %d\n", 621 off, i, size); 622 return -EACCES; 623 } 624 } 625 if (value_regno >= 0) 626 /* have read misc data from the stack */ 627 mark_reg_unknown_value_and_range(state->regs, 628 value_regno); 629 return 0; 630 } 631 } 632 633 /* check read/write into map element returned by bpf_map_lookup_elem() */ 634 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off, 635 int size) 636 { 637 struct bpf_map *map = env->cur_state.regs[regno].map_ptr; 638 639 if (off < 0 || size <= 0 || off + size > map->value_size) { 640 verbose("invalid access to map value, value_size=%d off=%d size=%d\n", 641 map->value_size, off, size); 642 return -EACCES; 643 } 644 return 0; 645 } 646 647 /* check read/write into an adjusted map element */ 648 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno, 649 int off, int size) 650 { 651 struct bpf_verifier_state *state = &env->cur_state; 652 struct bpf_reg_state *reg = &state->regs[regno]; 653 int err; 654 655 /* We adjusted the register to this map value, so we 656 * need to change off and size to min_value and max_value 657 * respectively to make sure our theoretical access will be 658 * safe. 659 */ 660 if (log_level) 661 print_verifier_state(state); 662 env->varlen_map_value_access = true; 663 /* The minimum value is only important with signed 664 * comparisons where we can't assume the floor of a 665 * value is 0. If we are using signed variables for our 666 * index'es we need to make sure that whatever we use 667 * will have a set floor within our range. 668 */ 669 if (reg->min_value < 0) { 670 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", 671 regno); 672 return -EACCES; 673 } 674 err = check_map_access(env, regno, reg->min_value + off, size); 675 if (err) { 676 verbose("R%d min value is outside of the array range\n", 677 regno); 678 return err; 679 } 680 681 /* If we haven't set a max value then we need to bail 682 * since we can't be sure we won't do bad things. 683 */ 684 if (reg->max_value == BPF_REGISTER_MAX_RANGE) { 685 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n", 686 regno); 687 return -EACCES; 688 } 689 return check_map_access(env, regno, reg->max_value + off, size); 690 } 691 692 #define MAX_PACKET_OFF 0xffff 693 694 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, 695 const struct bpf_call_arg_meta *meta, 696 enum bpf_access_type t) 697 { 698 switch (env->prog->type) { 699 case BPF_PROG_TYPE_LWT_IN: 700 case BPF_PROG_TYPE_LWT_OUT: 701 /* dst_input() and dst_output() can't write for now */ 702 if (t == BPF_WRITE) 703 return false; 704 /* fallthrough */ 705 case BPF_PROG_TYPE_SCHED_CLS: 706 case BPF_PROG_TYPE_SCHED_ACT: 707 case BPF_PROG_TYPE_XDP: 708 case BPF_PROG_TYPE_LWT_XMIT: 709 if (meta) 710 return meta->pkt_access; 711 712 env->seen_direct_write = true; 713 return true; 714 default: 715 return false; 716 } 717 } 718 719 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, 720 int size) 721 { 722 struct bpf_reg_state *regs = env->cur_state.regs; 723 struct bpf_reg_state *reg = ®s[regno]; 724 725 off += reg->off; 726 if (off < 0 || size <= 0 || off + size > reg->range) { 727 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", 728 off, size, regno, reg->id, reg->off, reg->range); 729 return -EACCES; 730 } 731 return 0; 732 } 733 734 /* check access to 'struct bpf_context' fields */ 735 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size, 736 enum bpf_access_type t, enum bpf_reg_type *reg_type) 737 { 738 /* for analyzer ctx accesses are already validated and converted */ 739 if (env->analyzer_ops) 740 return 0; 741 742 if (env->prog->aux->ops->is_valid_access && 743 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) { 744 /* remember the offset of last byte accessed in ctx */ 745 if (env->prog->aux->max_ctx_offset < off + size) 746 env->prog->aux->max_ctx_offset = off + size; 747 return 0; 748 } 749 750 verbose("invalid bpf_context access off=%d size=%d\n", off, size); 751 return -EACCES; 752 } 753 754 static bool is_pointer_value(struct bpf_verifier_env *env, int regno) 755 { 756 if (env->allow_ptr_leaks) 757 return false; 758 759 switch (env->cur_state.regs[regno].type) { 760 case UNKNOWN_VALUE: 761 case CONST_IMM: 762 return false; 763 default: 764 return true; 765 } 766 } 767 768 static int check_ptr_alignment(struct bpf_verifier_env *env, 769 struct bpf_reg_state *reg, int off, int size) 770 { 771 if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) { 772 if (off % size != 0) { 773 verbose("misaligned access off %d size %d\n", 774 off, size); 775 return -EACCES; 776 } else { 777 return 0; 778 } 779 } 780 781 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 782 /* misaligned access to packet is ok on x86,arm,arm64 */ 783 return 0; 784 785 if (reg->id && size != 1) { 786 verbose("Unknown packet alignment. Only byte-sized access allowed\n"); 787 return -EACCES; 788 } 789 790 /* skb->data is NET_IP_ALIGN-ed */ 791 if (reg->type == PTR_TO_PACKET && 792 (NET_IP_ALIGN + reg->off + off) % size != 0) { 793 verbose("misaligned packet access off %d+%d+%d size %d\n", 794 NET_IP_ALIGN, reg->off, off, size); 795 return -EACCES; 796 } 797 return 0; 798 } 799 800 /* check whether memory at (regno + off) is accessible for t = (read | write) 801 * if t==write, value_regno is a register which value is stored into memory 802 * if t==read, value_regno is a register which will receive the value from memory 803 * if t==write && value_regno==-1, some unknown value is stored into memory 804 * if t==read && value_regno==-1, don't care what we read from memory 805 */ 806 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off, 807 int bpf_size, enum bpf_access_type t, 808 int value_regno) 809 { 810 struct bpf_verifier_state *state = &env->cur_state; 811 struct bpf_reg_state *reg = &state->regs[regno]; 812 int size, err = 0; 813 814 if (reg->type == PTR_TO_STACK) 815 off += reg->imm; 816 817 size = bpf_size_to_bytes(bpf_size); 818 if (size < 0) 819 return size; 820 821 err = check_ptr_alignment(env, reg, off, size); 822 if (err) 823 return err; 824 825 if (reg->type == PTR_TO_MAP_VALUE || 826 reg->type == PTR_TO_MAP_VALUE_ADJ) { 827 if (t == BPF_WRITE && value_regno >= 0 && 828 is_pointer_value(env, value_regno)) { 829 verbose("R%d leaks addr into map\n", value_regno); 830 return -EACCES; 831 } 832 833 if (reg->type == PTR_TO_MAP_VALUE_ADJ) 834 err = check_map_access_adj(env, regno, off, size); 835 else 836 err = check_map_access(env, regno, off, size); 837 if (!err && t == BPF_READ && value_regno >= 0) 838 mark_reg_unknown_value_and_range(state->regs, 839 value_regno); 840 841 } else if (reg->type == PTR_TO_CTX) { 842 enum bpf_reg_type reg_type = UNKNOWN_VALUE; 843 844 if (t == BPF_WRITE && value_regno >= 0 && 845 is_pointer_value(env, value_regno)) { 846 verbose("R%d leaks addr into ctx\n", value_regno); 847 return -EACCES; 848 } 849 err = check_ctx_access(env, off, size, t, ®_type); 850 if (!err && t == BPF_READ && value_regno >= 0) { 851 mark_reg_unknown_value_and_range(state->regs, 852 value_regno); 853 /* note that reg.[id|off|range] == 0 */ 854 state->regs[value_regno].type = reg_type; 855 } 856 857 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { 858 if (off >= 0 || off < -MAX_BPF_STACK) { 859 verbose("invalid stack off=%d size=%d\n", off, size); 860 return -EACCES; 861 } 862 if (t == BPF_WRITE) { 863 if (!env->allow_ptr_leaks && 864 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && 865 size != BPF_REG_SIZE) { 866 verbose("attempt to corrupt spilled pointer on stack\n"); 867 return -EACCES; 868 } 869 err = check_stack_write(state, off, size, value_regno); 870 } else { 871 err = check_stack_read(state, off, size, value_regno); 872 } 873 } else if (state->regs[regno].type == PTR_TO_PACKET) { 874 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { 875 verbose("cannot write into packet\n"); 876 return -EACCES; 877 } 878 if (t == BPF_WRITE && value_regno >= 0 && 879 is_pointer_value(env, value_regno)) { 880 verbose("R%d leaks addr into packet\n", value_regno); 881 return -EACCES; 882 } 883 err = check_packet_access(env, regno, off, size); 884 if (!err && t == BPF_READ && value_regno >= 0) 885 mark_reg_unknown_value_and_range(state->regs, 886 value_regno); 887 } else { 888 verbose("R%d invalid mem access '%s'\n", 889 regno, reg_type_str[reg->type]); 890 return -EACCES; 891 } 892 893 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && 894 state->regs[value_regno].type == UNKNOWN_VALUE) { 895 /* 1 or 2 byte load zero-extends, determine the number of 896 * zero upper bits. Not doing it fo 4 byte load, since 897 * such values cannot be added to ptr_to_packet anyway. 898 */ 899 state->regs[value_regno].imm = 64 - size * 8; 900 } 901 return err; 902 } 903 904 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn) 905 { 906 struct bpf_reg_state *regs = env->cur_state.regs; 907 int err; 908 909 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || 910 insn->imm != 0) { 911 verbose("BPF_XADD uses reserved fields\n"); 912 return -EINVAL; 913 } 914 915 /* check src1 operand */ 916 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 917 if (err) 918 return err; 919 920 /* check src2 operand */ 921 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 922 if (err) 923 return err; 924 925 /* check whether atomic_add can read the memory */ 926 err = check_mem_access(env, insn->dst_reg, insn->off, 927 BPF_SIZE(insn->code), BPF_READ, -1); 928 if (err) 929 return err; 930 931 /* check whether atomic_add can write into the same memory */ 932 return check_mem_access(env, insn->dst_reg, insn->off, 933 BPF_SIZE(insn->code), BPF_WRITE, -1); 934 } 935 936 /* when register 'regno' is passed into function that will read 'access_size' 937 * bytes from that pointer, make sure that it's within stack boundary 938 * and all elements of stack are initialized 939 */ 940 static int check_stack_boundary(struct bpf_verifier_env *env, int regno, 941 int access_size, bool zero_size_allowed, 942 struct bpf_call_arg_meta *meta) 943 { 944 struct bpf_verifier_state *state = &env->cur_state; 945 struct bpf_reg_state *regs = state->regs; 946 int off, i; 947 948 if (regs[regno].type != PTR_TO_STACK) { 949 if (zero_size_allowed && access_size == 0 && 950 regs[regno].type == CONST_IMM && 951 regs[regno].imm == 0) 952 return 0; 953 954 verbose("R%d type=%s expected=%s\n", regno, 955 reg_type_str[regs[regno].type], 956 reg_type_str[PTR_TO_STACK]); 957 return -EACCES; 958 } 959 960 off = regs[regno].imm; 961 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || 962 access_size <= 0) { 963 verbose("invalid stack type R%d off=%d access_size=%d\n", 964 regno, off, access_size); 965 return -EACCES; 966 } 967 968 if (meta && meta->raw_mode) { 969 meta->access_size = access_size; 970 meta->regno = regno; 971 return 0; 972 } 973 974 for (i = 0; i < access_size; i++) { 975 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { 976 verbose("invalid indirect read from stack off %d+%d size %d\n", 977 off, i, access_size); 978 return -EACCES; 979 } 980 } 981 return 0; 982 } 983 984 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, 985 int access_size, bool zero_size_allowed, 986 struct bpf_call_arg_meta *meta) 987 { 988 struct bpf_reg_state *regs = env->cur_state.regs; 989 990 switch (regs[regno].type) { 991 case PTR_TO_PACKET: 992 return check_packet_access(env, regno, 0, access_size); 993 case PTR_TO_MAP_VALUE: 994 return check_map_access(env, regno, 0, access_size); 995 case PTR_TO_MAP_VALUE_ADJ: 996 return check_map_access_adj(env, regno, 0, access_size); 997 default: /* const_imm|ptr_to_stack or invalid ptr */ 998 return check_stack_boundary(env, regno, access_size, 999 zero_size_allowed, meta); 1000 } 1001 } 1002 1003 static int check_func_arg(struct bpf_verifier_env *env, u32 regno, 1004 enum bpf_arg_type arg_type, 1005 struct bpf_call_arg_meta *meta) 1006 { 1007 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; 1008 enum bpf_reg_type expected_type, type = reg->type; 1009 int err = 0; 1010 1011 if (arg_type == ARG_DONTCARE) 1012 return 0; 1013 1014 if (type == NOT_INIT) { 1015 verbose("R%d !read_ok\n", regno); 1016 return -EACCES; 1017 } 1018 1019 if (arg_type == ARG_ANYTHING) { 1020 if (is_pointer_value(env, regno)) { 1021 verbose("R%d leaks addr into helper function\n", regno); 1022 return -EACCES; 1023 } 1024 return 0; 1025 } 1026 1027 if (type == PTR_TO_PACKET && 1028 !may_access_direct_pkt_data(env, meta, BPF_READ)) { 1029 verbose("helper access to the packet is not allowed\n"); 1030 return -EACCES; 1031 } 1032 1033 if (arg_type == ARG_PTR_TO_MAP_KEY || 1034 arg_type == ARG_PTR_TO_MAP_VALUE) { 1035 expected_type = PTR_TO_STACK; 1036 if (type != PTR_TO_PACKET && type != expected_type) 1037 goto err_type; 1038 } else if (arg_type == ARG_CONST_SIZE || 1039 arg_type == ARG_CONST_SIZE_OR_ZERO) { 1040 expected_type = CONST_IMM; 1041 /* One exception. Allow UNKNOWN_VALUE registers when the 1042 * boundaries are known and don't cause unsafe memory accesses 1043 */ 1044 if (type != UNKNOWN_VALUE && type != expected_type) 1045 goto err_type; 1046 } else if (arg_type == ARG_CONST_MAP_PTR) { 1047 expected_type = CONST_PTR_TO_MAP; 1048 if (type != expected_type) 1049 goto err_type; 1050 } else if (arg_type == ARG_PTR_TO_CTX) { 1051 expected_type = PTR_TO_CTX; 1052 if (type != expected_type) 1053 goto err_type; 1054 } else if (arg_type == ARG_PTR_TO_MEM || 1055 arg_type == ARG_PTR_TO_UNINIT_MEM) { 1056 expected_type = PTR_TO_STACK; 1057 /* One exception here. In case function allows for NULL to be 1058 * passed in as argument, it's a CONST_IMM type. Final test 1059 * happens during stack boundary checking. 1060 */ 1061 if (type == CONST_IMM && reg->imm == 0) 1062 /* final test in check_stack_boundary() */; 1063 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE && 1064 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type) 1065 goto err_type; 1066 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; 1067 } else { 1068 verbose("unsupported arg_type %d\n", arg_type); 1069 return -EFAULT; 1070 } 1071 1072 if (arg_type == ARG_CONST_MAP_PTR) { 1073 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ 1074 meta->map_ptr = reg->map_ptr; 1075 } else if (arg_type == ARG_PTR_TO_MAP_KEY) { 1076 /* bpf_map_xxx(..., map_ptr, ..., key) call: 1077 * check that [key, key + map->key_size) are within 1078 * stack limits and initialized 1079 */ 1080 if (!meta->map_ptr) { 1081 /* in function declaration map_ptr must come before 1082 * map_key, so that it's verified and known before 1083 * we have to check map_key here. Otherwise it means 1084 * that kernel subsystem misconfigured verifier 1085 */ 1086 verbose("invalid map_ptr to access map->key\n"); 1087 return -EACCES; 1088 } 1089 if (type == PTR_TO_PACKET) 1090 err = check_packet_access(env, regno, 0, 1091 meta->map_ptr->key_size); 1092 else 1093 err = check_stack_boundary(env, regno, 1094 meta->map_ptr->key_size, 1095 false, NULL); 1096 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { 1097 /* bpf_map_xxx(..., map_ptr, ..., value) call: 1098 * check [value, value + map->value_size) validity 1099 */ 1100 if (!meta->map_ptr) { 1101 /* kernel subsystem misconfigured verifier */ 1102 verbose("invalid map_ptr to access map->value\n"); 1103 return -EACCES; 1104 } 1105 if (type == PTR_TO_PACKET) 1106 err = check_packet_access(env, regno, 0, 1107 meta->map_ptr->value_size); 1108 else 1109 err = check_stack_boundary(env, regno, 1110 meta->map_ptr->value_size, 1111 false, NULL); 1112 } else if (arg_type == ARG_CONST_SIZE || 1113 arg_type == ARG_CONST_SIZE_OR_ZERO) { 1114 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); 1115 1116 /* bpf_xxx(..., buf, len) call will access 'len' bytes 1117 * from stack pointer 'buf'. Check it 1118 * note: regno == len, regno - 1 == buf 1119 */ 1120 if (regno == 0) { 1121 /* kernel subsystem misconfigured verifier */ 1122 verbose("ARG_CONST_SIZE cannot be first argument\n"); 1123 return -EACCES; 1124 } 1125 1126 /* If the register is UNKNOWN_VALUE, the access check happens 1127 * using its boundaries. Otherwise, just use its imm 1128 */ 1129 if (type == UNKNOWN_VALUE) { 1130 /* For unprivileged variable accesses, disable raw 1131 * mode so that the program is required to 1132 * initialize all the memory that the helper could 1133 * just partially fill up. 1134 */ 1135 meta = NULL; 1136 1137 if (reg->min_value < 0) { 1138 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n", 1139 regno); 1140 return -EACCES; 1141 } 1142 1143 if (reg->min_value == 0) { 1144 err = check_helper_mem_access(env, regno - 1, 0, 1145 zero_size_allowed, 1146 meta); 1147 if (err) 1148 return err; 1149 } 1150 1151 if (reg->max_value == BPF_REGISTER_MAX_RANGE) { 1152 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", 1153 regno); 1154 return -EACCES; 1155 } 1156 err = check_helper_mem_access(env, regno - 1, 1157 reg->max_value, 1158 zero_size_allowed, meta); 1159 if (err) 1160 return err; 1161 } else { 1162 /* register is CONST_IMM */ 1163 err = check_helper_mem_access(env, regno - 1, reg->imm, 1164 zero_size_allowed, meta); 1165 } 1166 } 1167 1168 return err; 1169 err_type: 1170 verbose("R%d type=%s expected=%s\n", regno, 1171 reg_type_str[type], reg_type_str[expected_type]); 1172 return -EACCES; 1173 } 1174 1175 static int check_map_func_compatibility(struct bpf_map *map, int func_id) 1176 { 1177 if (!map) 1178 return 0; 1179 1180 /* We need a two way check, first is from map perspective ... */ 1181 switch (map->map_type) { 1182 case BPF_MAP_TYPE_PROG_ARRAY: 1183 if (func_id != BPF_FUNC_tail_call) 1184 goto error; 1185 break; 1186 case BPF_MAP_TYPE_PERF_EVENT_ARRAY: 1187 if (func_id != BPF_FUNC_perf_event_read && 1188 func_id != BPF_FUNC_perf_event_output) 1189 goto error; 1190 break; 1191 case BPF_MAP_TYPE_STACK_TRACE: 1192 if (func_id != BPF_FUNC_get_stackid) 1193 goto error; 1194 break; 1195 case BPF_MAP_TYPE_CGROUP_ARRAY: 1196 if (func_id != BPF_FUNC_skb_under_cgroup && 1197 func_id != BPF_FUNC_current_task_under_cgroup) 1198 goto error; 1199 break; 1200 default: 1201 break; 1202 } 1203 1204 /* ... and second from the function itself. */ 1205 switch (func_id) { 1206 case BPF_FUNC_tail_call: 1207 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 1208 goto error; 1209 break; 1210 case BPF_FUNC_perf_event_read: 1211 case BPF_FUNC_perf_event_output: 1212 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) 1213 goto error; 1214 break; 1215 case BPF_FUNC_get_stackid: 1216 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) 1217 goto error; 1218 break; 1219 case BPF_FUNC_current_task_under_cgroup: 1220 case BPF_FUNC_skb_under_cgroup: 1221 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) 1222 goto error; 1223 break; 1224 default: 1225 break; 1226 } 1227 1228 return 0; 1229 error: 1230 verbose("cannot pass map_type %d into func %s#%d\n", 1231 map->map_type, func_id_name(func_id), func_id); 1232 return -EINVAL; 1233 } 1234 1235 static int check_raw_mode(const struct bpf_func_proto *fn) 1236 { 1237 int count = 0; 1238 1239 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) 1240 count++; 1241 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) 1242 count++; 1243 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) 1244 count++; 1245 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) 1246 count++; 1247 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) 1248 count++; 1249 1250 return count > 1 ? -EINVAL : 0; 1251 } 1252 1253 static void clear_all_pkt_pointers(struct bpf_verifier_env *env) 1254 { 1255 struct bpf_verifier_state *state = &env->cur_state; 1256 struct bpf_reg_state *regs = state->regs, *reg; 1257 int i; 1258 1259 for (i = 0; i < MAX_BPF_REG; i++) 1260 if (regs[i].type == PTR_TO_PACKET || 1261 regs[i].type == PTR_TO_PACKET_END) 1262 mark_reg_unknown_value(regs, i); 1263 1264 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1265 if (state->stack_slot_type[i] != STACK_SPILL) 1266 continue; 1267 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1268 if (reg->type != PTR_TO_PACKET && 1269 reg->type != PTR_TO_PACKET_END) 1270 continue; 1271 reg->type = UNKNOWN_VALUE; 1272 reg->imm = 0; 1273 } 1274 } 1275 1276 static int check_call(struct bpf_verifier_env *env, int func_id) 1277 { 1278 struct bpf_verifier_state *state = &env->cur_state; 1279 const struct bpf_func_proto *fn = NULL; 1280 struct bpf_reg_state *regs = state->regs; 1281 struct bpf_reg_state *reg; 1282 struct bpf_call_arg_meta meta; 1283 bool changes_data; 1284 int i, err; 1285 1286 /* find function prototype */ 1287 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { 1288 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id); 1289 return -EINVAL; 1290 } 1291 1292 if (env->prog->aux->ops->get_func_proto) 1293 fn = env->prog->aux->ops->get_func_proto(func_id); 1294 1295 if (!fn) { 1296 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id); 1297 return -EINVAL; 1298 } 1299 1300 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1301 if (!env->prog->gpl_compatible && fn->gpl_only) { 1302 verbose("cannot call GPL only function from proprietary program\n"); 1303 return -EINVAL; 1304 } 1305 1306 changes_data = bpf_helper_changes_pkt_data(fn->func); 1307 1308 memset(&meta, 0, sizeof(meta)); 1309 meta.pkt_access = fn->pkt_access; 1310 1311 /* We only support one arg being in raw mode at the moment, which 1312 * is sufficient for the helper functions we have right now. 1313 */ 1314 err = check_raw_mode(fn); 1315 if (err) { 1316 verbose("kernel subsystem misconfigured func %s#%d\n", 1317 func_id_name(func_id), func_id); 1318 return err; 1319 } 1320 1321 /* check args */ 1322 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); 1323 if (err) 1324 return err; 1325 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); 1326 if (err) 1327 return err; 1328 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); 1329 if (err) 1330 return err; 1331 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); 1332 if (err) 1333 return err; 1334 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); 1335 if (err) 1336 return err; 1337 1338 /* Mark slots with STACK_MISC in case of raw mode, stack offset 1339 * is inferred from register state. 1340 */ 1341 for (i = 0; i < meta.access_size; i++) { 1342 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1); 1343 if (err) 1344 return err; 1345 } 1346 1347 /* reset caller saved regs */ 1348 for (i = 0; i < CALLER_SAVED_REGS; i++) { 1349 reg = regs + caller_saved[i]; 1350 reg->type = NOT_INIT; 1351 reg->imm = 0; 1352 } 1353 1354 /* update return register */ 1355 if (fn->ret_type == RET_INTEGER) { 1356 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1357 } else if (fn->ret_type == RET_VOID) { 1358 regs[BPF_REG_0].type = NOT_INIT; 1359 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { 1360 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; 1361 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0; 1362 /* remember map_ptr, so that check_map_access() 1363 * can check 'value_size' boundary of memory access 1364 * to map element returned from bpf_map_lookup_elem() 1365 */ 1366 if (meta.map_ptr == NULL) { 1367 verbose("kernel subsystem misconfigured verifier\n"); 1368 return -EINVAL; 1369 } 1370 regs[BPF_REG_0].map_ptr = meta.map_ptr; 1371 regs[BPF_REG_0].id = ++env->id_gen; 1372 } else { 1373 verbose("unknown return type %d of func %s#%d\n", 1374 fn->ret_type, func_id_name(func_id), func_id); 1375 return -EINVAL; 1376 } 1377 1378 err = check_map_func_compatibility(meta.map_ptr, func_id); 1379 if (err) 1380 return err; 1381 1382 if (changes_data) 1383 clear_all_pkt_pointers(env); 1384 return 0; 1385 } 1386 1387 static int check_packet_ptr_add(struct bpf_verifier_env *env, 1388 struct bpf_insn *insn) 1389 { 1390 struct bpf_reg_state *regs = env->cur_state.regs; 1391 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1392 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1393 struct bpf_reg_state tmp_reg; 1394 s32 imm; 1395 1396 if (BPF_SRC(insn->code) == BPF_K) { 1397 /* pkt_ptr += imm */ 1398 imm = insn->imm; 1399 1400 add_imm: 1401 if (imm < 0) { 1402 verbose("addition of negative constant to packet pointer is not allowed\n"); 1403 return -EACCES; 1404 } 1405 if (imm >= MAX_PACKET_OFF || 1406 imm + dst_reg->off >= MAX_PACKET_OFF) { 1407 verbose("constant %d is too large to add to packet pointer\n", 1408 imm); 1409 return -EACCES; 1410 } 1411 /* a constant was added to pkt_ptr. 1412 * Remember it while keeping the same 'id' 1413 */ 1414 dst_reg->off += imm; 1415 } else { 1416 if (src_reg->type == PTR_TO_PACKET) { 1417 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ 1418 tmp_reg = *dst_reg; /* save r7 state */ 1419 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ 1420 src_reg = &tmp_reg; /* pretend it's src_reg state */ 1421 /* if the checks below reject it, the copy won't matter, 1422 * since we're rejecting the whole program. If all ok, 1423 * then imm22 state will be added to r7 1424 * and r7 will be pkt(id=0,off=22,r=62) while 1425 * r6 will stay as pkt(id=0,off=0,r=62) 1426 */ 1427 } 1428 1429 if (src_reg->type == CONST_IMM) { 1430 /* pkt_ptr += reg where reg is known constant */ 1431 imm = src_reg->imm; 1432 goto add_imm; 1433 } 1434 /* disallow pkt_ptr += reg 1435 * if reg is not uknown_value with guaranteed zero upper bits 1436 * otherwise pkt_ptr may overflow and addition will become 1437 * subtraction which is not allowed 1438 */ 1439 if (src_reg->type != UNKNOWN_VALUE) { 1440 verbose("cannot add '%s' to ptr_to_packet\n", 1441 reg_type_str[src_reg->type]); 1442 return -EACCES; 1443 } 1444 if (src_reg->imm < 48) { 1445 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", 1446 src_reg->imm); 1447 return -EACCES; 1448 } 1449 /* dst_reg stays as pkt_ptr type and since some positive 1450 * integer value was added to the pointer, increment its 'id' 1451 */ 1452 dst_reg->id = ++env->id_gen; 1453 1454 /* something was added to pkt_ptr, set range and off to zero */ 1455 dst_reg->off = 0; 1456 dst_reg->range = 0; 1457 } 1458 return 0; 1459 } 1460 1461 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) 1462 { 1463 struct bpf_reg_state *regs = env->cur_state.regs; 1464 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1465 u8 opcode = BPF_OP(insn->code); 1466 s64 imm_log2; 1467 1468 /* for type == UNKNOWN_VALUE: 1469 * imm > 0 -> number of zero upper bits 1470 * imm == 0 -> don't track which is the same as all bits can be non-zero 1471 */ 1472 1473 if (BPF_SRC(insn->code) == BPF_X) { 1474 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1475 1476 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && 1477 dst_reg->imm && opcode == BPF_ADD) { 1478 /* dreg += sreg 1479 * where both have zero upper bits. Adding them 1480 * can only result making one more bit non-zero 1481 * in the larger value. 1482 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1483 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1484 */ 1485 dst_reg->imm = min(dst_reg->imm, src_reg->imm); 1486 dst_reg->imm--; 1487 return 0; 1488 } 1489 if (src_reg->type == CONST_IMM && src_reg->imm > 0 && 1490 dst_reg->imm && opcode == BPF_ADD) { 1491 /* dreg += sreg 1492 * where dreg has zero upper bits and sreg is const. 1493 * Adding them can only result making one more bit 1494 * non-zero in the larger value. 1495 */ 1496 imm_log2 = __ilog2_u64((long long)src_reg->imm); 1497 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1498 dst_reg->imm--; 1499 return 0; 1500 } 1501 /* all other cases non supported yet, just mark dst_reg */ 1502 dst_reg->imm = 0; 1503 return 0; 1504 } 1505 1506 /* sign extend 32-bit imm into 64-bit to make sure that 1507 * negative values occupy bit 63. Note ilog2() would have 1508 * been incorrect, since sizeof(insn->imm) == 4 1509 */ 1510 imm_log2 = __ilog2_u64((long long)insn->imm); 1511 1512 if (dst_reg->imm && opcode == BPF_LSH) { 1513 /* reg <<= imm 1514 * if reg was a result of 2 byte load, then its imm == 48 1515 * which means that upper 48 bits are zero and shifting this reg 1516 * left by 4 would mean that upper 44 bits are still zero 1517 */ 1518 dst_reg->imm -= insn->imm; 1519 } else if (dst_reg->imm && opcode == BPF_MUL) { 1520 /* reg *= imm 1521 * if multiplying by 14 subtract 4 1522 * This is conservative calculation of upper zero bits. 1523 * It's not trying to special case insn->imm == 1 or 0 cases 1524 */ 1525 dst_reg->imm -= imm_log2 + 1; 1526 } else if (opcode == BPF_AND) { 1527 /* reg &= imm */ 1528 dst_reg->imm = 63 - imm_log2; 1529 } else if (dst_reg->imm && opcode == BPF_ADD) { 1530 /* reg += imm */ 1531 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1532 dst_reg->imm--; 1533 } else if (opcode == BPF_RSH) { 1534 /* reg >>= imm 1535 * which means that after right shift, upper bits will be zero 1536 * note that verifier already checked that 1537 * 0 <= imm < 64 for shift insn 1538 */ 1539 dst_reg->imm += insn->imm; 1540 if (unlikely(dst_reg->imm > 64)) 1541 /* some dumb code did: 1542 * r2 = *(u32 *)mem; 1543 * r2 >>= 32; 1544 * and all bits are zero now */ 1545 dst_reg->imm = 64; 1546 } else { 1547 /* all other alu ops, means that we don't know what will 1548 * happen to the value, mark it with unknown number of zero bits 1549 */ 1550 dst_reg->imm = 0; 1551 } 1552 1553 if (dst_reg->imm < 0) { 1554 /* all 64 bits of the register can contain non-zero bits 1555 * and such value cannot be added to ptr_to_packet, since it 1556 * may overflow, mark it as unknown to avoid further eval 1557 */ 1558 dst_reg->imm = 0; 1559 } 1560 return 0; 1561 } 1562 1563 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, 1564 struct bpf_insn *insn) 1565 { 1566 struct bpf_reg_state *regs = env->cur_state.regs; 1567 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; 1568 struct bpf_reg_state *src_reg = ®s[insn->src_reg]; 1569 u8 opcode = BPF_OP(insn->code); 1570 u64 dst_imm = dst_reg->imm; 1571 1572 /* dst_reg->type == CONST_IMM here. Simulate execution of insns 1573 * containing ALU ops. Don't care about overflow or negative 1574 * values, just add/sub/... them; registers are in u64. 1575 */ 1576 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) { 1577 dst_imm += insn->imm; 1578 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && 1579 src_reg->type == CONST_IMM) { 1580 dst_imm += src_reg->imm; 1581 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) { 1582 dst_imm -= insn->imm; 1583 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X && 1584 src_reg->type == CONST_IMM) { 1585 dst_imm -= src_reg->imm; 1586 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) { 1587 dst_imm *= insn->imm; 1588 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X && 1589 src_reg->type == CONST_IMM) { 1590 dst_imm *= src_reg->imm; 1591 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) { 1592 dst_imm |= insn->imm; 1593 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X && 1594 src_reg->type == CONST_IMM) { 1595 dst_imm |= src_reg->imm; 1596 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) { 1597 dst_imm &= insn->imm; 1598 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X && 1599 src_reg->type == CONST_IMM) { 1600 dst_imm &= src_reg->imm; 1601 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) { 1602 dst_imm >>= insn->imm; 1603 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X && 1604 src_reg->type == CONST_IMM) { 1605 dst_imm >>= src_reg->imm; 1606 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) { 1607 dst_imm <<= insn->imm; 1608 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X && 1609 src_reg->type == CONST_IMM) { 1610 dst_imm <<= src_reg->imm; 1611 } else { 1612 mark_reg_unknown_value(regs, insn->dst_reg); 1613 goto out; 1614 } 1615 1616 dst_reg->imm = dst_imm; 1617 out: 1618 return 0; 1619 } 1620 1621 static void check_reg_overflow(struct bpf_reg_state *reg) 1622 { 1623 if (reg->max_value > BPF_REGISTER_MAX_RANGE) 1624 reg->max_value = BPF_REGISTER_MAX_RANGE; 1625 if (reg->min_value < BPF_REGISTER_MIN_RANGE || 1626 reg->min_value > BPF_REGISTER_MAX_RANGE) 1627 reg->min_value = BPF_REGISTER_MIN_RANGE; 1628 } 1629 1630 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env, 1631 struct bpf_insn *insn) 1632 { 1633 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1634 s64 min_val = BPF_REGISTER_MIN_RANGE; 1635 u64 max_val = BPF_REGISTER_MAX_RANGE; 1636 u8 opcode = BPF_OP(insn->code); 1637 1638 dst_reg = ®s[insn->dst_reg]; 1639 if (BPF_SRC(insn->code) == BPF_X) { 1640 check_reg_overflow(®s[insn->src_reg]); 1641 min_val = regs[insn->src_reg].min_value; 1642 max_val = regs[insn->src_reg].max_value; 1643 1644 /* If the source register is a random pointer then the 1645 * min_value/max_value values represent the range of the known 1646 * accesses into that value, not the actual min/max value of the 1647 * register itself. In this case we have to reset the reg range 1648 * values so we know it is not safe to look at. 1649 */ 1650 if (regs[insn->src_reg].type != CONST_IMM && 1651 regs[insn->src_reg].type != UNKNOWN_VALUE) { 1652 min_val = BPF_REGISTER_MIN_RANGE; 1653 max_val = BPF_REGISTER_MAX_RANGE; 1654 } 1655 } else if (insn->imm < BPF_REGISTER_MAX_RANGE && 1656 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) { 1657 min_val = max_val = insn->imm; 1658 } 1659 1660 /* We don't know anything about what was done to this register, mark it 1661 * as unknown. 1662 */ 1663 if (min_val == BPF_REGISTER_MIN_RANGE && 1664 max_val == BPF_REGISTER_MAX_RANGE) { 1665 reset_reg_range_values(regs, insn->dst_reg); 1666 return; 1667 } 1668 1669 /* If one of our values was at the end of our ranges then we can't just 1670 * do our normal operations to the register, we need to set the values 1671 * to the min/max since they are undefined. 1672 */ 1673 if (min_val == BPF_REGISTER_MIN_RANGE) 1674 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1675 if (max_val == BPF_REGISTER_MAX_RANGE) 1676 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1677 1678 switch (opcode) { 1679 case BPF_ADD: 1680 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1681 dst_reg->min_value += min_val; 1682 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1683 dst_reg->max_value += max_val; 1684 break; 1685 case BPF_SUB: 1686 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1687 dst_reg->min_value -= min_val; 1688 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1689 dst_reg->max_value -= max_val; 1690 break; 1691 case BPF_MUL: 1692 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1693 dst_reg->min_value *= min_val; 1694 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1695 dst_reg->max_value *= max_val; 1696 break; 1697 case BPF_AND: 1698 /* Disallow AND'ing of negative numbers, ain't nobody got time 1699 * for that. Otherwise the minimum is 0 and the max is the max 1700 * value we could AND against. 1701 */ 1702 if (min_val < 0) 1703 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1704 else 1705 dst_reg->min_value = 0; 1706 dst_reg->max_value = max_val; 1707 break; 1708 case BPF_LSH: 1709 /* Gotta have special overflow logic here, if we're shifting 1710 * more than MAX_RANGE then just assume we have an invalid 1711 * range. 1712 */ 1713 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) 1714 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1715 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) 1716 dst_reg->min_value <<= min_val; 1717 1718 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) 1719 dst_reg->max_value = BPF_REGISTER_MAX_RANGE; 1720 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1721 dst_reg->max_value <<= max_val; 1722 break; 1723 case BPF_RSH: 1724 /* RSH by a negative number is undefined, and the BPF_RSH is an 1725 * unsigned shift, so make the appropriate casts. 1726 */ 1727 if (min_val < 0 || dst_reg->min_value < 0) 1728 dst_reg->min_value = BPF_REGISTER_MIN_RANGE; 1729 else 1730 dst_reg->min_value = 1731 (u64)(dst_reg->min_value) >> min_val; 1732 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) 1733 dst_reg->max_value >>= max_val; 1734 break; 1735 default: 1736 reset_reg_range_values(regs, insn->dst_reg); 1737 break; 1738 } 1739 1740 check_reg_overflow(dst_reg); 1741 } 1742 1743 /* check validity of 32-bit and 64-bit arithmetic operations */ 1744 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) 1745 { 1746 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; 1747 u8 opcode = BPF_OP(insn->code); 1748 int err; 1749 1750 if (opcode == BPF_END || opcode == BPF_NEG) { 1751 if (opcode == BPF_NEG) { 1752 if (BPF_SRC(insn->code) != 0 || 1753 insn->src_reg != BPF_REG_0 || 1754 insn->off != 0 || insn->imm != 0) { 1755 verbose("BPF_NEG uses reserved fields\n"); 1756 return -EINVAL; 1757 } 1758 } else { 1759 if (insn->src_reg != BPF_REG_0 || insn->off != 0 || 1760 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { 1761 verbose("BPF_END uses reserved fields\n"); 1762 return -EINVAL; 1763 } 1764 } 1765 1766 /* check src operand */ 1767 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1768 if (err) 1769 return err; 1770 1771 if (is_pointer_value(env, insn->dst_reg)) { 1772 verbose("R%d pointer arithmetic prohibited\n", 1773 insn->dst_reg); 1774 return -EACCES; 1775 } 1776 1777 /* check dest operand */ 1778 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1779 if (err) 1780 return err; 1781 1782 } else if (opcode == BPF_MOV) { 1783 1784 if (BPF_SRC(insn->code) == BPF_X) { 1785 if (insn->imm != 0 || insn->off != 0) { 1786 verbose("BPF_MOV uses reserved fields\n"); 1787 return -EINVAL; 1788 } 1789 1790 /* check src operand */ 1791 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1792 if (err) 1793 return err; 1794 } else { 1795 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 1796 verbose("BPF_MOV uses reserved fields\n"); 1797 return -EINVAL; 1798 } 1799 } 1800 1801 /* check dest operand */ 1802 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1803 if (err) 1804 return err; 1805 1806 /* we are setting our register to something new, we need to 1807 * reset its range values. 1808 */ 1809 reset_reg_range_values(regs, insn->dst_reg); 1810 1811 if (BPF_SRC(insn->code) == BPF_X) { 1812 if (BPF_CLASS(insn->code) == BPF_ALU64) { 1813 /* case: R1 = R2 1814 * copy register state to dest reg 1815 */ 1816 regs[insn->dst_reg] = regs[insn->src_reg]; 1817 } else { 1818 if (is_pointer_value(env, insn->src_reg)) { 1819 verbose("R%d partial copy of pointer\n", 1820 insn->src_reg); 1821 return -EACCES; 1822 } 1823 mark_reg_unknown_value(regs, insn->dst_reg); 1824 } 1825 } else { 1826 /* case: R = imm 1827 * remember the value we stored into this reg 1828 */ 1829 regs[insn->dst_reg].type = CONST_IMM; 1830 regs[insn->dst_reg].imm = insn->imm; 1831 regs[insn->dst_reg].max_value = insn->imm; 1832 regs[insn->dst_reg].min_value = insn->imm; 1833 } 1834 1835 } else if (opcode > BPF_END) { 1836 verbose("invalid BPF_ALU opcode %x\n", opcode); 1837 return -EINVAL; 1838 1839 } else { /* all other ALU ops: and, sub, xor, add, ... */ 1840 1841 if (BPF_SRC(insn->code) == BPF_X) { 1842 if (insn->imm != 0 || insn->off != 0) { 1843 verbose("BPF_ALU uses reserved fields\n"); 1844 return -EINVAL; 1845 } 1846 /* check src1 operand */ 1847 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1848 if (err) 1849 return err; 1850 } else { 1851 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 1852 verbose("BPF_ALU uses reserved fields\n"); 1853 return -EINVAL; 1854 } 1855 } 1856 1857 /* check src2 operand */ 1858 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1859 if (err) 1860 return err; 1861 1862 if ((opcode == BPF_MOD || opcode == BPF_DIV) && 1863 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { 1864 verbose("div by zero\n"); 1865 return -EINVAL; 1866 } 1867 1868 if ((opcode == BPF_LSH || opcode == BPF_RSH || 1869 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { 1870 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; 1871 1872 if (insn->imm < 0 || insn->imm >= size) { 1873 verbose("invalid shift %d\n", insn->imm); 1874 return -EINVAL; 1875 } 1876 } 1877 1878 /* check dest operand */ 1879 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 1880 if (err) 1881 return err; 1882 1883 dst_reg = ®s[insn->dst_reg]; 1884 1885 /* first we want to adjust our ranges. */ 1886 adjust_reg_min_max_vals(env, insn); 1887 1888 /* pattern match 'bpf_add Rx, imm' instruction */ 1889 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && 1890 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { 1891 dst_reg->type = PTR_TO_STACK; 1892 dst_reg->imm = insn->imm; 1893 return 0; 1894 } else if (opcode == BPF_ADD && 1895 BPF_CLASS(insn->code) == BPF_ALU64 && 1896 (dst_reg->type == PTR_TO_PACKET || 1897 (BPF_SRC(insn->code) == BPF_X && 1898 regs[insn->src_reg].type == PTR_TO_PACKET))) { 1899 /* ptr_to_packet += K|X */ 1900 return check_packet_ptr_add(env, insn); 1901 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 1902 dst_reg->type == UNKNOWN_VALUE && 1903 env->allow_ptr_leaks) { 1904 /* unknown += K|X */ 1905 return evaluate_reg_alu(env, insn); 1906 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 1907 dst_reg->type == CONST_IMM && 1908 env->allow_ptr_leaks) { 1909 /* reg_imm += K|X */ 1910 return evaluate_reg_imm_alu(env, insn); 1911 } else if (is_pointer_value(env, insn->dst_reg)) { 1912 verbose("R%d pointer arithmetic prohibited\n", 1913 insn->dst_reg); 1914 return -EACCES; 1915 } else if (BPF_SRC(insn->code) == BPF_X && 1916 is_pointer_value(env, insn->src_reg)) { 1917 verbose("R%d pointer arithmetic prohibited\n", 1918 insn->src_reg); 1919 return -EACCES; 1920 } 1921 1922 /* If we did pointer math on a map value then just set it to our 1923 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or 1924 * loads to this register appropriately, otherwise just mark the 1925 * register as unknown. 1926 */ 1927 if (env->allow_ptr_leaks && 1928 (dst_reg->type == PTR_TO_MAP_VALUE || 1929 dst_reg->type == PTR_TO_MAP_VALUE_ADJ)) 1930 dst_reg->type = PTR_TO_MAP_VALUE_ADJ; 1931 else 1932 mark_reg_unknown_value(regs, insn->dst_reg); 1933 } 1934 1935 return 0; 1936 } 1937 1938 static void find_good_pkt_pointers(struct bpf_verifier_state *state, 1939 struct bpf_reg_state *dst_reg) 1940 { 1941 struct bpf_reg_state *regs = state->regs, *reg; 1942 int i; 1943 1944 /* LLVM can generate two kind of checks: 1945 * 1946 * Type 1: 1947 * 1948 * r2 = r3; 1949 * r2 += 8; 1950 * if (r2 > pkt_end) goto <handle exception> 1951 * <access okay> 1952 * 1953 * Where: 1954 * r2 == dst_reg, pkt_end == src_reg 1955 * r2=pkt(id=n,off=8,r=0) 1956 * r3=pkt(id=n,off=0,r=0) 1957 * 1958 * Type 2: 1959 * 1960 * r2 = r3; 1961 * r2 += 8; 1962 * if (pkt_end >= r2) goto <access okay> 1963 * <handle exception> 1964 * 1965 * Where: 1966 * pkt_end == dst_reg, r2 == src_reg 1967 * r2=pkt(id=n,off=8,r=0) 1968 * r3=pkt(id=n,off=0,r=0) 1969 * 1970 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) 1971 * so that range of bytes [r3, r3 + 8) is safe to access. 1972 */ 1973 1974 for (i = 0; i < MAX_BPF_REG; i++) 1975 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) 1976 regs[i].range = dst_reg->off; 1977 1978 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1979 if (state->stack_slot_type[i] != STACK_SPILL) 1980 continue; 1981 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1982 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) 1983 reg->range = dst_reg->off; 1984 } 1985 } 1986 1987 /* Adjusts the register min/max values in the case that the dst_reg is the 1988 * variable register that we are working on, and src_reg is a constant or we're 1989 * simply doing a BPF_K check. 1990 */ 1991 static void reg_set_min_max(struct bpf_reg_state *true_reg, 1992 struct bpf_reg_state *false_reg, u64 val, 1993 u8 opcode) 1994 { 1995 switch (opcode) { 1996 case BPF_JEQ: 1997 /* If this is false then we know nothing Jon Snow, but if it is 1998 * true then we know for sure. 1999 */ 2000 true_reg->max_value = true_reg->min_value = val; 2001 break; 2002 case BPF_JNE: 2003 /* If this is true we know nothing Jon Snow, but if it is false 2004 * we know the value for sure; 2005 */ 2006 false_reg->max_value = false_reg->min_value = val; 2007 break; 2008 case BPF_JGT: 2009 /* Unsigned comparison, the minimum value is 0. */ 2010 false_reg->min_value = 0; 2011 /* fallthrough */ 2012 case BPF_JSGT: 2013 /* If this is false then we know the maximum val is val, 2014 * otherwise we know the min val is val+1. 2015 */ 2016 false_reg->max_value = val; 2017 true_reg->min_value = val + 1; 2018 break; 2019 case BPF_JGE: 2020 /* Unsigned comparison, the minimum value is 0. */ 2021 false_reg->min_value = 0; 2022 /* fallthrough */ 2023 case BPF_JSGE: 2024 /* If this is false then we know the maximum value is val - 1, 2025 * otherwise we know the mimimum value is val. 2026 */ 2027 false_reg->max_value = val - 1; 2028 true_reg->min_value = val; 2029 break; 2030 default: 2031 break; 2032 } 2033 2034 check_reg_overflow(false_reg); 2035 check_reg_overflow(true_reg); 2036 } 2037 2038 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg 2039 * is the variable reg. 2040 */ 2041 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, 2042 struct bpf_reg_state *false_reg, u64 val, 2043 u8 opcode) 2044 { 2045 switch (opcode) { 2046 case BPF_JEQ: 2047 /* If this is false then we know nothing Jon Snow, but if it is 2048 * true then we know for sure. 2049 */ 2050 true_reg->max_value = true_reg->min_value = val; 2051 break; 2052 case BPF_JNE: 2053 /* If this is true we know nothing Jon Snow, but if it is false 2054 * we know the value for sure; 2055 */ 2056 false_reg->max_value = false_reg->min_value = val; 2057 break; 2058 case BPF_JGT: 2059 /* Unsigned comparison, the minimum value is 0. */ 2060 true_reg->min_value = 0; 2061 /* fallthrough */ 2062 case BPF_JSGT: 2063 /* 2064 * If this is false, then the val is <= the register, if it is 2065 * true the register <= to the val. 2066 */ 2067 false_reg->min_value = val; 2068 true_reg->max_value = val - 1; 2069 break; 2070 case BPF_JGE: 2071 /* Unsigned comparison, the minimum value is 0. */ 2072 true_reg->min_value = 0; 2073 /* fallthrough */ 2074 case BPF_JSGE: 2075 /* If this is false then constant < register, if it is true then 2076 * the register < constant. 2077 */ 2078 false_reg->min_value = val + 1; 2079 true_reg->max_value = val; 2080 break; 2081 default: 2082 break; 2083 } 2084 2085 check_reg_overflow(false_reg); 2086 check_reg_overflow(true_reg); 2087 } 2088 2089 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, 2090 enum bpf_reg_type type) 2091 { 2092 struct bpf_reg_state *reg = ®s[regno]; 2093 2094 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { 2095 reg->type = type; 2096 /* We don't need id from this point onwards anymore, thus we 2097 * should better reset it, so that state pruning has chances 2098 * to take effect. 2099 */ 2100 reg->id = 0; 2101 if (type == UNKNOWN_VALUE) 2102 __mark_reg_unknown_value(regs, regno); 2103 } 2104 } 2105 2106 /* The logic is similar to find_good_pkt_pointers(), both could eventually 2107 * be folded together at some point. 2108 */ 2109 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, 2110 enum bpf_reg_type type) 2111 { 2112 struct bpf_reg_state *regs = state->regs; 2113 u32 id = regs[regno].id; 2114 int i; 2115 2116 for (i = 0; i < MAX_BPF_REG; i++) 2117 mark_map_reg(regs, i, id, type); 2118 2119 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 2120 if (state->stack_slot_type[i] != STACK_SPILL) 2121 continue; 2122 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type); 2123 } 2124 } 2125 2126 static int check_cond_jmp_op(struct bpf_verifier_env *env, 2127 struct bpf_insn *insn, int *insn_idx) 2128 { 2129 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; 2130 struct bpf_reg_state *regs = this_branch->regs, *dst_reg; 2131 u8 opcode = BPF_OP(insn->code); 2132 int err; 2133 2134 if (opcode > BPF_EXIT) { 2135 verbose("invalid BPF_JMP opcode %x\n", opcode); 2136 return -EINVAL; 2137 } 2138 2139 if (BPF_SRC(insn->code) == BPF_X) { 2140 if (insn->imm != 0) { 2141 verbose("BPF_JMP uses reserved fields\n"); 2142 return -EINVAL; 2143 } 2144 2145 /* check src1 operand */ 2146 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2147 if (err) 2148 return err; 2149 2150 if (is_pointer_value(env, insn->src_reg)) { 2151 verbose("R%d pointer comparison prohibited\n", 2152 insn->src_reg); 2153 return -EACCES; 2154 } 2155 } else { 2156 if (insn->src_reg != BPF_REG_0) { 2157 verbose("BPF_JMP uses reserved fields\n"); 2158 return -EINVAL; 2159 } 2160 } 2161 2162 /* check src2 operand */ 2163 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2164 if (err) 2165 return err; 2166 2167 dst_reg = ®s[insn->dst_reg]; 2168 2169 /* detect if R == 0 where R was initialized to zero earlier */ 2170 if (BPF_SRC(insn->code) == BPF_K && 2171 (opcode == BPF_JEQ || opcode == BPF_JNE) && 2172 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { 2173 if (opcode == BPF_JEQ) { 2174 /* if (imm == imm) goto pc+off; 2175 * only follow the goto, ignore fall-through 2176 */ 2177 *insn_idx += insn->off; 2178 return 0; 2179 } else { 2180 /* if (imm != imm) goto pc+off; 2181 * only follow fall-through branch, since 2182 * that's where the program will go 2183 */ 2184 return 0; 2185 } 2186 } 2187 2188 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); 2189 if (!other_branch) 2190 return -EFAULT; 2191 2192 /* detect if we are comparing against a constant value so we can adjust 2193 * our min/max values for our dst register. 2194 */ 2195 if (BPF_SRC(insn->code) == BPF_X) { 2196 if (regs[insn->src_reg].type == CONST_IMM) 2197 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2198 dst_reg, regs[insn->src_reg].imm, 2199 opcode); 2200 else if (dst_reg->type == CONST_IMM) 2201 reg_set_min_max_inv(&other_branch->regs[insn->src_reg], 2202 ®s[insn->src_reg], dst_reg->imm, 2203 opcode); 2204 } else { 2205 reg_set_min_max(&other_branch->regs[insn->dst_reg], 2206 dst_reg, insn->imm, opcode); 2207 } 2208 2209 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ 2210 if (BPF_SRC(insn->code) == BPF_K && 2211 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && 2212 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { 2213 /* Mark all identical map registers in each branch as either 2214 * safe or unknown depending R == 0 or R != 0 conditional. 2215 */ 2216 mark_map_regs(this_branch, insn->dst_reg, 2217 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE); 2218 mark_map_regs(other_branch, insn->dst_reg, 2219 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE); 2220 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && 2221 dst_reg->type == PTR_TO_PACKET && 2222 regs[insn->src_reg].type == PTR_TO_PACKET_END) { 2223 find_good_pkt_pointers(this_branch, dst_reg); 2224 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && 2225 dst_reg->type == PTR_TO_PACKET_END && 2226 regs[insn->src_reg].type == PTR_TO_PACKET) { 2227 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); 2228 } else if (is_pointer_value(env, insn->dst_reg)) { 2229 verbose("R%d pointer comparison prohibited\n", insn->dst_reg); 2230 return -EACCES; 2231 } 2232 if (log_level) 2233 print_verifier_state(this_branch); 2234 return 0; 2235 } 2236 2237 /* return the map pointer stored inside BPF_LD_IMM64 instruction */ 2238 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) 2239 { 2240 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; 2241 2242 return (struct bpf_map *) (unsigned long) imm64; 2243 } 2244 2245 /* verify BPF_LD_IMM64 instruction */ 2246 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) 2247 { 2248 struct bpf_reg_state *regs = env->cur_state.regs; 2249 int err; 2250 2251 if (BPF_SIZE(insn->code) != BPF_DW) { 2252 verbose("invalid BPF_LD_IMM insn\n"); 2253 return -EINVAL; 2254 } 2255 if (insn->off != 0) { 2256 verbose("BPF_LD_IMM64 uses reserved fields\n"); 2257 return -EINVAL; 2258 } 2259 2260 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 2261 if (err) 2262 return err; 2263 2264 if (insn->src_reg == 0) { 2265 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; 2266 2267 regs[insn->dst_reg].type = CONST_IMM; 2268 regs[insn->dst_reg].imm = imm; 2269 return 0; 2270 } 2271 2272 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ 2273 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); 2274 2275 regs[insn->dst_reg].type = CONST_PTR_TO_MAP; 2276 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); 2277 return 0; 2278 } 2279 2280 static bool may_access_skb(enum bpf_prog_type type) 2281 { 2282 switch (type) { 2283 case BPF_PROG_TYPE_SOCKET_FILTER: 2284 case BPF_PROG_TYPE_SCHED_CLS: 2285 case BPF_PROG_TYPE_SCHED_ACT: 2286 return true; 2287 default: 2288 return false; 2289 } 2290 } 2291 2292 /* verify safety of LD_ABS|LD_IND instructions: 2293 * - they can only appear in the programs where ctx == skb 2294 * - since they are wrappers of function calls, they scratch R1-R5 registers, 2295 * preserve R6-R9, and store return value into R0 2296 * 2297 * Implicit input: 2298 * ctx == skb == R6 == CTX 2299 * 2300 * Explicit input: 2301 * SRC == any register 2302 * IMM == 32-bit immediate 2303 * 2304 * Output: 2305 * R0 - 8/16/32-bit skb data converted to cpu endianness 2306 */ 2307 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) 2308 { 2309 struct bpf_reg_state *regs = env->cur_state.regs; 2310 u8 mode = BPF_MODE(insn->code); 2311 struct bpf_reg_state *reg; 2312 int i, err; 2313 2314 if (!may_access_skb(env->prog->type)) { 2315 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); 2316 return -EINVAL; 2317 } 2318 2319 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || 2320 BPF_SIZE(insn->code) == BPF_DW || 2321 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { 2322 verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); 2323 return -EINVAL; 2324 } 2325 2326 /* check whether implicit source operand (register R6) is readable */ 2327 err = check_reg_arg(regs, BPF_REG_6, SRC_OP); 2328 if (err) 2329 return err; 2330 2331 if (regs[BPF_REG_6].type != PTR_TO_CTX) { 2332 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); 2333 return -EINVAL; 2334 } 2335 2336 if (mode == BPF_IND) { 2337 /* check explicit source operand */ 2338 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2339 if (err) 2340 return err; 2341 } 2342 2343 /* reset caller saved regs to unreadable */ 2344 for (i = 0; i < CALLER_SAVED_REGS; i++) { 2345 reg = regs + caller_saved[i]; 2346 reg->type = NOT_INIT; 2347 reg->imm = 0; 2348 } 2349 2350 /* mark destination R0 register as readable, since it contains 2351 * the value fetched from the packet 2352 */ 2353 regs[BPF_REG_0].type = UNKNOWN_VALUE; 2354 return 0; 2355 } 2356 2357 /* non-recursive DFS pseudo code 2358 * 1 procedure DFS-iterative(G,v): 2359 * 2 label v as discovered 2360 * 3 let S be a stack 2361 * 4 S.push(v) 2362 * 5 while S is not empty 2363 * 6 t <- S.pop() 2364 * 7 if t is what we're looking for: 2365 * 8 return t 2366 * 9 for all edges e in G.adjacentEdges(t) do 2367 * 10 if edge e is already labelled 2368 * 11 continue with the next edge 2369 * 12 w <- G.adjacentVertex(t,e) 2370 * 13 if vertex w is not discovered and not explored 2371 * 14 label e as tree-edge 2372 * 15 label w as discovered 2373 * 16 S.push(w) 2374 * 17 continue at 5 2375 * 18 else if vertex w is discovered 2376 * 19 label e as back-edge 2377 * 20 else 2378 * 21 // vertex w is explored 2379 * 22 label e as forward- or cross-edge 2380 * 23 label t as explored 2381 * 24 S.pop() 2382 * 2383 * convention: 2384 * 0x10 - discovered 2385 * 0x11 - discovered and fall-through edge labelled 2386 * 0x12 - discovered and fall-through and branch edges labelled 2387 * 0x20 - explored 2388 */ 2389 2390 enum { 2391 DISCOVERED = 0x10, 2392 EXPLORED = 0x20, 2393 FALLTHROUGH = 1, 2394 BRANCH = 2, 2395 }; 2396 2397 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) 2398 2399 static int *insn_stack; /* stack of insns to process */ 2400 static int cur_stack; /* current stack index */ 2401 static int *insn_state; 2402 2403 /* t, w, e - match pseudo-code above: 2404 * t - index of current instruction 2405 * w - next instruction 2406 * e - edge 2407 */ 2408 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) 2409 { 2410 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) 2411 return 0; 2412 2413 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) 2414 return 0; 2415 2416 if (w < 0 || w >= env->prog->len) { 2417 verbose("jump out of range from insn %d to %d\n", t, w); 2418 return -EINVAL; 2419 } 2420 2421 if (e == BRANCH) 2422 /* mark branch target for state pruning */ 2423 env->explored_states[w] = STATE_LIST_MARK; 2424 2425 if (insn_state[w] == 0) { 2426 /* tree-edge */ 2427 insn_state[t] = DISCOVERED | e; 2428 insn_state[w] = DISCOVERED; 2429 if (cur_stack >= env->prog->len) 2430 return -E2BIG; 2431 insn_stack[cur_stack++] = w; 2432 return 1; 2433 } else if ((insn_state[w] & 0xF0) == DISCOVERED) { 2434 verbose("back-edge from insn %d to %d\n", t, w); 2435 return -EINVAL; 2436 } else if (insn_state[w] == EXPLORED) { 2437 /* forward- or cross-edge */ 2438 insn_state[t] = DISCOVERED | e; 2439 } else { 2440 verbose("insn state internal bug\n"); 2441 return -EFAULT; 2442 } 2443 return 0; 2444 } 2445 2446 /* non-recursive depth-first-search to detect loops in BPF program 2447 * loop == back-edge in directed graph 2448 */ 2449 static int check_cfg(struct bpf_verifier_env *env) 2450 { 2451 struct bpf_insn *insns = env->prog->insnsi; 2452 int insn_cnt = env->prog->len; 2453 int ret = 0; 2454 int i, t; 2455 2456 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2457 if (!insn_state) 2458 return -ENOMEM; 2459 2460 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 2461 if (!insn_stack) { 2462 kfree(insn_state); 2463 return -ENOMEM; 2464 } 2465 2466 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ 2467 insn_stack[0] = 0; /* 0 is the first instruction */ 2468 cur_stack = 1; 2469 2470 peek_stack: 2471 if (cur_stack == 0) 2472 goto check_state; 2473 t = insn_stack[cur_stack - 1]; 2474 2475 if (BPF_CLASS(insns[t].code) == BPF_JMP) { 2476 u8 opcode = BPF_OP(insns[t].code); 2477 2478 if (opcode == BPF_EXIT) { 2479 goto mark_explored; 2480 } else if (opcode == BPF_CALL) { 2481 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2482 if (ret == 1) 2483 goto peek_stack; 2484 else if (ret < 0) 2485 goto err_free; 2486 if (t + 1 < insn_cnt) 2487 env->explored_states[t + 1] = STATE_LIST_MARK; 2488 } else if (opcode == BPF_JA) { 2489 if (BPF_SRC(insns[t].code) != BPF_K) { 2490 ret = -EINVAL; 2491 goto err_free; 2492 } 2493 /* unconditional jump with single edge */ 2494 ret = push_insn(t, t + insns[t].off + 1, 2495 FALLTHROUGH, env); 2496 if (ret == 1) 2497 goto peek_stack; 2498 else if (ret < 0) 2499 goto err_free; 2500 /* tell verifier to check for equivalent states 2501 * after every call and jump 2502 */ 2503 if (t + 1 < insn_cnt) 2504 env->explored_states[t + 1] = STATE_LIST_MARK; 2505 } else { 2506 /* conditional jump with two edges */ 2507 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2508 if (ret == 1) 2509 goto peek_stack; 2510 else if (ret < 0) 2511 goto err_free; 2512 2513 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); 2514 if (ret == 1) 2515 goto peek_stack; 2516 else if (ret < 0) 2517 goto err_free; 2518 } 2519 } else { 2520 /* all other non-branch instructions with single 2521 * fall-through edge 2522 */ 2523 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2524 if (ret == 1) 2525 goto peek_stack; 2526 else if (ret < 0) 2527 goto err_free; 2528 } 2529 2530 mark_explored: 2531 insn_state[t] = EXPLORED; 2532 if (cur_stack-- <= 0) { 2533 verbose("pop stack internal bug\n"); 2534 ret = -EFAULT; 2535 goto err_free; 2536 } 2537 goto peek_stack; 2538 2539 check_state: 2540 for (i = 0; i < insn_cnt; i++) { 2541 if (insn_state[i] != EXPLORED) { 2542 verbose("unreachable insn %d\n", i); 2543 ret = -EINVAL; 2544 goto err_free; 2545 } 2546 } 2547 ret = 0; /* cfg looks good */ 2548 2549 err_free: 2550 kfree(insn_state); 2551 kfree(insn_stack); 2552 return ret; 2553 } 2554 2555 /* the following conditions reduce the number of explored insns 2556 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet 2557 */ 2558 static bool compare_ptrs_to_packet(struct bpf_reg_state *old, 2559 struct bpf_reg_state *cur) 2560 { 2561 if (old->id != cur->id) 2562 return false; 2563 2564 /* old ptr_to_packet is more conservative, since it allows smaller 2565 * range. Ex: 2566 * old(off=0,r=10) is equal to cur(off=0,r=20), because 2567 * old(off=0,r=10) means that with range=10 the verifier proceeded 2568 * further and found no issues with the program. Now we're in the same 2569 * spot with cur(off=0,r=20), so we're safe too, since anything further 2570 * will only be looking at most 10 bytes after this pointer. 2571 */ 2572 if (old->off == cur->off && old->range < cur->range) 2573 return true; 2574 2575 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) 2576 * since both cannot be used for packet access and safe(old) 2577 * pointer has smaller off that could be used for further 2578 * 'if (ptr > data_end)' check 2579 * Ex: 2580 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean 2581 * that we cannot access the packet. 2582 * The safe range is: 2583 * [ptr, ptr + range - off) 2584 * so whenever off >=range, it means no safe bytes from this pointer. 2585 * When comparing old->off <= cur->off, it means that older code 2586 * went with smaller offset and that offset was later 2587 * used to figure out the safe range after 'if (ptr > data_end)' check 2588 * Say, 'old' state was explored like: 2589 * ... R3(off=0, r=0) 2590 * R4 = R3 + 20 2591 * ... now R4(off=20,r=0) <-- here 2592 * if (R4 > data_end) 2593 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. 2594 * ... the code further went all the way to bpf_exit. 2595 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). 2596 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier 2597 * goes further, such cur_R4 will give larger safe packet range after 2598 * 'if (R4 > data_end)' and all further insn were already good with r=20, 2599 * so they will be good with r=30 and we can prune the search. 2600 */ 2601 if (old->off <= cur->off && 2602 old->off >= old->range && cur->off >= cur->range) 2603 return true; 2604 2605 return false; 2606 } 2607 2608 /* compare two verifier states 2609 * 2610 * all states stored in state_list are known to be valid, since 2611 * verifier reached 'bpf_exit' instruction through them 2612 * 2613 * this function is called when verifier exploring different branches of 2614 * execution popped from the state stack. If it sees an old state that has 2615 * more strict register state and more strict stack state then this execution 2616 * branch doesn't need to be explored further, since verifier already 2617 * concluded that more strict state leads to valid finish. 2618 * 2619 * Therefore two states are equivalent if register state is more conservative 2620 * and explored stack state is more conservative than the current one. 2621 * Example: 2622 * explored current 2623 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) 2624 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) 2625 * 2626 * In other words if current stack state (one being explored) has more 2627 * valid slots than old one that already passed validation, it means 2628 * the verifier can stop exploring and conclude that current state is valid too 2629 * 2630 * Similarly with registers. If explored state has register type as invalid 2631 * whereas register type in current state is meaningful, it means that 2632 * the current state will reach 'bpf_exit' instruction safely 2633 */ 2634 static bool states_equal(struct bpf_verifier_env *env, 2635 struct bpf_verifier_state *old, 2636 struct bpf_verifier_state *cur) 2637 { 2638 bool varlen_map_access = env->varlen_map_value_access; 2639 struct bpf_reg_state *rold, *rcur; 2640 int i; 2641 2642 for (i = 0; i < MAX_BPF_REG; i++) { 2643 rold = &old->regs[i]; 2644 rcur = &cur->regs[i]; 2645 2646 if (memcmp(rold, rcur, sizeof(*rold)) == 0) 2647 continue; 2648 2649 /* If the ranges were not the same, but everything else was and 2650 * we didn't do a variable access into a map then we are a-ok. 2651 */ 2652 if (!varlen_map_access && 2653 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0) 2654 continue; 2655 2656 /* If we didn't map access then again we don't care about the 2657 * mismatched range values and it's ok if our old type was 2658 * UNKNOWN and we didn't go to a NOT_INIT'ed reg. 2659 */ 2660 if (rold->type == NOT_INIT || 2661 (!varlen_map_access && rold->type == UNKNOWN_VALUE && 2662 rcur->type != NOT_INIT)) 2663 continue; 2664 2665 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && 2666 compare_ptrs_to_packet(rold, rcur)) 2667 continue; 2668 2669 return false; 2670 } 2671 2672 for (i = 0; i < MAX_BPF_STACK; i++) { 2673 if (old->stack_slot_type[i] == STACK_INVALID) 2674 continue; 2675 if (old->stack_slot_type[i] != cur->stack_slot_type[i]) 2676 /* Ex: old explored (safe) state has STACK_SPILL in 2677 * this stack slot, but current has has STACK_MISC -> 2678 * this verifier states are not equivalent, 2679 * return false to continue verification of this path 2680 */ 2681 return false; 2682 if (i % BPF_REG_SIZE) 2683 continue; 2684 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], 2685 &cur->spilled_regs[i / BPF_REG_SIZE], 2686 sizeof(old->spilled_regs[0]))) 2687 /* when explored and current stack slot types are 2688 * the same, check that stored pointers types 2689 * are the same as well. 2690 * Ex: explored safe path could have stored 2691 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} 2692 * but current path has stored: 2693 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} 2694 * such verifier states are not equivalent. 2695 * return false to continue verification of this path 2696 */ 2697 return false; 2698 else 2699 continue; 2700 } 2701 return true; 2702 } 2703 2704 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) 2705 { 2706 struct bpf_verifier_state_list *new_sl; 2707 struct bpf_verifier_state_list *sl; 2708 2709 sl = env->explored_states[insn_idx]; 2710 if (!sl) 2711 /* this 'insn_idx' instruction wasn't marked, so we will not 2712 * be doing state search here 2713 */ 2714 return 0; 2715 2716 while (sl != STATE_LIST_MARK) { 2717 if (states_equal(env, &sl->state, &env->cur_state)) 2718 /* reached equivalent register/stack state, 2719 * prune the search 2720 */ 2721 return 1; 2722 sl = sl->next; 2723 } 2724 2725 /* there were no equivalent states, remember current one. 2726 * technically the current state is not proven to be safe yet, 2727 * but it will either reach bpf_exit (which means it's safe) or 2728 * it will be rejected. Since there are no loops, we won't be 2729 * seeing this 'insn_idx' instruction again on the way to bpf_exit 2730 */ 2731 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); 2732 if (!new_sl) 2733 return -ENOMEM; 2734 2735 /* add new state to the head of linked list */ 2736 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); 2737 new_sl->next = env->explored_states[insn_idx]; 2738 env->explored_states[insn_idx] = new_sl; 2739 return 0; 2740 } 2741 2742 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, 2743 int insn_idx, int prev_insn_idx) 2744 { 2745 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) 2746 return 0; 2747 2748 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); 2749 } 2750 2751 static int do_check(struct bpf_verifier_env *env) 2752 { 2753 struct bpf_verifier_state *state = &env->cur_state; 2754 struct bpf_insn *insns = env->prog->insnsi; 2755 struct bpf_reg_state *regs = state->regs; 2756 int insn_cnt = env->prog->len; 2757 int insn_idx, prev_insn_idx = 0; 2758 int insn_processed = 0; 2759 bool do_print_state = false; 2760 2761 init_reg_state(regs); 2762 insn_idx = 0; 2763 env->varlen_map_value_access = false; 2764 for (;;) { 2765 struct bpf_insn *insn; 2766 u8 class; 2767 int err; 2768 2769 if (insn_idx >= insn_cnt) { 2770 verbose("invalid insn idx %d insn_cnt %d\n", 2771 insn_idx, insn_cnt); 2772 return -EFAULT; 2773 } 2774 2775 insn = &insns[insn_idx]; 2776 class = BPF_CLASS(insn->code); 2777 2778 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { 2779 verbose("BPF program is too large. Processed %d insn\n", 2780 insn_processed); 2781 return -E2BIG; 2782 } 2783 2784 err = is_state_visited(env, insn_idx); 2785 if (err < 0) 2786 return err; 2787 if (err == 1) { 2788 /* found equivalent state, can prune the search */ 2789 if (log_level) { 2790 if (do_print_state) 2791 verbose("\nfrom %d to %d: safe\n", 2792 prev_insn_idx, insn_idx); 2793 else 2794 verbose("%d: safe\n", insn_idx); 2795 } 2796 goto process_bpf_exit; 2797 } 2798 2799 if (log_level && do_print_state) { 2800 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); 2801 print_verifier_state(&env->cur_state); 2802 do_print_state = false; 2803 } 2804 2805 if (log_level) { 2806 verbose("%d: ", insn_idx); 2807 print_bpf_insn(insn); 2808 } 2809 2810 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); 2811 if (err) 2812 return err; 2813 2814 if (class == BPF_ALU || class == BPF_ALU64) { 2815 err = check_alu_op(env, insn); 2816 if (err) 2817 return err; 2818 2819 } else if (class == BPF_LDX) { 2820 enum bpf_reg_type *prev_src_type, src_reg_type; 2821 2822 /* check for reserved fields is already done */ 2823 2824 /* check src operand */ 2825 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2826 if (err) 2827 return err; 2828 2829 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 2830 if (err) 2831 return err; 2832 2833 src_reg_type = regs[insn->src_reg].type; 2834 2835 /* check that memory (src_reg + off) is readable, 2836 * the state of dst_reg will be updated by this func 2837 */ 2838 err = check_mem_access(env, insn->src_reg, insn->off, 2839 BPF_SIZE(insn->code), BPF_READ, 2840 insn->dst_reg); 2841 if (err) 2842 return err; 2843 2844 if (BPF_SIZE(insn->code) != BPF_W && 2845 BPF_SIZE(insn->code) != BPF_DW) { 2846 insn_idx++; 2847 continue; 2848 } 2849 2850 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; 2851 2852 if (*prev_src_type == NOT_INIT) { 2853 /* saw a valid insn 2854 * dst_reg = *(u32 *)(src_reg + off) 2855 * save type to validate intersecting paths 2856 */ 2857 *prev_src_type = src_reg_type; 2858 2859 } else if (src_reg_type != *prev_src_type && 2860 (src_reg_type == PTR_TO_CTX || 2861 *prev_src_type == PTR_TO_CTX)) { 2862 /* ABuser program is trying to use the same insn 2863 * dst_reg = *(u32*) (src_reg + off) 2864 * with different pointer types: 2865 * src_reg == ctx in one branch and 2866 * src_reg == stack|map in some other branch. 2867 * Reject it. 2868 */ 2869 verbose("same insn cannot be used with different pointers\n"); 2870 return -EINVAL; 2871 } 2872 2873 } else if (class == BPF_STX) { 2874 enum bpf_reg_type *prev_dst_type, dst_reg_type; 2875 2876 if (BPF_MODE(insn->code) == BPF_XADD) { 2877 err = check_xadd(env, insn); 2878 if (err) 2879 return err; 2880 insn_idx++; 2881 continue; 2882 } 2883 2884 /* check src1 operand */ 2885 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2886 if (err) 2887 return err; 2888 /* check src2 operand */ 2889 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2890 if (err) 2891 return err; 2892 2893 dst_reg_type = regs[insn->dst_reg].type; 2894 2895 /* check that memory (dst_reg + off) is writeable */ 2896 err = check_mem_access(env, insn->dst_reg, insn->off, 2897 BPF_SIZE(insn->code), BPF_WRITE, 2898 insn->src_reg); 2899 if (err) 2900 return err; 2901 2902 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; 2903 2904 if (*prev_dst_type == NOT_INIT) { 2905 *prev_dst_type = dst_reg_type; 2906 } else if (dst_reg_type != *prev_dst_type && 2907 (dst_reg_type == PTR_TO_CTX || 2908 *prev_dst_type == PTR_TO_CTX)) { 2909 verbose("same insn cannot be used with different pointers\n"); 2910 return -EINVAL; 2911 } 2912 2913 } else if (class == BPF_ST) { 2914 if (BPF_MODE(insn->code) != BPF_MEM || 2915 insn->src_reg != BPF_REG_0) { 2916 verbose("BPF_ST uses reserved fields\n"); 2917 return -EINVAL; 2918 } 2919 /* check src operand */ 2920 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2921 if (err) 2922 return err; 2923 2924 /* check that memory (dst_reg + off) is writeable */ 2925 err = check_mem_access(env, insn->dst_reg, insn->off, 2926 BPF_SIZE(insn->code), BPF_WRITE, 2927 -1); 2928 if (err) 2929 return err; 2930 2931 } else if (class == BPF_JMP) { 2932 u8 opcode = BPF_OP(insn->code); 2933 2934 if (opcode == BPF_CALL) { 2935 if (BPF_SRC(insn->code) != BPF_K || 2936 insn->off != 0 || 2937 insn->src_reg != BPF_REG_0 || 2938 insn->dst_reg != BPF_REG_0) { 2939 verbose("BPF_CALL uses reserved fields\n"); 2940 return -EINVAL; 2941 } 2942 2943 err = check_call(env, insn->imm); 2944 if (err) 2945 return err; 2946 2947 } else if (opcode == BPF_JA) { 2948 if (BPF_SRC(insn->code) != BPF_K || 2949 insn->imm != 0 || 2950 insn->src_reg != BPF_REG_0 || 2951 insn->dst_reg != BPF_REG_0) { 2952 verbose("BPF_JA uses reserved fields\n"); 2953 return -EINVAL; 2954 } 2955 2956 insn_idx += insn->off + 1; 2957 continue; 2958 2959 } else if (opcode == BPF_EXIT) { 2960 if (BPF_SRC(insn->code) != BPF_K || 2961 insn->imm != 0 || 2962 insn->src_reg != BPF_REG_0 || 2963 insn->dst_reg != BPF_REG_0) { 2964 verbose("BPF_EXIT uses reserved fields\n"); 2965 return -EINVAL; 2966 } 2967 2968 /* eBPF calling convetion is such that R0 is used 2969 * to return the value from eBPF program. 2970 * Make sure that it's readable at this time 2971 * of bpf_exit, which means that program wrote 2972 * something into it earlier 2973 */ 2974 err = check_reg_arg(regs, BPF_REG_0, SRC_OP); 2975 if (err) 2976 return err; 2977 2978 if (is_pointer_value(env, BPF_REG_0)) { 2979 verbose("R0 leaks addr as return value\n"); 2980 return -EACCES; 2981 } 2982 2983 process_bpf_exit: 2984 insn_idx = pop_stack(env, &prev_insn_idx); 2985 if (insn_idx < 0) { 2986 break; 2987 } else { 2988 do_print_state = true; 2989 continue; 2990 } 2991 } else { 2992 err = check_cond_jmp_op(env, insn, &insn_idx); 2993 if (err) 2994 return err; 2995 } 2996 } else if (class == BPF_LD) { 2997 u8 mode = BPF_MODE(insn->code); 2998 2999 if (mode == BPF_ABS || mode == BPF_IND) { 3000 err = check_ld_abs(env, insn); 3001 if (err) 3002 return err; 3003 3004 } else if (mode == BPF_IMM) { 3005 err = check_ld_imm(env, insn); 3006 if (err) 3007 return err; 3008 3009 insn_idx++; 3010 } else { 3011 verbose("invalid BPF_LD mode\n"); 3012 return -EINVAL; 3013 } 3014 reset_reg_range_values(regs, insn->dst_reg); 3015 } else { 3016 verbose("unknown insn class %d\n", class); 3017 return -EINVAL; 3018 } 3019 3020 insn_idx++; 3021 } 3022 3023 verbose("processed %d insns\n", insn_processed); 3024 return 0; 3025 } 3026 3027 static int check_map_prog_compatibility(struct bpf_map *map, 3028 struct bpf_prog *prog) 3029 3030 { 3031 if (prog->type == BPF_PROG_TYPE_PERF_EVENT && 3032 (map->map_type == BPF_MAP_TYPE_HASH || 3033 map->map_type == BPF_MAP_TYPE_PERCPU_HASH) && 3034 (map->map_flags & BPF_F_NO_PREALLOC)) { 3035 verbose("perf_event programs can only use preallocated hash map\n"); 3036 return -EINVAL; 3037 } 3038 return 0; 3039 } 3040 3041 /* look for pseudo eBPF instructions that access map FDs and 3042 * replace them with actual map pointers 3043 */ 3044 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) 3045 { 3046 struct bpf_insn *insn = env->prog->insnsi; 3047 int insn_cnt = env->prog->len; 3048 int i, j, err; 3049 3050 err = bpf_prog_calc_tag(env->prog); 3051 if (err) 3052 return err; 3053 3054 for (i = 0; i < insn_cnt; i++, insn++) { 3055 if (BPF_CLASS(insn->code) == BPF_LDX && 3056 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { 3057 verbose("BPF_LDX uses reserved fields\n"); 3058 return -EINVAL; 3059 } 3060 3061 if (BPF_CLASS(insn->code) == BPF_STX && 3062 ((BPF_MODE(insn->code) != BPF_MEM && 3063 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { 3064 verbose("BPF_STX uses reserved fields\n"); 3065 return -EINVAL; 3066 } 3067 3068 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { 3069 struct bpf_map *map; 3070 struct fd f; 3071 3072 if (i == insn_cnt - 1 || insn[1].code != 0 || 3073 insn[1].dst_reg != 0 || insn[1].src_reg != 0 || 3074 insn[1].off != 0) { 3075 verbose("invalid bpf_ld_imm64 insn\n"); 3076 return -EINVAL; 3077 } 3078 3079 if (insn->src_reg == 0) 3080 /* valid generic load 64-bit imm */ 3081 goto next_insn; 3082 3083 if (insn->src_reg != BPF_PSEUDO_MAP_FD) { 3084 verbose("unrecognized bpf_ld_imm64 insn\n"); 3085 return -EINVAL; 3086 } 3087 3088 f = fdget(insn->imm); 3089 map = __bpf_map_get(f); 3090 if (IS_ERR(map)) { 3091 verbose("fd %d is not pointing to valid bpf_map\n", 3092 insn->imm); 3093 return PTR_ERR(map); 3094 } 3095 3096 err = check_map_prog_compatibility(map, env->prog); 3097 if (err) { 3098 fdput(f); 3099 return err; 3100 } 3101 3102 /* store map pointer inside BPF_LD_IMM64 instruction */ 3103 insn[0].imm = (u32) (unsigned long) map; 3104 insn[1].imm = ((u64) (unsigned long) map) >> 32; 3105 3106 /* check whether we recorded this map already */ 3107 for (j = 0; j < env->used_map_cnt; j++) 3108 if (env->used_maps[j] == map) { 3109 fdput(f); 3110 goto next_insn; 3111 } 3112 3113 if (env->used_map_cnt >= MAX_USED_MAPS) { 3114 fdput(f); 3115 return -E2BIG; 3116 } 3117 3118 /* hold the map. If the program is rejected by verifier, 3119 * the map will be released by release_maps() or it 3120 * will be used by the valid program until it's unloaded 3121 * and all maps are released in free_bpf_prog_info() 3122 */ 3123 map = bpf_map_inc(map, false); 3124 if (IS_ERR(map)) { 3125 fdput(f); 3126 return PTR_ERR(map); 3127 } 3128 env->used_maps[env->used_map_cnt++] = map; 3129 3130 fdput(f); 3131 next_insn: 3132 insn++; 3133 i++; 3134 } 3135 } 3136 3137 /* now all pseudo BPF_LD_IMM64 instructions load valid 3138 * 'struct bpf_map *' into a register instead of user map_fd. 3139 * These pointers will be used later by verifier to validate map access. 3140 */ 3141 return 0; 3142 } 3143 3144 /* drop refcnt of maps used by the rejected program */ 3145 static void release_maps(struct bpf_verifier_env *env) 3146 { 3147 int i; 3148 3149 for (i = 0; i < env->used_map_cnt; i++) 3150 bpf_map_put(env->used_maps[i]); 3151 } 3152 3153 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ 3154 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) 3155 { 3156 struct bpf_insn *insn = env->prog->insnsi; 3157 int insn_cnt = env->prog->len; 3158 int i; 3159 3160 for (i = 0; i < insn_cnt; i++, insn++) 3161 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) 3162 insn->src_reg = 0; 3163 } 3164 3165 /* convert load instructions that access fields of 'struct __sk_buff' 3166 * into sequence of instructions that access fields of 'struct sk_buff' 3167 */ 3168 static int convert_ctx_accesses(struct bpf_verifier_env *env) 3169 { 3170 const struct bpf_verifier_ops *ops = env->prog->aux->ops; 3171 const int insn_cnt = env->prog->len; 3172 struct bpf_insn insn_buf[16], *insn; 3173 struct bpf_prog *new_prog; 3174 enum bpf_access_type type; 3175 int i, cnt, delta = 0; 3176 3177 if (ops->gen_prologue) { 3178 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, 3179 env->prog); 3180 if (cnt >= ARRAY_SIZE(insn_buf)) { 3181 verbose("bpf verifier is misconfigured\n"); 3182 return -EINVAL; 3183 } else if (cnt) { 3184 new_prog = bpf_patch_insn_single(env->prog, 0, 3185 insn_buf, cnt); 3186 if (!new_prog) 3187 return -ENOMEM; 3188 env->prog = new_prog; 3189 delta += cnt - 1; 3190 } 3191 } 3192 3193 if (!ops->convert_ctx_access) 3194 return 0; 3195 3196 insn = env->prog->insnsi + delta; 3197 3198 for (i = 0; i < insn_cnt; i++, insn++) { 3199 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || 3200 insn->code == (BPF_LDX | BPF_MEM | BPF_H) || 3201 insn->code == (BPF_LDX | BPF_MEM | BPF_W) || 3202 insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) 3203 type = BPF_READ; 3204 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || 3205 insn->code == (BPF_STX | BPF_MEM | BPF_H) || 3206 insn->code == (BPF_STX | BPF_MEM | BPF_W) || 3207 insn->code == (BPF_STX | BPF_MEM | BPF_DW)) 3208 type = BPF_WRITE; 3209 else 3210 continue; 3211 3212 if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX) 3213 continue; 3214 3215 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog); 3216 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { 3217 verbose("bpf verifier is misconfigured\n"); 3218 return -EINVAL; 3219 } 3220 3221 new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf, 3222 cnt); 3223 if (!new_prog) 3224 return -ENOMEM; 3225 3226 delta += cnt - 1; 3227 3228 /* keep walking new program and skip insns we just inserted */ 3229 env->prog = new_prog; 3230 insn = new_prog->insnsi + i + delta; 3231 } 3232 3233 return 0; 3234 } 3235 3236 static void free_states(struct bpf_verifier_env *env) 3237 { 3238 struct bpf_verifier_state_list *sl, *sln; 3239 int i; 3240 3241 if (!env->explored_states) 3242 return; 3243 3244 for (i = 0; i < env->prog->len; i++) { 3245 sl = env->explored_states[i]; 3246 3247 if (sl) 3248 while (sl != STATE_LIST_MARK) { 3249 sln = sl->next; 3250 kfree(sl); 3251 sl = sln; 3252 } 3253 } 3254 3255 kfree(env->explored_states); 3256 } 3257 3258 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) 3259 { 3260 char __user *log_ubuf = NULL; 3261 struct bpf_verifier_env *env; 3262 int ret = -EINVAL; 3263 3264 /* 'struct bpf_verifier_env' can be global, but since it's not small, 3265 * allocate/free it every time bpf_check() is called 3266 */ 3267 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3268 if (!env) 3269 return -ENOMEM; 3270 3271 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3272 (*prog)->len); 3273 ret = -ENOMEM; 3274 if (!env->insn_aux_data) 3275 goto err_free_env; 3276 env->prog = *prog; 3277 3278 /* grab the mutex to protect few globals used by verifier */ 3279 mutex_lock(&bpf_verifier_lock); 3280 3281 if (attr->log_level || attr->log_buf || attr->log_size) { 3282 /* user requested verbose verifier output 3283 * and supplied buffer to store the verification trace 3284 */ 3285 log_level = attr->log_level; 3286 log_ubuf = (char __user *) (unsigned long) attr->log_buf; 3287 log_size = attr->log_size; 3288 log_len = 0; 3289 3290 ret = -EINVAL; 3291 /* log_* values have to be sane */ 3292 if (log_size < 128 || log_size > UINT_MAX >> 8 || 3293 log_level == 0 || log_ubuf == NULL) 3294 goto err_unlock; 3295 3296 ret = -ENOMEM; 3297 log_buf = vmalloc(log_size); 3298 if (!log_buf) 3299 goto err_unlock; 3300 } else { 3301 log_level = 0; 3302 } 3303 3304 ret = replace_map_fd_with_map_ptr(env); 3305 if (ret < 0) 3306 goto skip_full_check; 3307 3308 env->explored_states = kcalloc(env->prog->len, 3309 sizeof(struct bpf_verifier_state_list *), 3310 GFP_USER); 3311 ret = -ENOMEM; 3312 if (!env->explored_states) 3313 goto skip_full_check; 3314 3315 ret = check_cfg(env); 3316 if (ret < 0) 3317 goto skip_full_check; 3318 3319 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3320 3321 ret = do_check(env); 3322 3323 skip_full_check: 3324 while (pop_stack(env, NULL) >= 0); 3325 free_states(env); 3326 3327 if (ret == 0) 3328 /* program is valid, convert *(u32*)(ctx + off) accesses */ 3329 ret = convert_ctx_accesses(env); 3330 3331 if (log_level && log_len >= log_size - 1) { 3332 BUG_ON(log_len >= log_size); 3333 /* verifier log exceeded user supplied buffer */ 3334 ret = -ENOSPC; 3335 /* fall through to return what was recorded */ 3336 } 3337 3338 /* copy verifier log back to user space including trailing zero */ 3339 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { 3340 ret = -EFAULT; 3341 goto free_log_buf; 3342 } 3343 3344 if (ret == 0 && env->used_map_cnt) { 3345 /* if program passed verifier, update used_maps in bpf_prog_info */ 3346 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, 3347 sizeof(env->used_maps[0]), 3348 GFP_KERNEL); 3349 3350 if (!env->prog->aux->used_maps) { 3351 ret = -ENOMEM; 3352 goto free_log_buf; 3353 } 3354 3355 memcpy(env->prog->aux->used_maps, env->used_maps, 3356 sizeof(env->used_maps[0]) * env->used_map_cnt); 3357 env->prog->aux->used_map_cnt = env->used_map_cnt; 3358 3359 /* program is valid. Convert pseudo bpf_ld_imm64 into generic 3360 * bpf_ld_imm64 instructions 3361 */ 3362 convert_pseudo_ld_imm64(env); 3363 } 3364 3365 free_log_buf: 3366 if (log_level) 3367 vfree(log_buf); 3368 if (!env->prog->aux->used_maps) 3369 /* if we didn't copy map pointers into bpf_prog_info, release 3370 * them now. Otherwise free_bpf_prog_info() will release them. 3371 */ 3372 release_maps(env); 3373 *prog = env->prog; 3374 err_unlock: 3375 mutex_unlock(&bpf_verifier_lock); 3376 vfree(env->insn_aux_data); 3377 err_free_env: 3378 kfree(env); 3379 return ret; 3380 } 3381 3382 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, 3383 void *priv) 3384 { 3385 struct bpf_verifier_env *env; 3386 int ret; 3387 3388 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); 3389 if (!env) 3390 return -ENOMEM; 3391 3392 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * 3393 prog->len); 3394 ret = -ENOMEM; 3395 if (!env->insn_aux_data) 3396 goto err_free_env; 3397 env->prog = prog; 3398 env->analyzer_ops = ops; 3399 env->analyzer_priv = priv; 3400 3401 /* grab the mutex to protect few globals used by verifier */ 3402 mutex_lock(&bpf_verifier_lock); 3403 3404 log_level = 0; 3405 3406 env->explored_states = kcalloc(env->prog->len, 3407 sizeof(struct bpf_verifier_state_list *), 3408 GFP_KERNEL); 3409 ret = -ENOMEM; 3410 if (!env->explored_states) 3411 goto skip_full_check; 3412 3413 ret = check_cfg(env); 3414 if (ret < 0) 3415 goto skip_full_check; 3416 3417 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 3418 3419 ret = do_check(env); 3420 3421 skip_full_check: 3422 while (pop_stack(env, NULL) >= 0); 3423 free_states(env); 3424 3425 mutex_unlock(&bpf_verifier_lock); 3426 vfree(env->insn_aux_data); 3427 err_free_env: 3428 kfree(env); 3429 return ret; 3430 } 3431 EXPORT_SYMBOL_GPL(bpf_analyzer); 3432