kernel/bpf/verifier.c

   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         bool allow_ptr_leaks;
 203 };
 204
 205 /* verbose verifier prints what it's seeing
 206  * bpf_check() is called under lock, so no race to access these global vars
 207  */
 208 static u32 log_level, log_size, log_len;
 209 static char *log_buf;
 210
 211 static DEFINE_MUTEX(bpf_verifier_lock);
 212
 213 /* log_level controls verbosity level of eBPF verifier.
 214  * verbose() is used to dump the verification trace to the log, so the user
 215  * can figure out what's wrong with the program
 216  */
 217 static __printf(1, 2) void verbose(const char *fmt, ...)
 218 {
 219         va_list args;
 220
 221         if (log_level == 0 || log_len >= log_size - 1)
 222                 return;
 223
 224         va_start(args, fmt);
 225         log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
 226         va_end(args);
 227 }
 228
 229 /* string representation of 'enum bpf_reg_type' */
 230 static const char * const reg_type_str[] = {
 231         [NOT_INIT]              = "?",
 232         [UNKNOWN_VALUE]         = "inv",
 233         [PTR_TO_CTX]            = "ctx",
 234         [CONST_PTR_TO_MAP]      = "map_ptr",
 235         [PTR_TO_MAP_VALUE]      = "map_value",
 236         [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
 237         [FRAME_PTR]             = "fp",
 238         [PTR_TO_STACK]          = "fp",
 239         [CONST_IMM]             = "imm",
 240 };
 241
 242 static const struct {
 243         int map_type;
 244         int func_id;
 245 } func_limit[] = {
 246         {BPF_MAP_TYPE_PROG_ARRAY, BPF_FUNC_tail_call},
 247         {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_read},
 248         {BPF_MAP_TYPE_PERF_EVENT_ARRAY, BPF_FUNC_perf_event_output},
 249         {BPF_MAP_TYPE_STACK_TRACE, BPF_FUNC_get_stackid},
 250 };
 251
 252 static void print_verifier_state(struct verifier_env *env)
 253 {
 254         enum bpf_reg_type t;
 255         int i;
 256
 257         for (i = 0; i < MAX_BPF_REG; i++) {
 258                 t = env->cur_state.regs[i].type;
 259                 if (t == NOT_INIT)
 260                         continue;
 261                 verbose(" R%d=%s", i, reg_type_str[t]);
 262                 if (t == CONST_IMM || t == PTR_TO_STACK)
 263                         verbose("%d", env->cur_state.regs[i].imm);
 264                 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
 265                          t == PTR_TO_MAP_VALUE_OR_NULL)
 266                         verbose("(ks=%d,vs=%d)",
 267                                 env->cur_state.regs[i].map_ptr->key_size,
 268                                 env->cur_state.regs[i].map_ptr->value_size);
 269         }
 270         for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
 271                 if (env->cur_state.stack_slot_type[i] == STACK_SPILL)
 272                         verbose(" fp%d=%s", -MAX_BPF_STACK + i,
 273                                 reg_type_str[env->cur_state.spilled_regs[i / BPF_REG_SIZE].type]);
 274         }
 275         verbose("\n");
 276 }
 277
 278 static const char *const bpf_class_string[] = {
 279         [BPF_LD]    = "ld",
 280         [BPF_LDX]   = "ldx",
 281         [BPF_ST]    = "st",
 282         [BPF_STX]   = "stx",
 283         [BPF_ALU]   = "alu",
 284         [BPF_JMP]   = "jmp",
 285         [BPF_RET]   = "BUG",
 286         [BPF_ALU64] = "alu64",
 287 };
 288
 289 static const char *const bpf_alu_string[16] = {
 290         [BPF_ADD >> 4]  = "+=",
 291         [BPF_SUB >> 4]  = "-=",
 292         [BPF_MUL >> 4]  = "*=",
 293         [BPF_DIV >> 4]  = "/=",
 294         [BPF_OR  >> 4]  = "|=",
 295         [BPF_AND >> 4]  = "&=",
 296         [BPF_LSH >> 4]  = "<<=",
 297         [BPF_RSH >> 4]  = ">>=",
 298         [BPF_NEG >> 4]  = "neg",
 299         [BPF_MOD >> 4]  = "%=",
 300         [BPF_XOR >> 4]  = "^=",
 301         [BPF_MOV >> 4]  = "=",
 302         [BPF_ARSH >> 4] = "s>>=",
 303         [BPF_END >> 4]  = "endian",
 304 };
 305
 306 static const char *const bpf_ldst_string[] = {
 307         [BPF_W >> 3]  = "u32",
 308         [BPF_H >> 3]  = "u16",
 309         [BPF_B >> 3]  = "u8",
 310         [BPF_DW >> 3] = "u64",
 311 };
 312
 313 static const char *const bpf_jmp_string[16] = {
 314         [BPF_JA >> 4]   = "jmp",
 315         [BPF_JEQ >> 4]  = "==",
 316         [BPF_JGT >> 4]  = ">",
 317         [BPF_JGE >> 4]  = ">=",
 318         [BPF_JSET >> 4] = "&",
 319         [BPF_JNE >> 4]  = "!=",
 320         [BPF_JSGT >> 4] = "s>",
 321         [BPF_JSGE >> 4] = "s>=",
 322         [BPF_CALL >> 4] = "call",
 323         [BPF_EXIT >> 4] = "exit",
 324 };
 325
 326 static void print_bpf_insn(struct bpf_insn *insn)
 327 {
 328         u8 class = BPF_CLASS(insn->code);
 329
 330         if (class == BPF_ALU || class == BPF_ALU64) {
 331                 if (BPF_SRC(insn->code) == BPF_X)
 332                         verbose("(%02x) %sr%d %s %sr%d\n",
 333                                 insn->code, class == BPF_ALU ? "(u32) " : "",
 334                                 insn->dst_reg,
 335                                 bpf_alu_string[BPF_OP(insn->code) >> 4],
 336                                 class == BPF_ALU ? "(u32) " : "",
 337                                 insn->src_reg);
 338                 else
 339                         verbose("(%02x) %sr%d %s %s%d\n",
 340                                 insn->code, class == BPF_ALU ? "(u32) " : "",
 341                                 insn->dst_reg,
 342                                 bpf_alu_string[BPF_OP(insn->code) >> 4],
 343                                 class == BPF_ALU ? "(u32) " : "",
 344                                 insn->imm);
 345         } else if (class == BPF_STX) {
 346                 if (BPF_MODE(insn->code) == BPF_MEM)
 347                         verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
 348                                 insn->code,
 349                                 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 350                                 insn->dst_reg,
 351                                 insn->off, insn->src_reg);
 352                 else if (BPF_MODE(insn->code) == BPF_XADD)
 353                         verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
 354                                 insn->code,
 355                                 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 356                                 insn->dst_reg, insn->off,
 357                                 insn->src_reg);
 358                 else
 359                         verbose("BUG_%02x\n", insn->code);
 360         } else if (class == BPF_ST) {
 361                 if (BPF_MODE(insn->code) != BPF_MEM) {
 362                         verbose("BUG_st_%02x\n", insn->code);
 363                         return;
 364                 }
 365                 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
 366                         insn->code,
 367                         bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 368                         insn->dst_reg,
 369                         insn->off, insn->imm);
 370         } else if (class == BPF_LDX) {
 371                 if (BPF_MODE(insn->code) != BPF_MEM) {
 372                         verbose("BUG_ldx_%02x\n", insn->code);
 373                         return;
 374                 }
 375                 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
 376                         insn->code, insn->dst_reg,
 377                         bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 378                         insn->src_reg, insn->off);
 379         } else if (class == BPF_LD) {
 380                 if (BPF_MODE(insn->code) == BPF_ABS) {
 381                         verbose("(%02x) r0 = *(%s *)skb[%d]\n",
 382                                 insn->code,
 383                                 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 384                                 insn->imm);
 385                 } else if (BPF_MODE(insn->code) == BPF_IND) {
 386                         verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
 387                                 insn->code,
 388                                 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
 389                                 insn->src_reg, insn->imm);
 390                 } else if (BPF_MODE(insn->code) == BPF_IMM) {
 391                         verbose("(%02x) r%d = 0x%x\n",
 392                                 insn->code, insn->dst_reg, insn->imm);
 393                 } else {
 394                         verbose("BUG_ld_%02x\n", insn->code);
 395                         return;
 396                 }
 397         } else if (class == BPF_JMP) {
 398                 u8 opcode = BPF_OP(insn->code);
 399
 400                 if (opcode == BPF_CALL) {
 401                         verbose("(%02x) call %d\n", insn->code, insn->imm);
 402                 } else if (insn->code == (BPF_JMP | BPF_JA)) {
 403                         verbose("(%02x) goto pc%+d\n",
 404                                 insn->code, insn->off);
 405                 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
 406                         verbose("(%02x) exit\n", insn->code);
 407                 } else if (BPF_SRC(insn->code) == BPF_X) {
 408                         verbose("(%02x) if r%d %s r%d goto pc%+d\n",
 409                                 insn->code, insn->dst_reg,
 410                                 bpf_jmp_string[BPF_OP(insn->code) >> 4],
 411                                 insn->src_reg, insn->off);
 412                 } else {
 413                         verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
 414                                 insn->code, insn->dst_reg,
 415                                 bpf_jmp_string[BPF_OP(insn->code) >> 4],
 416                                 insn->imm, insn->off);
 417                 }
 418         } else {
 419                 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
 420         }
 421 }
 422
 423 static int pop_stack(struct verifier_env *env, int *prev_insn_idx)
 424 {
 425         struct verifier_stack_elem *elem;
 426         int insn_idx;
 427
 428         if (env->head == NULL)
 429                 return -1;
 430
 431         memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
 432         insn_idx = env->head->insn_idx;
 433         if (prev_insn_idx)
 434                 *prev_insn_idx = env->head->prev_insn_idx;
 435         elem = env->head->next;
 436         kfree(env->head);
 437         env->head = elem;
 438         env->stack_size--;
 439         return insn_idx;
 440 }
 441
 442 static struct verifier_state *push_stack(struct verifier_env *env, int insn_idx,
 443                                          int prev_insn_idx)
 444 {
 445         struct verifier_stack_elem *elem;
 446
 447         elem = kmalloc(sizeof(struct verifier_stack_elem), GFP_KERNEL);
 448         if (!elem)
 449                 goto err;
 450
 451         memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
 452         elem->insn_idx = insn_idx;
 453         elem->prev_insn_idx = prev_insn_idx;
 454         elem->next = env->head;
 455         env->head = elem;
 456         env->stack_size++;
 457         if (env->stack_size > 1024) {
 458                 verbose("BPF program is too complex\n");
 459                 goto err;
 460         }
 461         return &elem->st;
 462 err:
 463         /* pop all elements and return */
 464         while (pop_stack(env, NULL) >= 0);
 465         return NULL;
 466 }
 467
 468 #define CALLER_SAVED_REGS 6
 469 static const int caller_saved[CALLER_SAVED_REGS] = {
 470         BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
 471 };
 472
 473 static void init_reg_state(struct reg_state *regs)
 474 {
 475         int i;
 476
 477         for (i = 0; i < MAX_BPF_REG; i++) {
 478                 regs[i].type = NOT_INIT;
 479                 regs[i].imm = 0;
 480                 regs[i].map_ptr = NULL;
 481         }
 482
 483         /* frame pointer */
 484         regs[BPF_REG_FP].type = FRAME_PTR;
 485
 486         /* 1st arg to a function */
 487         regs[BPF_REG_1].type = PTR_TO_CTX;
 488 }
 489
 490 static void mark_reg_unknown_value(struct reg_state *regs, u32 regno)
 491 {
 492         BUG_ON(regno >= MAX_BPF_REG);
 493         regs[regno].type = UNKNOWN_VALUE;
 494         regs[regno].imm = 0;
 495         regs[regno].map_ptr = NULL;
 496 }
 497
 498 enum reg_arg_type {
 499         SRC_OP,         /* register is used as source operand */
 500         DST_OP,         /* register is used as destination operand */
 501         DST_OP_NO_MARK  /* same as above, check only, don't mark */
 502 };
 503
 504 static int check_reg_arg(struct reg_state *regs, u32 regno,
 505                          enum reg_arg_type t)
 506 {
 507         if (regno >= MAX_BPF_REG) {
 508                 verbose("R%d is invalid\n", regno);
 509                 return -EINVAL;
 510         }
 511
 512         if (t == SRC_OP) {
 513                 /* check whether register used as source operand can be read */
 514                 if (regs[regno].type == NOT_INIT) {
 515                         verbose("R%d !read_ok\n", regno);
 516                         return -EACCES;
 517                 }
 518         } else {
 519                 /* check whether register used as dest operand can be written to */
 520                 if (regno == BPF_REG_FP) {
 521                         verbose("frame pointer is read only\n");
 522                         return -EACCES;
 523                 }
 524                 if (t == DST_OP)
 525                         mark_reg_unknown_value(regs, regno);
 526         }
 527         return 0;
 528 }
 529
 530 static int bpf_size_to_bytes(int bpf_size)
 531 {
 532         if (bpf_size == BPF_W)
 533                 return 4;
 534         else if (bpf_size == BPF_H)
 535                 return 2;
 536         else if (bpf_size == BPF_B)
 537                 return 1;
 538         else if (bpf_size == BPF_DW)
 539                 return 8;
 540         else
 541                 return -EINVAL;
 542 }
 543
 544 static bool is_spillable_regtype(enum bpf_reg_type type)
 545 {
 546         switch (type) {
 547         case PTR_TO_MAP_VALUE:
 548         case PTR_TO_MAP_VALUE_OR_NULL:
 549         case PTR_TO_STACK:
 550         case PTR_TO_CTX:
 551         case FRAME_PTR:
 552         case CONST_PTR_TO_MAP:
 553                 return true;
 554         default:
 555                 return false;
 556         }
 557 }
 558
 559 /* check_stack_read/write functions track spill/fill of registers,
 560  * stack boundary and alignment are checked in check_mem_access()
 561  */
 562 static int check_stack_write(struct verifier_state *state, int off, int size,
 563                              int value_regno)
 564 {
 565         int i;
 566         /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
 567          * so it's aligned access and [off, off + size) are within stack limits
 568          */
 569
 570         if (value_regno >= 0 &&
 571             is_spillable_regtype(state->regs[value_regno].type)) {
 572
 573                 /* register containing pointer is being spilled into stack */
 574                 if (size != BPF_REG_SIZE) {
 575                         verbose("invalid size of register spill\n");
 576                         return -EACCES;
 577                 }
 578
 579                 /* save register state */
 580                 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
 581                         state->regs[value_regno];
 582
 583                 for (i = 0; i < BPF_REG_SIZE; i++)
 584                         state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
 585         } else {
 586                 /* regular write of data into stack */
 587                 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
 588                         (struct reg_state) {};
 589
 590                 for (i = 0; i < size; i++)
 591                         state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
 592         }
 593         return 0;
 594 }
 595
 596 static int check_stack_read(struct verifier_state *state, int off, int size,
 597                             int value_regno)
 598 {
 599         u8 *slot_type;
 600         int i;
 601
 602         slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
 603
 604         if (slot_type[0] == STACK_SPILL) {
 605                 if (size != BPF_REG_SIZE) {
 606                         verbose("invalid size of register spill\n");
 607                         return -EACCES;
 608                 }
 609                 for (i = 1; i < BPF_REG_SIZE; i++) {
 610                         if (slot_type[i] != STACK_SPILL) {
 611                                 verbose("corrupted spill memory\n");
 612                                 return -EACCES;
 613                         }
 614                 }
 615
 616                 if (value_regno >= 0)
 617                         /* restore register state from stack */
 618                         state->regs[value_regno] =
 619                                 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
 620                 return 0;
 621         } else {
 622                 for (i = 0; i < size; i++) {
 623                         if (slot_type[i] != STACK_MISC) {
 624                                 verbose("invalid read from stack off %d+%d size %d\n",
 625                                         off, i, size);
 626                                 return -EACCES;
 627                         }
 628                 }
 629                 if (value_regno >= 0)
 630                         /* have read misc data from the stack */
 631                         mark_reg_unknown_value(state->regs, value_regno);
 632                 return 0;
 633         }
 634 }
 635
 636 /* check read/write into map element returned by bpf_map_lookup_elem() */
 637 static int check_map_access(struct verifier_env *env, u32 regno, int off,
 638                             int size)
 639 {
 640         struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
 641
 642         if (off < 0 || off + size > map->value_size) {
 643                 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
 644                         map->value_size, off, size);
 645                 return -EACCES;
 646         }
 647         return 0;
 648 }
 649
 650 /* check access to 'struct bpf_context' fields */
 651 static int check_ctx_access(struct verifier_env *env, int off, int size,
 652                             enum bpf_access_type t)
 653 {
 654         if (env->prog->aux->ops->is_valid_access &&
 655             env->prog->aux->ops->is_valid_access(off, size, t))
 656                 return 0;
 657
 658         verbose("invalid bpf_context access off=%d size=%d\n", off, size);
 659         return -EACCES;
 660 }
 661
 662 static bool is_pointer_value(struct verifier_env *env, int regno)
 663 {
 664         if (env->allow_ptr_leaks)
 665                 return false;
 666
 667         switch (env->cur_state.regs[regno].type) {
 668         case UNKNOWN_VALUE:
 669         case CONST_IMM:
 670                 return false;
 671         default:
 672                 return true;
 673         }
 674 }
 675
 676 /* check whether memory at (regno + off) is accessible for t = (read | write)
 677  * if t==write, value_regno is a register which value is stored into memory
 678  * if t==read, value_regno is a register which will receive the value from memory
 679  * if t==write && value_regno==-1, some unknown value is stored into memory
 680  * if t==read && value_regno==-1, don't care what we read from memory
 681  */
 682 static int check_mem_access(struct verifier_env *env, u32 regno, int off,
 683                             int bpf_size, enum bpf_access_type t,
 684                             int value_regno)
 685 {
 686         struct verifier_state *state = &env->cur_state;
 687         int size, err = 0;
 688
 689         if (state->regs[regno].type == PTR_TO_STACK)
 690                 off += state->regs[regno].imm;
 691
 692         size = bpf_size_to_bytes(bpf_size);
 693         if (size < 0)
 694                 return size;
 695
 696         if (off % size != 0) {
 697                 verbose("misaligned access off %d size %d\n", off, size);
 698                 return -EACCES;
 699         }
 700
 701         if (state->regs[regno].type == PTR_TO_MAP_VALUE) {
 702                 if (t == BPF_WRITE && value_regno >= 0 &&
 703                     is_pointer_value(env, value_regno)) {
 704                         verbose("R%d leaks addr into map\n", value_regno);
 705                         return -EACCES;
 706                 }
 707                 err = check_map_access(env, regno, off, size);
 708                 if (!err && t == BPF_READ && value_regno >= 0)
 709                         mark_reg_unknown_value(state->regs, value_regno);
 710
 711         } else if (state->regs[regno].type == PTR_TO_CTX) {
 712                 if (t == BPF_WRITE && value_regno >= 0 &&
 713                     is_pointer_value(env, value_regno)) {
 714                         verbose("R%d leaks addr into ctx\n", value_regno);
 715                         return -EACCES;
 716                 }
 717                 err = check_ctx_access(env, off, size, t);
 718                 if (!err && t == BPF_READ && value_regno >= 0)
 719                         mark_reg_unknown_value(state->regs, value_regno);
 720
 721         } else if (state->regs[regno].type == FRAME_PTR ||
 722                    state->regs[regno].type == PTR_TO_STACK) {
 723                 if (off >= 0 || off < -MAX_BPF_STACK) {
 724                         verbose("invalid stack off=%d size=%d\n", off, size);
 725                         return -EACCES;
 726                 }
 727                 if (t == BPF_WRITE) {
 728                         if (!env->allow_ptr_leaks &&
 729                             state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
 730                             size != BPF_REG_SIZE) {
 731                                 verbose("attempt to corrupt spilled pointer on stack\n");
 732                                 return -EACCES;
 733                         }
 734                         err = check_stack_write(state, off, size, value_regno);
 735                 } else {
 736                         err = check_stack_read(state, off, size, value_regno);
 737                 }
 738         } else {
 739                 verbose("R%d invalid mem access '%s'\n",
 740                         regno, reg_type_str[state->regs[regno].type]);
 741                 return -EACCES;
 742         }
 743         return err;
 744 }
 745
 746 static int check_xadd(struct verifier_env *env, struct bpf_insn *insn)
 747 {
 748         struct reg_state *regs = env->cur_state.regs;
 749         int err;
 750
 751         if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
 752             insn->imm != 0) {
 753                 verbose("BPF_XADD uses reserved fields\n");
 754                 return -EINVAL;
 755         }
 756
 757         /* check src1 operand */
 758         err = check_reg_arg(regs, insn->src_reg, SRC_OP);
 759         if (err)
 760                 return err;
 761
 762         /* check src2 operand */
 763         err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
 764         if (err)
 765                 return err;
 766
 767         /* check whether atomic_add can read the memory */
 768         err = check_mem_access(env, insn->dst_reg, insn->off,
 769                                BPF_SIZE(insn->code), BPF_READ, -1);
 770         if (err)
 771                 return err;
 772
 773         /* check whether atomic_add can write into the same memory */
 774         return check_mem_access(env, insn->dst_reg, insn->off,
 775                                 BPF_SIZE(insn->code), BPF_WRITE, -1);
 776 }
 777
 778 /* when register 'regno' is passed into function that will read 'access_size'
 779  * bytes from that pointer, make sure that it's within stack boundary
 780  * and all elements of stack are initialized
 781  */
 782 static int check_stack_boundary(struct verifier_env *env, int regno,
 783                                 int access_size, bool zero_size_allowed)
 784 {
 785         struct verifier_state *state = &env->cur_state;
 786         struct reg_state *regs = state->regs;
 787         int off, i;
 788
 789         if (regs[regno].type != PTR_TO_STACK) {
 790                 if (zero_size_allowed && access_size == 0 &&
 791                     regs[regno].type == CONST_IMM &&
 792                     regs[regno].imm  == 0)
 793                         return 0;
 794
 795                 verbose("R%d type=%s expected=%s\n", regno,
 796                         reg_type_str[regs[regno].type],
 797                         reg_type_str[PTR_TO_STACK]);
 798                 return -EACCES;
 799         }
 800
 801         off = regs[regno].imm;
 802         if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
 803             access_size <= 0) {
 804                 verbose("invalid stack type R%d off=%d access_size=%d\n",
 805                         regno, off, access_size);
 806                 return -EACCES;
 807         }
 808
 809         for (i = 0; i < access_size; i++) {
 810                 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
 811                         verbose("invalid indirect read from stack off %d+%d size %d\n",
 812                                 off, i, access_size);
 813                         return -EACCES;
 814                 }
 815         }
 816         return 0;
 817 }
 818
 819 static int check_func_arg(struct verifier_env *env, u32 regno,
 820                           enum bpf_arg_type arg_type, struct bpf_map **mapp)
 821 {
 822         struct reg_state *reg = env->cur_state.regs + regno;
 823         enum bpf_reg_type expected_type;
 824         int err = 0;
 825
 826         if (arg_type == ARG_DONTCARE)
 827                 return 0;
 828
 829         if (reg->type == NOT_INIT) {
 830                 verbose("R%d !read_ok\n", regno);
 831                 return -EACCES;
 832         }
 833
 834         if (arg_type == ARG_ANYTHING) {
 835                 if (is_pointer_value(env, regno)) {
 836                         verbose("R%d leaks addr into helper function\n", regno);
 837                         return -EACCES;
 838                 }
 839                 return 0;
 840         }
 841
 842         if (arg_type == ARG_PTR_TO_MAP_KEY ||
 843             arg_type == ARG_PTR_TO_MAP_VALUE) {
 844                 expected_type = PTR_TO_STACK;
 845         } else if (arg_type == ARG_CONST_STACK_SIZE ||
 846                    arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
 847                 expected_type = CONST_IMM;
 848         } else if (arg_type == ARG_CONST_MAP_PTR) {
 849                 expected_type = CONST_PTR_TO_MAP;
 850         } else if (arg_type == ARG_PTR_TO_CTX) {
 851                 expected_type = PTR_TO_CTX;
 852         } else if (arg_type == ARG_PTR_TO_STACK) {
 853                 expected_type = PTR_TO_STACK;
 854                 /* One exception here. In case function allows for NULL to be
 855                  * passed in as argument, it's a CONST_IMM type. Final test
 856                  * happens during stack boundary checking.
 857                  */
 858                 if (reg->type == CONST_IMM && reg->imm == 0)
 859                         expected_type = CONST_IMM;
 860         } else {
 861                 verbose("unsupported arg_type %d\n", arg_type);
 862                 return -EFAULT;
 863         }
 864
 865         if (reg->type != expected_type) {
 866                 verbose("R%d type=%s expected=%s\n", regno,
 867                         reg_type_str[reg->type], reg_type_str[expected_type]);
 868                 return -EACCES;
 869         }
 870
 871         if (arg_type == ARG_CONST_MAP_PTR) {
 872                 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
 873                 *mapp = reg->map_ptr;
 874
 875         } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
 876                 /* bpf_map_xxx(..., map_ptr, ..., key) call:
 877                  * check that [key, key + map->key_size) are within
 878                  * stack limits and initialized
 879                  */
 880                 if (!*mapp) {
 881                         /* in function declaration map_ptr must come before
 882                          * map_key, so that it's verified and known before
 883                          * we have to check map_key here. Otherwise it means
 884                          * that kernel subsystem misconfigured verifier
 885                          */
 886                         verbose("invalid map_ptr to access map->key\n");
 887                         return -EACCES;
 888                 }
 889                 err = check_stack_boundary(env, regno, (*mapp)->key_size,
 890                                            false);
 891         } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
 892                 /* bpf_map_xxx(..., map_ptr, ..., value) call:
 893                  * check [value, value + map->value_size) validity
 894                  */
 895                 if (!*mapp) {
 896                         /* kernel subsystem misconfigured verifier */
 897                         verbose("invalid map_ptr to access map->value\n");
 898                         return -EACCES;
 899                 }
 900                 err = check_stack_boundary(env, regno, (*mapp)->value_size,
 901                                            false);
 902         } else if (arg_type == ARG_CONST_STACK_SIZE ||
 903                    arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) {
 904                 bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO);
 905
 906                 /* bpf_xxx(..., buf, len) call will access 'len' bytes
 907                  * from stack pointer 'buf'. Check it
 908                  * note: regno == len, regno - 1 == buf
 909                  */
 910                 if (regno == 0) {
 911                         /* kernel subsystem misconfigured verifier */
 912                         verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
 913                         return -EACCES;
 914                 }
 915                 err = check_stack_boundary(env, regno - 1, reg->imm,
 916                                            zero_size_allowed);
 917         }
 918
 919         return err;
 920 }
 921
 922 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
 923 {
 924         bool bool_map, bool_func;
 925         int i;
 926
 927         if (!map)
 928                 return 0;
 929
 930         for (i = 0; i < ARRAY_SIZE(func_limit); i++) {
 931                 bool_map = (map->map_type == func_limit[i].map_type);
 932                 bool_func = (func_id == func_limit[i].func_id);
 933                 /* only when map & func pair match it can continue.
 934                  * don't allow any other map type to be passed into
 935                  * the special func;
 936                  */
 937                 if (bool_func && bool_map != bool_func) {
 938                         verbose("cannot pass map_type %d into func %d\n",
 939                                 map->map_type, func_id);
 940                         return -EINVAL;
 941                 }
 942         }
 943
 944         return 0;
 945 }
 946
 947 static int check_call(struct verifier_env *env, int func_id)
 948 {
 949         struct verifier_state *state = &env->cur_state;
 950         const struct bpf_func_proto *fn = NULL;
 951         struct reg_state *regs = state->regs;
 952         struct bpf_map *map = NULL;
 953         struct reg_state *reg;
 954         int i, err;
 955
 956         /* find function prototype */
 957         if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
 958                 verbose("invalid func %d\n", func_id);
 959                 return -EINVAL;
 960         }
 961
 962         if (env->prog->aux->ops->get_func_proto)
 963                 fn = env->prog->aux->ops->get_func_proto(func_id);
 964
 965         if (!fn) {
 966                 verbose("unknown func %d\n", func_id);
 967                 return -EINVAL;
 968         }
 969
 970         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 971         if (!env->prog->gpl_compatible && fn->gpl_only) {
 972                 verbose("cannot call GPL only function from proprietary program\n");
 973                 return -EINVAL;
 974         }
 975
 976         /* check args */
 977         err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &map);
 978         if (err)
 979                 return err;
 980         err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &map);
 981         if (err)
 982                 return err;
 983         err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &map);
 984         if (err)
 985                 return err;
 986         err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &map);
 987         if (err)
 988                 return err;
 989         err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &map);
 990         if (err)
 991                 return err;
 992
 993         /* reset caller saved regs */
 994         for (i = 0; i < CALLER_SAVED_REGS; i++) {
 995                 reg = regs + caller_saved[i];
 996                 reg->type = NOT_INIT;
 997                 reg->imm = 0;
 998         }
 999
1000         /* update return register */
1001         if (fn->ret_type == RET_INTEGER) {
1002                 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1003         } else if (fn->ret_type == RET_VOID) {
1004                 regs[BPF_REG_0].type = NOT_INIT;
1005         } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1006                 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1007                 /* remember map_ptr, so that check_map_access()
1008                  * can check 'value_size' boundary of memory access
1009                  * to map element returned from bpf_map_lookup_elem()
1010                  */
1011                 if (map == NULL) {
1012                         verbose("kernel subsystem misconfigured verifier\n");
1013                         return -EINVAL;
1014                 }
1015                 regs[BPF_REG_0].map_ptr = map;
1016         } else {
1017                 verbose("unknown return type %d of func %d\n",
1018                         fn->ret_type, func_id);
1019                 return -EINVAL;
1020         }
1021
1022         err = check_map_func_compatibility(map, func_id);
1023         if (err)
1024                 return err;
1025
1026         return 0;
1027 }
1028
1029 /* check validity of 32-bit and 64-bit arithmetic operations */
1030 static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
1031 {
1032         struct reg_state *regs = env->cur_state.regs;
1033         u8 opcode = BPF_OP(insn->code);
1034         int err;
1035
1036         if (opcode == BPF_END || opcode == BPF_NEG) {
1037                 if (opcode == BPF_NEG) {
1038                         if (BPF_SRC(insn->code) != 0 ||
1039                             insn->src_reg != BPF_REG_0 ||
1040                             insn->off != 0 || insn->imm != 0) {
1041                                 verbose("BPF_NEG uses reserved fields\n");
1042                                 return -EINVAL;
1043                         }
1044                 } else {
1045                         if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1046                             (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1047                                 verbose("BPF_END uses reserved fields\n");
1048                                 return -EINVAL;
1049                         }
1050                 }
1051
1052                 /* check src operand */
1053                 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1054                 if (err)
1055                         return err;
1056
1057                 if (is_pointer_value(env, insn->dst_reg)) {
1058                         verbose("R%d pointer arithmetic prohibited\n",
1059                                 insn->dst_reg);
1060                         return -EACCES;
1061                 }
1062
1063                 /* check dest operand */
1064                 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1065                 if (err)
1066                         return err;
1067
1068         } else if (opcode == BPF_MOV) {
1069
1070                 if (BPF_SRC(insn->code) == BPF_X) {
1071                         if (insn->imm != 0 || insn->off != 0) {
1072                                 verbose("BPF_MOV uses reserved fields\n");
1073                                 return -EINVAL;
1074                         }
1075
1076                         /* check src operand */
1077                         err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1078                         if (err)
1079                                 return err;
1080                 } else {
1081                         if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1082                                 verbose("BPF_MOV uses reserved fields\n");
1083                                 return -EINVAL;
1084                         }
1085                 }
1086
1087                 /* check dest operand */
1088                 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1089                 if (err)
1090                         return err;
1091
1092                 if (BPF_SRC(insn->code) == BPF_X) {
1093                         if (BPF_CLASS(insn->code) == BPF_ALU64) {
1094                                 /* case: R1 = R2
1095                                  * copy register state to dest reg
1096                                  */
1097                                 regs[insn->dst_reg] = regs[insn->src_reg];
1098                         } else {
1099                                 if (is_pointer_value(env, insn->src_reg)) {
1100                                         verbose("R%d partial copy of pointer\n",
1101                                                 insn->src_reg);
1102                                         return -EACCES;
1103                                 }
1104                                 regs[insn->dst_reg].type = UNKNOWN_VALUE;
1105                                 regs[insn->dst_reg].map_ptr = NULL;
1106                         }
1107                 } else {
1108                         /* case: R = imm
1109                          * remember the value we stored into this reg
1110                          */
1111                         regs[insn->dst_reg].type = CONST_IMM;
1112                         regs[insn->dst_reg].imm = insn->imm;
1113                 }
1114
1115         } else if (opcode > BPF_END) {
1116                 verbose("invalid BPF_ALU opcode %x\n", opcode);
1117                 return -EINVAL;
1118
1119         } else {        /* all other ALU ops: and, sub, xor, add, ... */
1120
1121                 bool stack_relative = false;
1122
1123                 if (BPF_SRC(insn->code) == BPF_X) {
1124                         if (insn->imm != 0 || insn->off != 0) {
1125                                 verbose("BPF_ALU uses reserved fields\n");
1126                                 return -EINVAL;
1127                         }
1128                         /* check src1 operand */
1129                         err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1130                         if (err)
1131                                 return err;
1132                 } else {
1133                         if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1134                                 verbose("BPF_ALU uses reserved fields\n");
1135                                 return -EINVAL;
1136                         }
1137                 }
1138
1139                 /* check src2 operand */
1140                 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1141                 if (err)
1142                         return err;
1143
1144                 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1145                     BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1146                         verbose("div by zero\n");
1147                         return -EINVAL;
1148                 }
1149
1150                 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1151                      opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1152                         int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1153
1154                         if (insn->imm < 0 || insn->imm >= size) {
1155                                 verbose("invalid shift %d\n", insn->imm);
1156                                 return -EINVAL;
1157                         }
1158                 }
1159
1160                 /* pattern match 'bpf_add Rx, imm' instruction */
1161                 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1162                     regs[insn->dst_reg].type == FRAME_PTR &&
1163                     BPF_SRC(insn->code) == BPF_K) {
1164                         stack_relative = true;
1165                 } else if (is_pointer_value(env, insn->dst_reg)) {
1166                         verbose("R%d pointer arithmetic prohibited\n",
1167                                 insn->dst_reg);
1168                         return -EACCES;
1169                 } else if (BPF_SRC(insn->code) == BPF_X &&
1170                            is_pointer_value(env, insn->src_reg)) {
1171                         verbose("R%d pointer arithmetic prohibited\n",
1172                                 insn->src_reg);
1173                         return -EACCES;
1174                 }
1175
1176                 /* check dest operand */
1177                 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1178                 if (err)
1179                         return err;
1180
1181                 if (stack_relative) {
1182                         regs[insn->dst_reg].type = PTR_TO_STACK;
1183                         regs[insn->dst_reg].imm = insn->imm;
1184                 }
1185         }
1186
1187         return 0;
1188 }
1189
1190 static int check_cond_jmp_op(struct verifier_env *env,
1191                              struct bpf_insn *insn, int *insn_idx)
1192 {
1193         struct reg_state *regs = env->cur_state.regs;
1194         struct verifier_state *other_branch;
1195         u8 opcode = BPF_OP(insn->code);
1196         int err;
1197
1198         if (opcode > BPF_EXIT) {
1199                 verbose("invalid BPF_JMP opcode %x\n", opcode);
1200                 return -EINVAL;
1201         }
1202
1203         if (BPF_SRC(insn->code) == BPF_X) {
1204                 if (insn->imm != 0) {
1205                         verbose("BPF_JMP uses reserved fields\n");
1206                         return -EINVAL;
1207                 }
1208
1209                 /* check src1 operand */
1210                 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1211                 if (err)
1212                         return err;
1213
1214                 if (is_pointer_value(env, insn->src_reg)) {
1215                         verbose("R%d pointer comparison prohibited\n",
1216                                 insn->src_reg);
1217                         return -EACCES;
1218                 }
1219         } else {
1220                 if (insn->src_reg != BPF_REG_0) {
1221                         verbose("BPF_JMP uses reserved fields\n");
1222                         return -EINVAL;
1223                 }
1224         }
1225
1226         /* check src2 operand */
1227         err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1228         if (err)
1229                 return err;
1230
1231         /* detect if R == 0 where R was initialized to zero earlier */
1232         if (BPF_SRC(insn->code) == BPF_K &&
1233             (opcode == BPF_JEQ || opcode == BPF_JNE) &&
1234             regs[insn->dst_reg].type == CONST_IMM &&
1235             regs[insn->dst_reg].imm == insn->imm) {
1236                 if (opcode == BPF_JEQ) {
1237                         /* if (imm == imm) goto pc+off;
1238                          * only follow the goto, ignore fall-through
1239                          */
1240                         *insn_idx += insn->off;
1241                         return 0;
1242                 } else {
1243                         /* if (imm != imm) goto pc+off;
1244                          * only follow fall-through branch, since
1245                          * that's where the program will go
1246                          */
1247                         return 0;
1248                 }
1249         }
1250
1251         other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
1252         if (!other_branch)
1253                 return -EFAULT;
1254
1255         /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1256         if (BPF_SRC(insn->code) == BPF_K &&
1257             insn->imm == 0 && (opcode == BPF_JEQ ||
1258                                opcode == BPF_JNE) &&
1259             regs[insn->dst_reg].type == PTR_TO_MAP_VALUE_OR_NULL) {
1260                 if (opcode == BPF_JEQ) {
1261                         /* next fallthrough insn can access memory via
1262                          * this register
1263                          */
1264                         regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1265                         /* branch targer cannot access it, since reg == 0 */
1266                         other_branch->regs[insn->dst_reg].type = CONST_IMM;
1267                         other_branch->regs[insn->dst_reg].imm = 0;
1268                 } else {
1269                         other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE;
1270                         regs[insn->dst_reg].type = CONST_IMM;
1271                         regs[insn->dst_reg].imm = 0;
1272                 }
1273         } else if (is_pointer_value(env, insn->dst_reg)) {
1274                 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
1275                 return -EACCES;
1276         } else if (BPF_SRC(insn->code) == BPF_K &&
1277                    (opcode == BPF_JEQ || opcode == BPF_JNE)) {
1278
1279                 if (opcode == BPF_JEQ) {
1280                         /* detect if (R == imm) goto
1281                          * and in the target state recognize that R = imm
1282                          */
1283                         other_branch->regs[insn->dst_reg].type = CONST_IMM;
1284                         other_branch->regs[insn->dst_reg].imm = insn->imm;
1285                 } else {
1286                         /* detect if (R != imm) goto
1287                          * and in the fall-through state recognize that R = imm
1288                          */
1289                         regs[insn->dst_reg].type = CONST_IMM;
1290                         regs[insn->dst_reg].imm = insn->imm;
1291                 }
1292         }
1293         if (log_level)
1294                 print_verifier_state(env);
1295         return 0;
1296 }
1297
1298 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1299 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
1300 {
1301         u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
1302
1303         return (struct bpf_map *) (unsigned long) imm64;
1304 }
1305
1306 /* verify BPF_LD_IMM64 instruction */
1307 static int check_ld_imm(struct verifier_env *env, struct bpf_insn *insn)
1308 {
1309         struct reg_state *regs = env->cur_state.regs;
1310         int err;
1311
1312         if (BPF_SIZE(insn->code) != BPF_DW) {
1313                 verbose("invalid BPF_LD_IMM insn\n");
1314                 return -EINVAL;
1315         }
1316         if (insn->off != 0) {
1317                 verbose("BPF_LD_IMM64 uses reserved fields\n");
1318                 return -EINVAL;
1319         }
1320
1321         err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1322         if (err)
1323                 return err;
1324
1325         if (insn->src_reg == 0)
1326                 /* generic move 64-bit immediate into a register */
1327                 return 0;
1328
1329         /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1330         BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
1331
1332         regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
1333         regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
1334         return 0;
1335 }
1336
1337 static bool may_access_skb(enum bpf_prog_type type)
1338 {
1339         switch (type) {
1340         case BPF_PROG_TYPE_SOCKET_FILTER:
1341         case BPF_PROG_TYPE_SCHED_CLS:
1342         case BPF_PROG_TYPE_SCHED_ACT:
1343                 return true;
1344         default:
1345                 return false;
1346         }
1347 }
1348
1349 /* verify safety of LD_ABS|LD_IND instructions:
1350  * - they can only appear in the programs where ctx == skb
1351  * - since they are wrappers of function calls, they scratch R1-R5 registers,
1352  *   preserve R6-R9, and store return value into R0
1353  *
1354  * Implicit input:
1355  *   ctx == skb == R6 == CTX
1356  *
1357  * Explicit input:
1358  *   SRC == any register
1359  *   IMM == 32-bit immediate
1360  *
1361  * Output:
1362  *   R0 - 8/16/32-bit skb data converted to cpu endianness
1363  */
1364 static int check_ld_abs(struct verifier_env *env, struct bpf_insn *insn)
1365 {
1366         struct reg_state *regs = env->cur_state.regs;
1367         u8 mode = BPF_MODE(insn->code);
1368         struct reg_state *reg;
1369         int i, err;
1370
1371         if (!may_access_skb(env->prog->type)) {
1372                 verbose("BPF_LD_ABS|IND instructions not allowed for this program type\n");
1373                 return -EINVAL;
1374         }
1375
1376         if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
1377             BPF_SIZE(insn->code) == BPF_DW ||
1378             (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
1379                 verbose("BPF_LD_ABS uses reserved fields\n");
1380                 return -EINVAL;
1381         }
1382
1383         /* check whether implicit source operand (register R6) is readable */
1384         err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
1385         if (err)
1386                 return err;
1387
1388         if (regs[BPF_REG_6].type != PTR_TO_CTX) {
1389                 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1390                 return -EINVAL;
1391         }
1392
1393         if (mode == BPF_IND) {
1394                 /* check explicit source operand */
1395                 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1396                 if (err)
1397                         return err;
1398         }
1399
1400         /* reset caller saved regs to unreadable */
1401         for (i = 0; i < CALLER_SAVED_REGS; i++) {
1402                 reg = regs + caller_saved[i];
1403                 reg->type = NOT_INIT;
1404                 reg->imm = 0;
1405         }
1406
1407         /* mark destination R0 register as readable, since it contains
1408          * the value fetched from the packet
1409          */
1410         regs[BPF_REG_0].type = UNKNOWN_VALUE;
1411         return 0;
1412 }
1413
1414 /* non-recursive DFS pseudo code
1415  * 1  procedure DFS-iterative(G,v):
1416  * 2      label v as discovered
1417  * 3      let S be a stack
1418  * 4      S.push(v)
1419  * 5      while S is not empty
1420  * 6            t <- S.pop()
1421  * 7            if t is what we're looking for:
1422  * 8                return t
1423  * 9            for all edges e in G.adjacentEdges(t) do
1424  * 10               if edge e is already labelled
1425  * 11                   continue with the next edge
1426  * 12               w <- G.adjacentVertex(t,e)
1427  * 13               if vertex w is not discovered and not explored
1428  * 14                   label e as tree-edge
1429  * 15                   label w as discovered
1430  * 16                   S.push(w)
1431  * 17                   continue at 5
1432  * 18               else if vertex w is discovered
1433  * 19                   label e as back-edge
1434  * 20               else
1435  * 21                   // vertex w is explored
1436  * 22                   label e as forward- or cross-edge
1437  * 23           label t as explored
1438  * 24           S.pop()
1439  *
1440  * convention:
1441  * 0x10 - discovered
1442  * 0x11 - discovered and fall-through edge labelled
1443  * 0x12 - discovered and fall-through and branch edges labelled
1444  * 0x20 - explored
1445  */
1446
1447 enum {
1448         DISCOVERED = 0x10,
1449         EXPLORED = 0x20,
1450         FALLTHROUGH = 1,
1451         BRANCH = 2,
1452 };
1453
1454 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1455
1456 static int *insn_stack; /* stack of insns to process */
1457 static int cur_stack;   /* current stack index */
1458 static int *insn_state;
1459
1460 /* t, w, e - match pseudo-code above:
1461  * t - index of current instruction
1462  * w - next instruction
1463  * e - edge
1464  */
1465 static int push_insn(int t, int w, int e, struct verifier_env *env)
1466 {
1467         if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
1468                 return 0;
1469
1470         if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
1471                 return 0;
1472
1473         if (w < 0 || w >= env->prog->len) {
1474                 verbose("jump out of range from insn %d to %d\n", t, w);
1475                 return -EINVAL;
1476         }
1477
1478         if (e == BRANCH)
1479                 /* mark branch target for state pruning */
1480                 env->explored_states[w] = STATE_LIST_MARK;
1481
1482         if (insn_state[w] == 0) {
1483                 /* tree-edge */
1484                 insn_state[t] = DISCOVERED | e;
1485                 insn_state[w] = DISCOVERED;
1486                 if (cur_stack >= env->prog->len)
1487                         return -E2BIG;
1488                 insn_stack[cur_stack++] = w;
1489                 return 1;
1490         } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
1491                 verbose("back-edge from insn %d to %d\n", t, w);
1492                 return -EINVAL;
1493         } else if (insn_state[w] == EXPLORED) {
1494                 /* forward- or cross-edge */
1495                 insn_state[t] = DISCOVERED | e;
1496         } else {
1497                 verbose("insn state internal bug\n");
1498                 return -EFAULT;
1499         }
1500         return 0;
1501 }
1502
1503 /* non-recursive depth-first-search to detect loops in BPF program
1504  * loop == back-edge in directed graph
1505  */
1506 static int check_cfg(struct verifier_env *env)
1507 {
1508         struct bpf_insn *insns = env->prog->insnsi;
1509         int insn_cnt = env->prog->len;
1510         int ret = 0;
1511         int i, t;
1512
1513         insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1514         if (!insn_state)
1515                 return -ENOMEM;
1516
1517         insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
1518         if (!insn_stack) {
1519                 kfree(insn_state);
1520                 return -ENOMEM;
1521         }
1522
1523         insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
1524         insn_stack[0] = 0; /* 0 is the first instruction */
1525         cur_stack = 1;
1526
1527 peek_stack:
1528         if (cur_stack == 0)
1529                 goto check_state;
1530         t = insn_stack[cur_stack - 1];
1531
1532         if (BPF_CLASS(insns[t].code) == BPF_JMP) {
1533                 u8 opcode = BPF_OP(insns[t].code);
1534
1535                 if (opcode == BPF_EXIT) {
1536                         goto mark_explored;
1537                 } else if (opcode == BPF_CALL) {
1538                         ret = push_insn(t, t + 1, FALLTHROUGH, env);
1539                         if (ret == 1)
1540                                 goto peek_stack;
1541                         else if (ret < 0)
1542                                 goto err_free;
1543                 } else if (opcode == BPF_JA) {
1544                         if (BPF_SRC(insns[t].code) != BPF_K) {
1545                                 ret = -EINVAL;
1546                                 goto err_free;
1547                         }
1548                         /* unconditional jump with single edge */
1549                         ret = push_insn(t, t + insns[t].off + 1,
1550                                         FALLTHROUGH, env);
1551                         if (ret == 1)
1552                                 goto peek_stack;
1553                         else if (ret < 0)
1554                                 goto err_free;
1555                         /* tell verifier to check for equivalent states
1556                          * after every call and jump
1557                          */
1558                         if (t + 1 < insn_cnt)
1559                                 env->explored_states[t + 1] = STATE_LIST_MARK;
1560                 } else {
1561                         /* conditional jump with two edges */
1562                         ret = push_insn(t, t + 1, FALLTHROUGH, env);
1563                         if (ret == 1)
1564                                 goto peek_stack;
1565                         else if (ret < 0)
1566                                 goto err_free;
1567
1568                         ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
1569                         if (ret == 1)
1570                                 goto peek_stack;
1571                         else if (ret < 0)
1572                                 goto err_free;
1573                 }
1574         } else {
1575                 /* all other non-branch instructions with single
1576                  * fall-through edge
1577                  */
1578                 ret = push_insn(t, t + 1, FALLTHROUGH, env);
1579                 if (ret == 1)
1580                         goto peek_stack;
1581                 else if (ret < 0)
1582                         goto err_free;
1583         }
1584
1585 mark_explored:
1586         insn_state[t] = EXPLORED;
1587         if (cur_stack-- <= 0) {
1588                 verbose("pop stack internal bug\n");
1589                 ret = -EFAULT;
1590                 goto err_free;
1591         }
1592         goto peek_stack;
1593
1594 check_state:
1595         for (i = 0; i < insn_cnt; i++) {
1596                 if (insn_state[i] != EXPLORED) {
1597                         verbose("unreachable insn %d\n", i);
1598                         ret = -EINVAL;
1599                         goto err_free;
1600                 }
1601         }
1602         ret = 0; /* cfg looks good */
1603
1604 err_free:
1605         kfree(insn_state);
1606         kfree(insn_stack);
1607         return ret;
1608 }
1609
1610 /* compare two verifier states
1611  *
1612  * all states stored in state_list are known to be valid, since
1613  * verifier reached 'bpf_exit' instruction through them
1614  *
1615  * this function is called when verifier exploring different branches of
1616  * execution popped from the state stack. If it sees an old state that has
1617  * more strict register state and more strict stack state then this execution
1618  * branch doesn't need to be explored further, since verifier already
1619  * concluded that more strict state leads to valid finish.
1620  *
1621  * Therefore two states are equivalent if register state is more conservative
1622  * and explored stack state is more conservative than the current one.
1623  * Example:
1624  *       explored                   current
1625  * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
1626  * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
1627  *
1628  * In other words if current stack state (one being explored) has more
1629  * valid slots than old one that already passed validation, it means
1630  * the verifier can stop exploring and conclude that current state is valid too
1631  *
1632  * Similarly with registers. If explored state has register type as invalid
1633  * whereas register type in current state is meaningful, it means that
1634  * the current state will reach 'bpf_exit' instruction safely
1635  */
1636 static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
1637 {
1638         int i;
1639
1640         for (i = 0; i < MAX_BPF_REG; i++) {
1641                 if (memcmp(&old->regs[i], &cur->regs[i],
1642                            sizeof(old->regs[0])) != 0) {
1643                         if (old->regs[i].type == NOT_INIT ||
1644                             (old->regs[i].type == UNKNOWN_VALUE &&
1645                              cur->regs[i].type != NOT_INIT))
1646                                 continue;
1647                         return false;
1648                 }
1649         }
1650
1651         for (i = 0; i < MAX_BPF_STACK; i++) {
1652                 if (old->stack_slot_type[i] == STACK_INVALID)
1653                         continue;
1654                 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
1655                         /* Ex: old explored (safe) state has STACK_SPILL in
1656                          * this stack slot, but current has has STACK_MISC ->
1657                          * this verifier states are not equivalent,
1658                          * return false to continue verification of this path
1659                          */
1660                         return false;
1661                 if (i % BPF_REG_SIZE)
1662                         continue;
1663                 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
1664                            &cur->spilled_regs[i / BPF_REG_SIZE],
1665                            sizeof(old->spilled_regs[0])))
1666                         /* when explored and current stack slot types are
1667                          * the same, check that stored pointers types
1668                          * are the same as well.
1669                          * Ex: explored safe path could have stored
1670                          * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
1671                          * but current path has stored:
1672                          * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
1673                          * such verifier states are not equivalent.
1674                          * return false to continue verification of this path
1675                          */
1676                         return false;
1677                 else
1678                         continue;
1679         }
1680         return true;
1681 }
1682
1683 static int is_state_visited(struct verifier_env *env, int insn_idx)
1684 {
1685         struct verifier_state_list *new_sl;
1686         struct verifier_state_list *sl;
1687
1688         sl = env->explored_states[insn_idx];
1689         if (!sl)
1690                 /* this 'insn_idx' instruction wasn't marked, so we will not
1691                  * be doing state search here
1692                  */
1693                 return 0;
1694
1695         while (sl != STATE_LIST_MARK) {
1696                 if (states_equal(&sl->state, &env->cur_state))
1697                         /* reached equivalent register/stack state,
1698                          * prune the search
1699                          */
1700                         return 1;
1701                 sl = sl->next;
1702         }
1703
1704         /* there were no equivalent states, remember current one.
1705          * technically the current state is not proven to be safe yet,
1706          * but it will either reach bpf_exit (which means it's safe) or
1707          * it will be rejected. Since there are no loops, we won't be
1708          * seeing this 'insn_idx' instruction again on the way to bpf_exit
1709          */
1710         new_sl = kmalloc(sizeof(struct verifier_state_list), GFP_USER);
1711         if (!new_sl)
1712                 return -ENOMEM;
1713
1714         /* add new state to the head of linked list */
1715         memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
1716         new_sl->next = env->explored_states[insn_idx];
1717         env->explored_states[insn_idx] = new_sl;
1718         return 0;
1719 }
1720
1721 static int do_check(struct verifier_env *env)
1722 {
1723         struct verifier_state *state = &env->cur_state;
1724         struct bpf_insn *insns = env->prog->insnsi;
1725         struct reg_state *regs = state->regs;
1726         int insn_cnt = env->prog->len;
1727         int insn_idx, prev_insn_idx = 0;
1728         int insn_processed = 0;
1729         bool do_print_state = false;
1730
1731         init_reg_state(regs);
1732         insn_idx = 0;
1733         for (;;) {
1734                 struct bpf_insn *insn;
1735                 u8 class;
1736                 int err;
1737
1738                 if (insn_idx >= insn_cnt) {
1739                         verbose("invalid insn idx %d insn_cnt %d\n",
1740                                 insn_idx, insn_cnt);
1741                         return -EFAULT;
1742                 }
1743
1744                 insn = &insns[insn_idx];
1745                 class = BPF_CLASS(insn->code);
1746
1747                 if (++insn_processed > 32768) {
1748                         verbose("BPF program is too large. Proccessed %d insn\n",
1749                                 insn_processed);
1750                         return -E2BIG;
1751                 }
1752
1753                 err = is_state_visited(env, insn_idx);
1754                 if (err < 0)
1755                         return err;
1756                 if (err == 1) {
1757                         /* found equivalent state, can prune the search */
1758                         if (log_level) {
1759                                 if (do_print_state)
1760                                         verbose("\nfrom %d to %d: safe\n",
1761                                                 prev_insn_idx, insn_idx);
1762                                 else
1763                                         verbose("%d: safe\n", insn_idx);
1764                         }
1765                         goto process_bpf_exit;
1766                 }
1767
1768                 if (log_level && do_print_state) {
1769                         verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
1770                         print_verifier_state(env);
1771                         do_print_state = false;
1772                 }
1773
1774                 if (log_level) {
1775                         verbose("%d: ", insn_idx);
1776                         print_bpf_insn(insn);
1777                 }
1778
1779                 if (class == BPF_ALU || class == BPF_ALU64) {
1780                         err = check_alu_op(env, insn);
1781                         if (err)
1782                                 return err;
1783
1784                 } else if (class == BPF_LDX) {
1785                         enum bpf_reg_type src_reg_type;
1786
1787                         /* check for reserved fields is already done */
1788
1789                         /* check src operand */
1790                         err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1791                         if (err)
1792                                 return err;
1793
1794                         err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1795                         if (err)
1796                                 return err;
1797
1798                         src_reg_type = regs[insn->src_reg].type;
1799
1800                         /* check that memory (src_reg + off) is readable,
1801                          * the state of dst_reg will be updated by this func
1802                          */
1803                         err = check_mem_access(env, insn->src_reg, insn->off,
1804                                                BPF_SIZE(insn->code), BPF_READ,
1805                                                insn->dst_reg);
1806                         if (err)
1807                                 return err;
1808
1809                         if (BPF_SIZE(insn->code) != BPF_W) {
1810                                 insn_idx++;
1811                                 continue;
1812                         }
1813
1814                         if (insn->imm == 0) {
1815                                 /* saw a valid insn
1816                                  * dst_reg = *(u32 *)(src_reg + off)
1817                                  * use reserved 'imm' field to mark this insn
1818                                  */
1819                                 insn->imm = src_reg_type;
1820
1821                         } else if (src_reg_type != insn->imm &&
1822                                    (src_reg_type == PTR_TO_CTX ||
1823                                     insn->imm == PTR_TO_CTX)) {
1824                                 /* ABuser program is trying to use the same insn
1825                                  * dst_reg = *(u32*) (src_reg + off)
1826                                  * with different pointer types:
1827                                  * src_reg == ctx in one branch and
1828                                  * src_reg == stack|map in some other branch.
1829                                  * Reject it.
1830                                  */
1831                                 verbose("same insn cannot be used with different pointers\n");
1832                                 return -EINVAL;
1833                         }
1834
1835                 } else if (class == BPF_STX) {
1836                         enum bpf_reg_type dst_reg_type;
1837
1838                         if (BPF_MODE(insn->code) == BPF_XADD) {
1839                                 err = check_xadd(env, insn);
1840                                 if (err)
1841                                         return err;
1842                                 insn_idx++;
1843                                 continue;
1844                         }
1845
1846                         /* check src1 operand */
1847                         err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1848                         if (err)
1849                                 return err;
1850                         /* check src2 operand */
1851                         err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1852                         if (err)
1853                                 return err;
1854
1855                         dst_reg_type = regs[insn->dst_reg].type;
1856
1857                         /* check that memory (dst_reg + off) is writeable */
1858                         err = check_mem_access(env, insn->dst_reg, insn->off,
1859                                                BPF_SIZE(insn->code), BPF_WRITE,
1860                                                insn->src_reg);
1861                         if (err)
1862                                 return err;
1863
1864                         if (insn->imm == 0) {
1865                                 insn->imm = dst_reg_type;
1866                         } else if (dst_reg_type != insn->imm &&
1867                                    (dst_reg_type == PTR_TO_CTX ||
1868                                     insn->imm == PTR_TO_CTX)) {
1869                                 verbose("same insn cannot be used with different pointers\n");
1870                                 return -EINVAL;
1871                         }
1872
1873                 } else if (class == BPF_ST) {
1874                         if (BPF_MODE(insn->code) != BPF_MEM ||
1875                             insn->src_reg != BPF_REG_0) {
1876                                 verbose("BPF_ST uses reserved fields\n");
1877                                 return -EINVAL;
1878                         }
1879                         /* check src operand */
1880                         err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1881                         if (err)
1882                                 return err;
1883
1884                         /* check that memory (dst_reg + off) is writeable */
1885                         err = check_mem_access(env, insn->dst_reg, insn->off,
1886                                                BPF_SIZE(insn->code), BPF_WRITE,
1887                                                -1);
1888                         if (err)
1889                                 return err;
1890
1891                 } else if (class == BPF_JMP) {
1892                         u8 opcode = BPF_OP(insn->code);
1893
1894                         if (opcode == BPF_CALL) {
1895                                 if (BPF_SRC(insn->code) != BPF_K ||
1896                                     insn->off != 0 ||
1897                                     insn->src_reg != BPF_REG_0 ||
1898                                     insn->dst_reg != BPF_REG_0) {
1899                                         verbose("BPF_CALL uses reserved fields\n");
1900                                         return -EINVAL;
1901                                 }
1902
1903                                 err = check_call(env, insn->imm);
1904                                 if (err)
1905                                         return err;
1906
1907                         } else if (opcode == BPF_JA) {
1908                                 if (BPF_SRC(insn->code) != BPF_K ||
1909                                     insn->imm != 0 ||
1910                                     insn->src_reg != BPF_REG_0 ||
1911                                     insn->dst_reg != BPF_REG_0) {
1912                                         verbose("BPF_JA uses reserved fields\n");
1913                                         return -EINVAL;
1914                                 }
1915
1916                                 insn_idx += insn->off + 1;
1917                                 continue;
1918
1919                         } else if (opcode == BPF_EXIT) {
1920                                 if (BPF_SRC(insn->code) != BPF_K ||
1921                                     insn->imm != 0 ||
1922                                     insn->src_reg != BPF_REG_0 ||
1923                                     insn->dst_reg != BPF_REG_0) {
1924                                         verbose("BPF_EXIT uses reserved fields\n");
1925                                         return -EINVAL;
1926                                 }
1927
1928                                 /* eBPF calling convetion is such that R0 is used
1929                                  * to return the value from eBPF program.
1930                                  * Make sure that it's readable at this time
1931                                  * of bpf_exit, which means that program wrote
1932                                  * something into it earlier
1933                                  */
1934                                 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
1935                                 if (err)
1936                                         return err;
1937
1938                                 if (is_pointer_value(env, BPF_REG_0)) {
1939                                         verbose("R0 leaks addr as return value\n");
1940                                         return -EACCES;
1941                                 }
1942
1943 process_bpf_exit:
1944                                 insn_idx = pop_stack(env, &prev_insn_idx);
1945                                 if (insn_idx < 0) {
1946                                         break;
1947                                 } else {
1948                                         do_print_state = true;
1949                                         continue;
1950                                 }
1951                         } else {
1952                                 err = check_cond_jmp_op(env, insn, &insn_idx);
1953                                 if (err)
1954                                         return err;
1955                         }
1956                 } else if (class == BPF_LD) {
1957                         u8 mode = BPF_MODE(insn->code);
1958
1959                         if (mode == BPF_ABS || mode == BPF_IND) {
1960                                 err = check_ld_abs(env, insn);
1961                                 if (err)
1962                                         return err;
1963
1964                         } else if (mode == BPF_IMM) {
1965                                 err = check_ld_imm(env, insn);
1966                                 if (err)
1967                                         return err;
1968
1969                                 insn_idx++;
1970                         } else {
1971                                 verbose("invalid BPF_LD mode\n");
1972                                 return -EINVAL;
1973                         }
1974                 } else {
1975                         verbose("unknown insn class %d\n", class);
1976                         return -EINVAL;
1977                 }
1978
1979                 insn_idx++;
1980         }
1981
1982         return 0;
1983 }
1984
1985 /* look for pseudo eBPF instructions that access map FDs and
1986  * replace them with actual map pointers
1987  */
1988 static int replace_map_fd_with_map_ptr(struct verifier_env *env)
1989 {
1990         struct bpf_insn *insn = env->prog->insnsi;
1991         int insn_cnt = env->prog->len;
1992         int i, j;
1993
1994         for (i = 0; i < insn_cnt; i++, insn++) {
1995                 if (BPF_CLASS(insn->code) == BPF_LDX &&
1996                     (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
1997                         verbose("BPF_LDX uses reserved fields\n");
1998                         return -EINVAL;
1999                 }
2000
2001                 if (BPF_CLASS(insn->code) == BPF_STX &&
2002                     ((BPF_MODE(insn->code) != BPF_MEM &&
2003                       BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
2004                         verbose("BPF_STX uses reserved fields\n");
2005                         return -EINVAL;
2006                 }
2007
2008                 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
2009                         struct bpf_map *map;
2010                         struct fd f;
2011
2012                         if (i == insn_cnt - 1 || insn[1].code != 0 ||
2013                             insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
2014                             insn[1].off != 0) {
2015                                 verbose("invalid bpf_ld_imm64 insn\n");
2016                                 return -EINVAL;
2017                         }
2018
2019                         if (insn->src_reg == 0)
2020                                 /* valid generic load 64-bit imm */
2021                                 goto next_insn;
2022
2023                         if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
2024                                 verbose("unrecognized bpf_ld_imm64 insn\n");
2025                                 return -EINVAL;
2026                         }
2027
2028                         f = fdget(insn->imm);
2029                         map = __bpf_map_get(f);
2030                         if (IS_ERR(map)) {
2031                                 verbose("fd %d is not pointing to valid bpf_map\n",
2032                                         insn->imm);
2033                                 return PTR_ERR(map);
2034                         }
2035
2036                         /* store map pointer inside BPF_LD_IMM64 instruction */
2037                         insn[0].imm = (u32) (unsigned long) map;
2038                         insn[1].imm = ((u64) (unsigned long) map) >> 32;
2039
2040                         /* check whether we recorded this map already */
2041                         for (j = 0; j < env->used_map_cnt; j++)
2042                                 if (env->used_maps[j] == map) {
2043                                         fdput(f);
2044                                         goto next_insn;
2045                                 }
2046
2047                         if (env->used_map_cnt >= MAX_USED_MAPS) {
2048                                 fdput(f);
2049                                 return -E2BIG;
2050                         }
2051
2052                         /* remember this map */
2053                         env->used_maps[env->used_map_cnt++] = map;
2054
2055                         /* hold the map. If the program is rejected by verifier,
2056                          * the map will be released by release_maps() or it
2057                          * will be used by the valid program until it's unloaded
2058                          * and all maps are released in free_bpf_prog_info()
2059                          */
2060                         bpf_map_inc(map, false);
2061                         fdput(f);
2062 next_insn:
2063                         insn++;
2064                         i++;
2065                 }
2066         }
2067
2068         /* now all pseudo BPF_LD_IMM64 instructions load valid
2069          * 'struct bpf_map *' into a register instead of user map_fd.
2070          * These pointers will be used later by verifier to validate map access.
2071          */
2072         return 0;
2073 }
2074
2075 /* drop refcnt of maps used by the rejected program */
2076 static void release_maps(struct verifier_env *env)
2077 {
2078         int i;
2079
2080         for (i = 0; i < env->used_map_cnt; i++)
2081                 bpf_map_put(env->used_maps[i]);
2082 }
2083
2084 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2085 static void convert_pseudo_ld_imm64(struct verifier_env *env)
2086 {
2087         struct bpf_insn *insn = env->prog->insnsi;
2088         int insn_cnt = env->prog->len;
2089         int i;
2090
2091         for (i = 0; i < insn_cnt; i++, insn++)
2092                 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
2093                         insn->src_reg = 0;
2094 }
2095
2096 static void adjust_branches(struct bpf_prog *prog, int pos, int delta)
2097 {
2098         struct bpf_insn *insn = prog->insnsi;
2099         int insn_cnt = prog->len;
2100         int i;
2101
2102         for (i = 0; i < insn_cnt; i++, insn++) {
2103                 if (BPF_CLASS(insn->code) != BPF_JMP ||
2104                     BPF_OP(insn->code) == BPF_CALL ||
2105                     BPF_OP(insn->code) == BPF_EXIT)
2106                         continue;
2107
2108                 /* adjust offset of jmps if necessary */
2109                 if (i < pos && i + insn->off + 1 > pos)
2110                         insn->off += delta;
2111                 else if (i > pos + delta && i + insn->off + 1 <= pos + delta)
2112                         insn->off -= delta;
2113         }
2114 }
2115
2116 /* convert load instructions that access fields of 'struct __sk_buff'
2117  * into sequence of instructions that access fields of 'struct sk_buff'
2118  */
2119 static int convert_ctx_accesses(struct verifier_env *env)
2120 {
2121         struct bpf_insn *insn = env->prog->insnsi;
2122         int insn_cnt = env->prog->len;
2123         struct bpf_insn insn_buf[16];
2124         struct bpf_prog *new_prog;
2125         u32 cnt;
2126         int i;
2127         enum bpf_access_type type;
2128
2129         if (!env->prog->aux->ops->convert_ctx_access)
2130                 return 0;
2131
2132         for (i = 0; i < insn_cnt; i++, insn++) {
2133                 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W))
2134                         type = BPF_READ;
2135                 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W))
2136                         type = BPF_WRITE;
2137                 else
2138                         continue;
2139
2140                 if (insn->imm != PTR_TO_CTX) {
2141                         /* clear internal mark */
2142                         insn->imm = 0;
2143                         continue;
2144                 }
2145
2146                 cnt = env->prog->aux->ops->
2147                         convert_ctx_access(type, insn->dst_reg, insn->src_reg,
2148                                            insn->off, insn_buf, env->prog);
2149                 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
2150                         verbose("bpf verifier is misconfigured\n");
2151                         return -EINVAL;
2152                 }
2153
2154                 if (cnt == 1) {
2155                         memcpy(insn, insn_buf, sizeof(*insn));
2156                         continue;
2157                 }
2158
2159                 /* several new insns need to be inserted. Make room for them */
2160                 insn_cnt += cnt - 1;
2161                 new_prog = bpf_prog_realloc(env->prog,
2162                                             bpf_prog_size(insn_cnt),
2163                                             GFP_USER);
2164                 if (!new_prog)
2165                         return -ENOMEM;
2166
2167                 new_prog->len = insn_cnt;
2168
2169                 memmove(new_prog->insnsi + i + cnt, new_prog->insns + i + 1,
2170                         sizeof(*insn) * (insn_cnt - i - cnt));
2171
2172                 /* copy substitute insns in place of load instruction */
2173                 memcpy(new_prog->insnsi + i, insn_buf, sizeof(*insn) * cnt);
2174
2175                 /* adjust branches in the whole program */
2176                 adjust_branches(new_prog, i, cnt - 1);
2177
2178                 /* keep walking new program and skip insns we just inserted */
2179                 env->prog = new_prog;
2180                 insn = new_prog->insnsi + i + cnt - 1;
2181                 i += cnt - 1;
2182         }
2183
2184         return 0;
2185 }
2186
2187 static void free_states(struct verifier_env *env)
2188 {
2189         struct verifier_state_list *sl, *sln;
2190         int i;
2191
2192         if (!env->explored_states)
2193                 return;
2194
2195         for (i = 0; i < env->prog->len; i++) {
2196                 sl = env->explored_states[i];
2197
2198                 if (sl)
2199                         while (sl != STATE_LIST_MARK) {
2200                                 sln = sl->next;
2201                                 kfree(sl);
2202                                 sl = sln;
2203                         }
2204         }
2205
2206         kfree(env->explored_states);
2207 }
2208
2209 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
2210 {
2211         char __user *log_ubuf = NULL;
2212         struct verifier_env *env;
2213         int ret = -EINVAL;
2214
2215         if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS)
2216                 return -E2BIG;
2217
2218         /* 'struct verifier_env' can be global, but since it's not small,
2219          * allocate/free it every time bpf_check() is called
2220          */
2221         env = kzalloc(sizeof(struct verifier_env), GFP_KERNEL);
2222         if (!env)
2223                 return -ENOMEM;
2224
2225         env->prog = *prog;
2226
2227         /* grab the mutex to protect few globals used by verifier */
2228         mutex_lock(&bpf_verifier_lock);
2229
2230         if (attr->log_level || attr->log_buf || attr->log_size) {
2231                 /* user requested verbose verifier output
2232                  * and supplied buffer to store the verification trace
2233                  */
2234                 log_level = attr->log_level;
2235                 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
2236                 log_size = attr->log_size;
2237                 log_len = 0;
2238
2239                 ret = -EINVAL;
2240                 /* log_* values have to be sane */
2241                 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
2242                     log_level == 0 || log_ubuf == NULL)
2243                         goto free_env;
2244
2245                 ret = -ENOMEM;
2246                 log_buf = vmalloc(log_size);
2247                 if (!log_buf)
2248                         goto free_env;
2249         } else {
2250                 log_level = 0;
2251         }
2252
2253         ret = replace_map_fd_with_map_ptr(env);
2254         if (ret < 0)
2255                 goto skip_full_check;
2256
2257         env->explored_states = kcalloc(env->prog->len,
2258                                        sizeof(struct verifier_state_list *),
2259                                        GFP_USER);
2260         ret = -ENOMEM;
2261         if (!env->explored_states)
2262                 goto skip_full_check;
2263
2264         ret = check_cfg(env);
2265         if (ret < 0)
2266                 goto skip_full_check;
2267
2268         env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
2269
2270         ret = do_check(env);
2271
2272 skip_full_check:
2273         while (pop_stack(env, NULL) >= 0);
2274         free_states(env);
2275
2276         if (ret == 0)
2277                 /* program is valid, convert *(u32*)(ctx + off) accesses */
2278                 ret = convert_ctx_accesses(env);
2279
2280         if (log_level && log_len >= log_size - 1) {
2281                 BUG_ON(log_len >= log_size);
2282                 /* verifier log exceeded user supplied buffer */
2283                 ret = -ENOSPC;
2284                 /* fall through to return what was recorded */
2285         }
2286
2287         /* copy verifier log back to user space including trailing zero */
2288         if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
2289                 ret = -EFAULT;
2290                 goto free_log_buf;
2291         }
2292
2293         if (ret == 0 && env->used_map_cnt) {
2294                 /* if program passed verifier, update used_maps in bpf_prog_info */
2295                 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
2296                                                           sizeof(env->used_maps[0]),
2297                                                           GFP_KERNEL);
2298
2299                 if (!env->prog->aux->used_maps) {
2300                         ret = -ENOMEM;
2301                         goto free_log_buf;
2302                 }
2303
2304                 memcpy(env->prog->aux->used_maps, env->used_maps,
2305                        sizeof(env->used_maps[0]) * env->used_map_cnt);
2306                 env->prog->aux->used_map_cnt = env->used_map_cnt;
2307
2308                 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2309                  * bpf_ld_imm64 instructions
2310                  */
2311                 convert_pseudo_ld_imm64(env);
2312         }
2313
2314 free_log_buf:
2315         if (log_level)
2316                 vfree(log_buf);
2317 free_env:
2318         if (!env->prog->aux->used_maps)
2319                 /* if we didn't copy map pointers into bpf_prog_info, release
2320                  * them now. Otherwise free_bpf_prog_info() will release them.
2321                  */
2322                 release_maps(env);
2323         *prog = env->prog;
2324         kfree(env);
2325         mutex_unlock(&bpf_verifier_lock);
2326         return ret;
2327 }