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