1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 2 * 3 * This program is free software; you can redistribute it and/or 4 * modify it under the terms of version 2 of the GNU General Public 5 * License as published by the Free Software Foundation. 6 * 7 * This program is distributed in the hope that it will be useful, but 8 * WITHOUT ANY WARRANTY; without even the implied warranty of 9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 10 * General Public License for more details. 11 */ 12 #include <linux/kernel.h> 13 #include <linux/types.h> 14 #include <linux/slab.h> 15 #include <linux/bpf.h> 16 #include <linux/filter.h> 17 #include <net/netlink.h> 18 #include <linux/file.h> 19 #include <linux/vmalloc.h> 20 21 /* bpf_check() is a static code analyzer that walks eBPF program 22 * instruction by instruction and updates register/stack state. 23 * All paths of conditional branches are analyzed until 'bpf_exit' insn. 24 * 25 * The first pass is depth-first-search to check that the program is a DAG. 26 * It rejects the following programs: 27 * - larger than BPF_MAXINSNS insns 28 * - if loop is present (detected via back-edge) 29 * - unreachable insns exist (shouldn't be a forest. program = one function) 30 * - out of bounds or malformed jumps 31 * The second pass is all possible path descent from the 1st insn. 32 * Since it's analyzing all pathes through the program, the length of the 33 * analysis is limited to 32k insn, which may be hit even if total number of 34 * insn is less then 4K, but there are too many branches that change stack/regs. 35 * Number of 'branches to be analyzed' is limited to 1k 36 * 37 * On entry to each instruction, each register has a type, and the instruction 38 * changes the types of the registers depending on instruction semantics. 39 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is 40 * copied to R1. 41 * 42 * All registers are 64-bit. 43 * R0 - return register 44 * R1-R5 argument passing registers 45 * R6-R9 callee saved registers 46 * R10 - frame pointer read-only 47 * 48 * At the start of BPF program the register R1 contains a pointer to bpf_context 49 * and has type PTR_TO_CTX. 50 * 51 * Verifier tracks arithmetic operations on pointers in case: 52 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), 53 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), 54 * 1st insn copies R10 (which has FRAME_PTR) type into R1 55 * and 2nd arithmetic instruction is pattern matched to recognize 56 * that it wants to construct a pointer to some element within stack. 57 * So after 2nd insn, the register R1 has type PTR_TO_STACK 58 * (and -20 constant is saved for further stack bounds checking). 59 * Meaning that this reg is a pointer to stack plus known immediate constant. 60 * 61 * Most of the time the registers have UNKNOWN_VALUE type, which 62 * means the register has some value, but it's not a valid pointer. 63 * (like pointer plus pointer becomes UNKNOWN_VALUE type) 64 * 65 * When verifier sees load or store instructions the type of base register 66 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer 67 * types recognized by check_mem_access() function. 68 * 69 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' 70 * and the range of [ptr, ptr + map's value_size) is accessible. 71 * 72 * registers used to pass values to function calls are checked against 73 * function argument constraints. 74 * 75 * ARG_PTR_TO_MAP_KEY is one of such argument constraints. 76 * It means that the register type passed to this function must be 77 * PTR_TO_STACK and it will be used inside the function as 78 * 'pointer to map element key' 79 * 80 * For example the argument constraints for bpf_map_lookup_elem(): 81 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, 82 * .arg1_type = ARG_CONST_MAP_PTR, 83 * .arg2_type = ARG_PTR_TO_MAP_KEY, 84 * 85 * ret_type says that this function returns 'pointer to map elem value or null' 86 * function expects 1st argument to be a const pointer to 'struct bpf_map' and 87 * 2nd argument should be a pointer to stack, which will be used inside 88 * the helper function as a pointer to map element key. 89 * 90 * On the kernel side the helper function looks like: 91 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) 92 * { 93 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; 94 * void *key = (void *) (unsigned long) r2; 95 * void *value; 96 * 97 * here kernel can access 'key' and 'map' pointers safely, knowing that 98 * [key, key + map->key_size) bytes are valid and were initialized on 99 * the stack of eBPF program. 100 * } 101 * 102 * Corresponding eBPF program may look like: 103 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR 104 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK 105 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP 106 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), 107 * here verifier looks at prototype of map_lookup_elem() and sees: 108 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, 109 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes 110 * 111 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, 112 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits 113 * and were initialized prior to this call. 114 * If it's ok, then verifier allows this BPF_CALL insn and looks at 115 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets 116 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function 117 * returns ether pointer to map value or NULL. 118 * 119 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' 120 * insn, the register holding that pointer in the true branch changes state to 121 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false 122 * branch. See check_cond_jmp_op(). 123 * 124 * After the call R0 is set to return type of the function and registers R1-R5 125 * are set to NOT_INIT to indicate that they are no longer readable. 126 */ 127 128 /* types of values stored in eBPF registers */ 129 enum bpf_reg_type { 130 NOT_INIT = 0, /* nothing was written into register */ 131 UNKNOWN_VALUE, /* reg doesn't contain a valid pointer */ 132 PTR_TO_CTX, /* reg points to bpf_context */ 133 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 134 PTR_TO_MAP_VALUE, /* reg points to map element value */ 135 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 136 FRAME_PTR, /* reg == frame_pointer */ 137 PTR_TO_STACK, /* reg == frame_pointer + imm */ 138 CONST_IMM, /* constant integer value */ 139 }; 140 141 struct reg_state { 142 enum bpf_reg_type type; 143 union { 144 /* valid when type == CONST_IMM | PTR_TO_STACK */ 145 int imm; 146 147 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE | 148 * PTR_TO_MAP_VALUE_OR_NULL 149 */ 150 struct bpf_map *map_ptr; 151 }; 152 }; 153 154 enum bpf_stack_slot_type { 155 STACK_INVALID, /* nothing was stored in this stack slot */ 156 STACK_SPILL, /* register spilled into stack */ 157 STACK_MISC /* BPF program wrote some data into this slot */ 158 }; 159 160 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */ 161 162 /* state of the program: 163 * type of all registers and stack info 164 */ 165 struct verifier_state { 166 struct reg_state regs[MAX_BPF_REG]; 167 u8 stack_slot_type[MAX_BPF_STACK]; 168 struct reg_state spilled_regs[MAX_BPF_STACK / BPF_REG_SIZE]; 169 }; 170 171 /* linked list of verifier states used to prune search */ 172 struct verifier_state_list { 173 struct verifier_state state; 174 struct verifier_state_list *next; 175 }; 176 177 /* verifier_state + insn_idx are pushed to stack when branch is encountered */ 178 struct verifier_stack_elem { 179 /* verifer state is 'st' 180 * before processing instruction 'insn_idx' 181 * and after processing instruction 'prev_insn_idx' 182 */ 183 struct verifier_state st; 184 int insn_idx; 185 int prev_insn_idx; 186 struct verifier_stack_elem *next; 187 }; 188 189 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */ 190 191 /* single container for all structs 192 * one verifier_env per bpf_check() call 193 */ 194 struct verifier_env { 195 struct bpf_prog *prog; /* eBPF program being verified */ 196 struct verifier_stack_elem *head; /* stack of verifier states to be processed */ 197 int stack_size; /* number of states to be processed */ 198 struct verifier_state cur_state; /* current verifier state */ 199 struct verifier_state_list **explored_states; /* search pruning optimization */ 200 struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */ 201 u32 used_map_cnt; /* number of used maps */ 202 }; 203 204 /* verbose verifier prints what it's seeing 205 * bpf_check() is called under lock, so no race to access these global vars 206 */ 207 static u32 log_level, log_size, log_len; 208 static char *log_buf; 209 210 static DEFINE_MUTEX(bpf_verifier_lock); 211 212 /* log_level controls verbosity level of eBPF verifier. 213 * verbose() is used to dump the verification trace to the log, so the user 214 * can figure out what's wrong with the program 215 */ 216 static void verbose(const char *fmt, ...) 217 { 218 va_list args; 219 220 if (log_level == 0 || log_len >= log_size - 1) 221 return; 222 223 va_start(args, fmt); 224 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); 225 va_end(args); 226 } 227 228 /* string representation of 'enum bpf_reg_type' */ 229 static const char * const reg_type_str[] = { 230 [NOT_INIT] = "?", 231 [UNKNOWN_VALUE] = "inv", 232 [PTR_TO_CTX] = "ctx", 233 [CONST_PTR_TO_MAP] = "map_ptr", 234 [PTR_TO_MAP_VALUE] = "map_value", 235 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", 236 [FRAME_PTR] = "fp", 237 [PTR_TO_STACK] = "fp", 238 [CONST_IMM] = "imm", 239 }; 240 241 static void print_verifier_state(struct verifier_env *env) 242 { 243 enum bpf_reg_type t; 244 int i; 245 246 for (i = 0; i < MAX_BPF_REG; i++) { 247 t = env->cur_state.regs[i].type; 248 if (t == NOT_INIT) 249 continue; 250 verbose(" R%d=%s", i, reg_type_str[t]); 251 if (t == CONST_IMM || t == PTR_TO_STACK) 252 verbose("%d", env->cur_state.regs[i].imm); 253 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || 254 t == PTR_TO_MAP_VALUE_OR_NULL) 255 verbose("(ks=%d,vs=%d)", 256 env->cur_state.regs[i].map_ptr->key_size, 257 env->cur_state.regs[i].map_ptr->value_size); 258 } 259 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { 260 if (env->cur_state.stack_slot_type[i] == STACK_SPILL) 261 verbose(" fp%d=%s", -MAX_BPF_STACK + i, 262 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]); 263 } 264 verbose("\n"); 265 } 266 267 static const char *const bpf_class_string[] = { 268 [BPF_LD] = "ld", 269 [BPF_LDX] = "ldx", 270 [BPF_ST] = "st", 271 [BPF_STX] = "stx", 272 [BPF_ALU] = "alu", 273 [BPF_JMP] = "jmp", 274 [BPF_RET] = "BUG", 275 [BPF_ALU64] = "alu64", 276 }; 277 278 static const char *const bpf_alu_string[] = { 279 [BPF_ADD >> 4] = "+=", 280 [BPF_SUB >> 4] = "-=", 281 [BPF_MUL >> 4] = "*=", 282 [BPF_DIV >> 4] = "/=", 283 [BPF_OR >> 4] = "|=", 284 [BPF_AND >> 4] = "&=", 285 [BPF_LSH >> 4] = "<<=", 286 [BPF_RSH >> 4] = ">>=", 287 [BPF_NEG >> 4] = "neg", 288 [BPF_MOD >> 4] = "%=", 289 [BPF_XOR >> 4] = "^=", 290 [BPF_MOV >> 4] = "=", 291 [BPF_ARSH >> 4] = "s>>=", 292 [BPF_END >> 4] = "endian", 293 }; 294 295 static const char *const bpf_ldst_string[] = { 296 [BPF_W >> 3] = "u32", 297 [BPF_H >> 3] = "u16", 298 [BPF_B >> 3] = "u8", 299 [BPF_DW >> 3] = "u64", 300 }; 301 302 static const char *const bpf_jmp_string[] = { 303 [BPF_JA >> 4] = "jmp", 304 [BPF_JEQ >> 4] = "==", 305 [BPF_JGT >> 4] = ">", 306 [BPF_JGE >> 4] = ">=", 307 [BPF_JSET >> 4] = "&", 308 [BPF_JNE >> 4] = "!=", 309 [BPF_JSGT >> 4] = "s>", 310 [BPF_JSGE >> 4] = "s>=", 311 [BPF_CALL >> 4] = "call", 312 [BPF_EXIT >> 4] = "exit", 313 }; 314 315 static void print_bpf_insn(struct bpf_insn *insn) 316 { 317 u8 class = BPF_CLASS(insn->code); 318 319 if (class == BPF_ALU || class == BPF_ALU64) { 320 if (BPF_SRC(insn->code) == BPF_X) 321 verbose("(%02x) %sr%d %s %sr%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->src_reg); 327 else 328 verbose("(%02x) %sr%d %s %s%d\n", 329 insn->code, class == BPF_ALU ? "(u32) " : "", 330 insn->dst_reg, 331 bpf_alu_string[BPF_OP(insn->code) >> 4], 332 class == BPF_ALU ? "(u32) " : "", 333 insn->imm); 334 } else if (class == BPF_STX) { 335 if (BPF_MODE(insn->code) == BPF_MEM) 336 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", 337 insn->code, 338 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 339 insn->dst_reg, 340 insn->off, insn->src_reg); 341 else if (BPF_MODE(insn->code) == BPF_XADD) 342 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", 343 insn->code, 344 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 345 insn->dst_reg, insn->off, 346 insn->src_reg); 347 else 348 verbose("BUG_%02x\n", insn->code); 349 } else if (class == BPF_ST) { 350 if (BPF_MODE(insn->code) != BPF_MEM) { 351 verbose("BUG_st_%02x\n", insn->code); 352 return; 353 } 354 verbose("(%02x) *(%s *)(r%d %+d) = %d\n", 355 insn->code, 356 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 357 insn->dst_reg, 358 insn->off, insn->imm); 359 } else if (class == BPF_LDX) { 360 if (BPF_MODE(insn->code) != BPF_MEM) { 361 verbose("BUG_ldx_%02x\n", insn->code); 362 return; 363 } 364 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", 365 insn->code, insn->dst_reg, 366 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 367 insn->src_reg, insn->off); 368 } else if (class == BPF_LD) { 369 if (BPF_MODE(insn->code) == BPF_ABS) { 370 verbose("(%02x) r0 = *(%s *)skb[%d]\n", 371 insn->code, 372 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 373 insn->imm); 374 } else if (BPF_MODE(insn->code) == BPF_IND) { 375 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", 376 insn->code, 377 bpf_ldst_string[BPF_SIZE(insn->code) >> 3], 378 insn->src_reg, insn->imm); 379 } else if (BPF_MODE(insn->code) == BPF_IMM) { 380 verbose("(%02x) r%d = 0x%x\n", 381 insn->code, insn->dst_reg, insn->imm); 382 } else { 383 verbose("BUG_ld_%02x\n", insn->code); 384 return; 385 } 386 } else if (class == BPF_JMP) { 387 u8 opcode = BPF_OP(insn->code); 388 389 if (opcode == BPF_CALL) { 390 verbose("(%02x) call %d\n", insn->code, insn->imm); 391 } else if (insn->code == (BPF_JMP | BPF_JA)) { 392 verbose("(%02x) goto pc%+d\n", 393 insn->code, insn->off); 394 } else if (insn->code == (BPF_JMP | BPF_EXIT)) { 395 verbose("(%02x) exit\n", insn->code); 396 } else if (BPF_SRC(insn->code) == BPF_X) { 397 verbose("(%02x) if r%d %s r%d goto pc%+d\n", 398 insn->code, insn->dst_reg, 399 bpf_jmp_string[BPF_OP(insn->code) >> 4], 400 insn->src_reg, insn->off); 401 } else { 402 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", 403 insn->code, insn->dst_reg, 404 bpf_jmp_string[BPF_OP(insn->code) >> 4], 405 insn->imm, insn->off); 406 } 407 } else { 408 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); 409 } 410 } 411 412 static int pop_stack(struct verifier_env *env, int *prev_insn_idx) 413 { 414 struct verifier_stack_elem *elem; 415 int insn_idx; 416 417 if (env->head == NULL) 418 return -1; 419 420 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); 421 insn_idx = env->head->insn_idx; 422 if (prev_insn_idx) 423 *prev_insn_idx = env->head->prev_insn_idx; 424 elem = env->head->next; 425 kfree(env->head); 426 env->head = elem; 427 env->stack_size--; 428 return insn_idx; 429 } 430 431 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx, 432 int prev_insn_idx) 433 { 434 struct verifier_stack_elem *elem; 435 436 elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL); 437 if (!elem) 438 goto err; 439 440 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); 441 elem->insn_idx = insn_idx; 442 elem->prev_insn_idx = prev_insn_idx; 443 elem->next = env->head; 444 env->head = elem; 445 env->stack_size++; 446 if (env->stack_size > 1024) { 447 verbose("BPF program is too complex\n"); 448 goto err; 449 } 450 return &elem->st; 451 err: 452 /* pop all elements and return */ 453 while (pop_stack(env, NULL) >= 0); 454 return NULL; 455 } 456 457 #define CALLER_SAVED_REGS 6 458 static const int caller_saved[CALLER_SAVED_REGS] = { 459 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 460 }; 461 462 static void init_reg_state(struct reg_state *regs) 463 { 464 int i; 465 466 for (i = 0; i < MAX_BPF_REG; i++) { 467 regs[i].type = NOT_INIT; 468 regs[i].imm = 0; 469 regs[i].map_ptr = NULL; 470 } 471 472 /* frame pointer */ 473 regs[BPF_REG_FP].type = FRAME_PTR; 474 475 /* 1st arg to a function */ 476 regs[BPF_REG_1].type = PTR_TO_CTX; 477 } 478 479 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno) 480 { 481 BUG_ON(regno >= MAX_BPF_REG); 482 regs[regno].type = UNKNOWN_VALUE; 483 regs[regno].imm = 0; 484 regs[regno].map_ptr = NULL; 485 } 486 487 enum reg_arg_type { 488 SRC_OP, /* register is used as source operand */ 489 DST_OP, /* register is used as destination operand */ 490 DST_OP_NO_MARK /* same as above, check only, don't mark */ 491 }; 492 493 static int check_reg_arg(struct reg_state *regs, u32 regno, 494 enum reg_arg_type t) 495 { 496 if (regno >= MAX_BPF_REG) { 497 verbose("R%d is invalid\n", regno); 498 return -EINVAL; 499 } 500 501 if (t == SRC_OP) { 502 /* check whether register used as source operand can be read */ 503 if (regs[regno].type == NOT_INIT) { 504 verbose("R%d !read_ok\n", regno); 505 return -EACCES; 506 } 507 } else { 508 /* check whether register used as dest operand can be written to */ 509 if (regno == BPF_REG_FP) { 510 verbose("frame pointer is read only\n"); 511 return -EACCES; 512 } 513 if (t == DST_OP) 514 mark_reg_unknown_value(regs, regno); 515 } 516 return 0; 517 } 518 519 static int bpf_size_to_bytes(int bpf_size) 520 { 521 if (bpf_size == BPF_W) 522 return 4; 523 else if (bpf_size == BPF_H) 524 return 2; 525 else if (bpf_size == BPF_B) 526 return 1; 527 else if (bpf_size == BPF_DW) 528 return 8; 529 else 530 return -EINVAL; 531 } 532 533 /* check_stack_read/write functions track spill/fill of registers, 534 * stack boundary and alignment are checked in check_mem_access() 535 */ 536 static int check_stack_write(struct verifier_state *state, int off, int size, 537 int value_regno) 538 { 539 int i; 540 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, 541 * so it's aligned access and [off, off + size) are within stack limits 542 */ 543 544 if (value_regno >= 0 && 545 (state->regs[value_regno].type == PTR_TO_MAP_VALUE || 546 state->regs[value_regno].type == PTR_TO_STACK || 547 state->regs[value_regno].type == PTR_TO_CTX)) { 548 549 /* register containing pointer is being spilled into stack */ 550 if (size != BPF_REG_SIZE) { 551 verbose("invalid size of register spill\n"); 552 return -EACCES; 553 } 554 555 /* save register state */ 556 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 557 state->regs[value_regno]; 558 559 for (i = 0; i < BPF_REG_SIZE; i++) 560 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; 561 } else { 562 /* regular write of data into stack */ 563 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = 564 (struct reg_state) {}; 565 566 for (i = 0; i < size; i++) 567 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; 568 } 569 return 0; 570 } 571 572 static int check_stack_read(struct verifier_state *state, int off, int size, 573 int value_regno) 574 { 575 u8 *slot_type; 576 int i; 577 578 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; 579 580 if (slot_type[0] == STACK_SPILL) { 581 if (size != BPF_REG_SIZE) { 582 verbose("invalid size of register spill\n"); 583 return -EACCES; 584 } 585 for (i = 1; i < BPF_REG_SIZE; i++) { 586 if (slot_type[i] != STACK_SPILL) { 587 verbose("corrupted spill memory\n"); 588 return -EACCES; 589 } 590 } 591 592 if (value_regno >= 0) 593 /* restore register state from stack */ 594 state->regs[value_regno] = 595 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; 596 return 0; 597 } else { 598 for (i = 0; i < size; i++) { 599 if (slot_type[i] != STACK_MISC) { 600 verbose("invalid read from stack off %d+%d size %d\n", 601 off, i, size); 602 return -EACCES; 603 } 604 } 605 if (value_regno >= 0) 606 /* have read misc data from the stack */ 607 mark_reg_unknown_value(state->regs, value_regno); 608 return 0; 609 } 610 } 611 612 /* check read/write into map element returned by bpf_map_lookup_elem() */ 613 static int check_map_access(struct verifier_env *env, u32 regno, int off, 614 int size) 615 { 616 struct bpf_map *map = env->cur_state.regs[regno].map_ptr; 617 618 if (off < 0 || off + size > map->value_size) { 619 verbose("invalid access to map value, value_size=%d off=%d size=%d\n", 620 map->value_size, off, size); 621 return -EACCES; 622 } 623 return 0; 624 } 625 626 /* check access to 'struct bpf_context' fields */ 627 static int check_ctx_access(struct verifier_env *env, int off, int size, 628 enum bpf_access_type t) 629 { 630 if (env->prog->aux->ops->is_valid_access && 631 env->prog->aux->ops->is_valid_access(off, size, t)) 632 return 0; 633 634 verbose("invalid bpf_context access off=%d size=%d\n", off, size); 635 return -EACCES; 636 } 637 638 /* check whether memory at (regno + off) is accessible for t = (read | write) 639 * if t==write, value_regno is a register which value is stored into memory 640 * if t==read, value_regno is a register which will receive the value from memory 641 * if t==write && value_regno==-1, some unknown value is stored into memory 642 * if t==read && value_regno==-1, don't care what we read from memory 643 */ 644 static int check_mem_access(struct verifier_env *env, u32 regno, int off, 645 int bpf_size, enum bpf_access_type t, 646 int value_regno) 647 { 648 struct verifier_state *state = &env->cur_state; 649 int size, err = 0; 650 651 size = bpf_size_to_bytes(bpf_size); 652 if (size < 0) 653 return size; 654 655 if (off % size != 0) { 656 verbose("misaligned access off %d size %d\n", off, size); 657 return -EACCES; 658 } 659 660 if (state->regs[regno].type == PTR_TO_MAP_VALUE) { 661 err = check_map_access(env, regno, off, size); 662 if (!err && t == BPF_READ && value_regno >= 0) 663 mark_reg_unknown_value(state->regs, value_regno); 664 665 } else if (state->regs[regno].type == PTR_TO_CTX) { 666 err = check_ctx_access(env, off, size, t); 667 if (!err && t == BPF_READ && value_regno >= 0) 668 mark_reg_unknown_value(state->regs, value_regno); 669 670 } else if (state->regs[regno].type == FRAME_PTR) { 671 if (off >= 0 || off < -MAX_BPF_STACK) { 672 verbose("invalid stack off=%d size=%d\n", off, size); 673 return -EACCES; 674 } 675 if (t == BPF_WRITE) 676 err = check_stack_write(state, off, size, value_regno); 677 else 678 err = check_stack_read(state, off, size, value_regno); 679 } else { 680 verbose("R%d invalid mem access '%s'\n", 681 regno, reg_type_str[state->regs[regno].type]); 682 return -EACCES; 683 } 684 return err; 685 } 686 687 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn) 688 { 689 struct reg_state *regs = env->cur_state.regs; 690 int err; 691 692 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || 693 insn->imm != 0) { 694 verbose("BPF_XADD uses reserved fields\n"); 695 return -EINVAL; 696 } 697 698 /* check src1 operand */ 699 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 700 if (err) 701 return err; 702 703 /* check src2 operand */ 704 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 705 if (err) 706 return err; 707 708 /* check whether atomic_add can read the memory */ 709 err = check_mem_access(env, insn->dst_reg, insn->off, 710 BPF_SIZE(insn->code), BPF_READ, -1); 711 if (err) 712 return err; 713 714 /* check whether atomic_add can write into the same memory */ 715 return check_mem_access(env, insn->dst_reg, insn->off, 716 BPF_SIZE(insn->code), BPF_WRITE, -1); 717 } 718 719 /* when register 'regno' is passed into function that will read 'access_size' 720 * bytes from that pointer, make sure that it's within stack boundary 721 * and all elements of stack are initialized 722 */ 723 static int check_stack_boundary(struct verifier_env *env, 724 int regno, int access_size) 725 { 726 struct verifier_state *state = &env->cur_state; 727 struct reg_state *regs = state->regs; 728 int off, i; 729 730 if (regs[regno].type != PTR_TO_STACK) 731 return -EACCES; 732 733 off = regs[regno].imm; 734 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || 735 access_size <= 0) { 736 verbose("invalid stack type R%d off=%d access_size=%d\n", 737 regno, off, access_size); 738 return -EACCES; 739 } 740 741 for (i = 0; i < access_size; i++) { 742 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { 743 verbose("invalid indirect read from stack off %d+%d size %d\n", 744 off, i, access_size); 745 return -EACCES; 746 } 747 } 748 return 0; 749 } 750 751 static int check_func_arg(struct verifier_env *env, u32 regno, 752 enum bpf_arg_type arg_type, struct bpf_map **mapp) 753 { 754 struct reg_state *reg = env->cur_state.regs + regno; 755 enum bpf_reg_type expected_type; 756 int err = 0; 757 758 if (arg_type == ARG_DONTCARE) 759 return 0; 760 761 if (reg->type == NOT_INIT) { 762 verbose("R%d !read_ok\n", regno); 763 return -EACCES; 764 } 765 766 if (arg_type == ARG_ANYTHING) 767 return 0; 768 769 if (arg_type == ARG_PTR_TO_STACK || arg_type == ARG_PTR_TO_MAP_KEY || 770 arg_type == ARG_PTR_TO_MAP_VALUE) { 771 expected_type = PTR_TO_STACK; 772 } else if (arg_type == ARG_CONST_STACK_SIZE) { 773 expected_type = CONST_IMM; 774 } else if (arg_type == ARG_CONST_MAP_PTR) { 775 expected_type = CONST_PTR_TO_MAP; 776 } else if (arg_type == ARG_PTR_TO_CTX) { 777 expected_type = PTR_TO_CTX; 778 } else { 779 verbose("unsupported arg_type %d\n", arg_type); 780 return -EFAULT; 781 } 782 783 if (reg->type != expected_type) { 784 verbose("R%d type=%s expected=%s\n", regno, 785 reg_type_str[reg->type], reg_type_str[expected_type]); 786 return -EACCES; 787 } 788 789 if (arg_type == ARG_CONST_MAP_PTR) { 790 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ 791 *mapp = reg->map_ptr; 792 793 } else if (arg_type == ARG_PTR_TO_MAP_KEY) { 794 /* bpf_map_xxx(..., map_ptr, ..., key) call: 795 * check that [key, key + map->key_size) are within 796 * stack limits and initialized 797 */ 798 if (!*mapp) { 799 /* in function declaration map_ptr must come before 800 * map_key, so that it's verified and known before 801 * we have to check map_key here. Otherwise it means 802 * that kernel subsystem misconfigured verifier 803 */ 804 verbose("invalid map_ptr to access map->key\n"); 805 return -EACCES; 806 } 807 err = check_stack_boundary(env, regno, (*mapp)->key_size); 808 809 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { 810 /* bpf_map_xxx(..., map_ptr, ..., value) call: 811 * check [value, value + map->value_size) validity 812 */ 813 if (!*mapp) { 814 /* kernel subsystem misconfigured verifier */ 815 verbose("invalid map_ptr to access map->value\n"); 816 return -EACCES; 817 } 818 err = check_stack_boundary(env, regno, (*mapp)->value_size); 819 820 } else if (arg_type == ARG_CONST_STACK_SIZE) { 821 /* bpf_xxx(..., buf, len) call will access 'len' bytes 822 * from stack pointer 'buf'. Check it 823 * note: regno == len, regno - 1 == buf 824 */ 825 if (regno == 0) { 826 /* kernel subsystem misconfigured verifier */ 827 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n"); 828 return -EACCES; 829 } 830 err = check_stack_boundary(env, regno - 1, reg->imm); 831 } 832 833 return err; 834 } 835 836 static int check_call(struct verifier_env *env, int func_id) 837 { 838 struct verifier_state *state = &env->cur_state; 839 const struct bpf_func_proto *fn = NULL; 840 struct reg_state *regs = state->regs; 841 struct bpf_map *map = NULL; 842 struct reg_state *reg; 843 int i, err; 844 845 /* find function prototype */ 846 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { 847 verbose("invalid func %d\n", func_id); 848 return -EINVAL; 849 } 850 851 if (env->prog->aux->ops->get_func_proto) 852 fn = env->prog->aux->ops->get_func_proto(func_id); 853 854 if (!fn) { 855 verbose("unknown func %d\n", func_id); 856 return -EINVAL; 857 } 858 859 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 860 if (!env->prog->gpl_compatible && fn->gpl_only) { 861 verbose("cannot call GPL only function from proprietary program\n"); 862 return -EINVAL; 863 } 864 865 /* check args */ 866 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map); 867 if (err) 868 return err; 869 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map); 870 if (err) 871 return err; 872 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map); 873 if (err) 874 return err; 875 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map); 876 if (err) 877 return err; 878 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map); 879 if (err) 880 return err; 881 882 /* reset caller saved regs */ 883 for (i = 0; i < CALLER_SAVED_REGS; i++) { 884 reg = regs + caller_saved[i]; 885 reg->type = NOT_INIT; 886 reg->imm = 0; 887 } 888 889 /* update return register */ 890 if (fn->ret_type == RET_INTEGER) { 891 regs[BPF_REG_0].type = UNKNOWN_VALUE; 892 } else if (fn->ret_type == RET_VOID) { 893 regs[BPF_REG_0].type = NOT_INIT; 894 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { 895 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; 896 /* remember map_ptr, so that check_map_access() 897 * can check 'value_size' boundary of memory access 898 * to map element returned from bpf_map_lookup_elem() 899 */ 900 if (map == NULL) { 901 verbose("kernel subsystem misconfigured verifier\n"); 902 return -EINVAL; 903 } 904 regs[BPF_REG_0].map_ptr = map; 905 } else { 906 verbose("unknown return type %d of func %d\n", 907 fn->ret_type, func_id); 908 return -EINVAL; 909 } 910 911 if (map && map->map_type == BPF_MAP_TYPE_PROG_ARRAY && 912 func_id != BPF_FUNC_tail_call) 913 /* prog_array map type needs extra care: 914 * only allow to pass it into bpf_tail_call() for now. 915 * bpf_map_delete_elem() can be allowed in the future, 916 * while bpf_map_update_elem() must only be done via syscall 917 */ 918 return -EINVAL; 919 920 if (func_id == BPF_FUNC_tail_call && 921 map->map_type != BPF_MAP_TYPE_PROG_ARRAY) 922 /* don't allow any other map type to be passed into 923 * bpf_tail_call() 924 */ 925 return -EINVAL; 926 927 return 0; 928 } 929 930 /* check validity of 32-bit and 64-bit arithmetic operations */ 931 static int check_alu_op(struct reg_state *regs, struct bpf_insn *insn) 932 { 933 u8 opcode = BPF_OP(insn->code); 934 int err; 935 936 if (opcode == BPF_END || opcode == BPF_NEG) { 937 if (opcode == BPF_NEG) { 938 if (BPF_SRC(insn->code) != 0 || 939 insn->src_reg != BPF_REG_0 || 940 insn->off != 0 || insn->imm != 0) { 941 verbose("BPF_NEG uses reserved fields\n"); 942 return -EINVAL; 943 } 944 } else { 945 if (insn->src_reg != BPF_REG_0 || insn->off != 0 || 946 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { 947 verbose("BPF_END uses reserved fields\n"); 948 return -EINVAL; 949 } 950 } 951 952 /* check src operand */ 953 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 954 if (err) 955 return err; 956 957 /* check dest operand */ 958 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 959 if (err) 960 return err; 961 962 } else if (opcode == BPF_MOV) { 963 964 if (BPF_SRC(insn->code) == BPF_X) { 965 if (insn->imm != 0 || insn->off != 0) { 966 verbose("BPF_MOV uses reserved fields\n"); 967 return -EINVAL; 968 } 969 970 /* check src operand */ 971 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 972 if (err) 973 return err; 974 } else { 975 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 976 verbose("BPF_MOV uses reserved fields\n"); 977 return -EINVAL; 978 } 979 } 980 981 /* check dest operand */ 982 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 983 if (err) 984 return err; 985 986 if (BPF_SRC(insn->code) == BPF_X) { 987 if (BPF_CLASS(insn->code) == BPF_ALU64) { 988 /* case: R1 = R2 989 * copy register state to dest reg 990 */ 991 regs[insn->dst_reg] = regs[insn->src_reg]; 992 } else { 993 regs[insn->dst_reg].type = UNKNOWN_VALUE; 994 regs[insn->dst_reg].map_ptr = NULL; 995 } 996 } else { 997 /* case: R = imm 998 * remember the value we stored into this reg 999 */ 1000 regs[insn->dst_reg].type = CONST_IMM; 1001 regs[insn->dst_reg].imm = insn->imm; 1002 } 1003 1004 } else if (opcode > BPF_END) { 1005 verbose("invalid BPF_ALU opcode %x\n", opcode); 1006 return -EINVAL; 1007 1008 } else { /* all other ALU ops: and, sub, xor, add, ... */ 1009 1010 bool stack_relative = false; 1011 1012 if (BPF_SRC(insn->code) == BPF_X) { 1013 if (insn->imm != 0 || insn->off != 0) { 1014 verbose("BPF_ALU uses reserved fields\n"); 1015 return -EINVAL; 1016 } 1017 /* check src1 operand */ 1018 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1019 if (err) 1020 return err; 1021 } else { 1022 if (insn->src_reg != BPF_REG_0 || insn->off != 0) { 1023 verbose("BPF_ALU uses reserved fields\n"); 1024 return -EINVAL; 1025 } 1026 } 1027 1028 /* check src2 operand */ 1029 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1030 if (err) 1031 return err; 1032 1033 if ((opcode == BPF_MOD || opcode == BPF_DIV) && 1034 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { 1035 verbose("div by zero\n"); 1036 return -EINVAL; 1037 } 1038 1039 /* pattern match 'bpf_add Rx, imm' instruction */ 1040 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && 1041 regs[insn->dst_reg].type == FRAME_PTR && 1042 BPF_SRC(insn->code) == BPF_K) 1043 stack_relative = true; 1044 1045 /* check dest operand */ 1046 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1047 if (err) 1048 return err; 1049 1050 if (stack_relative) { 1051 regs[insn->dst_reg].type = PTR_TO_STACK; 1052 regs[insn->dst_reg].imm = insn->imm; 1053 } 1054 } 1055 1056 return 0; 1057 } 1058 1059 static int check_cond_jmp_op(struct verifier_env *env, 1060 struct bpf_insn *insn, int *insn_idx) 1061 { 1062 struct reg_state *regs = env->cur_state.regs; 1063 struct verifier_state *other_branch; 1064 u8 opcode = BPF_OP(insn->code); 1065 int err; 1066 1067 if (opcode > BPF_EXIT) { 1068 verbose("invalid BPF_JMP opcode %x\n", opcode); 1069 return -EINVAL; 1070 } 1071 1072 if (BPF_SRC(insn->code) == BPF_X) { 1073 if (insn->imm != 0) { 1074 verbose("BPF_JMP uses reserved fields\n"); 1075 return -EINVAL; 1076 } 1077 1078 /* check src1 operand */ 1079 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1080 if (err) 1081 return err; 1082 } else { 1083 if (insn->src_reg != BPF_REG_0) { 1084 verbose("BPF_JMP uses reserved fields\n"); 1085 return -EINVAL; 1086 } 1087 } 1088 1089 /* check src2 operand */ 1090 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1091 if (err) 1092 return err; 1093 1094 /* detect if R == 0 where R was initialized to zero earlier */ 1095 if (BPF_SRC(insn->code) == BPF_K && 1096 (opcode == BPF_JEQ || opcode == BPF_JNE) && 1097 regs[insn->dst_reg].type == CONST_IMM && 1098 regs[insn->dst_reg].imm == insn->imm) { 1099 if (opcode == BPF_JEQ) { 1100 /* if (imm == imm) goto pc+off; 1101 * only follow the goto, ignore fall-through 1102 */ 1103 *insn_idx += insn->off; 1104 return 0; 1105 } else { 1106 /* if (imm != imm) goto pc+off; 1107 * only follow fall-through branch, since 1108 * that's where the program will go 1109 */ 1110 return 0; 1111 } 1112 } 1113 1114 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); 1115 if (!other_branch) 1116 return -EFAULT; 1117 1118 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */ 1119 if (BPF_SRC(insn->code) == BPF_K && 1120 insn->imm == 0 && (opcode == BPF_JEQ || 1121 opcode == BPF_JNE) && 1122 regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) { 1123 if (opcode == BPF_JEQ) { 1124 /* next fallthrough insn can access memory via 1125 * this register 1126 */ 1127 regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; 1128 /* branch targer cannot access it, since reg == 0 */ 1129 other_branch->regs[insn->dst_reg].type = CONST_IMM; 1130 other_branch->regs[insn->dst_reg].imm = 0; 1131 } else { 1132 other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; 1133 regs[insn->dst_reg].type = CONST_IMM; 1134 regs[insn->dst_reg].imm = 0; 1135 } 1136 } else if (BPF_SRC(insn->code) == BPF_K && 1137 (opcode == BPF_JEQ || opcode == BPF_JNE)) { 1138 1139 if (opcode == BPF_JEQ) { 1140 /* detect if (R == imm) goto 1141 * and in the target state recognize that R = imm 1142 */ 1143 other_branch->regs[insn->dst_reg].type = CONST_IMM; 1144 other_branch->regs[insn->dst_reg].imm = insn->imm; 1145 } else { 1146 /* detect if (R != imm) goto 1147 * and in the fall-through state recognize that R = imm 1148 */ 1149 regs[insn->dst_reg].type = CONST_IMM; 1150 regs[insn->dst_reg].imm = insn->imm; 1151 } 1152 } 1153 if (log_level) 1154 print_verifier_state(env); 1155 return 0; 1156 } 1157 1158 /* return the map pointer stored inside BPF_LD_IMM64 instruction */ 1159 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) 1160 { 1161 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; 1162 1163 return (struct bpf_map *) (unsigned long) imm64; 1164 } 1165 1166 /* verify BPF_LD_IMM64 instruction */ 1167 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn) 1168 { 1169 struct reg_state *regs = env->cur_state.regs; 1170 int err; 1171 1172 if (BPF_SIZE(insn->code) != BPF_DW) { 1173 verbose("invalid BPF_LD_IMM insn\n"); 1174 return -EINVAL; 1175 } 1176 if (insn->off != 0) { 1177 verbose("BPF_LD_IMM64 uses reserved fields\n"); 1178 return -EINVAL; 1179 } 1180 1181 err = check_reg_arg(regs, insn->dst_reg, DST_OP); 1182 if (err) 1183 return err; 1184 1185 if (insn->src_reg == 0) 1186 /* generic move 64-bit immediate into a register */ 1187 return 0; 1188 1189 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ 1190 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); 1191 1192 regs[insn->dst_reg].type = CONST_PTR_TO_MAP; 1193 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); 1194 return 0; 1195 } 1196 1197 static bool may_access_skb(enum bpf_prog_type type) 1198 { 1199 switch (type) { 1200 case BPF_PROG_TYPE_SOCKET_FILTER: 1201 case BPF_PROG_TYPE_SCHED_CLS: 1202 case BPF_PROG_TYPE_SCHED_ACT: 1203 return true; 1204 default: 1205 return false; 1206 } 1207 } 1208 1209 /* verify safety of LD_ABS|LD_IND instructions: 1210 * - they can only appear in the programs where ctx == skb 1211 * - since they are wrappers of function calls, they scratch R1-R5 registers, 1212 * preserve R6-R9, and store return value into R0 1213 * 1214 * Implicit input: 1215 * ctx == skb == R6 == CTX 1216 * 1217 * Explicit input: 1218 * SRC == any register 1219 * IMM == 32-bit immediate 1220 * 1221 * Output: 1222 * R0 - 8/16/32-bit skb data converted to cpu endianness 1223 */ 1224 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn) 1225 { 1226 struct reg_state *regs = env->cur_state.regs; 1227 u8 mode = BPF_MODE(insn->code); 1228 struct reg_state *reg; 1229 int i, err; 1230 1231 if (!may_access_skb(env->prog->type)) { 1232 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n"); 1233 return -EINVAL; 1234 } 1235 1236 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || 1237 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { 1238 verbose("BPF_LD_ABS uses reserved fields\n"); 1239 return -EINVAL; 1240 } 1241 1242 /* check whether implicit source operand (register R6) is readable */ 1243 err = check_reg_arg(regs, BPF_REG_6, SRC_OP); 1244 if (err) 1245 return err; 1246 1247 if (regs[BPF_REG_6].type != PTR_TO_CTX) { 1248 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); 1249 return -EINVAL; 1250 } 1251 1252 if (mode == BPF_IND) { 1253 /* check explicit source operand */ 1254 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1255 if (err) 1256 return err; 1257 } 1258 1259 /* reset caller saved regs to unreadable */ 1260 for (i = 0; i < CALLER_SAVED_REGS; i++) { 1261 reg = regs + caller_saved[i]; 1262 reg->type = NOT_INIT; 1263 reg->imm = 0; 1264 } 1265 1266 /* mark destination R0 register as readable, since it contains 1267 * the value fetched from the packet 1268 */ 1269 regs[BPF_REG_0].type = UNKNOWN_VALUE; 1270 return 0; 1271 } 1272 1273 /* non-recursive DFS pseudo code 1274 * 1 procedure DFS-iterative(G,v): 1275 * 2 label v as discovered 1276 * 3 let S be a stack 1277 * 4 S.push(v) 1278 * 5 while S is not empty 1279 * 6 t <- S.pop() 1280 * 7 if t is what we're looking for: 1281 * 8 return t 1282 * 9 for all edges e in G.adjacentEdges(t) do 1283 * 10 if edge e is already labelled 1284 * 11 continue with the next edge 1285 * 12 w <- G.adjacentVertex(t,e) 1286 * 13 if vertex w is not discovered and not explored 1287 * 14 label e as tree-edge 1288 * 15 label w as discovered 1289 * 16 S.push(w) 1290 * 17 continue at 5 1291 * 18 else if vertex w is discovered 1292 * 19 label e as back-edge 1293 * 20 else 1294 * 21 // vertex w is explored 1295 * 22 label e as forward- or cross-edge 1296 * 23 label t as explored 1297 * 24 S.pop() 1298 * 1299 * convention: 1300 * 0x10 - discovered 1301 * 0x11 - discovered and fall-through edge labelled 1302 * 0x12 - discovered and fall-through and branch edges labelled 1303 * 0x20 - explored 1304 */ 1305 1306 enum { 1307 DISCOVERED = 0x10, 1308 EXPLORED = 0x20, 1309 FALLTHROUGH = 1, 1310 BRANCH = 2, 1311 }; 1312 1313 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L) 1314 1315 static int *insn_stack; /* stack of insns to process */ 1316 static int cur_stack; /* current stack index */ 1317 static int *insn_state; 1318 1319 /* t, w, e - match pseudo-code above: 1320 * t - index of current instruction 1321 * w - next instruction 1322 * e - edge 1323 */ 1324 static int push_insn(int t, int w, int e, struct verifier_env *env) 1325 { 1326 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) 1327 return 0; 1328 1329 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) 1330 return 0; 1331 1332 if (w < 0 || w >= env->prog->len) { 1333 verbose("jump out of range from insn %d to %d\n", t, w); 1334 return -EINVAL; 1335 } 1336 1337 if (e == BRANCH) 1338 /* mark branch target for state pruning */ 1339 env->explored_states[w] = STATE_LIST_MARK; 1340 1341 if (insn_state[w] == 0) { 1342 /* tree-edge */ 1343 insn_state[t] = DISCOVERED | e; 1344 insn_state[w] = DISCOVERED; 1345 if (cur_stack >= env->prog->len) 1346 return -E2BIG; 1347 insn_stack[cur_stack++] = w; 1348 return 1; 1349 } else if ((insn_state[w] & 0xF0) == DISCOVERED) { 1350 verbose("back-edge from insn %d to %d\n", t, w); 1351 return -EINVAL; 1352 } else if (insn_state[w] == EXPLORED) { 1353 /* forward- or cross-edge */ 1354 insn_state[t] = DISCOVERED | e; 1355 } else { 1356 verbose("insn state internal bug\n"); 1357 return -EFAULT; 1358 } 1359 return 0; 1360 } 1361 1362 /* non-recursive depth-first-search to detect loops in BPF program 1363 * loop == back-edge in directed graph 1364 */ 1365 static int check_cfg(struct verifier_env *env) 1366 { 1367 struct bpf_insn *insns = env->prog->insnsi; 1368 int insn_cnt = env->prog->len; 1369 int ret = 0; 1370 int i, t; 1371 1372 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 1373 if (!insn_state) 1374 return -ENOMEM; 1375 1376 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); 1377 if (!insn_stack) { 1378 kfree(insn_state); 1379 return -ENOMEM; 1380 } 1381 1382 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ 1383 insn_stack[0] = 0; /* 0 is the first instruction */ 1384 cur_stack = 1; 1385 1386 peek_stack: 1387 if (cur_stack == 0) 1388 goto check_state; 1389 t = insn_stack[cur_stack - 1]; 1390 1391 if (BPF_CLASS(insns[t].code) == BPF_JMP) { 1392 u8 opcode = BPF_OP(insns[t].code); 1393 1394 if (opcode == BPF_EXIT) { 1395 goto mark_explored; 1396 } else if (opcode == BPF_CALL) { 1397 ret = push_insn(t, t + 1, FALLTHROUGH, env); 1398 if (ret == 1) 1399 goto peek_stack; 1400 else if (ret < 0) 1401 goto err_free; 1402 } else if (opcode == BPF_JA) { 1403 if (BPF_SRC(insns[t].code) != BPF_K) { 1404 ret = -EINVAL; 1405 goto err_free; 1406 } 1407 /* unconditional jump with single edge */ 1408 ret = push_insn(t, t + insns[t].off + 1, 1409 FALLTHROUGH, env); 1410 if (ret == 1) 1411 goto peek_stack; 1412 else if (ret < 0) 1413 goto err_free; 1414 /* tell verifier to check for equivalent states 1415 * after every call and jump 1416 */ 1417 if (t + 1 < insn_cnt) 1418 env->explored_states[t + 1] = STATE_LIST_MARK; 1419 } else { 1420 /* conditional jump with two edges */ 1421 ret = push_insn(t, t + 1, FALLTHROUGH, env); 1422 if (ret == 1) 1423 goto peek_stack; 1424 else if (ret < 0) 1425 goto err_free; 1426 1427 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); 1428 if (ret == 1) 1429 goto peek_stack; 1430 else if (ret < 0) 1431 goto err_free; 1432 } 1433 } else { 1434 /* all other non-branch instructions with single 1435 * fall-through edge 1436 */ 1437 ret = push_insn(t, t + 1, FALLTHROUGH, env); 1438 if (ret == 1) 1439 goto peek_stack; 1440 else if (ret < 0) 1441 goto err_free; 1442 } 1443 1444 mark_explored: 1445 insn_state[t] = EXPLORED; 1446 if (cur_stack-- <= 0) { 1447 verbose("pop stack internal bug\n"); 1448 ret = -EFAULT; 1449 goto err_free; 1450 } 1451 goto peek_stack; 1452 1453 check_state: 1454 for (i = 0; i < insn_cnt; i++) { 1455 if (insn_state[i] != EXPLORED) { 1456 verbose("unreachable insn %d\n", i); 1457 ret = -EINVAL; 1458 goto err_free; 1459 } 1460 } 1461 ret = 0; /* cfg looks good */ 1462 1463 err_free: 1464 kfree(insn_state); 1465 kfree(insn_stack); 1466 return ret; 1467 } 1468 1469 /* compare two verifier states 1470 * 1471 * all states stored in state_list are known to be valid, since 1472 * verifier reached 'bpf_exit' instruction through them 1473 * 1474 * this function is called when verifier exploring different branches of 1475 * execution popped from the state stack. If it sees an old state that has 1476 * more strict register state and more strict stack state then this execution 1477 * branch doesn't need to be explored further, since verifier already 1478 * concluded that more strict state leads to valid finish. 1479 * 1480 * Therefore two states are equivalent if register state is more conservative 1481 * and explored stack state is more conservative than the current one. 1482 * Example: 1483 * explored current 1484 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) 1485 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) 1486 * 1487 * In other words if current stack state (one being explored) has more 1488 * valid slots than old one that already passed validation, it means 1489 * the verifier can stop exploring and conclude that current state is valid too 1490 * 1491 * Similarly with registers. If explored state has register type as invalid 1492 * whereas register type in current state is meaningful, it means that 1493 * the current state will reach 'bpf_exit' instruction safely 1494 */ 1495 static bool states_equal(struct verifier_state *old, struct verifier_state *cur) 1496 { 1497 int i; 1498 1499 for (i = 0; i < MAX_BPF_REG; i++) { 1500 if (memcmp(&old->regs[i], &cur->regs[i], 1501 sizeof(old->regs[0])) != 0) { 1502 if (old->regs[i].type == NOT_INIT || 1503 (old->regs[i].type == UNKNOWN_VALUE && 1504 cur->regs[i].type != NOT_INIT)) 1505 continue; 1506 return false; 1507 } 1508 } 1509 1510 for (i = 0; i < MAX_BPF_STACK; i++) { 1511 if (old->stack_slot_type[i] == STACK_INVALID) 1512 continue; 1513 if (old->stack_slot_type[i] != cur->stack_slot_type[i]) 1514 /* Ex: old explored (safe) state has STACK_SPILL in 1515 * this stack slot, but current has has STACK_MISC -> 1516 * this verifier states are not equivalent, 1517 * return false to continue verification of this path 1518 */ 1519 return false; 1520 if (i % BPF_REG_SIZE) 1521 continue; 1522 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], 1523 &cur->spilled_regs[i / BPF_REG_SIZE], 1524 sizeof(old->spilled_regs[0]))) 1525 /* when explored and current stack slot types are 1526 * the same, check that stored pointers types 1527 * are the same as well. 1528 * Ex: explored safe path could have stored 1529 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8} 1530 * but current path has stored: 1531 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16} 1532 * such verifier states are not equivalent. 1533 * return false to continue verification of this path 1534 */ 1535 return false; 1536 else 1537 continue; 1538 } 1539 return true; 1540 } 1541 1542 static int is_state_visited(struct verifier_env *env, int insn_idx) 1543 { 1544 struct verifier_state_list *new_sl; 1545 struct verifier_state_list *sl; 1546 1547 sl = env->explored_states[insn_idx]; 1548 if (!sl) 1549 /* this 'insn_idx' instruction wasn't marked, so we will not 1550 * be doing state search here 1551 */ 1552 return 0; 1553 1554 while (sl != STATE_LIST_MARK) { 1555 if (states_equal(&sl->state, &env->cur_state)) 1556 /* reached equivalent register/stack state, 1557 * prune the search 1558 */ 1559 return 1; 1560 sl = sl->next; 1561 } 1562 1563 /* there were no equivalent states, remember current one. 1564 * technically the current state is not proven to be safe yet, 1565 * but it will either reach bpf_exit (which means it's safe) or 1566 * it will be rejected. Since there are no loops, we won't be 1567 * seeing this 'insn_idx' instruction again on the way to bpf_exit 1568 */ 1569 new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER); 1570 if (!new_sl) 1571 return -ENOMEM; 1572 1573 /* add new state to the head of linked list */ 1574 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); 1575 new_sl->next = env->explored_states[insn_idx]; 1576 env->explored_states[insn_idx] = new_sl; 1577 return 0; 1578 } 1579 1580 static int do_check(struct verifier_env *env) 1581 { 1582 struct verifier_state *state = &env->cur_state; 1583 struct bpf_insn *insns = env->prog->insnsi; 1584 struct reg_state *regs = state->regs; 1585 int insn_cnt = env->prog->len; 1586 int insn_idx, prev_insn_idx = 0; 1587 int insn_processed = 0; 1588 bool do_print_state = false; 1589 1590 init_reg_state(regs); 1591 insn_idx = 0; 1592 for (;;) { 1593 struct bpf_insn *insn; 1594 u8 class; 1595 int err; 1596 1597 if (insn_idx >= insn_cnt) { 1598 verbose("invalid insn idx %d insn_cnt %d\n", 1599 insn_idx, insn_cnt); 1600 return -EFAULT; 1601 } 1602 1603 insn = &insns[insn_idx]; 1604 class = BPF_CLASS(insn->code); 1605 1606 if (++insn_processed > 32768) { 1607 verbose("BPF program is too large. Proccessed %d insn\n", 1608 insn_processed); 1609 return -E2BIG; 1610 } 1611 1612 err = is_state_visited(env, insn_idx); 1613 if (err < 0) 1614 return err; 1615 if (err == 1) { 1616 /* found equivalent state, can prune the search */ 1617 if (log_level) { 1618 if (do_print_state) 1619 verbose("\nfrom %d to %d: safe\n", 1620 prev_insn_idx, insn_idx); 1621 else 1622 verbose("%d: safe\n", insn_idx); 1623 } 1624 goto process_bpf_exit; 1625 } 1626 1627 if (log_level && do_print_state) { 1628 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); 1629 print_verifier_state(env); 1630 do_print_state = false; 1631 } 1632 1633 if (log_level) { 1634 verbose("%d: ", insn_idx); 1635 print_bpf_insn(insn); 1636 } 1637 1638 if (class == BPF_ALU || class == BPF_ALU64) { 1639 err = check_alu_op(regs, insn); 1640 if (err) 1641 return err; 1642 1643 } else if (class == BPF_LDX) { 1644 enum bpf_reg_type src_reg_type; 1645 1646 /* check for reserved fields is already done */ 1647 1648 /* check src operand */ 1649 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1650 if (err) 1651 return err; 1652 1653 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); 1654 if (err) 1655 return err; 1656 1657 src_reg_type = regs[insn->src_reg].type; 1658 1659 /* check that memory (src_reg + off) is readable, 1660 * the state of dst_reg will be updated by this func 1661 */ 1662 err = check_mem_access(env, insn->src_reg, insn->off, 1663 BPF_SIZE(insn->code), BPF_READ, 1664 insn->dst_reg); 1665 if (err) 1666 return err; 1667 1668 if (BPF_SIZE(insn->code) != BPF_W) { 1669 insn_idx++; 1670 continue; 1671 } 1672 1673 if (insn->imm == 0) { 1674 /* saw a valid insn 1675 * dst_reg = *(u32 *)(src_reg + off) 1676 * use reserved 'imm' field to mark this insn 1677 */ 1678 insn->imm = src_reg_type; 1679 1680 } else if (src_reg_type != insn->imm && 1681 (src_reg_type == PTR_TO_CTX || 1682 insn->imm == PTR_TO_CTX)) { 1683 /* ABuser program is trying to use the same insn 1684 * dst_reg = *(u32*) (src_reg + off) 1685 * with different pointer types: 1686 * src_reg == ctx in one branch and 1687 * src_reg == stack|map in some other branch. 1688 * Reject it. 1689 */ 1690 verbose("same insn cannot be used with different pointers\n"); 1691 return -EINVAL; 1692 } 1693 1694 } else if (class == BPF_STX) { 1695 enum bpf_reg_type dst_reg_type; 1696 1697 if (BPF_MODE(insn->code) == BPF_XADD) { 1698 err = check_xadd(env, insn); 1699 if (err) 1700 return err; 1701 insn_idx++; 1702 continue; 1703 } 1704 1705 /* check src1 operand */ 1706 err = check_reg_arg(regs, insn->src_reg, SRC_OP); 1707 if (err) 1708 return err; 1709 /* check src2 operand */ 1710 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1711 if (err) 1712 return err; 1713 1714 dst_reg_type = regs[insn->dst_reg].type; 1715 1716 /* check that memory (dst_reg + off) is writeable */ 1717 err = check_mem_access(env, insn->dst_reg, insn->off, 1718 BPF_SIZE(insn->code), BPF_WRITE, 1719 insn->src_reg); 1720 if (err) 1721 return err; 1722 1723 if (insn->imm == 0) { 1724 insn->imm = dst_reg_type; 1725 } else if (dst_reg_type != insn->imm && 1726 (dst_reg_type == PTR_TO_CTX || 1727 insn->imm == PTR_TO_CTX)) { 1728 verbose("same insn cannot be used with different pointers\n"); 1729 return -EINVAL; 1730 } 1731 1732 } else if (class == BPF_ST) { 1733 if (BPF_MODE(insn->code) != BPF_MEM || 1734 insn->src_reg != BPF_REG_0) { 1735 verbose("BPF_ST uses reserved fields\n"); 1736 return -EINVAL; 1737 } 1738 /* check src operand */ 1739 err = check_reg_arg(regs, insn->dst_reg, SRC_OP); 1740 if (err) 1741 return err; 1742 1743 /* check that memory (dst_reg + off) is writeable */ 1744 err = check_mem_access(env, insn->dst_reg, insn->off, 1745 BPF_SIZE(insn->code), BPF_WRITE, 1746 -1); 1747 if (err) 1748 return err; 1749 1750 } else if (class == BPF_JMP) { 1751 u8 opcode = BPF_OP(insn->code); 1752 1753 if (opcode == BPF_CALL) { 1754 if (BPF_SRC(insn->code) != BPF_K || 1755 insn->off != 0 || 1756 insn->src_reg != BPF_REG_0 || 1757 insn->dst_reg != BPF_REG_0) { 1758 verbose("BPF_CALL uses reserved fields\n"); 1759 return -EINVAL; 1760 } 1761 1762 err = check_call(env, insn->imm); 1763 if (err) 1764 return err; 1765 1766 } else if (opcode == BPF_JA) { 1767 if (BPF_SRC(insn->code) != BPF_K || 1768 insn->imm != 0 || 1769 insn->src_reg != BPF_REG_0 || 1770 insn->dst_reg != BPF_REG_0) { 1771 verbose("BPF_JA uses reserved fields\n"); 1772 return -EINVAL; 1773 } 1774 1775 insn_idx += insn->off + 1; 1776 continue; 1777 1778 } else if (opcode == BPF_EXIT) { 1779 if (BPF_SRC(insn->code) != BPF_K || 1780 insn->imm != 0 || 1781 insn->src_reg != BPF_REG_0 || 1782 insn->dst_reg != BPF_REG_0) { 1783 verbose("BPF_EXIT uses reserved fields\n"); 1784 return -EINVAL; 1785 } 1786 1787 /* eBPF calling convetion is such that R0 is used 1788 * to return the value from eBPF program. 1789 * Make sure that it's readable at this time 1790 * of bpf_exit, which means that program wrote 1791 * something into it earlier 1792 */ 1793 err = check_reg_arg(regs, BPF_REG_0, SRC_OP); 1794 if (err) 1795 return err; 1796 1797 process_bpf_exit: 1798 insn_idx = pop_stack(env, &prev_insn_idx); 1799 if (insn_idx < 0) { 1800 break; 1801 } else { 1802 do_print_state = true; 1803 continue; 1804 } 1805 } else { 1806 err = check_cond_jmp_op(env, insn, &insn_idx); 1807 if (err) 1808 return err; 1809 } 1810 } else if (class == BPF_LD) { 1811 u8 mode = BPF_MODE(insn->code); 1812 1813 if (mode == BPF_ABS || mode == BPF_IND) { 1814 err = check_ld_abs(env, insn); 1815 if (err) 1816 return err; 1817 1818 } else if (mode == BPF_IMM) { 1819 err = check_ld_imm(env, insn); 1820 if (err) 1821 return err; 1822 1823 insn_idx++; 1824 } else { 1825 verbose("invalid BPF_LD mode\n"); 1826 return -EINVAL; 1827 } 1828 } else { 1829 verbose("unknown insn class %d\n", class); 1830 return -EINVAL; 1831 } 1832 1833 insn_idx++; 1834 } 1835 1836 return 0; 1837 } 1838 1839 /* look for pseudo eBPF instructions that access map FDs and 1840 * replace them with actual map pointers 1841 */ 1842 static int replace_map_fd_with_map_ptr(struct verifier_env *env) 1843 { 1844 struct bpf_insn *insn = env->prog->insnsi; 1845 int insn_cnt = env->prog->len; 1846 int i, j; 1847 1848 for (i = 0; i < insn_cnt; i++, insn++) { 1849 if (BPF_CLASS(insn->code) == BPF_LDX && 1850 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { 1851 verbose("BPF_LDX uses reserved fields\n"); 1852 return -EINVAL; 1853 } 1854 1855 if (BPF_CLASS(insn->code) == BPF_STX && 1856 ((BPF_MODE(insn->code) != BPF_MEM && 1857 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { 1858 verbose("BPF_STX uses reserved fields\n"); 1859 return -EINVAL; 1860 } 1861 1862 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { 1863 struct bpf_map *map; 1864 struct fd f; 1865 1866 if (i == insn_cnt - 1 || insn[1].code != 0 || 1867 insn[1].dst_reg != 0 || insn[1].src_reg != 0 || 1868 insn[1].off != 0) { 1869 verbose("invalid bpf_ld_imm64 insn\n"); 1870 return -EINVAL; 1871 } 1872 1873 if (insn->src_reg == 0) 1874 /* valid generic load 64-bit imm */ 1875 goto next_insn; 1876 1877 if (insn->src_reg != BPF_PSEUDO_MAP_FD) { 1878 verbose("unrecognized bpf_ld_imm64 insn\n"); 1879 return -EINVAL; 1880 } 1881 1882 f = fdget(insn->imm); 1883 1884 map = bpf_map_get(f); 1885 if (IS_ERR(map)) { 1886 verbose("fd %d is not pointing to valid bpf_map\n", 1887 insn->imm); 1888 fdput(f); 1889 return PTR_ERR(map); 1890 } 1891 1892 /* store map pointer inside BPF_LD_IMM64 instruction */ 1893 insn[0].imm = (u32) (unsigned long) map; 1894 insn[1].imm = ((u64) (unsigned long) map) >> 32; 1895 1896 /* check whether we recorded this map already */ 1897 for (j = 0; j < env->used_map_cnt; j++) 1898 if (env->used_maps[j] == map) { 1899 fdput(f); 1900 goto next_insn; 1901 } 1902 1903 if (env->used_map_cnt >= MAX_USED_MAPS) { 1904 fdput(f); 1905 return -E2BIG; 1906 } 1907 1908 /* remember this map */ 1909 env->used_maps[env->used_map_cnt++] = map; 1910 1911 /* hold the map. If the program is rejected by verifier, 1912 * the map will be released by release_maps() or it 1913 * will be used by the valid program until it's unloaded 1914 * and all maps are released in free_bpf_prog_info() 1915 */ 1916 atomic_inc(&map->refcnt); 1917 1918 fdput(f); 1919 next_insn: 1920 insn++; 1921 i++; 1922 } 1923 } 1924 1925 /* now all pseudo BPF_LD_IMM64 instructions load valid 1926 * 'struct bpf_map *' into a register instead of user map_fd. 1927 * These pointers will be used later by verifier to validate map access. 1928 */ 1929 return 0; 1930 } 1931 1932 /* drop refcnt of maps used by the rejected program */ 1933 static void release_maps(struct verifier_env *env) 1934 { 1935 int i; 1936 1937 for (i = 0; i < env->used_map_cnt; i++) 1938 bpf_map_put(env->used_maps[i]); 1939 } 1940 1941 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ 1942 static void convert_pseudo_ld_imm64(struct verifier_env *env) 1943 { 1944 struct bpf_insn *insn = env->prog->insnsi; 1945 int insn_cnt = env->prog->len; 1946 int i; 1947 1948 for (i = 0; i < insn_cnt; i++, insn++) 1949 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) 1950 insn->src_reg = 0; 1951 } 1952 1953 static void adjust_branches(struct bpf_prog *prog, int pos, int delta) 1954 { 1955 struct bpf_insn *insn = prog->insnsi; 1956 int insn_cnt = prog->len; 1957 int i; 1958 1959 for (i = 0; i < insn_cnt; i++, insn++) { 1960 if (BPF_CLASS(insn->code) != BPF_JMP || 1961 BPF_OP(insn->code) == BPF_CALL || 1962 BPF_OP(insn->code) == BPF_EXIT) 1963 continue; 1964 1965 /* adjust offset of jmps if necessary */ 1966 if (i < pos && i + insn->off + 1 > pos) 1967 insn->off += delta; 1968 else if (i > pos && i + insn->off + 1 < pos) 1969 insn->off -= delta; 1970 } 1971 } 1972 1973 /* convert load instructions that access fields of 'struct __sk_buff' 1974 * into sequence of instructions that access fields of 'struct sk_buff' 1975 */ 1976 static int convert_ctx_accesses(struct verifier_env *env) 1977 { 1978 struct bpf_insn *insn = env->prog->insnsi; 1979 int insn_cnt = env->prog->len; 1980 struct bpf_insn insn_buf[16]; 1981 struct bpf_prog *new_prog; 1982 u32 cnt; 1983 int i; 1984 enum bpf_access_type type; 1985 1986 if (!env->prog->aux->ops->convert_ctx_access) 1987 return 0; 1988 1989 for (i = 0; i < insn_cnt; i++, insn++) { 1990 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W)) 1991 type = BPF_READ; 1992 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W)) 1993 type = BPF_WRITE; 1994 else 1995 continue; 1996 1997 if (insn->imm != PTR_TO_CTX) { 1998 /* clear internal mark */ 1999 insn->imm = 0; 2000 continue; 2001 } 2002 2003 cnt = env->prog->aux->ops-> 2004 convert_ctx_access(type, insn->dst_reg, insn->src_reg, 2005 insn->off, insn_buf); 2006 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { 2007 verbose("bpf verifier is misconfigured\n"); 2008 return -EINVAL; 2009 } 2010 2011 if (cnt == 1) { 2012 memcpy(insn, insn_buf, sizeof(*insn)); 2013 continue; 2014 } 2015 2016 /* several new insns need to be inserted. Make room for them */ 2017 insn_cnt += cnt - 1; 2018 new_prog = bpf_prog_realloc(env->prog, 2019 bpf_prog_size(insn_cnt), 2020 GFP_USER); 2021 if (!new_prog) 2022 return -ENOMEM; 2023 2024 new_prog->len = insn_cnt; 2025 2026 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1, 2027 sizeof(*insn) * (insn_cnt - i - cnt)); 2028 2029 /* copy substitute insns in place of load instruction */ 2030 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt); 2031 2032 /* adjust branches in the whole program */ 2033 adjust_branches(new_prog, i, cnt - 1); 2034 2035 /* keep walking new program and skip insns we just inserted */ 2036 env->prog = new_prog; 2037 insn = new_prog->insnsi + i + cnt - 1; 2038 i += cnt - 1; 2039 } 2040 2041 return 0; 2042 } 2043 2044 static void free_states(struct verifier_env *env) 2045 { 2046 struct verifier_state_list *sl, *sln; 2047 int i; 2048 2049 if (!env->explored_states) 2050 return; 2051 2052 for (i = 0; i < env->prog->len; i++) { 2053 sl = env->explored_states[i]; 2054 2055 if (sl) 2056 while (sl != STATE_LIST_MARK) { 2057 sln = sl->next; 2058 kfree(sl); 2059 sl = sln; 2060 } 2061 } 2062 2063 kfree(env->explored_states); 2064 } 2065 2066 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) 2067 { 2068 char __user *log_ubuf = NULL; 2069 struct verifier_env *env; 2070 int ret = -EINVAL; 2071 2072 if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS) 2073 return -E2BIG; 2074 2075 /* 'struct verifier_env' can be global, but since it's not small, 2076 * allocate/free it every time bpf_check() is called 2077 */ 2078 env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL); 2079 if (!env) 2080 return -ENOMEM; 2081 2082 env->prog = *prog; 2083 2084 /* grab the mutex to protect few globals used by verifier */ 2085 mutex_lock(&bpf_verifier_lock); 2086 2087 if (attr->log_level || attr->log_buf || attr->log_size) { 2088 /* user requested verbose verifier output 2089 * and supplied buffer to store the verification trace 2090 */ 2091 log_level = attr->log_level; 2092 log_ubuf = (char __user *) (unsigned long) attr->log_buf; 2093 log_size = attr->log_size; 2094 log_len = 0; 2095 2096 ret = -EINVAL; 2097 /* log_* values have to be sane */ 2098 if (log_size < 128 || log_size > UINT_MAX >> 8 || 2099 log_level == 0 || log_ubuf == NULL) 2100 goto free_env; 2101 2102 ret = -ENOMEM; 2103 log_buf = vmalloc(log_size); 2104 if (!log_buf) 2105 goto free_env; 2106 } else { 2107 log_level = 0; 2108 } 2109 2110 ret = replace_map_fd_with_map_ptr(env); 2111 if (ret < 0) 2112 goto skip_full_check; 2113 2114 env->explored_states = kcalloc(env->prog->len, 2115 sizeof(struct verifier_state_list *), 2116 GFP_USER); 2117 ret = -ENOMEM; 2118 if (!env->explored_states) 2119 goto skip_full_check; 2120 2121 ret = check_cfg(env); 2122 if (ret < 0) 2123 goto skip_full_check; 2124 2125 ret = do_check(env); 2126 2127 skip_full_check: 2128 while (pop_stack(env, NULL) >= 0); 2129 free_states(env); 2130 2131 if (ret == 0) 2132 /* program is valid, convert *(u32*)(ctx + off) accesses */ 2133 ret = convert_ctx_accesses(env); 2134 2135 if (log_level && log_len >= log_size - 1) { 2136 BUG_ON(log_len >= log_size); 2137 /* verifier log exceeded user supplied buffer */ 2138 ret = -ENOSPC; 2139 /* fall through to return what was recorded */ 2140 } 2141 2142 /* copy verifier log back to user space including trailing zero */ 2143 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { 2144 ret = -EFAULT; 2145 goto free_log_buf; 2146 } 2147 2148 if (ret == 0 && env->used_map_cnt) { 2149 /* if program passed verifier, update used_maps in bpf_prog_info */ 2150 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, 2151 sizeof(env->used_maps[0]), 2152 GFP_KERNEL); 2153 2154 if (!env->prog->aux->used_maps) { 2155 ret = -ENOMEM; 2156 goto free_log_buf; 2157 } 2158 2159 memcpy(env->prog->aux->used_maps, env->used_maps, 2160 sizeof(env->used_maps[0]) * env->used_map_cnt); 2161 env->prog->aux->used_map_cnt = env->used_map_cnt; 2162 2163 /* program is valid. Convert pseudo bpf_ld_imm64 into generic 2164 * bpf_ld_imm64 instructions 2165 */ 2166 convert_pseudo_ld_imm64(env); 2167 } 2168 2169 free_log_buf: 2170 if (log_level) 2171 vfree(log_buf); 2172 free_env: 2173 if (!env->prog->aux->used_maps) 2174 /* if we didn't copy map pointers into bpf_prog_info, release 2175 * them now. Otherwise free_bpf_prog_info() will release them. 2176 */ 2177 release_maps(env); 2178 *prog = env->prog; 2179 kfree(env); 2180 mutex_unlock(&bpf_verifier_lock); 2181 return ret; 2182 } 2183