1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
24 /* bpf_check() is a static code analyzer that walks eBPF program
25 * instruction by instruction and updates register/stack state.
26 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
28 * The first pass is depth-first-search to check that the program is a DAG.
29 * It rejects the following programs:
30 * - larger than BPF_MAXINSNS insns
31 * - if loop is present (detected via back-edge)
32 * - unreachable insns exist (shouldn't be a forest. program = one function)
33 * - out of bounds or malformed jumps
34 * The second pass is all possible path descent from the 1st insn.
35 * Since it's analyzing all pathes through the program, the length of the
36 * analysis is limited to 64k insn, which may be hit even if total number of
37 * insn is less then 4K, but there are too many branches that change stack/regs.
38 * Number of 'branches to be analyzed' is limited to 1k
40 * On entry to each instruction, each register has a type, and the instruction
41 * changes the types of the registers depending on instruction semantics.
42 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
45 * All registers are 64-bit.
46 * R0 - return register
47 * R1-R5 argument passing registers
48 * R6-R9 callee saved registers
49 * R10 - frame pointer read-only
51 * At the start of BPF program the register R1 contains a pointer to bpf_context
52 * and has type PTR_TO_CTX.
54 * Verifier tracks arithmetic operations on pointers in case:
55 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57 * 1st insn copies R10 (which has FRAME_PTR) type into R1
58 * and 2nd arithmetic instruction is pattern matched to recognize
59 * that it wants to construct a pointer to some element within stack.
60 * So after 2nd insn, the register R1 has type PTR_TO_STACK
61 * (and -20 constant is saved for further stack bounds checking).
62 * Meaning that this reg is a pointer to stack plus known immediate constant.
64 * Most of the time the registers have UNKNOWN_VALUE type, which
65 * means the register has some value, but it's not a valid pointer.
66 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
68 * When verifier sees load or store instructions the type of base register
69 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70 * types recognized by check_mem_access() function.
72 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73 * and the range of [ptr, ptr + map's value_size) is accessible.
75 * registers used to pass values to function calls are checked against
76 * function argument constraints.
78 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79 * It means that the register type passed to this function must be
80 * PTR_TO_STACK and it will be used inside the function as
81 * 'pointer to map element key'
83 * For example the argument constraints for bpf_map_lookup_elem():
84 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85 * .arg1_type = ARG_CONST_MAP_PTR,
86 * .arg2_type = ARG_PTR_TO_MAP_KEY,
88 * ret_type says that this function returns 'pointer to map elem value or null'
89 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90 * 2nd argument should be a pointer to stack, which will be used inside
91 * the helper function as a pointer to map element key.
93 * On the kernel side the helper function looks like:
94 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
96 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97 * void *key = (void *) (unsigned long) r2;
100 * here kernel can access 'key' and 'map' pointers safely, knowing that
101 * [key, key + map->key_size) bytes are valid and were initialized on
102 * the stack of eBPF program.
105 * Corresponding eBPF program may look like:
106 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
107 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
109 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110 * here verifier looks at prototype of map_lookup_elem() and sees:
111 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
114 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116 * and were initialized prior to this call.
117 * If it's ok, then verifier allows this BPF_CALL insn and looks at
118 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120 * returns ether pointer to map value or NULL.
122 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123 * insn, the register holding that pointer in the true branch changes state to
124 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125 * branch. See check_cond_jmp_op().
127 * After the call R0 is set to return type of the function and registers R1-R5
128 * are set to NOT_INIT to indicate that they are no longer readable.
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 /* verifer state is 'st'
134 * before processing instruction 'insn_idx'
135 * and after processing instruction 'prev_insn_idx'
137 struct bpf_verifier_state st;
140 struct bpf_verifier_stack_elem *next;
143 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
148 struct bpf_call_arg_meta {
149 struct bpf_map *map_ptr;
156 /* verbose verifier prints what it's seeing
157 * bpf_check() is called under lock, so no race to access these global vars
159 static u32 log_level, log_size, log_len;
160 static char *log_buf;
162 static DEFINE_MUTEX(bpf_verifier_lock);
164 /* log_level controls verbosity level of eBPF verifier.
165 * verbose() is used to dump the verification trace to the log, so the user
166 * can figure out what's wrong with the program
168 static __printf(1, 2) void verbose(const char *fmt, ...)
172 if (log_level == 0 || log_len >= log_size - 1)
176 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
180 /* string representation of 'enum bpf_reg_type' */
181 static const char * const reg_type_str[] = {
183 [UNKNOWN_VALUE] = "inv",
184 [PTR_TO_CTX] = "ctx",
185 [CONST_PTR_TO_MAP] = "map_ptr",
186 [PTR_TO_MAP_VALUE] = "map_value",
187 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
188 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj",
190 [PTR_TO_STACK] = "fp",
192 [PTR_TO_PACKET] = "pkt",
193 [PTR_TO_PACKET_END] = "pkt_end",
196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
197 static const char * const func_id_str[] = {
198 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
200 #undef __BPF_FUNC_STR_FN
202 static const char *func_id_name(int id)
204 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
206 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
207 return func_id_str[id];
212 static void print_verifier_state(struct bpf_verifier_state *state)
214 struct bpf_reg_state *reg;
218 for (i = 0; i < MAX_BPF_REG; i++) {
219 reg = &state->regs[i];
223 verbose(" R%d=%s", i, reg_type_str[t]);
224 if (t == CONST_IMM || t == PTR_TO_STACK)
225 verbose("%lld", reg->imm);
226 else if (t == PTR_TO_PACKET)
227 verbose("(id=%d,off=%d,r=%d)",
228 reg->id, reg->off, reg->range);
229 else if (t == UNKNOWN_VALUE && reg->imm)
230 verbose("%lld", reg->imm);
231 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
232 t == PTR_TO_MAP_VALUE_OR_NULL ||
233 t == PTR_TO_MAP_VALUE_ADJ)
234 verbose("(ks=%d,vs=%d,id=%u)",
235 reg->map_ptr->key_size,
236 reg->map_ptr->value_size,
238 if (reg->min_value != BPF_REGISTER_MIN_RANGE)
239 verbose(",min_value=%lld",
240 (long long)reg->min_value);
241 if (reg->max_value != BPF_REGISTER_MAX_RANGE)
242 verbose(",max_value=%llu",
243 (unsigned long long)reg->max_value);
245 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
246 if (state->stack_slot_type[i] == STACK_SPILL)
247 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
248 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
253 static const char *const bpf_class_string[] = {
261 [BPF_ALU64] = "alu64",
264 static const char *const bpf_alu_string[16] = {
265 [BPF_ADD >> 4] = "+=",
266 [BPF_SUB >> 4] = "-=",
267 [BPF_MUL >> 4] = "*=",
268 [BPF_DIV >> 4] = "/=",
269 [BPF_OR >> 4] = "|=",
270 [BPF_AND >> 4] = "&=",
271 [BPF_LSH >> 4] = "<<=",
272 [BPF_RSH >> 4] = ">>=",
273 [BPF_NEG >> 4] = "neg",
274 [BPF_MOD >> 4] = "%=",
275 [BPF_XOR >> 4] = "^=",
276 [BPF_MOV >> 4] = "=",
277 [BPF_ARSH >> 4] = "s>>=",
278 [BPF_END >> 4] = "endian",
281 static const char *const bpf_ldst_string[] = {
282 [BPF_W >> 3] = "u32",
283 [BPF_H >> 3] = "u16",
285 [BPF_DW >> 3] = "u64",
288 static const char *const bpf_jmp_string[16] = {
289 [BPF_JA >> 4] = "jmp",
290 [BPF_JEQ >> 4] = "==",
291 [BPF_JGT >> 4] = ">",
292 [BPF_JGE >> 4] = ">=",
293 [BPF_JSET >> 4] = "&",
294 [BPF_JNE >> 4] = "!=",
295 [BPF_JSGT >> 4] = "s>",
296 [BPF_JSGE >> 4] = "s>=",
297 [BPF_CALL >> 4] = "call",
298 [BPF_EXIT >> 4] = "exit",
301 static void print_bpf_insn(struct bpf_insn *insn)
303 u8 class = BPF_CLASS(insn->code);
305 if (class == BPF_ALU || class == BPF_ALU64) {
306 if (BPF_SRC(insn->code) == BPF_X)
307 verbose("(%02x) %sr%d %s %sr%d\n",
308 insn->code, class == BPF_ALU ? "(u32) " : "",
310 bpf_alu_string[BPF_OP(insn->code) >> 4],
311 class == BPF_ALU ? "(u32) " : "",
314 verbose("(%02x) %sr%d %s %s%d\n",
315 insn->code, class == BPF_ALU ? "(u32) " : "",
317 bpf_alu_string[BPF_OP(insn->code) >> 4],
318 class == BPF_ALU ? "(u32) " : "",
320 } else if (class == BPF_STX) {
321 if (BPF_MODE(insn->code) == BPF_MEM)
322 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
324 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
326 insn->off, insn->src_reg);
327 else if (BPF_MODE(insn->code) == BPF_XADD)
328 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
330 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
331 insn->dst_reg, insn->off,
334 verbose("BUG_%02x\n", insn->code);
335 } else if (class == BPF_ST) {
336 if (BPF_MODE(insn->code) != BPF_MEM) {
337 verbose("BUG_st_%02x\n", insn->code);
340 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
342 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
344 insn->off, insn->imm);
345 } else if (class == BPF_LDX) {
346 if (BPF_MODE(insn->code) != BPF_MEM) {
347 verbose("BUG_ldx_%02x\n", insn->code);
350 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
351 insn->code, insn->dst_reg,
352 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
353 insn->src_reg, insn->off);
354 } else if (class == BPF_LD) {
355 if (BPF_MODE(insn->code) == BPF_ABS) {
356 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
358 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
360 } else if (BPF_MODE(insn->code) == BPF_IND) {
361 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
363 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
364 insn->src_reg, insn->imm);
365 } else if (BPF_MODE(insn->code) == BPF_IMM) {
366 verbose("(%02x) r%d = 0x%x\n",
367 insn->code, insn->dst_reg, insn->imm);
369 verbose("BUG_ld_%02x\n", insn->code);
372 } else if (class == BPF_JMP) {
373 u8 opcode = BPF_OP(insn->code);
375 if (opcode == BPF_CALL) {
376 verbose("(%02x) call %s#%d\n", insn->code,
377 func_id_name(insn->imm), insn->imm);
378 } else if (insn->code == (BPF_JMP | BPF_JA)) {
379 verbose("(%02x) goto pc%+d\n",
380 insn->code, insn->off);
381 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
382 verbose("(%02x) exit\n", insn->code);
383 } else if (BPF_SRC(insn->code) == BPF_X) {
384 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
385 insn->code, insn->dst_reg,
386 bpf_jmp_string[BPF_OP(insn->code) >> 4],
387 insn->src_reg, insn->off);
389 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
390 insn->code, insn->dst_reg,
391 bpf_jmp_string[BPF_OP(insn->code) >> 4],
392 insn->imm, insn->off);
395 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
399 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
401 struct bpf_verifier_stack_elem *elem;
404 if (env->head == NULL)
407 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
408 insn_idx = env->head->insn_idx;
410 *prev_insn_idx = env->head->prev_insn_idx;
411 elem = env->head->next;
418 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
419 int insn_idx, int prev_insn_idx)
421 struct bpf_verifier_stack_elem *elem;
423 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
427 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
428 elem->insn_idx = insn_idx;
429 elem->prev_insn_idx = prev_insn_idx;
430 elem->next = env->head;
433 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
434 verbose("BPF program is too complex\n");
439 /* pop all elements and return */
440 while (pop_stack(env, NULL) >= 0);
444 #define CALLER_SAVED_REGS 6
445 static const int caller_saved[CALLER_SAVED_REGS] = {
446 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
449 static void init_reg_state(struct bpf_reg_state *regs)
453 for (i = 0; i < MAX_BPF_REG; i++) {
454 regs[i].type = NOT_INIT;
456 regs[i].min_value = BPF_REGISTER_MIN_RANGE;
457 regs[i].max_value = BPF_REGISTER_MAX_RANGE;
461 regs[BPF_REG_FP].type = FRAME_PTR;
463 /* 1st arg to a function */
464 regs[BPF_REG_1].type = PTR_TO_CTX;
467 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
469 regs[regno].type = UNKNOWN_VALUE;
474 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
476 BUG_ON(regno >= MAX_BPF_REG);
477 __mark_reg_unknown_value(regs, regno);
480 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
482 regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
483 regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
486 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
489 mark_reg_unknown_value(regs, regno);
490 reset_reg_range_values(regs, regno);
494 SRC_OP, /* register is used as source operand */
495 DST_OP, /* register is used as destination operand */
496 DST_OP_NO_MARK /* same as above, check only, don't mark */
499 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
502 if (regno >= MAX_BPF_REG) {
503 verbose("R%d is invalid\n", regno);
508 /* check whether register used as source operand can be read */
509 if (regs[regno].type == NOT_INIT) {
510 verbose("R%d !read_ok\n", regno);
514 /* check whether register used as dest operand can be written to */
515 if (regno == BPF_REG_FP) {
516 verbose("frame pointer is read only\n");
520 mark_reg_unknown_value(regs, regno);
525 static int bpf_size_to_bytes(int bpf_size)
527 if (bpf_size == BPF_W)
529 else if (bpf_size == BPF_H)
531 else if (bpf_size == BPF_B)
533 else if (bpf_size == BPF_DW)
539 static bool is_spillable_regtype(enum bpf_reg_type type)
542 case PTR_TO_MAP_VALUE:
543 case PTR_TO_MAP_VALUE_OR_NULL:
544 case PTR_TO_MAP_VALUE_ADJ:
548 case PTR_TO_PACKET_END:
550 case CONST_PTR_TO_MAP:
557 /* check_stack_read/write functions track spill/fill of registers,
558 * stack boundary and alignment are checked in check_mem_access()
560 static int check_stack_write(struct bpf_verifier_state *state, int off,
561 int size, int value_regno)
564 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
565 * so it's aligned access and [off, off + size) are within stack limits
568 if (value_regno >= 0 &&
569 is_spillable_regtype(state->regs[value_regno].type)) {
571 /* register containing pointer is being spilled into stack */
572 if (size != BPF_REG_SIZE) {
573 verbose("invalid size of register spill\n");
577 /* save register state */
578 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
579 state->regs[value_regno];
581 for (i = 0; i < BPF_REG_SIZE; i++)
582 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
584 /* regular write of data into stack */
585 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
586 (struct bpf_reg_state) {};
588 for (i = 0; i < size; i++)
589 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
594 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
600 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
602 if (slot_type[0] == STACK_SPILL) {
603 if (size != BPF_REG_SIZE) {
604 verbose("invalid size of register spill\n");
607 for (i = 1; i < BPF_REG_SIZE; i++) {
608 if (slot_type[i] != STACK_SPILL) {
609 verbose("corrupted spill memory\n");
614 if (value_regno >= 0)
615 /* restore register state from stack */
616 state->regs[value_regno] =
617 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
620 for (i = 0; i < size; i++) {
621 if (slot_type[i] != STACK_MISC) {
622 verbose("invalid read from stack off %d+%d size %d\n",
627 if (value_regno >= 0)
628 /* have read misc data from the stack */
629 mark_reg_unknown_value_and_range(state->regs,
635 /* check read/write into map element returned by bpf_map_lookup_elem() */
636 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
639 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
641 if (off < 0 || size <= 0 || off + size > map->value_size) {
642 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
643 map->value_size, off, size);
649 /* check read/write into an adjusted map element */
650 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
653 struct bpf_verifier_state *state = &env->cur_state;
654 struct bpf_reg_state *reg = &state->regs[regno];
657 /* We adjusted the register to this map value, so we
658 * need to change off and size to min_value and max_value
659 * respectively to make sure our theoretical access will be
663 print_verifier_state(state);
664 env->varlen_map_value_access = true;
665 /* The minimum value is only important with signed
666 * comparisons where we can't assume the floor of a
667 * value is 0. If we are using signed variables for our
668 * index'es we need to make sure that whatever we use
669 * will have a set floor within our range.
671 if (reg->min_value < 0) {
672 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
676 err = check_map_access(env, regno, reg->min_value + off, size);
678 verbose("R%d min value is outside of the array range\n",
683 /* If we haven't set a max value then we need to bail
684 * since we can't be sure we won't do bad things.
686 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
687 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
691 return check_map_access(env, regno, reg->max_value + off, size);
694 #define MAX_PACKET_OFF 0xffff
696 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
697 const struct bpf_call_arg_meta *meta,
698 enum bpf_access_type t)
700 switch (env->prog->type) {
701 case BPF_PROG_TYPE_LWT_IN:
702 case BPF_PROG_TYPE_LWT_OUT:
703 /* dst_input() and dst_output() can't write for now */
707 case BPF_PROG_TYPE_SCHED_CLS:
708 case BPF_PROG_TYPE_SCHED_ACT:
709 case BPF_PROG_TYPE_XDP:
710 case BPF_PROG_TYPE_LWT_XMIT:
712 return meta->pkt_access;
714 env->seen_direct_write = true;
721 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
724 struct bpf_reg_state *regs = env->cur_state.regs;
725 struct bpf_reg_state *reg = ®s[regno];
728 if (off < 0 || size <= 0 || off + size > reg->range) {
729 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
730 off, size, regno, reg->id, reg->off, reg->range);
736 /* check access to 'struct bpf_context' fields */
737 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
738 enum bpf_access_type t, enum bpf_reg_type *reg_type)
740 /* for analyzer ctx accesses are already validated and converted */
741 if (env->analyzer_ops)
744 if (env->prog->aux->ops->is_valid_access &&
745 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
746 /* remember the offset of last byte accessed in ctx */
747 if (env->prog->aux->max_ctx_offset < off + size)
748 env->prog->aux->max_ctx_offset = off + size;
752 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
756 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
758 if (env->allow_ptr_leaks)
761 switch (env->cur_state.regs[regno].type) {
770 static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg,
773 if (reg->id && size != 1) {
774 verbose("Unknown alignment. Only byte-sized access allowed in packet access.\n");
778 /* skb->data is NET_IP_ALIGN-ed */
779 if ((NET_IP_ALIGN + reg->off + off) % size != 0) {
780 verbose("misaligned packet access off %d+%d+%d size %d\n",
781 NET_IP_ALIGN, reg->off, off, size);
788 static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
792 verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
799 static int check_ptr_alignment(const struct bpf_reg_state *reg,
804 return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 :
805 check_pkt_ptr_alignment(reg, off, size);
806 case PTR_TO_MAP_VALUE_ADJ:
807 return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 :
808 check_val_ptr_alignment(reg, size);
810 if (off % size != 0) {
811 verbose("misaligned access off %d size %d\n",
820 /* check whether memory at (regno + off) is accessible for t = (read | write)
821 * if t==write, value_regno is a register which value is stored into memory
822 * if t==read, value_regno is a register which will receive the value from memory
823 * if t==write && value_regno==-1, some unknown value is stored into memory
824 * if t==read && value_regno==-1, don't care what we read from memory
826 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
827 int bpf_size, enum bpf_access_type t,
830 struct bpf_verifier_state *state = &env->cur_state;
831 struct bpf_reg_state *reg = &state->regs[regno];
834 if (reg->type == PTR_TO_STACK)
837 size = bpf_size_to_bytes(bpf_size);
841 err = check_ptr_alignment(reg, off, size);
845 if (reg->type == PTR_TO_MAP_VALUE ||
846 reg->type == PTR_TO_MAP_VALUE_ADJ) {
847 if (t == BPF_WRITE && value_regno >= 0 &&
848 is_pointer_value(env, value_regno)) {
849 verbose("R%d leaks addr into map\n", value_regno);
853 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
854 err = check_map_access_adj(env, regno, off, size);
856 err = check_map_access(env, regno, off, size);
857 if (!err && t == BPF_READ && value_regno >= 0)
858 mark_reg_unknown_value_and_range(state->regs,
861 } else if (reg->type == PTR_TO_CTX) {
862 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
864 if (t == BPF_WRITE && value_regno >= 0 &&
865 is_pointer_value(env, value_regno)) {
866 verbose("R%d leaks addr into ctx\n", value_regno);
869 err = check_ctx_access(env, off, size, t, ®_type);
870 if (!err && t == BPF_READ && value_regno >= 0) {
871 mark_reg_unknown_value_and_range(state->regs,
873 /* note that reg.[id|off|range] == 0 */
874 state->regs[value_regno].type = reg_type;
877 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
878 if (off >= 0 || off < -MAX_BPF_STACK) {
879 verbose("invalid stack off=%d size=%d\n", off, size);
882 if (t == BPF_WRITE) {
883 if (!env->allow_ptr_leaks &&
884 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
885 size != BPF_REG_SIZE) {
886 verbose("attempt to corrupt spilled pointer on stack\n");
889 err = check_stack_write(state, off, size, value_regno);
891 err = check_stack_read(state, off, size, value_regno);
893 } else if (state->regs[regno].type == PTR_TO_PACKET) {
894 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
895 verbose("cannot write into packet\n");
898 if (t == BPF_WRITE && value_regno >= 0 &&
899 is_pointer_value(env, value_regno)) {
900 verbose("R%d leaks addr into packet\n", value_regno);
903 err = check_packet_access(env, regno, off, size);
904 if (!err && t == BPF_READ && value_regno >= 0)
905 mark_reg_unknown_value_and_range(state->regs,
908 verbose("R%d invalid mem access '%s'\n",
909 regno, reg_type_str[reg->type]);
913 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
914 state->regs[value_regno].type == UNKNOWN_VALUE) {
915 /* 1 or 2 byte load zero-extends, determine the number of
916 * zero upper bits. Not doing it fo 4 byte load, since
917 * such values cannot be added to ptr_to_packet anyway.
919 state->regs[value_regno].imm = 64 - size * 8;
924 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
926 struct bpf_reg_state *regs = env->cur_state.regs;
929 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
931 verbose("BPF_XADD uses reserved fields\n");
935 /* check src1 operand */
936 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
940 /* check src2 operand */
941 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
945 /* check whether atomic_add can read the memory */
946 err = check_mem_access(env, insn->dst_reg, insn->off,
947 BPF_SIZE(insn->code), BPF_READ, -1);
951 /* check whether atomic_add can write into the same memory */
952 return check_mem_access(env, insn->dst_reg, insn->off,
953 BPF_SIZE(insn->code), BPF_WRITE, -1);
956 /* when register 'regno' is passed into function that will read 'access_size'
957 * bytes from that pointer, make sure that it's within stack boundary
958 * and all elements of stack are initialized
960 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
961 int access_size, bool zero_size_allowed,
962 struct bpf_call_arg_meta *meta)
964 struct bpf_verifier_state *state = &env->cur_state;
965 struct bpf_reg_state *regs = state->regs;
968 if (regs[regno].type != PTR_TO_STACK) {
969 if (zero_size_allowed && access_size == 0 &&
970 regs[regno].type == CONST_IMM &&
971 regs[regno].imm == 0)
974 verbose("R%d type=%s expected=%s\n", regno,
975 reg_type_str[regs[regno].type],
976 reg_type_str[PTR_TO_STACK]);
980 off = regs[regno].imm;
981 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
983 verbose("invalid stack type R%d off=%d access_size=%d\n",
984 regno, off, access_size);
988 if (meta && meta->raw_mode) {
989 meta->access_size = access_size;
994 for (i = 0; i < access_size; i++) {
995 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
996 verbose("invalid indirect read from stack off %d+%d size %d\n",
997 off, i, access_size);
1004 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1005 int access_size, bool zero_size_allowed,
1006 struct bpf_call_arg_meta *meta)
1008 struct bpf_reg_state *regs = env->cur_state.regs;
1010 switch (regs[regno].type) {
1012 return check_packet_access(env, regno, 0, access_size);
1013 case PTR_TO_MAP_VALUE:
1014 return check_map_access(env, regno, 0, access_size);
1015 case PTR_TO_MAP_VALUE_ADJ:
1016 return check_map_access_adj(env, regno, 0, access_size);
1017 default: /* const_imm|ptr_to_stack or invalid ptr */
1018 return check_stack_boundary(env, regno, access_size,
1019 zero_size_allowed, meta);
1023 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1024 enum bpf_arg_type arg_type,
1025 struct bpf_call_arg_meta *meta)
1027 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno];
1028 enum bpf_reg_type expected_type, type = reg->type;
1031 if (arg_type == ARG_DONTCARE)
1034 if (type == NOT_INIT) {
1035 verbose("R%d !read_ok\n", regno);
1039 if (arg_type == ARG_ANYTHING) {
1040 if (is_pointer_value(env, regno)) {
1041 verbose("R%d leaks addr into helper function\n", regno);
1047 if (type == PTR_TO_PACKET &&
1048 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1049 verbose("helper access to the packet is not allowed\n");
1053 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1054 arg_type == ARG_PTR_TO_MAP_VALUE) {
1055 expected_type = PTR_TO_STACK;
1056 if (type != PTR_TO_PACKET && type != expected_type)
1058 } else if (arg_type == ARG_CONST_SIZE ||
1059 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1060 expected_type = CONST_IMM;
1061 /* One exception. Allow UNKNOWN_VALUE registers when the
1062 * boundaries are known and don't cause unsafe memory accesses
1064 if (type != UNKNOWN_VALUE && type != expected_type)
1066 } else if (arg_type == ARG_CONST_MAP_PTR) {
1067 expected_type = CONST_PTR_TO_MAP;
1068 if (type != expected_type)
1070 } else if (arg_type == ARG_PTR_TO_CTX) {
1071 expected_type = PTR_TO_CTX;
1072 if (type != expected_type)
1074 } else if (arg_type == ARG_PTR_TO_MEM ||
1075 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1076 expected_type = PTR_TO_STACK;
1077 /* One exception here. In case function allows for NULL to be
1078 * passed in as argument, it's a CONST_IMM type. Final test
1079 * happens during stack boundary checking.
1081 if (type == CONST_IMM && reg->imm == 0)
1082 /* final test in check_stack_boundary() */;
1083 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1084 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1086 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1088 verbose("unsupported arg_type %d\n", arg_type);
1092 if (arg_type == ARG_CONST_MAP_PTR) {
1093 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1094 meta->map_ptr = reg->map_ptr;
1095 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1096 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1097 * check that [key, key + map->key_size) are within
1098 * stack limits and initialized
1100 if (!meta->map_ptr) {
1101 /* in function declaration map_ptr must come before
1102 * map_key, so that it's verified and known before
1103 * we have to check map_key here. Otherwise it means
1104 * that kernel subsystem misconfigured verifier
1106 verbose("invalid map_ptr to access map->key\n");
1109 if (type == PTR_TO_PACKET)
1110 err = check_packet_access(env, regno, 0,
1111 meta->map_ptr->key_size);
1113 err = check_stack_boundary(env, regno,
1114 meta->map_ptr->key_size,
1116 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1117 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1118 * check [value, value + map->value_size) validity
1120 if (!meta->map_ptr) {
1121 /* kernel subsystem misconfigured verifier */
1122 verbose("invalid map_ptr to access map->value\n");
1125 if (type == PTR_TO_PACKET)
1126 err = check_packet_access(env, regno, 0,
1127 meta->map_ptr->value_size);
1129 err = check_stack_boundary(env, regno,
1130 meta->map_ptr->value_size,
1132 } else if (arg_type == ARG_CONST_SIZE ||
1133 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1134 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1136 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1137 * from stack pointer 'buf'. Check it
1138 * note: regno == len, regno - 1 == buf
1141 /* kernel subsystem misconfigured verifier */
1142 verbose("ARG_CONST_SIZE cannot be first argument\n");
1146 /* If the register is UNKNOWN_VALUE, the access check happens
1147 * using its boundaries. Otherwise, just use its imm
1149 if (type == UNKNOWN_VALUE) {
1150 /* For unprivileged variable accesses, disable raw
1151 * mode so that the program is required to
1152 * initialize all the memory that the helper could
1153 * just partially fill up.
1157 if (reg->min_value < 0) {
1158 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1163 if (reg->min_value == 0) {
1164 err = check_helper_mem_access(env, regno - 1, 0,
1171 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1172 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1176 err = check_helper_mem_access(env, regno - 1,
1178 zero_size_allowed, meta);
1182 /* register is CONST_IMM */
1183 err = check_helper_mem_access(env, regno - 1, reg->imm,
1184 zero_size_allowed, meta);
1190 verbose("R%d type=%s expected=%s\n", regno,
1191 reg_type_str[type], reg_type_str[expected_type]);
1195 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1200 /* We need a two way check, first is from map perspective ... */
1201 switch (map->map_type) {
1202 case BPF_MAP_TYPE_PROG_ARRAY:
1203 if (func_id != BPF_FUNC_tail_call)
1206 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1207 if (func_id != BPF_FUNC_perf_event_read &&
1208 func_id != BPF_FUNC_perf_event_output)
1211 case BPF_MAP_TYPE_STACK_TRACE:
1212 if (func_id != BPF_FUNC_get_stackid)
1215 case BPF_MAP_TYPE_CGROUP_ARRAY:
1216 if (func_id != BPF_FUNC_skb_under_cgroup &&
1217 func_id != BPF_FUNC_current_task_under_cgroup)
1220 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1221 case BPF_MAP_TYPE_HASH_OF_MAPS:
1222 if (func_id != BPF_FUNC_map_lookup_elem)
1228 /* ... and second from the function itself. */
1230 case BPF_FUNC_tail_call:
1231 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1234 case BPF_FUNC_perf_event_read:
1235 case BPF_FUNC_perf_event_output:
1236 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1239 case BPF_FUNC_get_stackid:
1240 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1243 case BPF_FUNC_current_task_under_cgroup:
1244 case BPF_FUNC_skb_under_cgroup:
1245 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1254 verbose("cannot pass map_type %d into func %s#%d\n",
1255 map->map_type, func_id_name(func_id), func_id);
1259 static int check_raw_mode(const struct bpf_func_proto *fn)
1263 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1265 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1267 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1269 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1271 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1274 return count > 1 ? -EINVAL : 0;
1277 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1279 struct bpf_verifier_state *state = &env->cur_state;
1280 struct bpf_reg_state *regs = state->regs, *reg;
1283 for (i = 0; i < MAX_BPF_REG; i++)
1284 if (regs[i].type == PTR_TO_PACKET ||
1285 regs[i].type == PTR_TO_PACKET_END)
1286 mark_reg_unknown_value(regs, i);
1288 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1289 if (state->stack_slot_type[i] != STACK_SPILL)
1291 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1292 if (reg->type != PTR_TO_PACKET &&
1293 reg->type != PTR_TO_PACKET_END)
1295 reg->type = UNKNOWN_VALUE;
1300 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1302 struct bpf_verifier_state *state = &env->cur_state;
1303 const struct bpf_func_proto *fn = NULL;
1304 struct bpf_reg_state *regs = state->regs;
1305 struct bpf_reg_state *reg;
1306 struct bpf_call_arg_meta meta;
1310 /* find function prototype */
1311 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1312 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1316 if (env->prog->aux->ops->get_func_proto)
1317 fn = env->prog->aux->ops->get_func_proto(func_id);
1320 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1324 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1325 if (!env->prog->gpl_compatible && fn->gpl_only) {
1326 verbose("cannot call GPL only function from proprietary program\n");
1330 changes_data = bpf_helper_changes_pkt_data(fn->func);
1332 memset(&meta, 0, sizeof(meta));
1333 meta.pkt_access = fn->pkt_access;
1335 /* We only support one arg being in raw mode at the moment, which
1336 * is sufficient for the helper functions we have right now.
1338 err = check_raw_mode(fn);
1340 verbose("kernel subsystem misconfigured func %s#%d\n",
1341 func_id_name(func_id), func_id);
1346 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1349 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1352 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1355 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1358 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1362 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1363 * is inferred from register state.
1365 for (i = 0; i < meta.access_size; i++) {
1366 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1371 /* reset caller saved regs */
1372 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1373 reg = regs + caller_saved[i];
1374 reg->type = NOT_INIT;
1378 /* update return register */
1379 if (fn->ret_type == RET_INTEGER) {
1380 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1381 } else if (fn->ret_type == RET_VOID) {
1382 regs[BPF_REG_0].type = NOT_INIT;
1383 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1384 struct bpf_insn_aux_data *insn_aux;
1386 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1387 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1388 /* remember map_ptr, so that check_map_access()
1389 * can check 'value_size' boundary of memory access
1390 * to map element returned from bpf_map_lookup_elem()
1392 if (meta.map_ptr == NULL) {
1393 verbose("kernel subsystem misconfigured verifier\n");
1396 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1397 regs[BPF_REG_0].id = ++env->id_gen;
1398 insn_aux = &env->insn_aux_data[insn_idx];
1399 if (!insn_aux->map_ptr)
1400 insn_aux->map_ptr = meta.map_ptr;
1401 else if (insn_aux->map_ptr != meta.map_ptr)
1402 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1404 verbose("unknown return type %d of func %s#%d\n",
1405 fn->ret_type, func_id_name(func_id), func_id);
1409 err = check_map_func_compatibility(meta.map_ptr, func_id);
1414 clear_all_pkt_pointers(env);
1418 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1419 struct bpf_insn *insn)
1421 struct bpf_reg_state *regs = env->cur_state.regs;
1422 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1423 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1424 struct bpf_reg_state tmp_reg;
1427 if (BPF_SRC(insn->code) == BPF_K) {
1428 /* pkt_ptr += imm */
1433 verbose("addition of negative constant to packet pointer is not allowed\n");
1436 if (imm >= MAX_PACKET_OFF ||
1437 imm + dst_reg->off >= MAX_PACKET_OFF) {
1438 verbose("constant %d is too large to add to packet pointer\n",
1442 /* a constant was added to pkt_ptr.
1443 * Remember it while keeping the same 'id'
1445 dst_reg->off += imm;
1447 if (src_reg->type == PTR_TO_PACKET) {
1448 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1449 tmp_reg = *dst_reg; /* save r7 state */
1450 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1451 src_reg = &tmp_reg; /* pretend it's src_reg state */
1452 /* if the checks below reject it, the copy won't matter,
1453 * since we're rejecting the whole program. If all ok,
1454 * then imm22 state will be added to r7
1455 * and r7 will be pkt(id=0,off=22,r=62) while
1456 * r6 will stay as pkt(id=0,off=0,r=62)
1460 if (src_reg->type == CONST_IMM) {
1461 /* pkt_ptr += reg where reg is known constant */
1465 /* disallow pkt_ptr += reg
1466 * if reg is not uknown_value with guaranteed zero upper bits
1467 * otherwise pkt_ptr may overflow and addition will become
1468 * subtraction which is not allowed
1470 if (src_reg->type != UNKNOWN_VALUE) {
1471 verbose("cannot add '%s' to ptr_to_packet\n",
1472 reg_type_str[src_reg->type]);
1475 if (src_reg->imm < 48) {
1476 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1480 /* dst_reg stays as pkt_ptr type and since some positive
1481 * integer value was added to the pointer, increment its 'id'
1483 dst_reg->id = ++env->id_gen;
1485 /* something was added to pkt_ptr, set range and off to zero */
1492 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1494 struct bpf_reg_state *regs = env->cur_state.regs;
1495 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1496 u8 opcode = BPF_OP(insn->code);
1499 /* for type == UNKNOWN_VALUE:
1500 * imm > 0 -> number of zero upper bits
1501 * imm == 0 -> don't track which is the same as all bits can be non-zero
1504 if (BPF_SRC(insn->code) == BPF_X) {
1505 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1507 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1508 dst_reg->imm && opcode == BPF_ADD) {
1510 * where both have zero upper bits. Adding them
1511 * can only result making one more bit non-zero
1512 * in the larger value.
1513 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1514 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1516 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1520 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1521 dst_reg->imm && opcode == BPF_ADD) {
1523 * where dreg has zero upper bits and sreg is const.
1524 * Adding them can only result making one more bit
1525 * non-zero in the larger value.
1527 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1528 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1532 /* all other cases non supported yet, just mark dst_reg */
1537 /* sign extend 32-bit imm into 64-bit to make sure that
1538 * negative values occupy bit 63. Note ilog2() would have
1539 * been incorrect, since sizeof(insn->imm) == 4
1541 imm_log2 = __ilog2_u64((long long)insn->imm);
1543 if (dst_reg->imm && opcode == BPF_LSH) {
1545 * if reg was a result of 2 byte load, then its imm == 48
1546 * which means that upper 48 bits are zero and shifting this reg
1547 * left by 4 would mean that upper 44 bits are still zero
1549 dst_reg->imm -= insn->imm;
1550 } else if (dst_reg->imm && opcode == BPF_MUL) {
1552 * if multiplying by 14 subtract 4
1553 * This is conservative calculation of upper zero bits.
1554 * It's not trying to special case insn->imm == 1 or 0 cases
1556 dst_reg->imm -= imm_log2 + 1;
1557 } else if (opcode == BPF_AND) {
1559 dst_reg->imm = 63 - imm_log2;
1560 } else if (dst_reg->imm && opcode == BPF_ADD) {
1562 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1564 } else if (opcode == BPF_RSH) {
1566 * which means that after right shift, upper bits will be zero
1567 * note that verifier already checked that
1568 * 0 <= imm < 64 for shift insn
1570 dst_reg->imm += insn->imm;
1571 if (unlikely(dst_reg->imm > 64))
1572 /* some dumb code did:
1575 * and all bits are zero now */
1578 /* all other alu ops, means that we don't know what will
1579 * happen to the value, mark it with unknown number of zero bits
1584 if (dst_reg->imm < 0) {
1585 /* all 64 bits of the register can contain non-zero bits
1586 * and such value cannot be added to ptr_to_packet, since it
1587 * may overflow, mark it as unknown to avoid further eval
1594 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1595 struct bpf_insn *insn)
1597 struct bpf_reg_state *regs = env->cur_state.regs;
1598 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1599 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1600 u8 opcode = BPF_OP(insn->code);
1601 u64 dst_imm = dst_reg->imm;
1603 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1604 * containing ALU ops. Don't care about overflow or negative
1605 * values, just add/sub/... them; registers are in u64.
1607 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1608 dst_imm += insn->imm;
1609 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1610 src_reg->type == CONST_IMM) {
1611 dst_imm += src_reg->imm;
1612 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1613 dst_imm -= insn->imm;
1614 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1615 src_reg->type == CONST_IMM) {
1616 dst_imm -= src_reg->imm;
1617 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1618 dst_imm *= insn->imm;
1619 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1620 src_reg->type == CONST_IMM) {
1621 dst_imm *= src_reg->imm;
1622 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1623 dst_imm |= insn->imm;
1624 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1625 src_reg->type == CONST_IMM) {
1626 dst_imm |= src_reg->imm;
1627 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1628 dst_imm &= insn->imm;
1629 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1630 src_reg->type == CONST_IMM) {
1631 dst_imm &= src_reg->imm;
1632 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1633 dst_imm >>= insn->imm;
1634 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1635 src_reg->type == CONST_IMM) {
1636 dst_imm >>= src_reg->imm;
1637 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1638 dst_imm <<= insn->imm;
1639 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1640 src_reg->type == CONST_IMM) {
1641 dst_imm <<= src_reg->imm;
1643 mark_reg_unknown_value(regs, insn->dst_reg);
1647 dst_reg->imm = dst_imm;
1652 static void check_reg_overflow(struct bpf_reg_state *reg)
1654 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1655 reg->max_value = BPF_REGISTER_MAX_RANGE;
1656 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1657 reg->min_value > BPF_REGISTER_MAX_RANGE)
1658 reg->min_value = BPF_REGISTER_MIN_RANGE;
1661 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1662 struct bpf_insn *insn)
1664 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1665 s64 min_val = BPF_REGISTER_MIN_RANGE;
1666 u64 max_val = BPF_REGISTER_MAX_RANGE;
1667 u8 opcode = BPF_OP(insn->code);
1669 dst_reg = ®s[insn->dst_reg];
1670 if (BPF_SRC(insn->code) == BPF_X) {
1671 check_reg_overflow(®s[insn->src_reg]);
1672 min_val = regs[insn->src_reg].min_value;
1673 max_val = regs[insn->src_reg].max_value;
1675 /* If the source register is a random pointer then the
1676 * min_value/max_value values represent the range of the known
1677 * accesses into that value, not the actual min/max value of the
1678 * register itself. In this case we have to reset the reg range
1679 * values so we know it is not safe to look at.
1681 if (regs[insn->src_reg].type != CONST_IMM &&
1682 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1683 min_val = BPF_REGISTER_MIN_RANGE;
1684 max_val = BPF_REGISTER_MAX_RANGE;
1686 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1687 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1688 min_val = max_val = insn->imm;
1691 /* We don't know anything about what was done to this register, mark it
1694 if (min_val == BPF_REGISTER_MIN_RANGE &&
1695 max_val == BPF_REGISTER_MAX_RANGE) {
1696 reset_reg_range_values(regs, insn->dst_reg);
1700 /* If one of our values was at the end of our ranges then we can't just
1701 * do our normal operations to the register, we need to set the values
1702 * to the min/max since they are undefined.
1704 if (min_val == BPF_REGISTER_MIN_RANGE)
1705 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1706 if (max_val == BPF_REGISTER_MAX_RANGE)
1707 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1711 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1712 dst_reg->min_value += min_val;
1713 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1714 dst_reg->max_value += max_val;
1717 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1718 dst_reg->min_value -= min_val;
1719 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1720 dst_reg->max_value -= max_val;
1723 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1724 dst_reg->min_value *= min_val;
1725 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1726 dst_reg->max_value *= max_val;
1729 /* Disallow AND'ing of negative numbers, ain't nobody got time
1730 * for that. Otherwise the minimum is 0 and the max is the max
1731 * value we could AND against.
1734 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1736 dst_reg->min_value = 0;
1737 dst_reg->max_value = max_val;
1740 /* Gotta have special overflow logic here, if we're shifting
1741 * more than MAX_RANGE then just assume we have an invalid
1744 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1745 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1746 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1747 dst_reg->min_value <<= min_val;
1749 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1750 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1751 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1752 dst_reg->max_value <<= max_val;
1755 /* RSH by a negative number is undefined, and the BPF_RSH is an
1756 * unsigned shift, so make the appropriate casts.
1758 if (min_val < 0 || dst_reg->min_value < 0)
1759 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1761 dst_reg->min_value =
1762 (u64)(dst_reg->min_value) >> min_val;
1763 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1764 dst_reg->max_value >>= max_val;
1767 reset_reg_range_values(regs, insn->dst_reg);
1771 check_reg_overflow(dst_reg);
1774 /* check validity of 32-bit and 64-bit arithmetic operations */
1775 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1777 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1778 u8 opcode = BPF_OP(insn->code);
1781 if (opcode == BPF_END || opcode == BPF_NEG) {
1782 if (opcode == BPF_NEG) {
1783 if (BPF_SRC(insn->code) != 0 ||
1784 insn->src_reg != BPF_REG_0 ||
1785 insn->off != 0 || insn->imm != 0) {
1786 verbose("BPF_NEG uses reserved fields\n");
1790 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1791 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1792 verbose("BPF_END uses reserved fields\n");
1797 /* check src operand */
1798 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1802 if (is_pointer_value(env, insn->dst_reg)) {
1803 verbose("R%d pointer arithmetic prohibited\n",
1808 /* check dest operand */
1809 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1813 } else if (opcode == BPF_MOV) {
1815 if (BPF_SRC(insn->code) == BPF_X) {
1816 if (insn->imm != 0 || insn->off != 0) {
1817 verbose("BPF_MOV uses reserved fields\n");
1821 /* check src operand */
1822 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1826 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1827 verbose("BPF_MOV uses reserved fields\n");
1832 /* check dest operand */
1833 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1837 /* we are setting our register to something new, we need to
1838 * reset its range values.
1840 reset_reg_range_values(regs, insn->dst_reg);
1842 if (BPF_SRC(insn->code) == BPF_X) {
1843 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1845 * copy register state to dest reg
1847 regs[insn->dst_reg] = regs[insn->src_reg];
1849 if (is_pointer_value(env, insn->src_reg)) {
1850 verbose("R%d partial copy of pointer\n",
1854 mark_reg_unknown_value(regs, insn->dst_reg);
1858 * remember the value we stored into this reg
1860 regs[insn->dst_reg].type = CONST_IMM;
1861 regs[insn->dst_reg].imm = insn->imm;
1862 regs[insn->dst_reg].max_value = insn->imm;
1863 regs[insn->dst_reg].min_value = insn->imm;
1866 } else if (opcode > BPF_END) {
1867 verbose("invalid BPF_ALU opcode %x\n", opcode);
1870 } else { /* all other ALU ops: and, sub, xor, add, ... */
1872 if (BPF_SRC(insn->code) == BPF_X) {
1873 if (insn->imm != 0 || insn->off != 0) {
1874 verbose("BPF_ALU uses reserved fields\n");
1877 /* check src1 operand */
1878 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1882 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1883 verbose("BPF_ALU uses reserved fields\n");
1888 /* check src2 operand */
1889 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1893 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1894 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1895 verbose("div by zero\n");
1899 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1900 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1901 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1903 if (insn->imm < 0 || insn->imm >= size) {
1904 verbose("invalid shift %d\n", insn->imm);
1909 /* check dest operand */
1910 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1914 dst_reg = ®s[insn->dst_reg];
1916 /* first we want to adjust our ranges. */
1917 adjust_reg_min_max_vals(env, insn);
1919 /* pattern match 'bpf_add Rx, imm' instruction */
1920 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1921 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1922 dst_reg->type = PTR_TO_STACK;
1923 dst_reg->imm = insn->imm;
1925 } else if (opcode == BPF_ADD &&
1926 BPF_CLASS(insn->code) == BPF_ALU64 &&
1927 dst_reg->type == PTR_TO_STACK &&
1928 ((BPF_SRC(insn->code) == BPF_X &&
1929 regs[insn->src_reg].type == CONST_IMM) ||
1930 BPF_SRC(insn->code) == BPF_K)) {
1931 if (BPF_SRC(insn->code) == BPF_X)
1932 dst_reg->imm += regs[insn->src_reg].imm;
1934 dst_reg->imm += insn->imm;
1936 } else if (opcode == BPF_ADD &&
1937 BPF_CLASS(insn->code) == BPF_ALU64 &&
1938 (dst_reg->type == PTR_TO_PACKET ||
1939 (BPF_SRC(insn->code) == BPF_X &&
1940 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1941 /* ptr_to_packet += K|X */
1942 return check_packet_ptr_add(env, insn);
1943 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1944 dst_reg->type == UNKNOWN_VALUE &&
1945 env->allow_ptr_leaks) {
1946 /* unknown += K|X */
1947 return evaluate_reg_alu(env, insn);
1948 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1949 dst_reg->type == CONST_IMM &&
1950 env->allow_ptr_leaks) {
1951 /* reg_imm += K|X */
1952 return evaluate_reg_imm_alu(env, insn);
1953 } else if (is_pointer_value(env, insn->dst_reg)) {
1954 verbose("R%d pointer arithmetic prohibited\n",
1957 } else if (BPF_SRC(insn->code) == BPF_X &&
1958 is_pointer_value(env, insn->src_reg)) {
1959 verbose("R%d pointer arithmetic prohibited\n",
1964 /* If we did pointer math on a map value then just set it to our
1965 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1966 * loads to this register appropriately, otherwise just mark the
1967 * register as unknown.
1969 if (env->allow_ptr_leaks &&
1970 BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD &&
1971 (dst_reg->type == PTR_TO_MAP_VALUE ||
1972 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1973 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1975 mark_reg_unknown_value(regs, insn->dst_reg);
1981 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1982 struct bpf_reg_state *dst_reg)
1984 struct bpf_reg_state *regs = state->regs, *reg;
1987 /* LLVM can generate two kind of checks:
1993 * if (r2 > pkt_end) goto <handle exception>
1997 * r2 == dst_reg, pkt_end == src_reg
1998 * r2=pkt(id=n,off=8,r=0)
1999 * r3=pkt(id=n,off=0,r=0)
2005 * if (pkt_end >= r2) goto <access okay>
2006 * <handle exception>
2009 * pkt_end == dst_reg, r2 == src_reg
2010 * r2=pkt(id=n,off=8,r=0)
2011 * r3=pkt(id=n,off=0,r=0)
2013 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2014 * so that range of bytes [r3, r3 + 8) is safe to access.
2017 for (i = 0; i < MAX_BPF_REG; i++)
2018 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
2019 /* keep the maximum range already checked */
2020 regs[i].range = max(regs[i].range, dst_reg->off);
2022 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2023 if (state->stack_slot_type[i] != STACK_SPILL)
2025 reg = &state->spilled_regs[i / BPF_REG_SIZE];
2026 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
2027 reg->range = max(reg->range, dst_reg->off);
2031 /* Adjusts the register min/max values in the case that the dst_reg is the
2032 * variable register that we are working on, and src_reg is a constant or we're
2033 * simply doing a BPF_K check.
2035 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2036 struct bpf_reg_state *false_reg, u64 val,
2041 /* If this is false then we know nothing Jon Snow, but if it is
2042 * true then we know for sure.
2044 true_reg->max_value = true_reg->min_value = val;
2047 /* If this is true we know nothing Jon Snow, but if it is false
2048 * we know the value for sure;
2050 false_reg->max_value = false_reg->min_value = val;
2053 /* Unsigned comparison, the minimum value is 0. */
2054 false_reg->min_value = 0;
2057 /* If this is false then we know the maximum val is val,
2058 * otherwise we know the min val is val+1.
2060 false_reg->max_value = val;
2061 true_reg->min_value = val + 1;
2064 /* Unsigned comparison, the minimum value is 0. */
2065 false_reg->min_value = 0;
2068 /* If this is false then we know the maximum value is val - 1,
2069 * otherwise we know the mimimum value is val.
2071 false_reg->max_value = val - 1;
2072 true_reg->min_value = val;
2078 check_reg_overflow(false_reg);
2079 check_reg_overflow(true_reg);
2082 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2083 * is the variable reg.
2085 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2086 struct bpf_reg_state *false_reg, u64 val,
2091 /* If this is false then we know nothing Jon Snow, but if it is
2092 * true then we know for sure.
2094 true_reg->max_value = true_reg->min_value = val;
2097 /* If this is true we know nothing Jon Snow, but if it is false
2098 * we know the value for sure;
2100 false_reg->max_value = false_reg->min_value = val;
2103 /* Unsigned comparison, the minimum value is 0. */
2104 true_reg->min_value = 0;
2108 * If this is false, then the val is <= the register, if it is
2109 * true the register <= to the val.
2111 false_reg->min_value = val;
2112 true_reg->max_value = val - 1;
2115 /* Unsigned comparison, the minimum value is 0. */
2116 true_reg->min_value = 0;
2119 /* If this is false then constant < register, if it is true then
2120 * the register < constant.
2122 false_reg->min_value = val + 1;
2123 true_reg->max_value = val;
2129 check_reg_overflow(false_reg);
2130 check_reg_overflow(true_reg);
2133 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2134 enum bpf_reg_type type)
2136 struct bpf_reg_state *reg = ®s[regno];
2138 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2139 if (type == UNKNOWN_VALUE) {
2140 __mark_reg_unknown_value(regs, regno);
2141 } else if (reg->map_ptr->inner_map_meta) {
2142 reg->type = CONST_PTR_TO_MAP;
2143 reg->map_ptr = reg->map_ptr->inner_map_meta;
2147 /* We don't need id from this point onwards anymore, thus we
2148 * should better reset it, so that state pruning has chances
2155 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2156 * be folded together at some point.
2158 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2159 enum bpf_reg_type type)
2161 struct bpf_reg_state *regs = state->regs;
2162 u32 id = regs[regno].id;
2165 for (i = 0; i < MAX_BPF_REG; i++)
2166 mark_map_reg(regs, i, id, type);
2168 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2169 if (state->stack_slot_type[i] != STACK_SPILL)
2171 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2175 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2176 struct bpf_insn *insn, int *insn_idx)
2178 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2179 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2180 u8 opcode = BPF_OP(insn->code);
2183 if (opcode > BPF_EXIT) {
2184 verbose("invalid BPF_JMP opcode %x\n", opcode);
2188 if (BPF_SRC(insn->code) == BPF_X) {
2189 if (insn->imm != 0) {
2190 verbose("BPF_JMP uses reserved fields\n");
2194 /* check src1 operand */
2195 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2199 if (is_pointer_value(env, insn->src_reg)) {
2200 verbose("R%d pointer comparison prohibited\n",
2205 if (insn->src_reg != BPF_REG_0) {
2206 verbose("BPF_JMP uses reserved fields\n");
2211 /* check src2 operand */
2212 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2216 dst_reg = ®s[insn->dst_reg];
2218 /* detect if R == 0 where R was initialized to zero earlier */
2219 if (BPF_SRC(insn->code) == BPF_K &&
2220 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2221 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2222 if (opcode == BPF_JEQ) {
2223 /* if (imm == imm) goto pc+off;
2224 * only follow the goto, ignore fall-through
2226 *insn_idx += insn->off;
2229 /* if (imm != imm) goto pc+off;
2230 * only follow fall-through branch, since
2231 * that's where the program will go
2237 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2241 /* detect if we are comparing against a constant value so we can adjust
2242 * our min/max values for our dst register.
2244 if (BPF_SRC(insn->code) == BPF_X) {
2245 if (regs[insn->src_reg].type == CONST_IMM)
2246 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2247 dst_reg, regs[insn->src_reg].imm,
2249 else if (dst_reg->type == CONST_IMM)
2250 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2251 ®s[insn->src_reg], dst_reg->imm,
2254 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2255 dst_reg, insn->imm, opcode);
2258 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2259 if (BPF_SRC(insn->code) == BPF_K &&
2260 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2261 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2262 /* Mark all identical map registers in each branch as either
2263 * safe or unknown depending R == 0 or R != 0 conditional.
2265 mark_map_regs(this_branch, insn->dst_reg,
2266 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2267 mark_map_regs(other_branch, insn->dst_reg,
2268 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2269 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2270 dst_reg->type == PTR_TO_PACKET &&
2271 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2272 find_good_pkt_pointers(this_branch, dst_reg);
2273 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2274 dst_reg->type == PTR_TO_PACKET_END &&
2275 regs[insn->src_reg].type == PTR_TO_PACKET) {
2276 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]);
2277 } else if (is_pointer_value(env, insn->dst_reg)) {
2278 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2282 print_verifier_state(this_branch);
2286 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2287 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2289 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2291 return (struct bpf_map *) (unsigned long) imm64;
2294 /* verify BPF_LD_IMM64 instruction */
2295 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2297 struct bpf_reg_state *regs = env->cur_state.regs;
2300 if (BPF_SIZE(insn->code) != BPF_DW) {
2301 verbose("invalid BPF_LD_IMM insn\n");
2304 if (insn->off != 0) {
2305 verbose("BPF_LD_IMM64 uses reserved fields\n");
2309 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2313 if (insn->src_reg == 0) {
2314 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2316 regs[insn->dst_reg].type = CONST_IMM;
2317 regs[insn->dst_reg].imm = imm;
2321 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2322 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2324 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2325 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2329 static bool may_access_skb(enum bpf_prog_type type)
2332 case BPF_PROG_TYPE_SOCKET_FILTER:
2333 case BPF_PROG_TYPE_SCHED_CLS:
2334 case BPF_PROG_TYPE_SCHED_ACT:
2341 /* verify safety of LD_ABS|LD_IND instructions:
2342 * - they can only appear in the programs where ctx == skb
2343 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2344 * preserve R6-R9, and store return value into R0
2347 * ctx == skb == R6 == CTX
2350 * SRC == any register
2351 * IMM == 32-bit immediate
2354 * R0 - 8/16/32-bit skb data converted to cpu endianness
2356 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2358 struct bpf_reg_state *regs = env->cur_state.regs;
2359 u8 mode = BPF_MODE(insn->code);
2360 struct bpf_reg_state *reg;
2363 if (!may_access_skb(env->prog->type)) {
2364 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2368 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2369 BPF_SIZE(insn->code) == BPF_DW ||
2370 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2371 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2375 /* check whether implicit source operand (register R6) is readable */
2376 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2380 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2381 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2385 if (mode == BPF_IND) {
2386 /* check explicit source operand */
2387 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2392 /* reset caller saved regs to unreadable */
2393 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2394 reg = regs + caller_saved[i];
2395 reg->type = NOT_INIT;
2399 /* mark destination R0 register as readable, since it contains
2400 * the value fetched from the packet
2402 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2406 /* non-recursive DFS pseudo code
2407 * 1 procedure DFS-iterative(G,v):
2408 * 2 label v as discovered
2409 * 3 let S be a stack
2411 * 5 while S is not empty
2413 * 7 if t is what we're looking for:
2415 * 9 for all edges e in G.adjacentEdges(t) do
2416 * 10 if edge e is already labelled
2417 * 11 continue with the next edge
2418 * 12 w <- G.adjacentVertex(t,e)
2419 * 13 if vertex w is not discovered and not explored
2420 * 14 label e as tree-edge
2421 * 15 label w as discovered
2424 * 18 else if vertex w is discovered
2425 * 19 label e as back-edge
2427 * 21 // vertex w is explored
2428 * 22 label e as forward- or cross-edge
2429 * 23 label t as explored
2434 * 0x11 - discovered and fall-through edge labelled
2435 * 0x12 - discovered and fall-through and branch edges labelled
2446 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2448 static int *insn_stack; /* stack of insns to process */
2449 static int cur_stack; /* current stack index */
2450 static int *insn_state;
2452 /* t, w, e - match pseudo-code above:
2453 * t - index of current instruction
2454 * w - next instruction
2457 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2459 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2462 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2465 if (w < 0 || w >= env->prog->len) {
2466 verbose("jump out of range from insn %d to %d\n", t, w);
2471 /* mark branch target for state pruning */
2472 env->explored_states[w] = STATE_LIST_MARK;
2474 if (insn_state[w] == 0) {
2476 insn_state[t] = DISCOVERED | e;
2477 insn_state[w] = DISCOVERED;
2478 if (cur_stack >= env->prog->len)
2480 insn_stack[cur_stack++] = w;
2482 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2483 verbose("back-edge from insn %d to %d\n", t, w);
2485 } else if (insn_state[w] == EXPLORED) {
2486 /* forward- or cross-edge */
2487 insn_state[t] = DISCOVERED | e;
2489 verbose("insn state internal bug\n");
2495 /* non-recursive depth-first-search to detect loops in BPF program
2496 * loop == back-edge in directed graph
2498 static int check_cfg(struct bpf_verifier_env *env)
2500 struct bpf_insn *insns = env->prog->insnsi;
2501 int insn_cnt = env->prog->len;
2505 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2509 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2515 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2516 insn_stack[0] = 0; /* 0 is the first instruction */
2522 t = insn_stack[cur_stack - 1];
2524 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2525 u8 opcode = BPF_OP(insns[t].code);
2527 if (opcode == BPF_EXIT) {
2529 } else if (opcode == BPF_CALL) {
2530 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2535 if (t + 1 < insn_cnt)
2536 env->explored_states[t + 1] = STATE_LIST_MARK;
2537 } else if (opcode == BPF_JA) {
2538 if (BPF_SRC(insns[t].code) != BPF_K) {
2542 /* unconditional jump with single edge */
2543 ret = push_insn(t, t + insns[t].off + 1,
2549 /* tell verifier to check for equivalent states
2550 * after every call and jump
2552 if (t + 1 < insn_cnt)
2553 env->explored_states[t + 1] = STATE_LIST_MARK;
2555 /* conditional jump with two edges */
2556 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2562 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2569 /* all other non-branch instructions with single
2572 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2580 insn_state[t] = EXPLORED;
2581 if (cur_stack-- <= 0) {
2582 verbose("pop stack internal bug\n");
2589 for (i = 0; i < insn_cnt; i++) {
2590 if (insn_state[i] != EXPLORED) {
2591 verbose("unreachable insn %d\n", i);
2596 ret = 0; /* cfg looks good */
2604 /* the following conditions reduce the number of explored insns
2605 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2607 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2608 struct bpf_reg_state *cur)
2610 if (old->id != cur->id)
2613 /* old ptr_to_packet is more conservative, since it allows smaller
2615 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2616 * old(off=0,r=10) means that with range=10 the verifier proceeded
2617 * further and found no issues with the program. Now we're in the same
2618 * spot with cur(off=0,r=20), so we're safe too, since anything further
2619 * will only be looking at most 10 bytes after this pointer.
2621 if (old->off == cur->off && old->range < cur->range)
2624 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2625 * since both cannot be used for packet access and safe(old)
2626 * pointer has smaller off that could be used for further
2627 * 'if (ptr > data_end)' check
2629 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2630 * that we cannot access the packet.
2631 * The safe range is:
2632 * [ptr, ptr + range - off)
2633 * so whenever off >=range, it means no safe bytes from this pointer.
2634 * When comparing old->off <= cur->off, it means that older code
2635 * went with smaller offset and that offset was later
2636 * used to figure out the safe range after 'if (ptr > data_end)' check
2637 * Say, 'old' state was explored like:
2638 * ... R3(off=0, r=0)
2640 * ... now R4(off=20,r=0) <-- here
2641 * if (R4 > data_end)
2642 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2643 * ... the code further went all the way to bpf_exit.
2644 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2645 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2646 * goes further, such cur_R4 will give larger safe packet range after
2647 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2648 * so they will be good with r=30 and we can prune the search.
2650 if (old->off <= cur->off &&
2651 old->off >= old->range && cur->off >= cur->range)
2657 /* compare two verifier states
2659 * all states stored in state_list are known to be valid, since
2660 * verifier reached 'bpf_exit' instruction through them
2662 * this function is called when verifier exploring different branches of
2663 * execution popped from the state stack. If it sees an old state that has
2664 * more strict register state and more strict stack state then this execution
2665 * branch doesn't need to be explored further, since verifier already
2666 * concluded that more strict state leads to valid finish.
2668 * Therefore two states are equivalent if register state is more conservative
2669 * and explored stack state is more conservative than the current one.
2672 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2673 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2675 * In other words if current stack state (one being explored) has more
2676 * valid slots than old one that already passed validation, it means
2677 * the verifier can stop exploring and conclude that current state is valid too
2679 * Similarly with registers. If explored state has register type as invalid
2680 * whereas register type in current state is meaningful, it means that
2681 * the current state will reach 'bpf_exit' instruction safely
2683 static bool states_equal(struct bpf_verifier_env *env,
2684 struct bpf_verifier_state *old,
2685 struct bpf_verifier_state *cur)
2687 bool varlen_map_access = env->varlen_map_value_access;
2688 struct bpf_reg_state *rold, *rcur;
2691 for (i = 0; i < MAX_BPF_REG; i++) {
2692 rold = &old->regs[i];
2693 rcur = &cur->regs[i];
2695 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2698 /* If the ranges were not the same, but everything else was and
2699 * we didn't do a variable access into a map then we are a-ok.
2701 if (!varlen_map_access &&
2702 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2705 /* If we didn't map access then again we don't care about the
2706 * mismatched range values and it's ok if our old type was
2707 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2709 if (rold->type == NOT_INIT ||
2710 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2711 rcur->type != NOT_INIT))
2714 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2715 compare_ptrs_to_packet(rold, rcur))
2721 for (i = 0; i < MAX_BPF_STACK; i++) {
2722 if (old->stack_slot_type[i] == STACK_INVALID)
2724 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2725 /* Ex: old explored (safe) state has STACK_SPILL in
2726 * this stack slot, but current has has STACK_MISC ->
2727 * this verifier states are not equivalent,
2728 * return false to continue verification of this path
2731 if (i % BPF_REG_SIZE)
2733 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2734 &cur->spilled_regs[i / BPF_REG_SIZE],
2735 sizeof(old->spilled_regs[0])))
2736 /* when explored and current stack slot types are
2737 * the same, check that stored pointers types
2738 * are the same as well.
2739 * Ex: explored safe path could have stored
2740 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2741 * but current path has stored:
2742 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2743 * such verifier states are not equivalent.
2744 * return false to continue verification of this path
2753 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2755 struct bpf_verifier_state_list *new_sl;
2756 struct bpf_verifier_state_list *sl;
2758 sl = env->explored_states[insn_idx];
2760 /* this 'insn_idx' instruction wasn't marked, so we will not
2761 * be doing state search here
2765 while (sl != STATE_LIST_MARK) {
2766 if (states_equal(env, &sl->state, &env->cur_state))
2767 /* reached equivalent register/stack state,
2774 /* there were no equivalent states, remember current one.
2775 * technically the current state is not proven to be safe yet,
2776 * but it will either reach bpf_exit (which means it's safe) or
2777 * it will be rejected. Since there are no loops, we won't be
2778 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2780 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2784 /* add new state to the head of linked list */
2785 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2786 new_sl->next = env->explored_states[insn_idx];
2787 env->explored_states[insn_idx] = new_sl;
2791 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2792 int insn_idx, int prev_insn_idx)
2794 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2797 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2800 static int do_check(struct bpf_verifier_env *env)
2802 struct bpf_verifier_state *state = &env->cur_state;
2803 struct bpf_insn *insns = env->prog->insnsi;
2804 struct bpf_reg_state *regs = state->regs;
2805 int insn_cnt = env->prog->len;
2806 int insn_idx, prev_insn_idx = 0;
2807 int insn_processed = 0;
2808 bool do_print_state = false;
2810 init_reg_state(regs);
2812 env->varlen_map_value_access = false;
2814 struct bpf_insn *insn;
2818 if (insn_idx >= insn_cnt) {
2819 verbose("invalid insn idx %d insn_cnt %d\n",
2820 insn_idx, insn_cnt);
2824 insn = &insns[insn_idx];
2825 class = BPF_CLASS(insn->code);
2827 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2828 verbose("BPF program is too large. Processed %d insn\n",
2833 err = is_state_visited(env, insn_idx);
2837 /* found equivalent state, can prune the search */
2840 verbose("\nfrom %d to %d: safe\n",
2841 prev_insn_idx, insn_idx);
2843 verbose("%d: safe\n", insn_idx);
2845 goto process_bpf_exit;
2848 if (log_level && do_print_state) {
2849 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2850 print_verifier_state(&env->cur_state);
2851 do_print_state = false;
2855 verbose("%d: ", insn_idx);
2856 print_bpf_insn(insn);
2859 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2863 if (class == BPF_ALU || class == BPF_ALU64) {
2864 err = check_alu_op(env, insn);
2868 } else if (class == BPF_LDX) {
2869 enum bpf_reg_type *prev_src_type, src_reg_type;
2871 /* check for reserved fields is already done */
2873 /* check src operand */
2874 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2878 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2882 src_reg_type = regs[insn->src_reg].type;
2884 /* check that memory (src_reg + off) is readable,
2885 * the state of dst_reg will be updated by this func
2887 err = check_mem_access(env, insn->src_reg, insn->off,
2888 BPF_SIZE(insn->code), BPF_READ,
2893 if (BPF_SIZE(insn->code) != BPF_W &&
2894 BPF_SIZE(insn->code) != BPF_DW) {
2899 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2901 if (*prev_src_type == NOT_INIT) {
2903 * dst_reg = *(u32 *)(src_reg + off)
2904 * save type to validate intersecting paths
2906 *prev_src_type = src_reg_type;
2908 } else if (src_reg_type != *prev_src_type &&
2909 (src_reg_type == PTR_TO_CTX ||
2910 *prev_src_type == PTR_TO_CTX)) {
2911 /* ABuser program is trying to use the same insn
2912 * dst_reg = *(u32*) (src_reg + off)
2913 * with different pointer types:
2914 * src_reg == ctx in one branch and
2915 * src_reg == stack|map in some other branch.
2918 verbose("same insn cannot be used with different pointers\n");
2922 } else if (class == BPF_STX) {
2923 enum bpf_reg_type *prev_dst_type, dst_reg_type;
2925 if (BPF_MODE(insn->code) == BPF_XADD) {
2926 err = check_xadd(env, insn);
2933 /* check src1 operand */
2934 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2937 /* check src2 operand */
2938 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2942 dst_reg_type = regs[insn->dst_reg].type;
2944 /* check that memory (dst_reg + off) is writeable */
2945 err = check_mem_access(env, insn->dst_reg, insn->off,
2946 BPF_SIZE(insn->code), BPF_WRITE,
2951 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2953 if (*prev_dst_type == NOT_INIT) {
2954 *prev_dst_type = dst_reg_type;
2955 } else if (dst_reg_type != *prev_dst_type &&
2956 (dst_reg_type == PTR_TO_CTX ||
2957 *prev_dst_type == PTR_TO_CTX)) {
2958 verbose("same insn cannot be used with different pointers\n");
2962 } else if (class == BPF_ST) {
2963 if (BPF_MODE(insn->code) != BPF_MEM ||
2964 insn->src_reg != BPF_REG_0) {
2965 verbose("BPF_ST uses reserved fields\n");
2968 /* check src operand */
2969 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2973 /* check that memory (dst_reg + off) is writeable */
2974 err = check_mem_access(env, insn->dst_reg, insn->off,
2975 BPF_SIZE(insn->code), BPF_WRITE,
2980 } else if (class == BPF_JMP) {
2981 u8 opcode = BPF_OP(insn->code);
2983 if (opcode == BPF_CALL) {
2984 if (BPF_SRC(insn->code) != BPF_K ||
2986 insn->src_reg != BPF_REG_0 ||
2987 insn->dst_reg != BPF_REG_0) {
2988 verbose("BPF_CALL uses reserved fields\n");
2992 err = check_call(env, insn->imm, insn_idx);
2996 } else if (opcode == BPF_JA) {
2997 if (BPF_SRC(insn->code) != BPF_K ||
2999 insn->src_reg != BPF_REG_0 ||
3000 insn->dst_reg != BPF_REG_0) {
3001 verbose("BPF_JA uses reserved fields\n");
3005 insn_idx += insn->off + 1;
3008 } else if (opcode == BPF_EXIT) {
3009 if (BPF_SRC(insn->code) != BPF_K ||
3011 insn->src_reg != BPF_REG_0 ||
3012 insn->dst_reg != BPF_REG_0) {
3013 verbose("BPF_EXIT uses reserved fields\n");
3017 /* eBPF calling convetion is such that R0 is used
3018 * to return the value from eBPF program.
3019 * Make sure that it's readable at this time
3020 * of bpf_exit, which means that program wrote
3021 * something into it earlier
3023 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
3027 if (is_pointer_value(env, BPF_REG_0)) {
3028 verbose("R0 leaks addr as return value\n");
3033 insn_idx = pop_stack(env, &prev_insn_idx);
3037 do_print_state = true;
3041 err = check_cond_jmp_op(env, insn, &insn_idx);
3045 } else if (class == BPF_LD) {
3046 u8 mode = BPF_MODE(insn->code);
3048 if (mode == BPF_ABS || mode == BPF_IND) {
3049 err = check_ld_abs(env, insn);
3053 } else if (mode == BPF_IMM) {
3054 err = check_ld_imm(env, insn);
3060 verbose("invalid BPF_LD mode\n");
3063 reset_reg_range_values(regs, insn->dst_reg);
3065 verbose("unknown insn class %d\n", class);
3072 verbose("processed %d insns\n", insn_processed);
3076 static int check_map_prealloc(struct bpf_map *map)
3078 return (map->map_type != BPF_MAP_TYPE_HASH &&
3079 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
3080 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
3081 !(map->map_flags & BPF_F_NO_PREALLOC);
3084 static int check_map_prog_compatibility(struct bpf_map *map,
3085 struct bpf_prog *prog)
3088 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3089 * preallocated hash maps, since doing memory allocation
3090 * in overflow_handler can crash depending on where nmi got
3093 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
3094 if (!check_map_prealloc(map)) {
3095 verbose("perf_event programs can only use preallocated hash map\n");
3098 if (map->inner_map_meta &&
3099 !check_map_prealloc(map->inner_map_meta)) {
3100 verbose("perf_event programs can only use preallocated inner hash map\n");
3107 /* look for pseudo eBPF instructions that access map FDs and
3108 * replace them with actual map pointers
3110 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3112 struct bpf_insn *insn = env->prog->insnsi;
3113 int insn_cnt = env->prog->len;
3116 err = bpf_prog_calc_tag(env->prog);
3120 for (i = 0; i < insn_cnt; i++, insn++) {
3121 if (BPF_CLASS(insn->code) == BPF_LDX &&
3122 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3123 verbose("BPF_LDX uses reserved fields\n");
3127 if (BPF_CLASS(insn->code) == BPF_STX &&
3128 ((BPF_MODE(insn->code) != BPF_MEM &&
3129 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3130 verbose("BPF_STX uses reserved fields\n");
3134 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3135 struct bpf_map *map;
3138 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3139 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3141 verbose("invalid bpf_ld_imm64 insn\n");
3145 if (insn->src_reg == 0)
3146 /* valid generic load 64-bit imm */
3149 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3150 verbose("unrecognized bpf_ld_imm64 insn\n");
3154 f = fdget(insn->imm);
3155 map = __bpf_map_get(f);
3157 verbose("fd %d is not pointing to valid bpf_map\n",
3159 return PTR_ERR(map);
3162 err = check_map_prog_compatibility(map, env->prog);
3168 /* store map pointer inside BPF_LD_IMM64 instruction */
3169 insn[0].imm = (u32) (unsigned long) map;
3170 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3172 /* check whether we recorded this map already */
3173 for (j = 0; j < env->used_map_cnt; j++)
3174 if (env->used_maps[j] == map) {
3179 if (env->used_map_cnt >= MAX_USED_MAPS) {
3184 /* hold the map. If the program is rejected by verifier,
3185 * the map will be released by release_maps() or it
3186 * will be used by the valid program until it's unloaded
3187 * and all maps are released in free_bpf_prog_info()
3189 map = bpf_map_inc(map, false);
3192 return PTR_ERR(map);
3194 env->used_maps[env->used_map_cnt++] = map;
3203 /* now all pseudo BPF_LD_IMM64 instructions load valid
3204 * 'struct bpf_map *' into a register instead of user map_fd.
3205 * These pointers will be used later by verifier to validate map access.
3210 /* drop refcnt of maps used by the rejected program */
3211 static void release_maps(struct bpf_verifier_env *env)
3215 for (i = 0; i < env->used_map_cnt; i++)
3216 bpf_map_put(env->used_maps[i]);
3219 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3220 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3222 struct bpf_insn *insn = env->prog->insnsi;
3223 int insn_cnt = env->prog->len;
3226 for (i = 0; i < insn_cnt; i++, insn++)
3227 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3231 /* single env->prog->insni[off] instruction was replaced with the range
3232 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3233 * [0, off) and [off, end) to new locations, so the patched range stays zero
3235 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3238 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3242 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3245 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3246 memcpy(new_data + off + cnt - 1, old_data + off,
3247 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3248 env->insn_aux_data = new_data;
3253 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3254 const struct bpf_insn *patch, u32 len)
3256 struct bpf_prog *new_prog;
3258 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3261 if (adjust_insn_aux_data(env, new_prog->len, off, len))
3266 /* convert load instructions that access fields of 'struct __sk_buff'
3267 * into sequence of instructions that access fields of 'struct sk_buff'
3269 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3271 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3272 const int insn_cnt = env->prog->len;
3273 struct bpf_insn insn_buf[16], *insn;
3274 struct bpf_prog *new_prog;
3275 enum bpf_access_type type;
3276 int i, cnt, delta = 0;
3278 if (ops->gen_prologue) {
3279 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3281 if (cnt >= ARRAY_SIZE(insn_buf)) {
3282 verbose("bpf verifier is misconfigured\n");
3285 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3289 env->prog = new_prog;
3294 if (!ops->convert_ctx_access)
3297 insn = env->prog->insnsi + delta;
3299 for (i = 0; i < insn_cnt; i++, insn++) {
3300 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3301 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3302 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3303 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3305 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3306 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3307 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3308 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3313 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3316 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3317 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3318 verbose("bpf verifier is misconfigured\n");
3322 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3328 /* keep walking new program and skip insns we just inserted */
3329 env->prog = new_prog;
3330 insn = new_prog->insnsi + i + delta;
3336 /* fixup insn->imm field of bpf_call instructions
3337 * and inline eligible helpers as explicit sequence of BPF instructions
3339 * this function is called after eBPF program passed verification
3341 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3343 struct bpf_prog *prog = env->prog;
3344 struct bpf_insn *insn = prog->insnsi;
3345 const struct bpf_func_proto *fn;
3346 const int insn_cnt = prog->len;
3347 struct bpf_insn insn_buf[16];
3348 struct bpf_prog *new_prog;
3349 struct bpf_map *map_ptr;
3350 int i, cnt, delta = 0;
3352 for (i = 0; i < insn_cnt; i++, insn++) {
3353 if (insn->code != (BPF_JMP | BPF_CALL))
3356 if (insn->imm == BPF_FUNC_get_route_realm)
3357 prog->dst_needed = 1;
3358 if (insn->imm == BPF_FUNC_get_prandom_u32)
3359 bpf_user_rnd_init_once();
3360 if (insn->imm == BPF_FUNC_tail_call) {
3361 /* If we tail call into other programs, we
3362 * cannot make any assumptions since they can
3363 * be replaced dynamically during runtime in
3364 * the program array.
3366 prog->cb_access = 1;
3368 /* mark bpf_tail_call as different opcode to avoid
3369 * conditional branch in the interpeter for every normal
3370 * call and to prevent accidental JITing by JIT compiler
3371 * that doesn't support bpf_tail_call yet
3374 insn->code |= BPF_X;
3378 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3379 map_ptr = env->insn_aux_data[i + delta].map_ptr;
3380 if (map_ptr == BPF_MAP_PTR_POISON ||
3381 !map_ptr->ops->map_gen_lookup)
3382 goto patch_call_imm;
3384 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3385 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3386 verbose("bpf verifier is misconfigured\n");
3390 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3397 /* keep walking new program and skip insns we just inserted */
3398 env->prog = prog = new_prog;
3399 insn = new_prog->insnsi + i + delta;
3404 fn = prog->aux->ops->get_func_proto(insn->imm);
3405 /* all functions that have prototype and verifier allowed
3406 * programs to call them, must be real in-kernel functions
3409 verbose("kernel subsystem misconfigured func %s#%d\n",
3410 func_id_name(insn->imm), insn->imm);
3413 insn->imm = fn->func - __bpf_call_base;
3419 static void free_states(struct bpf_verifier_env *env)
3421 struct bpf_verifier_state_list *sl, *sln;
3424 if (!env->explored_states)
3427 for (i = 0; i < env->prog->len; i++) {
3428 sl = env->explored_states[i];
3431 while (sl != STATE_LIST_MARK) {
3438 kfree(env->explored_states);
3441 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3443 char __user *log_ubuf = NULL;
3444 struct bpf_verifier_env *env;
3447 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3448 * allocate/free it every time bpf_check() is called
3450 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3454 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3457 if (!env->insn_aux_data)
3461 /* grab the mutex to protect few globals used by verifier */
3462 mutex_lock(&bpf_verifier_lock);
3464 if (attr->log_level || attr->log_buf || attr->log_size) {
3465 /* user requested verbose verifier output
3466 * and supplied buffer to store the verification trace
3468 log_level = attr->log_level;
3469 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3470 log_size = attr->log_size;
3474 /* log_* values have to be sane */
3475 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3476 log_level == 0 || log_ubuf == NULL)
3480 log_buf = vmalloc(log_size);
3487 ret = replace_map_fd_with_map_ptr(env);
3489 goto skip_full_check;
3491 env->explored_states = kcalloc(env->prog->len,
3492 sizeof(struct bpf_verifier_state_list *),
3495 if (!env->explored_states)
3496 goto skip_full_check;
3498 ret = check_cfg(env);
3500 goto skip_full_check;
3502 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3504 ret = do_check(env);
3507 while (pop_stack(env, NULL) >= 0);
3511 /* program is valid, convert *(u32*)(ctx + off) accesses */
3512 ret = convert_ctx_accesses(env);
3515 ret = fixup_bpf_calls(env);
3517 if (log_level && log_len >= log_size - 1) {
3518 BUG_ON(log_len >= log_size);
3519 /* verifier log exceeded user supplied buffer */
3521 /* fall through to return what was recorded */
3524 /* copy verifier log back to user space including trailing zero */
3525 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3530 if (ret == 0 && env->used_map_cnt) {
3531 /* if program passed verifier, update used_maps in bpf_prog_info */
3532 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3533 sizeof(env->used_maps[0]),
3536 if (!env->prog->aux->used_maps) {
3541 memcpy(env->prog->aux->used_maps, env->used_maps,
3542 sizeof(env->used_maps[0]) * env->used_map_cnt);
3543 env->prog->aux->used_map_cnt = env->used_map_cnt;
3545 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3546 * bpf_ld_imm64 instructions
3548 convert_pseudo_ld_imm64(env);
3554 if (!env->prog->aux->used_maps)
3555 /* if we didn't copy map pointers into bpf_prog_info, release
3556 * them now. Otherwise free_bpf_prog_info() will release them.
3561 mutex_unlock(&bpf_verifier_lock);
3562 vfree(env->insn_aux_data);
3568 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3571 struct bpf_verifier_env *env;
3574 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3578 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3581 if (!env->insn_aux_data)
3584 env->analyzer_ops = ops;
3585 env->analyzer_priv = priv;
3587 /* grab the mutex to protect few globals used by verifier */
3588 mutex_lock(&bpf_verifier_lock);
3592 env->explored_states = kcalloc(env->prog->len,
3593 sizeof(struct bpf_verifier_state_list *),
3596 if (!env->explored_states)
3597 goto skip_full_check;
3599 ret = check_cfg(env);
3601 goto skip_full_check;
3603 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3605 ret = do_check(env);
3608 while (pop_stack(env, NULL) >= 0);
3611 mutex_unlock(&bpf_verifier_lock);
3612 vfree(env->insn_aux_data);
3617 EXPORT_SYMBOL_GPL(bpf_analyzer);