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/filter.h> 18 #include <net/netlink.h> 19 #include <linux/file.h> 20 #include <linux/vmalloc.h> 21 22 /* bpf_check() is a static code analyzer that walks eBPF program 23 * instruction by instruction and updates register/stack state. 24 * All paths of conditional branches are analyzed until 'bpf_exit' insn. 25 * 26 * The first pass is depth-first-search to check that the program is a DAG. 27 * It rejects the following programs: 28 * - larger than BPF_MAXINSNS insns 29 * - if loop is present (detected via back-edge) 30 * - unreachable insns exist (shouldn't be a forest. program = one function) 31 * - out of bounds or malformed jumps 32 * The second pass is all possible path descent from the 1st insn. 33 * Since it's analyzing all pathes through the program, the length of the 34 * analysis is limited to 32k insn, which may be hit even if total number of 35 * insn is less then 4K, but there are too many branches that change stack/regs. 36 * Number of 'branches to be analyzed' is limited to 1k 37 * 38 * On entry to each instruction, each register has a type, and the instruction 39 * changes the types of the registers depending on instruction semantics. 40 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is 41 * copied to R1. 42 * 43 * All registers are 64-bit. 44 * R0 - return register 45 * R1-R5 argument passing registers 46 * R6-R9 callee saved registers 47 * R10 - frame pointer read-only 48 * 49 * At the start of BPF program the register R1 contains a pointer to bpf_context 50 * and has type PTR_TO_CTX. 51 * 52 * Verifier tracks arithmetic operations on pointers in case: 53 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), 54 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), 55 * 1st insn copies R10 (which has FRAME_PTR) type into R1 56 * and 2nd arithmetic instruction is pattern matched to recognize 57 * that it wants to construct a pointer to some element within stack. 58 * So after 2nd insn, the register R1 has type PTR_TO_STACK 59 * (and -20 constant is saved for further stack bounds checking). 60 * Meaning that this reg is a pointer to stack plus known immediate constant. 61 * 62 * Most of the time the registers have UNKNOWN_VALUE type, which 63 * means the register has some value, but it's not a valid pointer. 64 * (like pointer plus pointer becomes UNKNOWN_VALUE type) 65 * 66 * When verifier sees load or store instructions the type of base register 67 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer 68 * types recognized by check_mem_access() function. 69 * 70 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' 71 * and the range of [ptr, ptr + map's value_size) is accessible. 72 * 73 * registers used to pass values to function calls are checked against 74 * function argument constraints. 75 * 76 * ARG_PTR_TO_MAP_KEY is one of such argument constraints. 77 * It means that the register type passed to this function must be 78 * PTR_TO_STACK and it will be used inside the function as 79 * 'pointer to map element key' 80 * 81 * For example the argument constraints for bpf_map_lookup_elem(): 82 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, 83 * .arg1_type = ARG_CONST_MAP_PTR, 84 * .arg2_type = ARG_PTR_TO_MAP_KEY, 85 * 86 * ret_type says that this function returns 'pointer to map elem value or null' 87 * function expects 1st argument to be a const pointer to 'struct bpf_map' and 88 * 2nd argument should be a pointer to stack, which will be used inside 89 * the helper function as a pointer to map element key. 90 * 91 * On the kernel side the helper function looks like: 92 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) 93 * { 94 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; 95 * void *key = (void *) (unsigned long) r2; 96 * void *value; 97 * 98 * here kernel can access 'key' and 'map' pointers safely, knowing that 99 * [key, key + map->key_size) bytes are valid and were initialized on 100 * the stack of eBPF program. 101 * } 102 * 103 * Corresponding eBPF program may look like: 104 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR 105 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK 106 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP 107 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), 108 * here verifier looks at prototype of map_lookup_elem() and sees: 109 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, 110 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes 111 * 112 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, 113 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits 114 * and were initialized prior to this call. 115 * If it's ok, then verifier allows this BPF_CALL insn and looks at 116 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets 117 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function 118 * returns ether pointer to map value or NULL. 119 * 120 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' 121 * insn, the register holding that pointer in the true branch changes state to 122 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false 123 * branch. See check_cond_jmp_op(). 124 * 125 * After the call R0 is set to return type of the function and registers R1-R5 126 * are set to NOT_INIT to indicate that they are no longer readable. 127 */ 128 129 /* types of values stored in eBPF registers */ 130 enum bpf_reg_type { 131 NOT_INIT = 0, /* nothing was written into register */ 132 UNKNOWN_VALUE, /* reg doesn't contain a valid pointer */ 133 PTR_TO_CTX, /* reg points to bpf_context */ 134 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 135 PTR_TO_MAP_VALUE, /* reg points to map element value */ 136 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 137 FRAME_PTR, /* reg == frame_pointer */ 138 PTR_TO_STACK, /* reg == frame_pointer + imm */ 139 CONST_IMM, /* constant integer value */ 140 141 /* PTR_TO_PACKET represents: 142 * skb->data 143 * skb->data + imm 144 * skb->data + (u16) var 145 * skb->data + (u16) var + imm 146 * if (range > 0) then [ptr, ptr + range - off) is safe to access 147 * if (id > 0) means that some 'var' was added 148 * if (off > 0) menas that 'imm' was added 149 */ 150 PTR_TO_PACKET, 151 PTR_TO_PACKET_END, /* skb->data + headlen */ 152 }; 153 154 struct reg_state { 155 enum bpf_reg_type type; 156 union { 157 /* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */ 158 s64 imm; 159 160 /* valid when type == PTR_TO_PACKET* */ 161 struct { 162 u32 id; 163 u16 off; 164 u16 range; 165 }; 166 167 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE | 168 * PTR_TO_MAP_VALUE_OR_NULL 169 */ 170 struct bpf_map *map_ptr; 171 }; 172 }; 173 174 enum bpf_stack_slot_type { 175 STACK_INVALID, /* nothing was stored in this stack slot */ 176 STACK_SPILL, /* register spilled into stack */ 177 STACK_MISC /* BPF program wrote some data into this slot */ 178 }; 179 180 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */ 181 182 /* state of the program: 183 * type of all registers and stack info 184 */ 185 struct verifier_state { 186 struct reg_state regs[MAX_BPF_REG]; 187 u8 stack_slot_type[MAX_BPF_STACK]; 188 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE]; 189 }; 190 191 /* linked list of verifier states used to prune search */ 192 struct verifier_state_list { 193 struct verifier_state state; 194 struct verifier_state_list *next; 195 }; 196 197 /* verifier_state + insn_idx are pushed to stack when branch is encountered */ 198 struct verifier_stack_elem { 199 /* verifer state is 'st' 200 * before processing instruction 'insn_idx' 201 * and after processing instruction 'prev_insn_idx' 202 */ 203 struct verifier_state st; 204 int insn_idx; 205 int prev_insn_idx; 206 struct verifier_stack_elem *next; 207 }; 208 209 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */ 210 211 /* single container for all structs 212 * one verifier_env per bpf_check() call 213 */ 214 struct verifier_env { 215 struct bpf_prog *prog; /* eBPF program being verified */ 216 struct verifier_stack_elem *head; /* stack of verifier states to be processed */ 217 int stack_size; /* number of states to be processed */ 218 struct verifier_state cur_state; /* current verifier state */ 219 struct verifier_state_list **explored_states; /* search pruning optimization */ 220 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */ 221 u32 used_map_cnt; /* number of used maps */ 222 bool allow_ptr_leaks; 223 }; 224 225 #define BPF_COMPLEXITY_LIMIT_INSNS 65536 226 #define BPF_COMPLEXITY_LIMIT_STACK 1024 227 228 struct bpf_call_arg_meta { 229 struct bpf_map *map_ptr; 230 bool raw_mode; 231 int regno; 232 int access_size; 233 }; 234 235 /* verbose verifier prints what it's seeing 236 * bpf_check() is called under lock, so no race to access these global vars 237 */ 238 static u32 log_level, log_size, log_len; 239 static char *log_buf; 240 241 static DEFINE_MUTEX(bpf_verifier_lock); 242 243 /* log_level controls verbosity level of eBPF verifier. 244 * verbose() is used to dump the verification trace to the log, so the user 245 * can figure out what's wrong with the program 246 */ 247 static __printf(1, 2) void verbose(const char *fmt, ...) 248 { 249 va_list args; 250 251 if (log_level == 0 || log_len >= log_size - 1) 252 return; 253 254 va_start(args, fmt); 255 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); 256 va_end(args); 257 } 258 259 /* string representation of 'enum bpf_reg_type' */ 260 static const char * const reg_type_str[] = { 261 [NOT_INIT] = "?", 262 [UNKNOWN_VALUE] = "inv", 263 [PTR_TO_CTX] = "ctx", 264 [CONST_PTR_TO_MAP] = "map_ptr", 265 [PTR_TO_MAP_VALUE] = "map_value", 266 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", 267 [FRAME_PTR] = "fp", 268 [PTR_TO_STACK] = "fp", 269 [CONST_IMM] = "imm", 270 [PTR_TO_PACKET] = "pkt", 271 [PTR_TO_PACKET_END] = "pkt_end", 272 }; 273 274 static void print_verifier_state(struct verifier_state *state) 275 { 276 struct reg_state *reg; 277 enum bpf_reg_type t; 278 int i; 279 280 for (i = 0; i < MAX_BPF_REG; i++) { 281 reg = &state->regs[i]; 282 t = reg->type; 283 if (t == NOT_INIT) 284 continue; 285 verbose(" R%d=%s", i, reg_type_str[t]); 286 if (t == CONST_IMM || t == PTR_TO_STACK) 287 verbose("%lld", reg->imm); 288 else if (t == PTR_TO_PACKET) 289 verbose("(id=%d,off=%d,r=%d)", 290 reg->id, reg->off, reg->range); 291 else if (t == UNKNOWN_VALUE && reg->imm) 292 verbose("%lld", reg->imm); 293 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || 294 t == PTR_TO_MAP_VALUE_OR_NULL) 295 verbose("(ks=%d,vs=%d)", 296 reg->map_ptr->key_size, 297 reg->map_ptr->value_size); 298 } 299 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 300 if (state->stack_slot_type[i] == STACK_SPILL) 301 verbose(" fp%d=%s", -MAX_BPF_STACK + i, 302 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); 303 } 304 verbose("\n"); 305 } 306 307 static const char *const bpf_class_string[] = { 308 [BPF_LD] = "ld", 309 [BPF_LDX] = "ldx", 310 [BPF_ST] = "st", 311 [BPF_STX] = "stx", 312 [BPF_ALU] = "alu", 313 [BPF_JMP] = "jmp", 314 [BPF_RET] = "BUG", 315 [BPF_ALU64] = "alu64", 316 }; 317 318 static const char *const bpf_alu_string[16] = { 319 [BPF_ADD >> 4] = "+=", 320 [BPF_SUB >> 4] = "-=", 321 [BPF_MUL >> 4] = "*=", 322 [BPF_DIV >> 4] = "/=", 323 [BPF_OR >> 4] = "|=", 324 [BPF_AND >> 4] = "&=", 325 [BPF_LSH >> 4] = "<<=", 326 [BPF_RSH >> 4] = ">>=", 327 [BPF_NEG >> 4] = "neg", 328 [BPF_MOD >> 4] = "%=", 329 [BPF_XOR >> 4] = "^=", 330 [BPF_MOV >> 4] = "=", 331 [BPF_ARSH >> 4] = "s>>=", 332 [BPF_END >> 4] = "endian", 333 }; 334 335 static const char *const bpf_ldst_string[] = { 336 [BPF_W >> 3] = "u32", 337 [BPF_H >> 3] = "u16", 338 [BPF_B >> 3] = "u8", 339 [BPF_DW >> 3] = "u64", 340 }; 341 342 static const char *const bpf_jmp_string[16] = { 343 [BPF_JA >> 4] = "jmp", 344 [BPF_JEQ >> 4] = "==", 345 [BPF_JGT >> 4] = ">", 346 [BPF_JGE >> 4] = ">=", 347 [BPF_JSET >> 4] = "&", 348 [BPF_JNE >> 4] = "!=", 349 [BPF_JSGT >> 4] = "s>", 350 [BPF_JSGE >> 4] = "s>=", 351 [BPF_CALL >> 4] = "call", 352 [BPF_EXIT >> 4] = "exit", 353 }; 354 355 static void print_bpf_insn(struct bpf_insn *insn) 356 { 357 u8 class = BPF_CLASS(insn->code); 358 359 if (class == BPF_ALU || class == BPF_ALU64) { 360 if (BPF_SRC(insn->code) == BPF_X) 361 verbose("(%02x) %sr%d %s %sr%d\n", 362 insn->code, class == BPF_ALU ? "(u32) " : "", 363 insn->dst_reg, 364 bpf_alu_string[BPF_OP(insn->code) >> 4], 365 class == BPF_ALU ? "(u32) " : "", 366 insn->src_reg); 367 else 368 verbose("(%02x) %sr%d %s %s%d\n", 369 insn->code, class == BPF_ALU ? "(u32) " : "", 370 insn->dst_reg, 371 bpf_alu_string[BPF_OP(insn->code) >> 4], 372 class == BPF_ALU ? "(u32) " : "", 373 insn->imm); 374 } else if (class == BPF_STX) { 375 if (BPF_MODE(insn->code) == BPF_MEM) 376 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", 377 insn->code, 378 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 379 insn->dst_reg, 380 insn->off, insn->src_reg); 381 else if (BPF_MODE(insn->code) == BPF_XADD) 382 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", 383 insn->code, 384 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 385 insn->dst_reg, insn->off, 386 insn->src_reg); 387 else 388 verbose("BUG_%02x\n", insn->code); 389 } else if (class == BPF_ST) { 390 if (BPF_MODE(insn->code) != BPF_MEM) { 391 verbose("BUG_st_%02x\n", insn->code); 392 return; 393 } 394 verbose("(%02x) *(%s *)(r%d %+d) = %d\n", 395 insn->code, 396 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 397 insn->dst_reg, 398 insn->off, insn->imm); 399 } else if (class == BPF_LDX) { 400 if (BPF_MODE(insn->code) != BPF_MEM) { 401 verbose("BUG_ldx_%02x\n", insn->code); 402 return; 403 } 404 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", 405 insn->code, insn->dst_reg, 406 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 407 insn->src_reg, insn->off); 408 } else if (class == BPF_LD) { 409 if (BPF_MODE(insn->code) == BPF_ABS) { 410 verbose("(%02x) r0 = *(%s *)skb[%d]\n", 411 insn->code, 412 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 413 insn->imm); 414 } else if (BPF_MODE(insn->code) == BPF_IND) { 415 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", 416 insn->code, 417 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 418 insn->src_reg, insn->imm); 419 } else if (BPF_MODE(insn->code) == BPF_IMM) { 420 verbose("(%02x) r%d = 0x%x\n", 421 insn->code, insn->dst_reg, insn->imm); 422 } else { 423 verbose("BUG_ld_%02x\n", insn->code); 424 return; 425 } 426 } else if (class == BPF_JMP) { 427 u8 opcode = BPF_OP(insn->code); 428 429 if (opcode == BPF_CALL) { 430 verbose("(%02x) call %d\n", insn->code, insn->imm); 431 } else if (insn->code == (BPF_JMP | BPF_JA)) { 432 verbose("(%02x) goto pc%+d\n", 433 insn->code, insn->off); 434 } else if (insn->code == (BPF_JMP | BPF_EXIT)) { 435 verbose("(%02x) exit\n", insn->code); 436 } else if (BPF_SRC(insn->code) == BPF_X) { 437 verbose("(%02x) if r%d %s r%d goto pc%+d\n", 438 insn->code, insn->dst_reg, 439 bpf_jmp_string[BPF_OP(insn->code) >> 4], 440 insn->src_reg, insn->off); 441 } else { 442 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", 443 insn->code, insn->dst_reg, 444 bpf_jmp_string[BPF_OP(insn->code) >> 4], 445 insn->imm, insn->off); 446 } 447 } else { 448 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); 449 } 450 } 451 452 static int pop_stack(struct verifier_env *env, int *prev_insn_idx) 453 { 454 struct verifier_stack_elem *elem; 455 int insn_idx; 456 457 if (env->head == NULL) 458 return -1; 459 460 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); 461 insn_idx = env->head->insn_idx; 462 if (prev_insn_idx) 463 *prev_insn_idx = env->head->prev_insn_idx; 464 elem = env->head->next; 465 kfree(env->head); 466 env->head = elem; 467 env->stack_size--; 468 return insn_idx; 469 } 470 471 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx, 472 int prev_insn_idx) 473 { 474 struct verifier_stack_elem *elem; 475 476 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL); 477 if (!elem) 478 goto err; 479 480 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); 481 elem->insn_idx = insn_idx; 482 elem->prev_insn_idx = prev_insn_idx; 483 elem->next = env->head; 484 env->head = elem; 485 env->stack_size++; 486 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { 487 verbose("BPF program is too complex\n"); 488 goto err; 489 } 490 return &elem->st; 491 err: 492 /* pop all elements and return */ 493 while (pop_stack(env, NULL) >= 0); 494 return NULL; 495 } 496 497 #define CALLER_SAVED_REGS 6 498 static const int caller_saved[CALLER_SAVED_REGS] = { 499 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 500 }; 501 502 static void init_reg_state(struct reg_state *regs) 503 { 504 int i; 505 506 for (i = 0; i < MAX_BPF_REG; i++) { 507 regs[i].type = NOT_INIT; 508 regs[i].imm = 0; 509 } 510 511 /* frame pointer */ 512 regs[BPF_REG_FP].type = FRAME_PTR; 513 514 /* 1st arg to a function */ 515 regs[BPF_REG_1].type = PTR_TO_CTX; 516 } 517 518 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno) 519 { 520 BUG_ON(regno >= MAX_BPF_REG); 521 regs[regno].type = UNKNOWN_VALUE; 522 regs[regno].imm = 0; 523 } 524 525 enum reg_arg_type { 526 SRC_OP, /* register is used as source operand */ 527 DST_OP, /* register is used as destination operand */ 528 DST_OP_NO_MARK /* same as above, check only, don't mark */ 529 }; 530 531 static int check_reg_arg(struct reg_state *regs, u32 regno, 532 enum reg_arg_type t) 533 { 534 if (regno >= MAX_BPF_REG) { 535 verbose("R%d is invalid\n", regno); 536 return -EINVAL; 537 } 538 539 if (t == SRC_OP) { 540 /* check whether register used as source operand can be read */ 541 if (regs[regno].type == NOT_INIT) { 542 verbose("R%d !read_ok\n", regno); 543 return -EACCES; 544 } 545 } else { 546 /* check whether register used as dest operand can be written to */ 547 if (regno == BPF_REG_FP) { 548 verbose("frame pointer is read only\n"); 549 return -EACCES; 550 } 551 if (t == DST_OP) 552 mark_reg_unknown_value(regs, regno); 553 } 554 return 0; 555 } 556 557 static int bpf_size_to_bytes(int bpf_size) 558 { 559 if (bpf_size == BPF_W) 560 return 4; 561 else if (bpf_size == BPF_H) 562 return 2; 563 else if (bpf_size == BPF_B) 564 return 1; 565 else if (bpf_size == BPF_DW) 566 return 8; 567 else 568 return -EINVAL; 569 } 570 571 static bool is_spillable_regtype(enum bpf_reg_type type) 572 { 573 switch (type) { 574 case PTR_TO_MAP_VALUE: 575 case PTR_TO_MAP_VALUE_OR_NULL: 576 case PTR_TO_STACK: 577 case PTR_TO_CTX: 578 case PTR_TO_PACKET: 579 case PTR_TO_PACKET_END: 580 case FRAME_PTR: 581 case CONST_PTR_TO_MAP: 582 return true; 583 default: 584 return false; 585 } 586 } 587 588 /* check_stack_read/write functions track spill/fill of registers, 589 * stack boundary and alignment are checked in check_mem_access() 590 */ 591 static int check_stack_write(struct verifier_state *state, int off, int size, 592 int value_regno) 593 { 594 int i; 595 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, 596 * so it's aligned access and [off, off + size) are within stack limits 597 */ 598 599 if (value_regno >= 0 && 600 is_spillable_regtype(state->regs[value_regno].type)) { 601 602 /* register containing pointer is being spilled into stack */ 603 if (size != BPF_REG_SIZE) { 604 verbose("invalid size of register spill\n"); 605 return -EACCES; 606 } 607 608 /* save register state */ 609 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 610 state->regs[value_regno]; 611 612 for (i = 0; i < BPF_REG_SIZE; i++) 613 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; 614 } else { 615 /* regular write of data into stack */ 616 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 617 (struct reg_state) {}; 618 619 for (i = 0; i < size; i++) 620 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; 621 } 622 return 0; 623 } 624 625 static int check_stack_read(struct verifier_state *state, int off, int size, 626 int value_regno) 627 { 628 u8 *slot_type; 629 int i; 630 631 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; 632 633 if (slot_type[0] == STACK_SPILL) { 634 if (size != BPF_REG_SIZE) { 635 verbose("invalid size of register spill\n"); 636 return -EACCES; 637 } 638 for (i = 1; i < BPF_REG_SIZE; i++) { 639 if (slot_type[i] != STACK_SPILL) { 640 verbose("corrupted spill memory\n"); 641 return -EACCES; 642 } 643 } 644 645 if (value_regno >= 0) 646 /* restore register state from stack */ 647 state->regs[value_regno] = 648 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; 649 return 0; 650 } else { 651 for (i = 0; i < size; i++) { 652 if (slot_type[i] != STACK_MISC) { 653 verbose("invalid read from stack off %d+%d size %d\n", 654 off, i, size); 655 return -EACCES; 656 } 657 } 658 if (value_regno >= 0) 659 /* have read misc data from the stack */ 660 mark_reg_unknown_value(state->regs, value_regno); 661 return 0; 662 } 663 } 664 665 /* check read/write into map element returned by bpf_map_lookup_elem() */ 666 static int check_map_access(struct verifier_env *env, u32 regno, int off, 667 int size) 668 { 669 struct bpf_map *map = env->cur_state.regs[regno].map_ptr; 670 671 if (off < 0 || off + size > map->value_size) { 672 verbose("invalid access to map value, value_size=%d off=%d size=%d\n", 673 map->value_size, off, size); 674 return -EACCES; 675 } 676 return 0; 677 } 678 679 #define MAX_PACKET_OFF 0xffff 680 681 static int check_packet_access(struct verifier_env *env, u32 regno, int off, 682 int size) 683 { 684 struct reg_state *regs = env->cur_state.regs; 685 struct reg_state *reg = ®s[regno]; 686 int linear_size = (int) reg->range - (int) reg->off; 687 688 if (linear_size < 0 || linear_size >= MAX_PACKET_OFF) { 689 verbose("verifier bug\n"); 690 return -EFAULT; 691 } 692 if (off < 0 || off + size > linear_size) { 693 verbose("invalid access to packet, off=%d size=%d, allowed=%d\n", 694 off, size, linear_size); 695 return -EACCES; 696 } 697 return 0; 698 } 699 700 /* check access to 'struct bpf_context' fields */ 701 static int check_ctx_access(struct verifier_env *env, int off, int size, 702 enum bpf_access_type t) 703 { 704 if (env->prog->aux->ops->is_valid_access && 705 env->prog->aux->ops->is_valid_access(off, size, t)) { 706 /* remember the offset of last byte accessed in ctx */ 707 if (env->prog->aux->max_ctx_offset < off + size) 708 env->prog->aux->max_ctx_offset = off + size; 709 return 0; 710 } 711 712 verbose("invalid bpf_context access off=%d size=%d\n", off, size); 713 return -EACCES; 714 } 715 716 static bool is_pointer_value(struct verifier_env *env, int regno) 717 { 718 if (env->allow_ptr_leaks) 719 return false; 720 721 switch (env->cur_state.regs[regno].type) { 722 case UNKNOWN_VALUE: 723 case CONST_IMM: 724 return false; 725 default: 726 return true; 727 } 728 } 729 730 static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg, 731 int off, int size) 732 { 733 if (reg->type != PTR_TO_PACKET) { 734 if (off % size != 0) { 735 verbose("misaligned access off %d size %d\n", off, size); 736 return -EACCES; 737 } else { 738 return 0; 739 } 740 } 741 742 switch (env->prog->type) { 743 case BPF_PROG_TYPE_SCHED_CLS: 744 case BPF_PROG_TYPE_SCHED_ACT: 745 break; 746 default: 747 verbose("verifier is misconfigured\n"); 748 return -EACCES; 749 } 750 751 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) 752 /* misaligned access to packet is ok on x86,arm,arm64 */ 753 return 0; 754 755 if (reg->id && size != 1) { 756 verbose("Unknown packet alignment. Only byte-sized access allowed\n"); 757 return -EACCES; 758 } 759 760 /* skb->data is NET_IP_ALIGN-ed */ 761 if ((NET_IP_ALIGN + reg->off + off) % size != 0) { 762 verbose("misaligned packet access off %d+%d+%d size %d\n", 763 NET_IP_ALIGN, reg->off, off, size); 764 return -EACCES; 765 } 766 return 0; 767 } 768 769 /* check whether memory at (regno + off) is accessible for t = (read | write) 770 * if t==write, value_regno is a register which value is stored into memory 771 * if t==read, value_regno is a register which will receive the value from memory 772 * if t==write && value_regno==-1, some unknown value is stored into memory 773 * if t==read && value_regno==-1, don't care what we read from memory 774 */ 775 static int check_mem_access(struct verifier_env *env, u32 regno, int off, 776 int bpf_size, enum bpf_access_type t, 777 int value_regno) 778 { 779 struct verifier_state *state = &env->cur_state; 780 struct reg_state *reg = &state->regs[regno]; 781 int size, err = 0; 782 783 if (reg->type == PTR_TO_STACK) 784 off += reg->imm; 785 786 size = bpf_size_to_bytes(bpf_size); 787 if (size < 0) 788 return size; 789 790 err = check_ptr_alignment(env, reg, off, size); 791 if (err) 792 return err; 793 794 if (reg->type == PTR_TO_MAP_VALUE) { 795 if (t == BPF_WRITE && value_regno >= 0 && 796 is_pointer_value(env, value_regno)) { 797 verbose("R%d leaks addr into map\n", value_regno); 798 return -EACCES; 799 } 800 err = check_map_access(env, regno, off, size); 801 if (!err && t == BPF_READ && value_regno >= 0) 802 mark_reg_unknown_value(state->regs, value_regno); 803 804 } else if (reg->type == PTR_TO_CTX) { 805 if (t == BPF_WRITE && value_regno >= 0 && 806 is_pointer_value(env, value_regno)) { 807 verbose("R%d leaks addr into ctx\n", value_regno); 808 return -EACCES; 809 } 810 err = check_ctx_access(env, off, size, t); 811 if (!err && t == BPF_READ && value_regno >= 0) { 812 mark_reg_unknown_value(state->regs, value_regno); 813 if (off == offsetof(struct __sk_buff, data) && 814 env->allow_ptr_leaks) 815 /* note that reg.[id|off|range] == 0 */ 816 state->regs[value_regno].type = PTR_TO_PACKET; 817 else if (off == offsetof(struct __sk_buff, data_end) && 818 env->allow_ptr_leaks) 819 state->regs[value_regno].type = PTR_TO_PACKET_END; 820 } 821 822 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { 823 if (off >= 0 || off < -MAX_BPF_STACK) { 824 verbose("invalid stack off=%d size=%d\n", off, size); 825 return -EACCES; 826 } 827 if (t == BPF_WRITE) { 828 if (!env->allow_ptr_leaks && 829 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && 830 size != BPF_REG_SIZE) { 831 verbose("attempt to corrupt spilled pointer on stack\n"); 832 return -EACCES; 833 } 834 err = check_stack_write(state, off, size, value_regno); 835 } else { 836 err = check_stack_read(state, off, size, value_regno); 837 } 838 } else if (state->regs[regno].type == PTR_TO_PACKET) { 839 if (t == BPF_WRITE) { 840 verbose("cannot write into packet\n"); 841 return -EACCES; 842 } 843 err = check_packet_access(env, regno, off, size); 844 if (!err && t == BPF_READ && value_regno >= 0) 845 mark_reg_unknown_value(state->regs, value_regno); 846 } else { 847 verbose("R%d invalid mem access '%s'\n", 848 regno, reg_type_str[reg->type]); 849 return -EACCES; 850 } 851 852 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && 853 state->regs[value_regno].type == UNKNOWN_VALUE) { 854 /* 1 or 2 byte load zero-extends, determine the number of 855 * zero upper bits. Not doing it fo 4 byte load, since 856 * such values cannot be added to ptr_to_packet anyway. 857 */ 858 state->regs[value_regno].imm = 64 - size * 8; 859 } 860 return err; 861 } 862 863 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn) 864 { 865 struct reg_state *regs = env->cur_state.regs; 866 int err; 867 868 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || 869 insn->imm != 0) { 870 verbose("BPF_XADD uses reserved fields\n"); 871 return -EINVAL; 872 } 873 874 /* check src1 operand */ 875 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 876 if (err) 877 return err; 878 879 /* check src2 operand */ 880 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 881 if (err) 882 return err; 883 884 /* check whether atomic_add can read the memory */ 885 err = check_mem_access(env, insn->dst_reg, insn->off, 886 BPF_SIZE(insn->code), BPF_READ, -1); 887 if (err) 888 return err; 889 890 /* check whether atomic_add can write into the same memory */ 891 return check_mem_access(env, insn->dst_reg, insn->off, 892 BPF_SIZE(insn->code), BPF_WRITE, -1); 893 } 894 895 /* when register 'regno' is passed into function that will read 'access_size' 896 * bytes from that pointer, make sure that it's within stack boundary 897 * and all elements of stack are initialized 898 */ 899 static int check_stack_boundary(struct verifier_env *env, int regno, 900 int access_size, bool zero_size_allowed, 901 struct bpf_call_arg_meta *meta) 902 { 903 struct verifier_state *state = &env->cur_state; 904 struct reg_state *regs = state->regs; 905 int off, i; 906 907 if (regs[regno].type != PTR_TO_STACK) { 908 if (zero_size_allowed && access_size == 0 && 909 regs[regno].type == CONST_IMM && 910 regs[regno].imm == 0) 911 return 0; 912 913 verbose("R%d type=%s expected=%s\n", regno, 914 reg_type_str[regs[regno].type], 915 reg_type_str[PTR_TO_STACK]); 916 return -EACCES; 917 } 918 919 off = regs[regno].imm; 920 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || 921 access_size <= 0) { 922 verbose("invalid stack type R%d off=%d access_size=%d\n", 923 regno, off, access_size); 924 return -EACCES; 925 } 926 927 if (meta && meta->raw_mode) { 928 meta->access_size = access_size; 929 meta->regno = regno; 930 return 0; 931 } 932 933 for (i = 0; i < access_size; i++) { 934 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { 935 verbose("invalid indirect read from stack off %d+%d size %d\n", 936 off, i, access_size); 937 return -EACCES; 938 } 939 } 940 return 0; 941 } 942 943 static int check_func_arg(struct verifier_env *env, u32 regno, 944 enum bpf_arg_type arg_type, 945 struct bpf_call_arg_meta *meta) 946 { 947 struct reg_state *reg = env->cur_state.regs + regno; 948 enum bpf_reg_type expected_type; 949 int err = 0; 950 951 if (arg_type == ARG_DONTCARE) 952 return 0; 953 954 if (reg->type == NOT_INIT) { 955 verbose("R%d !read_ok\n", regno); 956 return -EACCES; 957 } 958 959 if (arg_type == ARG_ANYTHING) { 960 if (is_pointer_value(env, regno)) { 961 verbose("R%d leaks addr into helper function\n", regno); 962 return -EACCES; 963 } 964 return 0; 965 } 966 967 if (arg_type == ARG_PTR_TO_MAP_KEY || 968 arg_type == ARG_PTR_TO_MAP_VALUE) { 969 expected_type = PTR_TO_STACK; 970 } else if (arg_type == ARG_CONST_STACK_SIZE || 971 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { 972 expected_type = CONST_IMM; 973 } else if (arg_type == ARG_CONST_MAP_PTR) { 974 expected_type = CONST_PTR_TO_MAP; 975 } else if (arg_type == ARG_PTR_TO_CTX) { 976 expected_type = PTR_TO_CTX; 977 } else if (arg_type == ARG_PTR_TO_STACK || 978 arg_type == ARG_PTR_TO_RAW_STACK) { 979 expected_type = PTR_TO_STACK; 980 /* One exception here. In case function allows for NULL to be 981 * passed in as argument, it's a CONST_IMM type. Final test 982 * happens during stack boundary checking. 983 */ 984 if (reg->type == CONST_IMM && reg->imm == 0) 985 expected_type = CONST_IMM; 986 meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK; 987 } else { 988 verbose("unsupported arg_type %d\n", arg_type); 989 return -EFAULT; 990 } 991 992 if (reg->type != expected_type) { 993 verbose("R%d type=%s expected=%s\n", regno, 994 reg_type_str[reg->type], reg_type_str[expected_type]); 995 return -EACCES; 996 } 997 998 if (arg_type == ARG_CONST_MAP_PTR) { 999 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ 1000 meta->map_ptr = reg->map_ptr; 1001 } else if (arg_type == ARG_PTR_TO_MAP_KEY) { 1002 /* bpf_map_xxx(..., map_ptr, ..., key) call: 1003 * check that [key, key + map->key_size) are within 1004 * stack limits and initialized 1005 */ 1006 if (!meta->map_ptr) { 1007 /* in function declaration map_ptr must come before 1008 * map_key, so that it's verified and known before 1009 * we have to check map_key here. Otherwise it means 1010 * that kernel subsystem misconfigured verifier 1011 */ 1012 verbose("invalid map_ptr to access map->key\n"); 1013 return -EACCES; 1014 } 1015 err = check_stack_boundary(env, regno, meta->map_ptr->key_size, 1016 false, NULL); 1017 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { 1018 /* bpf_map_xxx(..., map_ptr, ..., value) call: 1019 * check [value, value + map->value_size) validity 1020 */ 1021 if (!meta->map_ptr) { 1022 /* kernel subsystem misconfigured verifier */ 1023 verbose("invalid map_ptr to access map->value\n"); 1024 return -EACCES; 1025 } 1026 err = check_stack_boundary(env, regno, 1027 meta->map_ptr->value_size, 1028 false, NULL); 1029 } else if (arg_type == ARG_CONST_STACK_SIZE || 1030 arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { 1031 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO); 1032 1033 /* bpf_xxx(..., buf, len) call will access 'len' bytes 1034 * from stack pointer 'buf'. Check it 1035 * note: regno == len, regno - 1 == buf 1036 */ 1037 if (regno == 0) { 1038 /* kernel subsystem misconfigured verifier */ 1039 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n"); 1040 return -EACCES; 1041 } 1042 err = check_stack_boundary(env, regno - 1, reg->imm, 1043 zero_size_allowed, meta); 1044 } 1045 1046 return err; 1047 } 1048 1049 static int check_map_func_compatibility(struct bpf_map *map, int func_id) 1050 { 1051 if (!map) 1052 return 0; 1053 1054 /* We need a two way check, first is from map perspective ... */ 1055 switch (map->map_type) { 1056 case BPF_MAP_TYPE_PROG_ARRAY: 1057 if (func_id != BPF_FUNC_tail_call) 1058 goto error; 1059 break; 1060 case BPF_MAP_TYPE_PERF_EVENT_ARRAY: 1061 if (func_id != BPF_FUNC_perf_event_read && 1062 func_id != BPF_FUNC_perf_event_output) 1063 goto error; 1064 break; 1065 case BPF_MAP_TYPE_STACK_TRACE: 1066 if (func_id != BPF_FUNC_get_stackid) 1067 goto error; 1068 break; 1069 default: 1070 break; 1071 } 1072 1073 /* ... and second from the function itself. */ 1074 switch (func_id) { 1075 case BPF_FUNC_tail_call: 1076 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 1077 goto error; 1078 break; 1079 case BPF_FUNC_perf_event_read: 1080 case BPF_FUNC_perf_event_output: 1081 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) 1082 goto error; 1083 break; 1084 case BPF_FUNC_get_stackid: 1085 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) 1086 goto error; 1087 break; 1088 default: 1089 break; 1090 } 1091 1092 return 0; 1093 error: 1094 verbose("cannot pass map_type %d into func %d\n", 1095 map->map_type, func_id); 1096 return -EINVAL; 1097 } 1098 1099 static int check_raw_mode(const struct bpf_func_proto *fn) 1100 { 1101 int count = 0; 1102 1103 if (fn->arg1_type == ARG_PTR_TO_RAW_STACK) 1104 count++; 1105 if (fn->arg2_type == ARG_PTR_TO_RAW_STACK) 1106 count++; 1107 if (fn->arg3_type == ARG_PTR_TO_RAW_STACK) 1108 count++; 1109 if (fn->arg4_type == ARG_PTR_TO_RAW_STACK) 1110 count++; 1111 if (fn->arg5_type == ARG_PTR_TO_RAW_STACK) 1112 count++; 1113 1114 return count > 1 ? -EINVAL : 0; 1115 } 1116 1117 static void clear_all_pkt_pointers(struct verifier_env *env) 1118 { 1119 struct verifier_state *state = &env->cur_state; 1120 struct reg_state *regs = state->regs, *reg; 1121 int i; 1122 1123 for (i = 0; i < MAX_BPF_REG; i++) 1124 if (regs[i].type == PTR_TO_PACKET || 1125 regs[i].type == PTR_TO_PACKET_END) 1126 mark_reg_unknown_value(regs, i); 1127 1128 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1129 if (state->stack_slot_type[i] != STACK_SPILL) 1130 continue; 1131 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1132 if (reg->type != PTR_TO_PACKET && 1133 reg->type != PTR_TO_PACKET_END) 1134 continue; 1135 reg->type = UNKNOWN_VALUE; 1136 reg->imm = 0; 1137 } 1138 } 1139 1140 static int check_call(struct verifier_env *env, int func_id) 1141 { 1142 struct verifier_state *state = &env->cur_state; 1143 const struct bpf_func_proto *fn = NULL; 1144 struct reg_state *regs = state->regs; 1145 struct reg_state *reg; 1146 struct bpf_call_arg_meta meta; 1147 bool changes_data; 1148 int i, err; 1149 1150 /* find function prototype */ 1151 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { 1152 verbose("invalid func %d\n", func_id); 1153 return -EINVAL; 1154 } 1155 1156 if (env->prog->aux->ops->get_func_proto) 1157 fn = env->prog->aux->ops->get_func_proto(func_id); 1158 1159 if (!fn) { 1160 verbose("unknown func %d\n", func_id); 1161 return -EINVAL; 1162 } 1163 1164 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1165 if (!env->prog->gpl_compatible && fn->gpl_only) { 1166 verbose("cannot call GPL only function from proprietary program\n"); 1167 return -EINVAL; 1168 } 1169 1170 changes_data = bpf_helper_changes_skb_data(fn->func); 1171 1172 memset(&meta, 0, sizeof(meta)); 1173 1174 /* We only support one arg being in raw mode at the moment, which 1175 * is sufficient for the helper functions we have right now. 1176 */ 1177 err = check_raw_mode(fn); 1178 if (err) { 1179 verbose("kernel subsystem misconfigured func %d\n", func_id); 1180 return err; 1181 } 1182 1183 /* check args */ 1184 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); 1185 if (err) 1186 return err; 1187 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); 1188 if (err) 1189 return err; 1190 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); 1191 if (err) 1192 return err; 1193 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); 1194 if (err) 1195 return err; 1196 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); 1197 if (err) 1198 return err; 1199 1200 /* Mark slots with STACK_MISC in case of raw mode, stack offset 1201 * is inferred from register state. 1202 */ 1203 for (i = 0; i < meta.access_size; i++) { 1204 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1); 1205 if (err) 1206 return err; 1207 } 1208 1209 /* reset caller saved regs */ 1210 for (i = 0; i < CALLER_SAVED_REGS; i++) { 1211 reg = regs + caller_saved[i]; 1212 reg->type = NOT_INIT; 1213 reg->imm = 0; 1214 } 1215 1216 /* update return register */ 1217 if (fn->ret_type == RET_INTEGER) { 1218 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1219 } else if (fn->ret_type == RET_VOID) { 1220 regs[BPF_REG_0].type = NOT_INIT; 1221 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { 1222 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; 1223 /* remember map_ptr, so that check_map_access() 1224 * can check 'value_size' boundary of memory access 1225 * to map element returned from bpf_map_lookup_elem() 1226 */ 1227 if (meta.map_ptr == NULL) { 1228 verbose("kernel subsystem misconfigured verifier\n"); 1229 return -EINVAL; 1230 } 1231 regs[BPF_REG_0].map_ptr = meta.map_ptr; 1232 } else { 1233 verbose("unknown return type %d of func %d\n", 1234 fn->ret_type, func_id); 1235 return -EINVAL; 1236 } 1237 1238 err = check_map_func_compatibility(meta.map_ptr, func_id); 1239 if (err) 1240 return err; 1241 1242 if (changes_data) 1243 clear_all_pkt_pointers(env); 1244 return 0; 1245 } 1246 1247 static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn) 1248 { 1249 struct reg_state *regs = env->cur_state.regs; 1250 struct reg_state *dst_reg = ®s[insn->dst_reg]; 1251 struct reg_state *src_reg = ®s[insn->src_reg]; 1252 s32 imm; 1253 1254 if (BPF_SRC(insn->code) == BPF_K) { 1255 /* pkt_ptr += imm */ 1256 imm = insn->imm; 1257 1258 add_imm: 1259 if (imm <= 0) { 1260 verbose("addition of negative constant to packet pointer is not allowed\n"); 1261 return -EACCES; 1262 } 1263 if (imm >= MAX_PACKET_OFF || 1264 imm + dst_reg->off >= MAX_PACKET_OFF) { 1265 verbose("constant %d is too large to add to packet pointer\n", 1266 imm); 1267 return -EACCES; 1268 } 1269 /* a constant was added to pkt_ptr. 1270 * Remember it while keeping the same 'id' 1271 */ 1272 dst_reg->off += imm; 1273 } else { 1274 if (src_reg->type == CONST_IMM) { 1275 /* pkt_ptr += reg where reg is known constant */ 1276 imm = src_reg->imm; 1277 goto add_imm; 1278 } 1279 /* disallow pkt_ptr += reg 1280 * if reg is not uknown_value with guaranteed zero upper bits 1281 * otherwise pkt_ptr may overflow and addition will become 1282 * subtraction which is not allowed 1283 */ 1284 if (src_reg->type != UNKNOWN_VALUE) { 1285 verbose("cannot add '%s' to ptr_to_packet\n", 1286 reg_type_str[src_reg->type]); 1287 return -EACCES; 1288 } 1289 if (src_reg->imm < 48) { 1290 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", 1291 src_reg->imm); 1292 return -EACCES; 1293 } 1294 /* dst_reg stays as pkt_ptr type and since some positive 1295 * integer value was added to the pointer, increment its 'id' 1296 */ 1297 dst_reg->id++; 1298 1299 /* something was added to pkt_ptr, set range and off to zero */ 1300 dst_reg->off = 0; 1301 dst_reg->range = 0; 1302 } 1303 return 0; 1304 } 1305 1306 static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn) 1307 { 1308 struct reg_state *regs = env->cur_state.regs; 1309 struct reg_state *dst_reg = ®s[insn->dst_reg]; 1310 u8 opcode = BPF_OP(insn->code); 1311 s64 imm_log2; 1312 1313 /* for type == UNKNOWN_VALUE: 1314 * imm > 0 -> number of zero upper bits 1315 * imm == 0 -> don't track which is the same as all bits can be non-zero 1316 */ 1317 1318 if (BPF_SRC(insn->code) == BPF_X) { 1319 struct reg_state *src_reg = ®s[insn->src_reg]; 1320 1321 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && 1322 dst_reg->imm && opcode == BPF_ADD) { 1323 /* dreg += sreg 1324 * where both have zero upper bits. Adding them 1325 * can only result making one more bit non-zero 1326 * in the larger value. 1327 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) 1328 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) 1329 */ 1330 dst_reg->imm = min(dst_reg->imm, src_reg->imm); 1331 dst_reg->imm--; 1332 return 0; 1333 } 1334 if (src_reg->type == CONST_IMM && src_reg->imm > 0 && 1335 dst_reg->imm && opcode == BPF_ADD) { 1336 /* dreg += sreg 1337 * where dreg has zero upper bits and sreg is const. 1338 * Adding them can only result making one more bit 1339 * non-zero in the larger value. 1340 */ 1341 imm_log2 = __ilog2_u64((long long)src_reg->imm); 1342 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1343 dst_reg->imm--; 1344 return 0; 1345 } 1346 /* all other cases non supported yet, just mark dst_reg */ 1347 dst_reg->imm = 0; 1348 return 0; 1349 } 1350 1351 /* sign extend 32-bit imm into 64-bit to make sure that 1352 * negative values occupy bit 63. Note ilog2() would have 1353 * been incorrect, since sizeof(insn->imm) == 4 1354 */ 1355 imm_log2 = __ilog2_u64((long long)insn->imm); 1356 1357 if (dst_reg->imm && opcode == BPF_LSH) { 1358 /* reg <<= imm 1359 * if reg was a result of 2 byte load, then its imm == 48 1360 * which means that upper 48 bits are zero and shifting this reg 1361 * left by 4 would mean that upper 44 bits are still zero 1362 */ 1363 dst_reg->imm -= insn->imm; 1364 } else if (dst_reg->imm && opcode == BPF_MUL) { 1365 /* reg *= imm 1366 * if multiplying by 14 subtract 4 1367 * This is conservative calculation of upper zero bits. 1368 * It's not trying to special case insn->imm == 1 or 0 cases 1369 */ 1370 dst_reg->imm -= imm_log2 + 1; 1371 } else if (opcode == BPF_AND) { 1372 /* reg &= imm */ 1373 dst_reg->imm = 63 - imm_log2; 1374 } else if (dst_reg->imm && opcode == BPF_ADD) { 1375 /* reg += imm */ 1376 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); 1377 dst_reg->imm--; 1378 } else if (opcode == BPF_RSH) { 1379 /* reg >>= imm 1380 * which means that after right shift, upper bits will be zero 1381 * note that verifier already checked that 1382 * 0 <= imm < 64 for shift insn 1383 */ 1384 dst_reg->imm += insn->imm; 1385 if (unlikely(dst_reg->imm > 64)) 1386 /* some dumb code did: 1387 * r2 = *(u32 *)mem; 1388 * r2 >>= 32; 1389 * and all bits are zero now */ 1390 dst_reg->imm = 64; 1391 } else { 1392 /* all other alu ops, means that we don't know what will 1393 * happen to the value, mark it with unknown number of zero bits 1394 */ 1395 dst_reg->imm = 0; 1396 } 1397 1398 if (dst_reg->imm < 0) { 1399 /* all 64 bits of the register can contain non-zero bits 1400 * and such value cannot be added to ptr_to_packet, since it 1401 * may overflow, mark it as unknown to avoid further eval 1402 */ 1403 dst_reg->imm = 0; 1404 } 1405 return 0; 1406 } 1407 1408 static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn) 1409 { 1410 struct reg_state *regs = env->cur_state.regs; 1411 struct reg_state *dst_reg = ®s[insn->dst_reg]; 1412 struct reg_state *src_reg = ®s[insn->src_reg]; 1413 u8 opcode = BPF_OP(insn->code); 1414 1415 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn. 1416 * Don't care about overflow or negative values, just add them 1417 */ 1418 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) 1419 dst_reg->imm += insn->imm; 1420 else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && 1421 src_reg->type == CONST_IMM) 1422 dst_reg->imm += src_reg->imm; 1423 else 1424 mark_reg_unknown_value(regs, insn->dst_reg); 1425 return 0; 1426 } 1427 1428 /* check validity of 32-bit and 64-bit arithmetic operations */ 1429 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn) 1430 { 1431 struct reg_state *regs = env->cur_state.regs, *dst_reg; 1432 u8 opcode = BPF_OP(insn->code); 1433 int err; 1434 1435 if (opcode == BPF_END || opcode == BPF_NEG) { 1436 if (opcode == BPF_NEG) { 1437 if (BPF_SRC(insn->code) != 0 || 1438 insn->src_reg != BPF_REG_0 || 1439 insn->off != 0 || insn->imm != 0) { 1440 verbose("BPF_NEG uses reserved fields\n"); 1441 return -EINVAL; 1442 } 1443 } else { 1444 if (insn->src_reg != BPF_REG_0 || insn->off != 0 || 1445 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { 1446 verbose("BPF_END uses reserved fields\n"); 1447 return -EINVAL; 1448 } 1449 } 1450 1451 /* check src operand */ 1452 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1453 if (err) 1454 return err; 1455 1456 if (is_pointer_value(env, insn->dst_reg)) { 1457 verbose("R%d pointer arithmetic prohibited\n", 1458 insn->dst_reg); 1459 return -EACCES; 1460 } 1461 1462 /* check dest operand */ 1463 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1464 if (err) 1465 return err; 1466 1467 } else if (opcode == BPF_MOV) { 1468 1469 if (BPF_SRC(insn->code) == BPF_X) { 1470 if (insn->imm != 0 || insn->off != 0) { 1471 verbose("BPF_MOV uses reserved fields\n"); 1472 return -EINVAL; 1473 } 1474 1475 /* check src operand */ 1476 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1477 if (err) 1478 return err; 1479 } else { 1480 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 1481 verbose("BPF_MOV uses reserved fields\n"); 1482 return -EINVAL; 1483 } 1484 } 1485 1486 /* check dest operand */ 1487 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1488 if (err) 1489 return err; 1490 1491 if (BPF_SRC(insn->code) == BPF_X) { 1492 if (BPF_CLASS(insn->code) == BPF_ALU64) { 1493 /* case: R1 = R2 1494 * copy register state to dest reg 1495 */ 1496 regs[insn->dst_reg] = regs[insn->src_reg]; 1497 } else { 1498 if (is_pointer_value(env, insn->src_reg)) { 1499 verbose("R%d partial copy of pointer\n", 1500 insn->src_reg); 1501 return -EACCES; 1502 } 1503 regs[insn->dst_reg].type = UNKNOWN_VALUE; 1504 regs[insn->dst_reg].map_ptr = NULL; 1505 } 1506 } else { 1507 /* case: R = imm 1508 * remember the value we stored into this reg 1509 */ 1510 regs[insn->dst_reg].type = CONST_IMM; 1511 regs[insn->dst_reg].imm = insn->imm; 1512 } 1513 1514 } else if (opcode > BPF_END) { 1515 verbose("invalid BPF_ALU opcode %x\n", opcode); 1516 return -EINVAL; 1517 1518 } else { /* all other ALU ops: and, sub, xor, add, ... */ 1519 1520 if (BPF_SRC(insn->code) == BPF_X) { 1521 if (insn->imm != 0 || insn->off != 0) { 1522 verbose("BPF_ALU uses reserved fields\n"); 1523 return -EINVAL; 1524 } 1525 /* check src1 operand */ 1526 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1527 if (err) 1528 return err; 1529 } else { 1530 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 1531 verbose("BPF_ALU uses reserved fields\n"); 1532 return -EINVAL; 1533 } 1534 } 1535 1536 /* check src2 operand */ 1537 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1538 if (err) 1539 return err; 1540 1541 if ((opcode == BPF_MOD || opcode == BPF_DIV) && 1542 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { 1543 verbose("div by zero\n"); 1544 return -EINVAL; 1545 } 1546 1547 if ((opcode == BPF_LSH || opcode == BPF_RSH || 1548 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { 1549 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; 1550 1551 if (insn->imm < 0 || insn->imm >= size) { 1552 verbose("invalid shift %d\n", insn->imm); 1553 return -EINVAL; 1554 } 1555 } 1556 1557 /* check dest operand */ 1558 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 1559 if (err) 1560 return err; 1561 1562 dst_reg = ®s[insn->dst_reg]; 1563 1564 /* pattern match 'bpf_add Rx, imm' instruction */ 1565 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && 1566 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { 1567 dst_reg->type = PTR_TO_STACK; 1568 dst_reg->imm = insn->imm; 1569 return 0; 1570 } else if (opcode == BPF_ADD && 1571 BPF_CLASS(insn->code) == BPF_ALU64 && 1572 dst_reg->type == PTR_TO_PACKET) { 1573 /* ptr_to_packet += K|X */ 1574 return check_packet_ptr_add(env, insn); 1575 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 1576 dst_reg->type == UNKNOWN_VALUE && 1577 env->allow_ptr_leaks) { 1578 /* unknown += K|X */ 1579 return evaluate_reg_alu(env, insn); 1580 } else if (BPF_CLASS(insn->code) == BPF_ALU64 && 1581 dst_reg->type == CONST_IMM && 1582 env->allow_ptr_leaks) { 1583 /* reg_imm += K|X */ 1584 return evaluate_reg_imm_alu(env, insn); 1585 } else if (is_pointer_value(env, insn->dst_reg)) { 1586 verbose("R%d pointer arithmetic prohibited\n", 1587 insn->dst_reg); 1588 return -EACCES; 1589 } else if (BPF_SRC(insn->code) == BPF_X && 1590 is_pointer_value(env, insn->src_reg)) { 1591 verbose("R%d pointer arithmetic prohibited\n", 1592 insn->src_reg); 1593 return -EACCES; 1594 } 1595 1596 /* mark dest operand */ 1597 mark_reg_unknown_value(regs, insn->dst_reg); 1598 } 1599 1600 return 0; 1601 } 1602 1603 static void find_good_pkt_pointers(struct verifier_env *env, 1604 struct reg_state *dst_reg) 1605 { 1606 struct verifier_state *state = &env->cur_state; 1607 struct reg_state *regs = state->regs, *reg; 1608 int i; 1609 /* r2 = r3; 1610 * r2 += 8 1611 * if (r2 > pkt_end) goto somewhere 1612 * r2 == dst_reg, pkt_end == src_reg, 1613 * r2=pkt(id=n,off=8,r=0) 1614 * r3=pkt(id=n,off=0,r=0) 1615 * find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) 1616 * so that range of bytes [r3, r3 + 8) is safe to access 1617 */ 1618 for (i = 0; i < MAX_BPF_REG; i++) 1619 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) 1620 regs[i].range = dst_reg->off; 1621 1622 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 1623 if (state->stack_slot_type[i] != STACK_SPILL) 1624 continue; 1625 reg = &state->spilled_regs[i / BPF_REG_SIZE]; 1626 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) 1627 reg->range = dst_reg->off; 1628 } 1629 } 1630 1631 static int check_cond_jmp_op(struct verifier_env *env, 1632 struct bpf_insn *insn, int *insn_idx) 1633 { 1634 struct reg_state *regs = env->cur_state.regs, *dst_reg; 1635 struct verifier_state *other_branch; 1636 u8 opcode = BPF_OP(insn->code); 1637 int err; 1638 1639 if (opcode > BPF_EXIT) { 1640 verbose("invalid BPF_JMP opcode %x\n", opcode); 1641 return -EINVAL; 1642 } 1643 1644 if (BPF_SRC(insn->code) == BPF_X) { 1645 if (insn->imm != 0) { 1646 verbose("BPF_JMP uses reserved fields\n"); 1647 return -EINVAL; 1648 } 1649 1650 /* check src1 operand */ 1651 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1652 if (err) 1653 return err; 1654 1655 if (is_pointer_value(env, insn->src_reg)) { 1656 verbose("R%d pointer comparison prohibited\n", 1657 insn->src_reg); 1658 return -EACCES; 1659 } 1660 } else { 1661 if (insn->src_reg != BPF_REG_0) { 1662 verbose("BPF_JMP uses reserved fields\n"); 1663 return -EINVAL; 1664 } 1665 } 1666 1667 /* check src2 operand */ 1668 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1669 if (err) 1670 return err; 1671 1672 dst_reg = ®s[insn->dst_reg]; 1673 1674 /* detect if R == 0 where R was initialized to zero earlier */ 1675 if (BPF_SRC(insn->code) == BPF_K && 1676 (opcode == BPF_JEQ || opcode == BPF_JNE) && 1677 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { 1678 if (opcode == BPF_JEQ) { 1679 /* if (imm == imm) goto pc+off; 1680 * only follow the goto, ignore fall-through 1681 */ 1682 *insn_idx += insn->off; 1683 return 0; 1684 } else { 1685 /* if (imm != imm) goto pc+off; 1686 * only follow fall-through branch, since 1687 * that's where the program will go 1688 */ 1689 return 0; 1690 } 1691 } 1692 1693 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); 1694 if (!other_branch) 1695 return -EFAULT; 1696 1697 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */ 1698 if (BPF_SRC(insn->code) == BPF_K && 1699 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && 1700 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { 1701 if (opcode == BPF_JEQ) { 1702 /* next fallthrough insn can access memory via 1703 * this register 1704 */ 1705 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; 1706 /* branch targer cannot access it, since reg == 0 */ 1707 mark_reg_unknown_value(other_branch->regs, 1708 insn->dst_reg); 1709 } else { 1710 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; 1711 mark_reg_unknown_value(regs, insn->dst_reg); 1712 } 1713 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && 1714 dst_reg->type == PTR_TO_PACKET && 1715 regs[insn->src_reg].type == PTR_TO_PACKET_END) { 1716 find_good_pkt_pointers(env, dst_reg); 1717 } else if (is_pointer_value(env, insn->dst_reg)) { 1718 verbose("R%d pointer comparison prohibited\n", insn->dst_reg); 1719 return -EACCES; 1720 } 1721 if (log_level) 1722 print_verifier_state(&env->cur_state); 1723 return 0; 1724 } 1725 1726 /* return the map pointer stored inside BPF_LD_IMM64 instruction */ 1727 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) 1728 { 1729 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; 1730 1731 return (struct bpf_map *) (unsigned long) imm64; 1732 } 1733 1734 /* verify BPF_LD_IMM64 instruction */ 1735 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn) 1736 { 1737 struct reg_state *regs = env->cur_state.regs; 1738 int err; 1739 1740 if (BPF_SIZE(insn->code) != BPF_DW) { 1741 verbose("invalid BPF_LD_IMM insn\n"); 1742 return -EINVAL; 1743 } 1744 if (insn->off != 0) { 1745 verbose("BPF_LD_IMM64 uses reserved fields\n"); 1746 return -EINVAL; 1747 } 1748 1749 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1750 if (err) 1751 return err; 1752 1753 if (insn->src_reg == 0) 1754 /* generic move 64-bit immediate into a register */ 1755 return 0; 1756 1757 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ 1758 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); 1759 1760 regs[insn->dst_reg].type = CONST_PTR_TO_MAP; 1761 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); 1762 return 0; 1763 } 1764 1765 static bool may_access_skb(enum bpf_prog_type type) 1766 { 1767 switch (type) { 1768 case BPF_PROG_TYPE_SOCKET_FILTER: 1769 case BPF_PROG_TYPE_SCHED_CLS: 1770 case BPF_PROG_TYPE_SCHED_ACT: 1771 return true; 1772 default: 1773 return false; 1774 } 1775 } 1776 1777 /* verify safety of LD_ABS|LD_IND instructions: 1778 * - they can only appear in the programs where ctx == skb 1779 * - since they are wrappers of function calls, they scratch R1-R5 registers, 1780 * preserve R6-R9, and store return value into R0 1781 * 1782 * Implicit input: 1783 * ctx == skb == R6 == CTX 1784 * 1785 * Explicit input: 1786 * SRC == any register 1787 * IMM == 32-bit immediate 1788 * 1789 * Output: 1790 * R0 - 8/16/32-bit skb data converted to cpu endianness 1791 */ 1792 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn) 1793 { 1794 struct reg_state *regs = env->cur_state.regs; 1795 u8 mode = BPF_MODE(insn->code); 1796 struct reg_state *reg; 1797 int i, err; 1798 1799 if (!may_access_skb(env->prog->type)) { 1800 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); 1801 return -EINVAL; 1802 } 1803 1804 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || 1805 BPF_SIZE(insn->code) == BPF_DW || 1806 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { 1807 verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); 1808 return -EINVAL; 1809 } 1810 1811 /* check whether implicit source operand (register R6) is readable */ 1812 err = check_reg_arg(regs, BPF_REG_6, SRC_OP); 1813 if (err) 1814 return err; 1815 1816 if (regs[BPF_REG_6].type != PTR_TO_CTX) { 1817 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); 1818 return -EINVAL; 1819 } 1820 1821 if (mode == BPF_IND) { 1822 /* check explicit source operand */ 1823 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1824 if (err) 1825 return err; 1826 } 1827 1828 /* reset caller saved regs to unreadable */ 1829 for (i = 0; i < CALLER_SAVED_REGS; i++) { 1830 reg = regs + caller_saved[i]; 1831 reg->type = NOT_INIT; 1832 reg->imm = 0; 1833 } 1834 1835 /* mark destination R0 register as readable, since it contains 1836 * the value fetched from the packet 1837 */ 1838 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1839 return 0; 1840 } 1841 1842 /* non-recursive DFS pseudo code 1843 * 1 procedure DFS-iterative(G,v): 1844 * 2 label v as discovered 1845 * 3 let S be a stack 1846 * 4 S.push(v) 1847 * 5 while S is not empty 1848 * 6 t <- S.pop() 1849 * 7 if t is what we're looking for: 1850 * 8 return t 1851 * 9 for all edges e in G.adjacentEdges(t) do 1852 * 10 if edge e is already labelled 1853 * 11 continue with the next edge 1854 * 12 w <- G.adjacentVertex(t,e) 1855 * 13 if vertex w is not discovered and not explored 1856 * 14 label e as tree-edge 1857 * 15 label w as discovered 1858 * 16 S.push(w) 1859 * 17 continue at 5 1860 * 18 else if vertex w is discovered 1861 * 19 label e as back-edge 1862 * 20 else 1863 * 21 // vertex w is explored 1864 * 22 label e as forward- or cross-edge 1865 * 23 label t as explored 1866 * 24 S.pop() 1867 * 1868 * convention: 1869 * 0x10 - discovered 1870 * 0x11 - discovered and fall-through edge labelled 1871 * 0x12 - discovered and fall-through and branch edges labelled 1872 * 0x20 - explored 1873 */ 1874 1875 enum { 1876 DISCOVERED = 0x10, 1877 EXPLORED = 0x20, 1878 FALLTHROUGH = 1, 1879 BRANCH = 2, 1880 }; 1881 1882 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L) 1883 1884 static int *insn_stack; /* stack of insns to process */ 1885 static int cur_stack; /* current stack index */ 1886 static int *insn_state; 1887 1888 /* t, w, e - match pseudo-code above: 1889 * t - index of current instruction 1890 * w - next instruction 1891 * e - edge 1892 */ 1893 static int push_insn(int t, int w, int e, struct verifier_env *env) 1894 { 1895 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) 1896 return 0; 1897 1898 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) 1899 return 0; 1900 1901 if (w < 0 || w >= env->prog->len) { 1902 verbose("jump out of range from insn %d to %d\n", t, w); 1903 return -EINVAL; 1904 } 1905 1906 if (e == BRANCH) 1907 /* mark branch target for state pruning */ 1908 env->explored_states[w] = STATE_LIST_MARK; 1909 1910 if (insn_state[w] == 0) { 1911 /* tree-edge */ 1912 insn_state[t] = DISCOVERED | e; 1913 insn_state[w] = DISCOVERED; 1914 if (cur_stack >= env->prog->len) 1915 return -E2BIG; 1916 insn_stack[cur_stack++] = w; 1917 return 1; 1918 } else if ((insn_state[w] & 0xF0) == DISCOVERED) { 1919 verbose("back-edge from insn %d to %d\n", t, w); 1920 return -EINVAL; 1921 } else if (insn_state[w] == EXPLORED) { 1922 /* forward- or cross-edge */ 1923 insn_state[t] = DISCOVERED | e; 1924 } else { 1925 verbose("insn state internal bug\n"); 1926 return -EFAULT; 1927 } 1928 return 0; 1929 } 1930 1931 /* non-recursive depth-first-search to detect loops in BPF program 1932 * loop == back-edge in directed graph 1933 */ 1934 static int check_cfg(struct verifier_env *env) 1935 { 1936 struct bpf_insn *insns = env->prog->insnsi; 1937 int insn_cnt = env->prog->len; 1938 int ret = 0; 1939 int i, t; 1940 1941 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 1942 if (!insn_state) 1943 return -ENOMEM; 1944 1945 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 1946 if (!insn_stack) { 1947 kfree(insn_state); 1948 return -ENOMEM; 1949 } 1950 1951 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ 1952 insn_stack[0] = 0; /* 0 is the first instruction */ 1953 cur_stack = 1; 1954 1955 peek_stack: 1956 if (cur_stack == 0) 1957 goto check_state; 1958 t = insn_stack[cur_stack - 1]; 1959 1960 if (BPF_CLASS(insns[t].code) == BPF_JMP) { 1961 u8 opcode = BPF_OP(insns[t].code); 1962 1963 if (opcode == BPF_EXIT) { 1964 goto mark_explored; 1965 } else if (opcode == BPF_CALL) { 1966 ret = push_insn(t, t + 1, FALLTHROUGH, env); 1967 if (ret == 1) 1968 goto peek_stack; 1969 else if (ret < 0) 1970 goto err_free; 1971 if (t + 1 < insn_cnt) 1972 env->explored_states[t + 1] = STATE_LIST_MARK; 1973 } else if (opcode == BPF_JA) { 1974 if (BPF_SRC(insns[t].code) != BPF_K) { 1975 ret = -EINVAL; 1976 goto err_free; 1977 } 1978 /* unconditional jump with single edge */ 1979 ret = push_insn(t, t + insns[t].off + 1, 1980 FALLTHROUGH, env); 1981 if (ret == 1) 1982 goto peek_stack; 1983 else if (ret < 0) 1984 goto err_free; 1985 /* tell verifier to check for equivalent states 1986 * after every call and jump 1987 */ 1988 if (t + 1 < insn_cnt) 1989 env->explored_states[t + 1] = STATE_LIST_MARK; 1990 } else { 1991 /* conditional jump with two edges */ 1992 ret = push_insn(t, t + 1, FALLTHROUGH, env); 1993 if (ret == 1) 1994 goto peek_stack; 1995 else if (ret < 0) 1996 goto err_free; 1997 1998 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); 1999 if (ret == 1) 2000 goto peek_stack; 2001 else if (ret < 0) 2002 goto err_free; 2003 } 2004 } else { 2005 /* all other non-branch instructions with single 2006 * fall-through edge 2007 */ 2008 ret = push_insn(t, t + 1, FALLTHROUGH, env); 2009 if (ret == 1) 2010 goto peek_stack; 2011 else if (ret < 0) 2012 goto err_free; 2013 } 2014 2015 mark_explored: 2016 insn_state[t] = EXPLORED; 2017 if (cur_stack-- <= 0) { 2018 verbose("pop stack internal bug\n"); 2019 ret = -EFAULT; 2020 goto err_free; 2021 } 2022 goto peek_stack; 2023 2024 check_state: 2025 for (i = 0; i < insn_cnt; i++) { 2026 if (insn_state[i] != EXPLORED) { 2027 verbose("unreachable insn %d\n", i); 2028 ret = -EINVAL; 2029 goto err_free; 2030 } 2031 } 2032 ret = 0; /* cfg looks good */ 2033 2034 err_free: 2035 kfree(insn_state); 2036 kfree(insn_stack); 2037 return ret; 2038 } 2039 2040 /* the following conditions reduce the number of explored insns 2041 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet 2042 */ 2043 static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur) 2044 { 2045 if (old->id != cur->id) 2046 return false; 2047 2048 /* old ptr_to_packet is more conservative, since it allows smaller 2049 * range. Ex: 2050 * old(off=0,r=10) is equal to cur(off=0,r=20), because 2051 * old(off=0,r=10) means that with range=10 the verifier proceeded 2052 * further and found no issues with the program. Now we're in the same 2053 * spot with cur(off=0,r=20), so we're safe too, since anything further 2054 * will only be looking at most 10 bytes after this pointer. 2055 */ 2056 if (old->off == cur->off && old->range < cur->range) 2057 return true; 2058 2059 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) 2060 * since both cannot be used for packet access and safe(old) 2061 * pointer has smaller off that could be used for further 2062 * 'if (ptr > data_end)' check 2063 * Ex: 2064 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean 2065 * that we cannot access the packet. 2066 * The safe range is: 2067 * [ptr, ptr + range - off) 2068 * so whenever off >=range, it means no safe bytes from this pointer. 2069 * When comparing old->off <= cur->off, it means that older code 2070 * went with smaller offset and that offset was later 2071 * used to figure out the safe range after 'if (ptr > data_end)' check 2072 * Say, 'old' state was explored like: 2073 * ... R3(off=0, r=0) 2074 * R4 = R3 + 20 2075 * ... now R4(off=20,r=0) <-- here 2076 * if (R4 > data_end) 2077 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. 2078 * ... the code further went all the way to bpf_exit. 2079 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). 2080 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier 2081 * goes further, such cur_R4 will give larger safe packet range after 2082 * 'if (R4 > data_end)' and all further insn were already good with r=20, 2083 * so they will be good with r=30 and we can prune the search. 2084 */ 2085 if (old->off <= cur->off && 2086 old->off >= old->range && cur->off >= cur->range) 2087 return true; 2088 2089 return false; 2090 } 2091 2092 /* compare two verifier states 2093 * 2094 * all states stored in state_list are known to be valid, since 2095 * verifier reached 'bpf_exit' instruction through them 2096 * 2097 * this function is called when verifier exploring different branches of 2098 * execution popped from the state stack. If it sees an old state that has 2099 * more strict register state and more strict stack state then this execution 2100 * branch doesn't need to be explored further, since verifier already 2101 * concluded that more strict state leads to valid finish. 2102 * 2103 * Therefore two states are equivalent if register state is more conservative 2104 * and explored stack state is more conservative than the current one. 2105 * Example: 2106 * explored current 2107 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) 2108 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) 2109 * 2110 * In other words if current stack state (one being explored) has more 2111 * valid slots than old one that already passed validation, it means 2112 * the verifier can stop exploring and conclude that current state is valid too 2113 * 2114 * Similarly with registers. If explored state has register type as invalid 2115 * whereas register type in current state is meaningful, it means that 2116 * the current state will reach 'bpf_exit' instruction safely 2117 */ 2118 static bool states_equal(struct verifier_state *old, struct verifier_state *cur) 2119 { 2120 struct reg_state *rold, *rcur; 2121 int i; 2122 2123 for (i = 0; i < MAX_BPF_REG; i++) { 2124 rold = &old->regs[i]; 2125 rcur = &cur->regs[i]; 2126 2127 if (memcmp(rold, rcur, sizeof(*rold)) == 0) 2128 continue; 2129 2130 if (rold->type == NOT_INIT || 2131 (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT)) 2132 continue; 2133 2134 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && 2135 compare_ptrs_to_packet(rold, rcur)) 2136 continue; 2137 2138 return false; 2139 } 2140 2141 for (i = 0; i < MAX_BPF_STACK; i++) { 2142 if (old->stack_slot_type[i] == STACK_INVALID) 2143 continue; 2144 if (old->stack_slot_type[i] != cur->stack_slot_type[i]) 2145 /* Ex: old explored (safe) state has STACK_SPILL in 2146 * this stack slot, but current has has STACK_MISC -> 2147 * this verifier states are not equivalent, 2148 * return false to continue verification of this path 2149 */ 2150 return false; 2151 if (i % BPF_REG_SIZE) 2152 continue; 2153 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], 2154 &cur->spilled_regs[i / BPF_REG_SIZE], 2155 sizeof(old->spilled_regs[0]))) 2156 /* when explored and current stack slot types are 2157 * the same, check that stored pointers types 2158 * are the same as well. 2159 * Ex: explored safe path could have stored 2160 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8} 2161 * but current path has stored: 2162 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16} 2163 * such verifier states are not equivalent. 2164 * return false to continue verification of this path 2165 */ 2166 return false; 2167 else 2168 continue; 2169 } 2170 return true; 2171 } 2172 2173 static int is_state_visited(struct verifier_env *env, int insn_idx) 2174 { 2175 struct verifier_state_list *new_sl; 2176 struct verifier_state_list *sl; 2177 2178 sl = env->explored_states[insn_idx]; 2179 if (!sl) 2180 /* this 'insn_idx' instruction wasn't marked, so we will not 2181 * be doing state search here 2182 */ 2183 return 0; 2184 2185 while (sl != STATE_LIST_MARK) { 2186 if (states_equal(&sl->state, &env->cur_state)) 2187 /* reached equivalent register/stack state, 2188 * prune the search 2189 */ 2190 return 1; 2191 sl = sl->next; 2192 } 2193 2194 /* there were no equivalent states, remember current one. 2195 * technically the current state is not proven to be safe yet, 2196 * but it will either reach bpf_exit (which means it's safe) or 2197 * it will be rejected. Since there are no loops, we won't be 2198 * seeing this 'insn_idx' instruction again on the way to bpf_exit 2199 */ 2200 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER); 2201 if (!new_sl) 2202 return -ENOMEM; 2203 2204 /* add new state to the head of linked list */ 2205 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); 2206 new_sl->next = env->explored_states[insn_idx]; 2207 env->explored_states[insn_idx] = new_sl; 2208 return 0; 2209 } 2210 2211 static int do_check(struct verifier_env *env) 2212 { 2213 struct verifier_state *state = &env->cur_state; 2214 struct bpf_insn *insns = env->prog->insnsi; 2215 struct reg_state *regs = state->regs; 2216 int insn_cnt = env->prog->len; 2217 int insn_idx, prev_insn_idx = 0; 2218 int insn_processed = 0; 2219 bool do_print_state = false; 2220 2221 init_reg_state(regs); 2222 insn_idx = 0; 2223 for (;;) { 2224 struct bpf_insn *insn; 2225 u8 class; 2226 int err; 2227 2228 if (insn_idx >= insn_cnt) { 2229 verbose("invalid insn idx %d insn_cnt %d\n", 2230 insn_idx, insn_cnt); 2231 return -EFAULT; 2232 } 2233 2234 insn = &insns[insn_idx]; 2235 class = BPF_CLASS(insn->code); 2236 2237 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { 2238 verbose("BPF program is too large. Proccessed %d insn\n", 2239 insn_processed); 2240 return -E2BIG; 2241 } 2242 2243 err = is_state_visited(env, insn_idx); 2244 if (err < 0) 2245 return err; 2246 if (err == 1) { 2247 /* found equivalent state, can prune the search */ 2248 if (log_level) { 2249 if (do_print_state) 2250 verbose("\nfrom %d to %d: safe\n", 2251 prev_insn_idx, insn_idx); 2252 else 2253 verbose("%d: safe\n", insn_idx); 2254 } 2255 goto process_bpf_exit; 2256 } 2257 2258 if (log_level && do_print_state) { 2259 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); 2260 print_verifier_state(&env->cur_state); 2261 do_print_state = false; 2262 } 2263 2264 if (log_level) { 2265 verbose("%d: ", insn_idx); 2266 print_bpf_insn(insn); 2267 } 2268 2269 if (class == BPF_ALU || class == BPF_ALU64) { 2270 err = check_alu_op(env, insn); 2271 if (err) 2272 return err; 2273 2274 } else if (class == BPF_LDX) { 2275 enum bpf_reg_type src_reg_type; 2276 2277 /* check for reserved fields is already done */ 2278 2279 /* check src operand */ 2280 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2281 if (err) 2282 return err; 2283 2284 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 2285 if (err) 2286 return err; 2287 2288 src_reg_type = regs[insn->src_reg].type; 2289 2290 /* check that memory (src_reg + off) is readable, 2291 * the state of dst_reg will be updated by this func 2292 */ 2293 err = check_mem_access(env, insn->src_reg, insn->off, 2294 BPF_SIZE(insn->code), BPF_READ, 2295 insn->dst_reg); 2296 if (err) 2297 return err; 2298 2299 if (BPF_SIZE(insn->code) != BPF_W) { 2300 insn_idx++; 2301 continue; 2302 } 2303 2304 if (insn->imm == 0) { 2305 /* saw a valid insn 2306 * dst_reg = *(u32 *)(src_reg + off) 2307 * use reserved 'imm' field to mark this insn 2308 */ 2309 insn->imm = src_reg_type; 2310 2311 } else if (src_reg_type != insn->imm && 2312 (src_reg_type == PTR_TO_CTX || 2313 insn->imm == PTR_TO_CTX)) { 2314 /* ABuser program is trying to use the same insn 2315 * dst_reg = *(u32*) (src_reg + off) 2316 * with different pointer types: 2317 * src_reg == ctx in one branch and 2318 * src_reg == stack|map in some other branch. 2319 * Reject it. 2320 */ 2321 verbose("same insn cannot be used with different pointers\n"); 2322 return -EINVAL; 2323 } 2324 2325 } else if (class == BPF_STX) { 2326 enum bpf_reg_type dst_reg_type; 2327 2328 if (BPF_MODE(insn->code) == BPF_XADD) { 2329 err = check_xadd(env, insn); 2330 if (err) 2331 return err; 2332 insn_idx++; 2333 continue; 2334 } 2335 2336 /* check src1 operand */ 2337 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 2338 if (err) 2339 return err; 2340 /* check src2 operand */ 2341 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2342 if (err) 2343 return err; 2344 2345 dst_reg_type = regs[insn->dst_reg].type; 2346 2347 /* check that memory (dst_reg + off) is writeable */ 2348 err = check_mem_access(env, insn->dst_reg, insn->off, 2349 BPF_SIZE(insn->code), BPF_WRITE, 2350 insn->src_reg); 2351 if (err) 2352 return err; 2353 2354 if (insn->imm == 0) { 2355 insn->imm = dst_reg_type; 2356 } else if (dst_reg_type != insn->imm && 2357 (dst_reg_type == PTR_TO_CTX || 2358 insn->imm == PTR_TO_CTX)) { 2359 verbose("same insn cannot be used with different pointers\n"); 2360 return -EINVAL; 2361 } 2362 2363 } else if (class == BPF_ST) { 2364 if (BPF_MODE(insn->code) != BPF_MEM || 2365 insn->src_reg != BPF_REG_0) { 2366 verbose("BPF_ST uses reserved fields\n"); 2367 return -EINVAL; 2368 } 2369 /* check src operand */ 2370 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 2371 if (err) 2372 return err; 2373 2374 /* check that memory (dst_reg + off) is writeable */ 2375 err = check_mem_access(env, insn->dst_reg, insn->off, 2376 BPF_SIZE(insn->code), BPF_WRITE, 2377 -1); 2378 if (err) 2379 return err; 2380 2381 } else if (class == BPF_JMP) { 2382 u8 opcode = BPF_OP(insn->code); 2383 2384 if (opcode == BPF_CALL) { 2385 if (BPF_SRC(insn->code) != BPF_K || 2386 insn->off != 0 || 2387 insn->src_reg != BPF_REG_0 || 2388 insn->dst_reg != BPF_REG_0) { 2389 verbose("BPF_CALL uses reserved fields\n"); 2390 return -EINVAL; 2391 } 2392 2393 err = check_call(env, insn->imm); 2394 if (err) 2395 return err; 2396 2397 } else if (opcode == BPF_JA) { 2398 if (BPF_SRC(insn->code) != BPF_K || 2399 insn->imm != 0 || 2400 insn->src_reg != BPF_REG_0 || 2401 insn->dst_reg != BPF_REG_0) { 2402 verbose("BPF_JA uses reserved fields\n"); 2403 return -EINVAL; 2404 } 2405 2406 insn_idx += insn->off + 1; 2407 continue; 2408 2409 } else if (opcode == BPF_EXIT) { 2410 if (BPF_SRC(insn->code) != BPF_K || 2411 insn->imm != 0 || 2412 insn->src_reg != BPF_REG_0 || 2413 insn->dst_reg != BPF_REG_0) { 2414 verbose("BPF_EXIT uses reserved fields\n"); 2415 return -EINVAL; 2416 } 2417 2418 /* eBPF calling convetion is such that R0 is used 2419 * to return the value from eBPF program. 2420 * Make sure that it's readable at this time 2421 * of bpf_exit, which means that program wrote 2422 * something into it earlier 2423 */ 2424 err = check_reg_arg(regs, BPF_REG_0, SRC_OP); 2425 if (err) 2426 return err; 2427 2428 if (is_pointer_value(env, BPF_REG_0)) { 2429 verbose("R0 leaks addr as return value\n"); 2430 return -EACCES; 2431 } 2432 2433 process_bpf_exit: 2434 insn_idx = pop_stack(env, &prev_insn_idx); 2435 if (insn_idx < 0) { 2436 break; 2437 } else { 2438 do_print_state = true; 2439 continue; 2440 } 2441 } else { 2442 err = check_cond_jmp_op(env, insn, &insn_idx); 2443 if (err) 2444 return err; 2445 } 2446 } else if (class == BPF_LD) { 2447 u8 mode = BPF_MODE(insn->code); 2448 2449 if (mode == BPF_ABS || mode == BPF_IND) { 2450 err = check_ld_abs(env, insn); 2451 if (err) 2452 return err; 2453 2454 } else if (mode == BPF_IMM) { 2455 err = check_ld_imm(env, insn); 2456 if (err) 2457 return err; 2458 2459 insn_idx++; 2460 } else { 2461 verbose("invalid BPF_LD mode\n"); 2462 return -EINVAL; 2463 } 2464 } else { 2465 verbose("unknown insn class %d\n", class); 2466 return -EINVAL; 2467 } 2468 2469 insn_idx++; 2470 } 2471 2472 verbose("processed %d insns\n", insn_processed); 2473 return 0; 2474 } 2475 2476 /* look for pseudo eBPF instructions that access map FDs and 2477 * replace them with actual map pointers 2478 */ 2479 static int replace_map_fd_with_map_ptr(struct verifier_env *env) 2480 { 2481 struct bpf_insn *insn = env->prog->insnsi; 2482 int insn_cnt = env->prog->len; 2483 int i, j; 2484 2485 for (i = 0; i < insn_cnt; i++, insn++) { 2486 if (BPF_CLASS(insn->code) == BPF_LDX && 2487 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { 2488 verbose("BPF_LDX uses reserved fields\n"); 2489 return -EINVAL; 2490 } 2491 2492 if (BPF_CLASS(insn->code) == BPF_STX && 2493 ((BPF_MODE(insn->code) != BPF_MEM && 2494 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { 2495 verbose("BPF_STX uses reserved fields\n"); 2496 return -EINVAL; 2497 } 2498 2499 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { 2500 struct bpf_map *map; 2501 struct fd f; 2502 2503 if (i == insn_cnt - 1 || insn[1].code != 0 || 2504 insn[1].dst_reg != 0 || insn[1].src_reg != 0 || 2505 insn[1].off != 0) { 2506 verbose("invalid bpf_ld_imm64 insn\n"); 2507 return -EINVAL; 2508 } 2509 2510 if (insn->src_reg == 0) 2511 /* valid generic load 64-bit imm */ 2512 goto next_insn; 2513 2514 if (insn->src_reg != BPF_PSEUDO_MAP_FD) { 2515 verbose("unrecognized bpf_ld_imm64 insn\n"); 2516 return -EINVAL; 2517 } 2518 2519 f = fdget(insn->imm); 2520 map = __bpf_map_get(f); 2521 if (IS_ERR(map)) { 2522 verbose("fd %d is not pointing to valid bpf_map\n", 2523 insn->imm); 2524 return PTR_ERR(map); 2525 } 2526 2527 /* store map pointer inside BPF_LD_IMM64 instruction */ 2528 insn[0].imm = (u32) (unsigned long) map; 2529 insn[1].imm = ((u64) (unsigned long) map) >> 32; 2530 2531 /* check whether we recorded this map already */ 2532 for (j = 0; j < env->used_map_cnt; j++) 2533 if (env->used_maps[j] == map) { 2534 fdput(f); 2535 goto next_insn; 2536 } 2537 2538 if (env->used_map_cnt >= MAX_USED_MAPS) { 2539 fdput(f); 2540 return -E2BIG; 2541 } 2542 2543 /* hold the map. If the program is rejected by verifier, 2544 * the map will be released by release_maps() or it 2545 * will be used by the valid program until it's unloaded 2546 * and all maps are released in free_bpf_prog_info() 2547 */ 2548 map = bpf_map_inc(map, false); 2549 if (IS_ERR(map)) { 2550 fdput(f); 2551 return PTR_ERR(map); 2552 } 2553 env->used_maps[env->used_map_cnt++] = map; 2554 2555 fdput(f); 2556 next_insn: 2557 insn++; 2558 i++; 2559 } 2560 } 2561 2562 /* now all pseudo BPF_LD_IMM64 instructions load valid 2563 * 'struct bpf_map *' into a register instead of user map_fd. 2564 * These pointers will be used later by verifier to validate map access. 2565 */ 2566 return 0; 2567 } 2568 2569 /* drop refcnt of maps used by the rejected program */ 2570 static void release_maps(struct verifier_env *env) 2571 { 2572 int i; 2573 2574 for (i = 0; i < env->used_map_cnt; i++) 2575 bpf_map_put(env->used_maps[i]); 2576 } 2577 2578 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ 2579 static void convert_pseudo_ld_imm64(struct verifier_env *env) 2580 { 2581 struct bpf_insn *insn = env->prog->insnsi; 2582 int insn_cnt = env->prog->len; 2583 int i; 2584 2585 for (i = 0; i < insn_cnt; i++, insn++) 2586 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) 2587 insn->src_reg = 0; 2588 } 2589 2590 /* convert load instructions that access fields of 'struct __sk_buff' 2591 * into sequence of instructions that access fields of 'struct sk_buff' 2592 */ 2593 static int convert_ctx_accesses(struct verifier_env *env) 2594 { 2595 struct bpf_insn *insn = env->prog->insnsi; 2596 int insn_cnt = env->prog->len; 2597 struct bpf_insn insn_buf[16]; 2598 struct bpf_prog *new_prog; 2599 enum bpf_access_type type; 2600 int i; 2601 2602 if (!env->prog->aux->ops->convert_ctx_access) 2603 return 0; 2604 2605 for (i = 0; i < insn_cnt; i++, insn++) { 2606 u32 insn_delta, cnt; 2607 2608 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W)) 2609 type = BPF_READ; 2610 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W)) 2611 type = BPF_WRITE; 2612 else 2613 continue; 2614 2615 if (insn->imm != PTR_TO_CTX) { 2616 /* clear internal mark */ 2617 insn->imm = 0; 2618 continue; 2619 } 2620 2621 cnt = env->prog->aux->ops-> 2622 convert_ctx_access(type, insn->dst_reg, insn->src_reg, 2623 insn->off, insn_buf, env->prog); 2624 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { 2625 verbose("bpf verifier is misconfigured\n"); 2626 return -EINVAL; 2627 } 2628 2629 new_prog = bpf_patch_insn_single(env->prog, i, insn_buf, cnt); 2630 if (!new_prog) 2631 return -ENOMEM; 2632 2633 insn_delta = cnt - 1; 2634 2635 /* keep walking new program and skip insns we just inserted */ 2636 env->prog = new_prog; 2637 insn = new_prog->insnsi + i + insn_delta; 2638 2639 insn_cnt += insn_delta; 2640 i += insn_delta; 2641 } 2642 2643 return 0; 2644 } 2645 2646 static void free_states(struct verifier_env *env) 2647 { 2648 struct verifier_state_list *sl, *sln; 2649 int i; 2650 2651 if (!env->explored_states) 2652 return; 2653 2654 for (i = 0; i < env->prog->len; i++) { 2655 sl = env->explored_states[i]; 2656 2657 if (sl) 2658 while (sl != STATE_LIST_MARK) { 2659 sln = sl->next; 2660 kfree(sl); 2661 sl = sln; 2662 } 2663 } 2664 2665 kfree(env->explored_states); 2666 } 2667 2668 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) 2669 { 2670 char __user *log_ubuf = NULL; 2671 struct verifier_env *env; 2672 int ret = -EINVAL; 2673 2674 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS) 2675 return -E2BIG; 2676 2677 /* 'struct verifier_env' can be global, but since it's not small, 2678 * allocate/free it every time bpf_check() is called 2679 */ 2680 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL); 2681 if (!env) 2682 return -ENOMEM; 2683 2684 env->prog = *prog; 2685 2686 /* grab the mutex to protect few globals used by verifier */ 2687 mutex_lock(&bpf_verifier_lock); 2688 2689 if (attr->log_level || attr->log_buf || attr->log_size) { 2690 /* user requested verbose verifier output 2691 * and supplied buffer to store the verification trace 2692 */ 2693 log_level = attr->log_level; 2694 log_ubuf = (char __user *) (unsigned long) attr->log_buf; 2695 log_size = attr->log_size; 2696 log_len = 0; 2697 2698 ret = -EINVAL; 2699 /* log_* values have to be sane */ 2700 if (log_size < 128 || log_size > UINT_MAX >> 8 || 2701 log_level == 0 || log_ubuf == NULL) 2702 goto free_env; 2703 2704 ret = -ENOMEM; 2705 log_buf = vmalloc(log_size); 2706 if (!log_buf) 2707 goto free_env; 2708 } else { 2709 log_level = 0; 2710 } 2711 2712 ret = replace_map_fd_with_map_ptr(env); 2713 if (ret < 0) 2714 goto skip_full_check; 2715 2716 env->explored_states = kcalloc(env->prog->len, 2717 sizeof(struct verifier_state_list *), 2718 GFP_USER); 2719 ret = -ENOMEM; 2720 if (!env->explored_states) 2721 goto skip_full_check; 2722 2723 ret = check_cfg(env); 2724 if (ret < 0) 2725 goto skip_full_check; 2726 2727 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); 2728 2729 ret = do_check(env); 2730 2731 skip_full_check: 2732 while (pop_stack(env, NULL) >= 0); 2733 free_states(env); 2734 2735 if (ret == 0) 2736 /* program is valid, convert *(u32*)(ctx + off) accesses */ 2737 ret = convert_ctx_accesses(env); 2738 2739 if (log_level && log_len >= log_size - 1) { 2740 BUG_ON(log_len >= log_size); 2741 /* verifier log exceeded user supplied buffer */ 2742 ret = -ENOSPC; 2743 /* fall through to return what was recorded */ 2744 } 2745 2746 /* copy verifier log back to user space including trailing zero */ 2747 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { 2748 ret = -EFAULT; 2749 goto free_log_buf; 2750 } 2751 2752 if (ret == 0 && env->used_map_cnt) { 2753 /* if program passed verifier, update used_maps in bpf_prog_info */ 2754 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, 2755 sizeof(env->used_maps[0]), 2756 GFP_KERNEL); 2757 2758 if (!env->prog->aux->used_maps) { 2759 ret = -ENOMEM; 2760 goto free_log_buf; 2761 } 2762 2763 memcpy(env->prog->aux->used_maps, env->used_maps, 2764 sizeof(env->used_maps[0]) * env->used_map_cnt); 2765 env->prog->aux->used_map_cnt = env->used_map_cnt; 2766 2767 /* program is valid. Convert pseudo bpf_ld_imm64 into generic 2768 * bpf_ld_imm64 instructions 2769 */ 2770 convert_pseudo_ld_imm64(env); 2771 } 2772 2773 free_log_buf: 2774 if (log_level) 2775 vfree(log_buf); 2776 free_env: 2777 if (!env->prog->aux->used_maps) 2778 /* if we didn't copy map pointers into bpf_prog_info, release 2779 * them now. Otherwise free_bpf_prog_info() will release them. 2780 */ 2781 release_maps(env); 2782 *prog = env->prog; 2783 kfree(env); 2784 mutex_unlock(&bpf_verifier_lock); 2785 return ret; 2786 } 2787