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_ptr_alignment(struct bpf_verifier_env *env,
771 struct bpf_reg_state *reg, int off, int size)
773 if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
774 if (off % size != 0) {
775 verbose("misaligned access off %d size %d\n",
783 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
784 /* misaligned access to packet is ok on x86,arm,arm64 */
787 if (reg->id && size != 1) {
788 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
792 /* skb->data is NET_IP_ALIGN-ed */
793 if (reg->type == PTR_TO_PACKET &&
794 (NET_IP_ALIGN + reg->off + off) % size != 0) {
795 verbose("misaligned packet access off %d+%d+%d size %d\n",
796 NET_IP_ALIGN, reg->off, off, size);
802 /* check whether memory at (regno + off) is accessible for t = (read | write)
803 * if t==write, value_regno is a register which value is stored into memory
804 * if t==read, value_regno is a register which will receive the value from memory
805 * if t==write && value_regno==-1, some unknown value is stored into memory
806 * if t==read && value_regno==-1, don't care what we read from memory
808 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
809 int bpf_size, enum bpf_access_type t,
812 struct bpf_verifier_state *state = &env->cur_state;
813 struct bpf_reg_state *reg = &state->regs[regno];
816 if (reg->type == PTR_TO_STACK)
819 size = bpf_size_to_bytes(bpf_size);
823 err = check_ptr_alignment(env, reg, off, size);
827 if (reg->type == PTR_TO_MAP_VALUE ||
828 reg->type == PTR_TO_MAP_VALUE_ADJ) {
829 if (t == BPF_WRITE && value_regno >= 0 &&
830 is_pointer_value(env, value_regno)) {
831 verbose("R%d leaks addr into map\n", value_regno);
835 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
836 err = check_map_access_adj(env, regno, off, size);
838 err = check_map_access(env, regno, off, size);
839 if (!err && t == BPF_READ && value_regno >= 0)
840 mark_reg_unknown_value_and_range(state->regs,
843 } else if (reg->type == PTR_TO_CTX) {
844 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
846 if (t == BPF_WRITE && value_regno >= 0 &&
847 is_pointer_value(env, value_regno)) {
848 verbose("R%d leaks addr into ctx\n", value_regno);
851 err = check_ctx_access(env, off, size, t, ®_type);
852 if (!err && t == BPF_READ && value_regno >= 0) {
853 mark_reg_unknown_value_and_range(state->regs,
855 /* note that reg.[id|off|range] == 0 */
856 state->regs[value_regno].type = reg_type;
859 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
860 if (off >= 0 || off < -MAX_BPF_STACK) {
861 verbose("invalid stack off=%d size=%d\n", off, size);
864 if (t == BPF_WRITE) {
865 if (!env->allow_ptr_leaks &&
866 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
867 size != BPF_REG_SIZE) {
868 verbose("attempt to corrupt spilled pointer on stack\n");
871 err = check_stack_write(state, off, size, value_regno);
873 err = check_stack_read(state, off, size, value_regno);
875 } else if (state->regs[regno].type == PTR_TO_PACKET) {
876 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
877 verbose("cannot write into packet\n");
880 if (t == BPF_WRITE && value_regno >= 0 &&
881 is_pointer_value(env, value_regno)) {
882 verbose("R%d leaks addr into packet\n", value_regno);
885 err = check_packet_access(env, regno, off, size);
886 if (!err && t == BPF_READ && value_regno >= 0)
887 mark_reg_unknown_value_and_range(state->regs,
890 verbose("R%d invalid mem access '%s'\n",
891 regno, reg_type_str[reg->type]);
895 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
896 state->regs[value_regno].type == UNKNOWN_VALUE) {
897 /* 1 or 2 byte load zero-extends, determine the number of
898 * zero upper bits. Not doing it fo 4 byte load, since
899 * such values cannot be added to ptr_to_packet anyway.
901 state->regs[value_regno].imm = 64 - size * 8;
906 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
908 struct bpf_reg_state *regs = env->cur_state.regs;
911 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
913 verbose("BPF_XADD uses reserved fields\n");
917 /* check src1 operand */
918 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
922 /* check src2 operand */
923 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
927 /* check whether atomic_add can read the memory */
928 err = check_mem_access(env, insn->dst_reg, insn->off,
929 BPF_SIZE(insn->code), BPF_READ, -1);
933 /* check whether atomic_add can write into the same memory */
934 return check_mem_access(env, insn->dst_reg, insn->off,
935 BPF_SIZE(insn->code), BPF_WRITE, -1);
938 /* when register 'regno' is passed into function that will read 'access_size'
939 * bytes from that pointer, make sure that it's within stack boundary
940 * and all elements of stack are initialized
942 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
943 int access_size, bool zero_size_allowed,
944 struct bpf_call_arg_meta *meta)
946 struct bpf_verifier_state *state = &env->cur_state;
947 struct bpf_reg_state *regs = state->regs;
950 if (regs[regno].type != PTR_TO_STACK) {
951 if (zero_size_allowed && access_size == 0 &&
952 regs[regno].type == CONST_IMM &&
953 regs[regno].imm == 0)
956 verbose("R%d type=%s expected=%s\n", regno,
957 reg_type_str[regs[regno].type],
958 reg_type_str[PTR_TO_STACK]);
962 off = regs[regno].imm;
963 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
965 verbose("invalid stack type R%d off=%d access_size=%d\n",
966 regno, off, access_size);
970 if (meta && meta->raw_mode) {
971 meta->access_size = access_size;
976 for (i = 0; i < access_size; i++) {
977 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
978 verbose("invalid indirect read from stack off %d+%d size %d\n",
979 off, i, access_size);
986 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
987 int access_size, bool zero_size_allowed,
988 struct bpf_call_arg_meta *meta)
990 struct bpf_reg_state *regs = env->cur_state.regs;
992 switch (regs[regno].type) {
994 return check_packet_access(env, regno, 0, access_size);
995 case PTR_TO_MAP_VALUE:
996 return check_map_access(env, regno, 0, access_size);
997 case PTR_TO_MAP_VALUE_ADJ:
998 return check_map_access_adj(env, regno, 0, access_size);
999 default: /* const_imm|ptr_to_stack or invalid ptr */
1000 return check_stack_boundary(env, regno, access_size,
1001 zero_size_allowed, meta);
1005 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1006 enum bpf_arg_type arg_type,
1007 struct bpf_call_arg_meta *meta)
1009 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno];
1010 enum bpf_reg_type expected_type, type = reg->type;
1013 if (arg_type == ARG_DONTCARE)
1016 if (type == NOT_INIT) {
1017 verbose("R%d !read_ok\n", regno);
1021 if (arg_type == ARG_ANYTHING) {
1022 if (is_pointer_value(env, regno)) {
1023 verbose("R%d leaks addr into helper function\n", regno);
1029 if (type == PTR_TO_PACKET &&
1030 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1031 verbose("helper access to the packet is not allowed\n");
1035 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1036 arg_type == ARG_PTR_TO_MAP_VALUE) {
1037 expected_type = PTR_TO_STACK;
1038 if (type != PTR_TO_PACKET && type != expected_type)
1040 } else if (arg_type == ARG_CONST_SIZE ||
1041 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1042 expected_type = CONST_IMM;
1043 /* One exception. Allow UNKNOWN_VALUE registers when the
1044 * boundaries are known and don't cause unsafe memory accesses
1046 if (type != UNKNOWN_VALUE && type != expected_type)
1048 } else if (arg_type == ARG_CONST_MAP_PTR) {
1049 expected_type = CONST_PTR_TO_MAP;
1050 if (type != expected_type)
1052 } else if (arg_type == ARG_PTR_TO_CTX) {
1053 expected_type = PTR_TO_CTX;
1054 if (type != expected_type)
1056 } else if (arg_type == ARG_PTR_TO_MEM ||
1057 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1058 expected_type = PTR_TO_STACK;
1059 /* One exception here. In case function allows for NULL to be
1060 * passed in as argument, it's a CONST_IMM type. Final test
1061 * happens during stack boundary checking.
1063 if (type == CONST_IMM && reg->imm == 0)
1064 /* final test in check_stack_boundary() */;
1065 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1066 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1068 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1070 verbose("unsupported arg_type %d\n", arg_type);
1074 if (arg_type == ARG_CONST_MAP_PTR) {
1075 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1076 meta->map_ptr = reg->map_ptr;
1077 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1078 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1079 * check that [key, key + map->key_size) are within
1080 * stack limits and initialized
1082 if (!meta->map_ptr) {
1083 /* in function declaration map_ptr must come before
1084 * map_key, so that it's verified and known before
1085 * we have to check map_key here. Otherwise it means
1086 * that kernel subsystem misconfigured verifier
1088 verbose("invalid map_ptr to access map->key\n");
1091 if (type == PTR_TO_PACKET)
1092 err = check_packet_access(env, regno, 0,
1093 meta->map_ptr->key_size);
1095 err = check_stack_boundary(env, regno,
1096 meta->map_ptr->key_size,
1098 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1099 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1100 * check [value, value + map->value_size) validity
1102 if (!meta->map_ptr) {
1103 /* kernel subsystem misconfigured verifier */
1104 verbose("invalid map_ptr to access map->value\n");
1107 if (type == PTR_TO_PACKET)
1108 err = check_packet_access(env, regno, 0,
1109 meta->map_ptr->value_size);
1111 err = check_stack_boundary(env, regno,
1112 meta->map_ptr->value_size,
1114 } else if (arg_type == ARG_CONST_SIZE ||
1115 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1116 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1118 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1119 * from stack pointer 'buf'. Check it
1120 * note: regno == len, regno - 1 == buf
1123 /* kernel subsystem misconfigured verifier */
1124 verbose("ARG_CONST_SIZE cannot be first argument\n");
1128 /* If the register is UNKNOWN_VALUE, the access check happens
1129 * using its boundaries. Otherwise, just use its imm
1131 if (type == UNKNOWN_VALUE) {
1132 /* For unprivileged variable accesses, disable raw
1133 * mode so that the program is required to
1134 * initialize all the memory that the helper could
1135 * just partially fill up.
1139 if (reg->min_value < 0) {
1140 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1145 if (reg->min_value == 0) {
1146 err = check_helper_mem_access(env, regno - 1, 0,
1153 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1154 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1158 err = check_helper_mem_access(env, regno - 1,
1160 zero_size_allowed, meta);
1164 /* register is CONST_IMM */
1165 err = check_helper_mem_access(env, regno - 1, reg->imm,
1166 zero_size_allowed, meta);
1172 verbose("R%d type=%s expected=%s\n", regno,
1173 reg_type_str[type], reg_type_str[expected_type]);
1177 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1182 /* We need a two way check, first is from map perspective ... */
1183 switch (map->map_type) {
1184 case BPF_MAP_TYPE_PROG_ARRAY:
1185 if (func_id != BPF_FUNC_tail_call)
1188 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1189 if (func_id != BPF_FUNC_perf_event_read &&
1190 func_id != BPF_FUNC_perf_event_output)
1193 case BPF_MAP_TYPE_STACK_TRACE:
1194 if (func_id != BPF_FUNC_get_stackid)
1197 case BPF_MAP_TYPE_CGROUP_ARRAY:
1198 if (func_id != BPF_FUNC_skb_under_cgroup &&
1199 func_id != BPF_FUNC_current_task_under_cgroup)
1206 /* ... and second from the function itself. */
1208 case BPF_FUNC_tail_call:
1209 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1212 case BPF_FUNC_perf_event_read:
1213 case BPF_FUNC_perf_event_output:
1214 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1217 case BPF_FUNC_get_stackid:
1218 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1221 case BPF_FUNC_current_task_under_cgroup:
1222 case BPF_FUNC_skb_under_cgroup:
1223 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1232 verbose("cannot pass map_type %d into func %s#%d\n",
1233 map->map_type, func_id_name(func_id), func_id);
1237 static int check_raw_mode(const struct bpf_func_proto *fn)
1241 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1243 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1245 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1247 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1249 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1252 return count > 1 ? -EINVAL : 0;
1255 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1257 struct bpf_verifier_state *state = &env->cur_state;
1258 struct bpf_reg_state *regs = state->regs, *reg;
1261 for (i = 0; i < MAX_BPF_REG; i++)
1262 if (regs[i].type == PTR_TO_PACKET ||
1263 regs[i].type == PTR_TO_PACKET_END)
1264 mark_reg_unknown_value(regs, i);
1266 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1267 if (state->stack_slot_type[i] != STACK_SPILL)
1269 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1270 if (reg->type != PTR_TO_PACKET &&
1271 reg->type != PTR_TO_PACKET_END)
1273 reg->type = UNKNOWN_VALUE;
1278 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1280 struct bpf_verifier_state *state = &env->cur_state;
1281 const struct bpf_func_proto *fn = NULL;
1282 struct bpf_reg_state *regs = state->regs;
1283 struct bpf_reg_state *reg;
1284 struct bpf_call_arg_meta meta;
1288 /* find function prototype */
1289 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1290 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1294 if (env->prog->aux->ops->get_func_proto)
1295 fn = env->prog->aux->ops->get_func_proto(func_id);
1298 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1302 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1303 if (!env->prog->gpl_compatible && fn->gpl_only) {
1304 verbose("cannot call GPL only function from proprietary program\n");
1308 changes_data = bpf_helper_changes_pkt_data(fn->func);
1310 memset(&meta, 0, sizeof(meta));
1311 meta.pkt_access = fn->pkt_access;
1313 /* We only support one arg being in raw mode at the moment, which
1314 * is sufficient for the helper functions we have right now.
1316 err = check_raw_mode(fn);
1318 verbose("kernel subsystem misconfigured func %s#%d\n",
1319 func_id_name(func_id), func_id);
1324 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1327 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1330 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1333 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1336 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1340 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1341 * is inferred from register state.
1343 for (i = 0; i < meta.access_size; i++) {
1344 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1349 /* reset caller saved regs */
1350 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1351 reg = regs + caller_saved[i];
1352 reg->type = NOT_INIT;
1356 /* update return register */
1357 if (fn->ret_type == RET_INTEGER) {
1358 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1359 } else if (fn->ret_type == RET_VOID) {
1360 regs[BPF_REG_0].type = NOT_INIT;
1361 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1362 struct bpf_insn_aux_data *insn_aux;
1364 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1365 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1366 /* remember map_ptr, so that check_map_access()
1367 * can check 'value_size' boundary of memory access
1368 * to map element returned from bpf_map_lookup_elem()
1370 if (meta.map_ptr == NULL) {
1371 verbose("kernel subsystem misconfigured verifier\n");
1374 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1375 regs[BPF_REG_0].id = ++env->id_gen;
1376 insn_aux = &env->insn_aux_data[insn_idx];
1377 if (!insn_aux->map_ptr)
1378 insn_aux->map_ptr = meta.map_ptr;
1379 else if (insn_aux->map_ptr != meta.map_ptr)
1380 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1382 verbose("unknown return type %d of func %s#%d\n",
1383 fn->ret_type, func_id_name(func_id), func_id);
1387 err = check_map_func_compatibility(meta.map_ptr, func_id);
1392 clear_all_pkt_pointers(env);
1396 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1397 struct bpf_insn *insn)
1399 struct bpf_reg_state *regs = env->cur_state.regs;
1400 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1401 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1402 struct bpf_reg_state tmp_reg;
1405 if (BPF_SRC(insn->code) == BPF_K) {
1406 /* pkt_ptr += imm */
1411 verbose("addition of negative constant to packet pointer is not allowed\n");
1414 if (imm >= MAX_PACKET_OFF ||
1415 imm + dst_reg->off >= MAX_PACKET_OFF) {
1416 verbose("constant %d is too large to add to packet pointer\n",
1420 /* a constant was added to pkt_ptr.
1421 * Remember it while keeping the same 'id'
1423 dst_reg->off += imm;
1425 if (src_reg->type == PTR_TO_PACKET) {
1426 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1427 tmp_reg = *dst_reg; /* save r7 state */
1428 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1429 src_reg = &tmp_reg; /* pretend it's src_reg state */
1430 /* if the checks below reject it, the copy won't matter,
1431 * since we're rejecting the whole program. If all ok,
1432 * then imm22 state will be added to r7
1433 * and r7 will be pkt(id=0,off=22,r=62) while
1434 * r6 will stay as pkt(id=0,off=0,r=62)
1438 if (src_reg->type == CONST_IMM) {
1439 /* pkt_ptr += reg where reg is known constant */
1443 /* disallow pkt_ptr += reg
1444 * if reg is not uknown_value with guaranteed zero upper bits
1445 * otherwise pkt_ptr may overflow and addition will become
1446 * subtraction which is not allowed
1448 if (src_reg->type != UNKNOWN_VALUE) {
1449 verbose("cannot add '%s' to ptr_to_packet\n",
1450 reg_type_str[src_reg->type]);
1453 if (src_reg->imm < 48) {
1454 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1458 /* dst_reg stays as pkt_ptr type and since some positive
1459 * integer value was added to the pointer, increment its 'id'
1461 dst_reg->id = ++env->id_gen;
1463 /* something was added to pkt_ptr, set range and off to zero */
1470 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1472 struct bpf_reg_state *regs = env->cur_state.regs;
1473 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1474 u8 opcode = BPF_OP(insn->code);
1477 /* for type == UNKNOWN_VALUE:
1478 * imm > 0 -> number of zero upper bits
1479 * imm == 0 -> don't track which is the same as all bits can be non-zero
1482 if (BPF_SRC(insn->code) == BPF_X) {
1483 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1485 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1486 dst_reg->imm && opcode == BPF_ADD) {
1488 * where both have zero upper bits. Adding them
1489 * can only result making one more bit non-zero
1490 * in the larger value.
1491 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1492 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1494 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1498 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1499 dst_reg->imm && opcode == BPF_ADD) {
1501 * where dreg has zero upper bits and sreg is const.
1502 * Adding them can only result making one more bit
1503 * non-zero in the larger value.
1505 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1506 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1510 /* all other cases non supported yet, just mark dst_reg */
1515 /* sign extend 32-bit imm into 64-bit to make sure that
1516 * negative values occupy bit 63. Note ilog2() would have
1517 * been incorrect, since sizeof(insn->imm) == 4
1519 imm_log2 = __ilog2_u64((long long)insn->imm);
1521 if (dst_reg->imm && opcode == BPF_LSH) {
1523 * if reg was a result of 2 byte load, then its imm == 48
1524 * which means that upper 48 bits are zero and shifting this reg
1525 * left by 4 would mean that upper 44 bits are still zero
1527 dst_reg->imm -= insn->imm;
1528 } else if (dst_reg->imm && opcode == BPF_MUL) {
1530 * if multiplying by 14 subtract 4
1531 * This is conservative calculation of upper zero bits.
1532 * It's not trying to special case insn->imm == 1 or 0 cases
1534 dst_reg->imm -= imm_log2 + 1;
1535 } else if (opcode == BPF_AND) {
1537 dst_reg->imm = 63 - imm_log2;
1538 } else if (dst_reg->imm && opcode == BPF_ADD) {
1540 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1542 } else if (opcode == BPF_RSH) {
1544 * which means that after right shift, upper bits will be zero
1545 * note that verifier already checked that
1546 * 0 <= imm < 64 for shift insn
1548 dst_reg->imm += insn->imm;
1549 if (unlikely(dst_reg->imm > 64))
1550 /* some dumb code did:
1553 * and all bits are zero now */
1556 /* all other alu ops, means that we don't know what will
1557 * happen to the value, mark it with unknown number of zero bits
1562 if (dst_reg->imm < 0) {
1563 /* all 64 bits of the register can contain non-zero bits
1564 * and such value cannot be added to ptr_to_packet, since it
1565 * may overflow, mark it as unknown to avoid further eval
1572 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1573 struct bpf_insn *insn)
1575 struct bpf_reg_state *regs = env->cur_state.regs;
1576 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1577 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1578 u8 opcode = BPF_OP(insn->code);
1579 u64 dst_imm = dst_reg->imm;
1581 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1582 * containing ALU ops. Don't care about overflow or negative
1583 * values, just add/sub/... them; registers are in u64.
1585 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1586 dst_imm += insn->imm;
1587 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1588 src_reg->type == CONST_IMM) {
1589 dst_imm += src_reg->imm;
1590 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1591 dst_imm -= insn->imm;
1592 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1593 src_reg->type == CONST_IMM) {
1594 dst_imm -= src_reg->imm;
1595 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1596 dst_imm *= insn->imm;
1597 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1598 src_reg->type == CONST_IMM) {
1599 dst_imm *= src_reg->imm;
1600 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1601 dst_imm |= insn->imm;
1602 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1603 src_reg->type == CONST_IMM) {
1604 dst_imm |= src_reg->imm;
1605 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1606 dst_imm &= insn->imm;
1607 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1608 src_reg->type == CONST_IMM) {
1609 dst_imm &= src_reg->imm;
1610 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1611 dst_imm >>= insn->imm;
1612 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1613 src_reg->type == CONST_IMM) {
1614 dst_imm >>= src_reg->imm;
1615 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1616 dst_imm <<= insn->imm;
1617 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1618 src_reg->type == CONST_IMM) {
1619 dst_imm <<= src_reg->imm;
1621 mark_reg_unknown_value(regs, insn->dst_reg);
1625 dst_reg->imm = dst_imm;
1630 static void check_reg_overflow(struct bpf_reg_state *reg)
1632 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1633 reg->max_value = BPF_REGISTER_MAX_RANGE;
1634 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1635 reg->min_value > BPF_REGISTER_MAX_RANGE)
1636 reg->min_value = BPF_REGISTER_MIN_RANGE;
1639 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1640 struct bpf_insn *insn)
1642 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1643 s64 min_val = BPF_REGISTER_MIN_RANGE;
1644 u64 max_val = BPF_REGISTER_MAX_RANGE;
1645 u8 opcode = BPF_OP(insn->code);
1647 dst_reg = ®s[insn->dst_reg];
1648 if (BPF_SRC(insn->code) == BPF_X) {
1649 check_reg_overflow(®s[insn->src_reg]);
1650 min_val = regs[insn->src_reg].min_value;
1651 max_val = regs[insn->src_reg].max_value;
1653 /* If the source register is a random pointer then the
1654 * min_value/max_value values represent the range of the known
1655 * accesses into that value, not the actual min/max value of the
1656 * register itself. In this case we have to reset the reg range
1657 * values so we know it is not safe to look at.
1659 if (regs[insn->src_reg].type != CONST_IMM &&
1660 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1661 min_val = BPF_REGISTER_MIN_RANGE;
1662 max_val = BPF_REGISTER_MAX_RANGE;
1664 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1665 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1666 min_val = max_val = insn->imm;
1669 /* We don't know anything about what was done to this register, mark it
1672 if (min_val == BPF_REGISTER_MIN_RANGE &&
1673 max_val == BPF_REGISTER_MAX_RANGE) {
1674 reset_reg_range_values(regs, insn->dst_reg);
1678 /* If one of our values was at the end of our ranges then we can't just
1679 * do our normal operations to the register, we need to set the values
1680 * to the min/max since they are undefined.
1682 if (min_val == BPF_REGISTER_MIN_RANGE)
1683 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1684 if (max_val == BPF_REGISTER_MAX_RANGE)
1685 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1689 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1690 dst_reg->min_value += min_val;
1691 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1692 dst_reg->max_value += max_val;
1695 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1696 dst_reg->min_value -= min_val;
1697 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1698 dst_reg->max_value -= max_val;
1701 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1702 dst_reg->min_value *= min_val;
1703 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1704 dst_reg->max_value *= max_val;
1707 /* Disallow AND'ing of negative numbers, ain't nobody got time
1708 * for that. Otherwise the minimum is 0 and the max is the max
1709 * value we could AND against.
1712 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1714 dst_reg->min_value = 0;
1715 dst_reg->max_value = max_val;
1718 /* Gotta have special overflow logic here, if we're shifting
1719 * more than MAX_RANGE then just assume we have an invalid
1722 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1723 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1724 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1725 dst_reg->min_value <<= min_val;
1727 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1728 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1729 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1730 dst_reg->max_value <<= max_val;
1733 /* RSH by a negative number is undefined, and the BPF_RSH is an
1734 * unsigned shift, so make the appropriate casts.
1736 if (min_val < 0 || dst_reg->min_value < 0)
1737 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1739 dst_reg->min_value =
1740 (u64)(dst_reg->min_value) >> min_val;
1741 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1742 dst_reg->max_value >>= max_val;
1745 reset_reg_range_values(regs, insn->dst_reg);
1749 check_reg_overflow(dst_reg);
1752 /* check validity of 32-bit and 64-bit arithmetic operations */
1753 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1755 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1756 u8 opcode = BPF_OP(insn->code);
1759 if (opcode == BPF_END || opcode == BPF_NEG) {
1760 if (opcode == BPF_NEG) {
1761 if (BPF_SRC(insn->code) != 0 ||
1762 insn->src_reg != BPF_REG_0 ||
1763 insn->off != 0 || insn->imm != 0) {
1764 verbose("BPF_NEG uses reserved fields\n");
1768 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1769 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1770 verbose("BPF_END uses reserved fields\n");
1775 /* check src operand */
1776 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1780 if (is_pointer_value(env, insn->dst_reg)) {
1781 verbose("R%d pointer arithmetic prohibited\n",
1786 /* check dest operand */
1787 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1791 } else if (opcode == BPF_MOV) {
1793 if (BPF_SRC(insn->code) == BPF_X) {
1794 if (insn->imm != 0 || insn->off != 0) {
1795 verbose("BPF_MOV uses reserved fields\n");
1799 /* check src operand */
1800 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1804 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1805 verbose("BPF_MOV uses reserved fields\n");
1810 /* check dest operand */
1811 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1815 /* we are setting our register to something new, we need to
1816 * reset its range values.
1818 reset_reg_range_values(regs, insn->dst_reg);
1820 if (BPF_SRC(insn->code) == BPF_X) {
1821 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1823 * copy register state to dest reg
1825 regs[insn->dst_reg] = regs[insn->src_reg];
1827 if (is_pointer_value(env, insn->src_reg)) {
1828 verbose("R%d partial copy of pointer\n",
1832 mark_reg_unknown_value(regs, insn->dst_reg);
1836 * remember the value we stored into this reg
1838 regs[insn->dst_reg].type = CONST_IMM;
1839 regs[insn->dst_reg].imm = insn->imm;
1840 regs[insn->dst_reg].max_value = insn->imm;
1841 regs[insn->dst_reg].min_value = insn->imm;
1844 } else if (opcode > BPF_END) {
1845 verbose("invalid BPF_ALU opcode %x\n", opcode);
1848 } else { /* all other ALU ops: and, sub, xor, add, ... */
1850 if (BPF_SRC(insn->code) == BPF_X) {
1851 if (insn->imm != 0 || insn->off != 0) {
1852 verbose("BPF_ALU uses reserved fields\n");
1855 /* check src1 operand */
1856 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1860 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1861 verbose("BPF_ALU uses reserved fields\n");
1866 /* check src2 operand */
1867 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1871 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1872 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1873 verbose("div by zero\n");
1877 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1878 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1879 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1881 if (insn->imm < 0 || insn->imm >= size) {
1882 verbose("invalid shift %d\n", insn->imm);
1887 /* check dest operand */
1888 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1892 dst_reg = ®s[insn->dst_reg];
1894 /* first we want to adjust our ranges. */
1895 adjust_reg_min_max_vals(env, insn);
1897 /* pattern match 'bpf_add Rx, imm' instruction */
1898 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1899 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1900 dst_reg->type = PTR_TO_STACK;
1901 dst_reg->imm = insn->imm;
1903 } else if (opcode == BPF_ADD &&
1904 BPF_CLASS(insn->code) == BPF_ALU64 &&
1905 (dst_reg->type == PTR_TO_PACKET ||
1906 (BPF_SRC(insn->code) == BPF_X &&
1907 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1908 /* ptr_to_packet += K|X */
1909 return check_packet_ptr_add(env, insn);
1910 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1911 dst_reg->type == UNKNOWN_VALUE &&
1912 env->allow_ptr_leaks) {
1913 /* unknown += K|X */
1914 return evaluate_reg_alu(env, insn);
1915 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1916 dst_reg->type == CONST_IMM &&
1917 env->allow_ptr_leaks) {
1918 /* reg_imm += K|X */
1919 return evaluate_reg_imm_alu(env, insn);
1920 } else if (is_pointer_value(env, insn->dst_reg)) {
1921 verbose("R%d pointer arithmetic prohibited\n",
1924 } else if (BPF_SRC(insn->code) == BPF_X &&
1925 is_pointer_value(env, insn->src_reg)) {
1926 verbose("R%d pointer arithmetic prohibited\n",
1931 /* If we did pointer math on a map value then just set it to our
1932 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1933 * loads to this register appropriately, otherwise just mark the
1934 * register as unknown.
1936 if (env->allow_ptr_leaks &&
1937 (dst_reg->type == PTR_TO_MAP_VALUE ||
1938 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1939 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1941 mark_reg_unknown_value(regs, insn->dst_reg);
1947 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1948 struct bpf_reg_state *dst_reg)
1950 struct bpf_reg_state *regs = state->regs, *reg;
1953 /* LLVM can generate two kind of checks:
1959 * if (r2 > pkt_end) goto <handle exception>
1963 * r2 == dst_reg, pkt_end == src_reg
1964 * r2=pkt(id=n,off=8,r=0)
1965 * r3=pkt(id=n,off=0,r=0)
1971 * if (pkt_end >= r2) goto <access okay>
1972 * <handle exception>
1975 * pkt_end == dst_reg, r2 == src_reg
1976 * r2=pkt(id=n,off=8,r=0)
1977 * r3=pkt(id=n,off=0,r=0)
1979 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1980 * so that range of bytes [r3, r3 + 8) is safe to access.
1983 for (i = 0; i < MAX_BPF_REG; i++)
1984 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1985 regs[i].range = dst_reg->off;
1987 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1988 if (state->stack_slot_type[i] != STACK_SPILL)
1990 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1991 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1992 reg->range = dst_reg->off;
1996 /* Adjusts the register min/max values in the case that the dst_reg is the
1997 * variable register that we are working on, and src_reg is a constant or we're
1998 * simply doing a BPF_K check.
2000 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2001 struct bpf_reg_state *false_reg, u64 val,
2006 /* If this is false then we know nothing Jon Snow, but if it is
2007 * true then we know for sure.
2009 true_reg->max_value = true_reg->min_value = val;
2012 /* If this is true we know nothing Jon Snow, but if it is false
2013 * we know the value for sure;
2015 false_reg->max_value = false_reg->min_value = val;
2018 /* Unsigned comparison, the minimum value is 0. */
2019 false_reg->min_value = 0;
2022 /* If this is false then we know the maximum val is val,
2023 * otherwise we know the min val is val+1.
2025 false_reg->max_value = val;
2026 true_reg->min_value = val + 1;
2029 /* Unsigned comparison, the minimum value is 0. */
2030 false_reg->min_value = 0;
2033 /* If this is false then we know the maximum value is val - 1,
2034 * otherwise we know the mimimum value is val.
2036 false_reg->max_value = val - 1;
2037 true_reg->min_value = val;
2043 check_reg_overflow(false_reg);
2044 check_reg_overflow(true_reg);
2047 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2048 * is the variable reg.
2050 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2051 struct bpf_reg_state *false_reg, u64 val,
2056 /* If this is false then we know nothing Jon Snow, but if it is
2057 * true then we know for sure.
2059 true_reg->max_value = true_reg->min_value = val;
2062 /* If this is true we know nothing Jon Snow, but if it is false
2063 * we know the value for sure;
2065 false_reg->max_value = false_reg->min_value = val;
2068 /* Unsigned comparison, the minimum value is 0. */
2069 true_reg->min_value = 0;
2073 * If this is false, then the val is <= the register, if it is
2074 * true the register <= to the val.
2076 false_reg->min_value = val;
2077 true_reg->max_value = val - 1;
2080 /* Unsigned comparison, the minimum value is 0. */
2081 true_reg->min_value = 0;
2084 /* If this is false then constant < register, if it is true then
2085 * the register < constant.
2087 false_reg->min_value = val + 1;
2088 true_reg->max_value = val;
2094 check_reg_overflow(false_reg);
2095 check_reg_overflow(true_reg);
2098 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2099 enum bpf_reg_type type)
2101 struct bpf_reg_state *reg = ®s[regno];
2103 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2105 /* We don't need id from this point onwards anymore, thus we
2106 * should better reset it, so that state pruning has chances
2110 if (type == UNKNOWN_VALUE)
2111 __mark_reg_unknown_value(regs, regno);
2115 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2116 * be folded together at some point.
2118 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2119 enum bpf_reg_type type)
2121 struct bpf_reg_state *regs = state->regs;
2122 u32 id = regs[regno].id;
2125 for (i = 0; i < MAX_BPF_REG; i++)
2126 mark_map_reg(regs, i, id, type);
2128 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2129 if (state->stack_slot_type[i] != STACK_SPILL)
2131 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2135 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2136 struct bpf_insn *insn, int *insn_idx)
2138 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2139 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2140 u8 opcode = BPF_OP(insn->code);
2143 if (opcode > BPF_EXIT) {
2144 verbose("invalid BPF_JMP opcode %x\n", opcode);
2148 if (BPF_SRC(insn->code) == BPF_X) {
2149 if (insn->imm != 0) {
2150 verbose("BPF_JMP uses reserved fields\n");
2154 /* check src1 operand */
2155 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2159 if (is_pointer_value(env, insn->src_reg)) {
2160 verbose("R%d pointer comparison prohibited\n",
2165 if (insn->src_reg != BPF_REG_0) {
2166 verbose("BPF_JMP uses reserved fields\n");
2171 /* check src2 operand */
2172 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2176 dst_reg = ®s[insn->dst_reg];
2178 /* detect if R == 0 where R was initialized to zero earlier */
2179 if (BPF_SRC(insn->code) == BPF_K &&
2180 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2181 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2182 if (opcode == BPF_JEQ) {
2183 /* if (imm == imm) goto pc+off;
2184 * only follow the goto, ignore fall-through
2186 *insn_idx += insn->off;
2189 /* if (imm != imm) goto pc+off;
2190 * only follow fall-through branch, since
2191 * that's where the program will go
2197 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2201 /* detect if we are comparing against a constant value so we can adjust
2202 * our min/max values for our dst register.
2204 if (BPF_SRC(insn->code) == BPF_X) {
2205 if (regs[insn->src_reg].type == CONST_IMM)
2206 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2207 dst_reg, regs[insn->src_reg].imm,
2209 else if (dst_reg->type == CONST_IMM)
2210 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2211 ®s[insn->src_reg], dst_reg->imm,
2214 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2215 dst_reg, insn->imm, opcode);
2218 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2219 if (BPF_SRC(insn->code) == BPF_K &&
2220 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2221 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2222 /* Mark all identical map registers in each branch as either
2223 * safe or unknown depending R == 0 or R != 0 conditional.
2225 mark_map_regs(this_branch, insn->dst_reg,
2226 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2227 mark_map_regs(other_branch, insn->dst_reg,
2228 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2229 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2230 dst_reg->type == PTR_TO_PACKET &&
2231 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2232 find_good_pkt_pointers(this_branch, dst_reg);
2233 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2234 dst_reg->type == PTR_TO_PACKET_END &&
2235 regs[insn->src_reg].type == PTR_TO_PACKET) {
2236 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]);
2237 } else if (is_pointer_value(env, insn->dst_reg)) {
2238 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2242 print_verifier_state(this_branch);
2246 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2247 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2249 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2251 return (struct bpf_map *) (unsigned long) imm64;
2254 /* verify BPF_LD_IMM64 instruction */
2255 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2257 struct bpf_reg_state *regs = env->cur_state.regs;
2260 if (BPF_SIZE(insn->code) != BPF_DW) {
2261 verbose("invalid BPF_LD_IMM insn\n");
2264 if (insn->off != 0) {
2265 verbose("BPF_LD_IMM64 uses reserved fields\n");
2269 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2273 if (insn->src_reg == 0) {
2274 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2276 regs[insn->dst_reg].type = CONST_IMM;
2277 regs[insn->dst_reg].imm = imm;
2281 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2282 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2284 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2285 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2289 static bool may_access_skb(enum bpf_prog_type type)
2292 case BPF_PROG_TYPE_SOCKET_FILTER:
2293 case BPF_PROG_TYPE_SCHED_CLS:
2294 case BPF_PROG_TYPE_SCHED_ACT:
2301 /* verify safety of LD_ABS|LD_IND instructions:
2302 * - they can only appear in the programs where ctx == skb
2303 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2304 * preserve R6-R9, and store return value into R0
2307 * ctx == skb == R6 == CTX
2310 * SRC == any register
2311 * IMM == 32-bit immediate
2314 * R0 - 8/16/32-bit skb data converted to cpu endianness
2316 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2318 struct bpf_reg_state *regs = env->cur_state.regs;
2319 u8 mode = BPF_MODE(insn->code);
2320 struct bpf_reg_state *reg;
2323 if (!may_access_skb(env->prog->type)) {
2324 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2328 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2329 BPF_SIZE(insn->code) == BPF_DW ||
2330 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2331 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2335 /* check whether implicit source operand (register R6) is readable */
2336 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2340 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2341 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2345 if (mode == BPF_IND) {
2346 /* check explicit source operand */
2347 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2352 /* reset caller saved regs to unreadable */
2353 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2354 reg = regs + caller_saved[i];
2355 reg->type = NOT_INIT;
2359 /* mark destination R0 register as readable, since it contains
2360 * the value fetched from the packet
2362 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2366 /* non-recursive DFS pseudo code
2367 * 1 procedure DFS-iterative(G,v):
2368 * 2 label v as discovered
2369 * 3 let S be a stack
2371 * 5 while S is not empty
2373 * 7 if t is what we're looking for:
2375 * 9 for all edges e in G.adjacentEdges(t) do
2376 * 10 if edge e is already labelled
2377 * 11 continue with the next edge
2378 * 12 w <- G.adjacentVertex(t,e)
2379 * 13 if vertex w is not discovered and not explored
2380 * 14 label e as tree-edge
2381 * 15 label w as discovered
2384 * 18 else if vertex w is discovered
2385 * 19 label e as back-edge
2387 * 21 // vertex w is explored
2388 * 22 label e as forward- or cross-edge
2389 * 23 label t as explored
2394 * 0x11 - discovered and fall-through edge labelled
2395 * 0x12 - discovered and fall-through and branch edges labelled
2406 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2408 static int *insn_stack; /* stack of insns to process */
2409 static int cur_stack; /* current stack index */
2410 static int *insn_state;
2412 /* t, w, e - match pseudo-code above:
2413 * t - index of current instruction
2414 * w - next instruction
2417 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2419 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2422 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2425 if (w < 0 || w >= env->prog->len) {
2426 verbose("jump out of range from insn %d to %d\n", t, w);
2431 /* mark branch target for state pruning */
2432 env->explored_states[w] = STATE_LIST_MARK;
2434 if (insn_state[w] == 0) {
2436 insn_state[t] = DISCOVERED | e;
2437 insn_state[w] = DISCOVERED;
2438 if (cur_stack >= env->prog->len)
2440 insn_stack[cur_stack++] = w;
2442 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2443 verbose("back-edge from insn %d to %d\n", t, w);
2445 } else if (insn_state[w] == EXPLORED) {
2446 /* forward- or cross-edge */
2447 insn_state[t] = DISCOVERED | e;
2449 verbose("insn state internal bug\n");
2455 /* non-recursive depth-first-search to detect loops in BPF program
2456 * loop == back-edge in directed graph
2458 static int check_cfg(struct bpf_verifier_env *env)
2460 struct bpf_insn *insns = env->prog->insnsi;
2461 int insn_cnt = env->prog->len;
2465 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2469 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2475 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2476 insn_stack[0] = 0; /* 0 is the first instruction */
2482 t = insn_stack[cur_stack - 1];
2484 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2485 u8 opcode = BPF_OP(insns[t].code);
2487 if (opcode == BPF_EXIT) {
2489 } else if (opcode == BPF_CALL) {
2490 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2495 if (t + 1 < insn_cnt)
2496 env->explored_states[t + 1] = STATE_LIST_MARK;
2497 } else if (opcode == BPF_JA) {
2498 if (BPF_SRC(insns[t].code) != BPF_K) {
2502 /* unconditional jump with single edge */
2503 ret = push_insn(t, t + insns[t].off + 1,
2509 /* tell verifier to check for equivalent states
2510 * after every call and jump
2512 if (t + 1 < insn_cnt)
2513 env->explored_states[t + 1] = STATE_LIST_MARK;
2515 /* conditional jump with two edges */
2516 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2522 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2529 /* all other non-branch instructions with single
2532 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2540 insn_state[t] = EXPLORED;
2541 if (cur_stack-- <= 0) {
2542 verbose("pop stack internal bug\n");
2549 for (i = 0; i < insn_cnt; i++) {
2550 if (insn_state[i] != EXPLORED) {
2551 verbose("unreachable insn %d\n", i);
2556 ret = 0; /* cfg looks good */
2564 /* the following conditions reduce the number of explored insns
2565 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2567 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2568 struct bpf_reg_state *cur)
2570 if (old->id != cur->id)
2573 /* old ptr_to_packet is more conservative, since it allows smaller
2575 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2576 * old(off=0,r=10) means that with range=10 the verifier proceeded
2577 * further and found no issues with the program. Now we're in the same
2578 * spot with cur(off=0,r=20), so we're safe too, since anything further
2579 * will only be looking at most 10 bytes after this pointer.
2581 if (old->off == cur->off && old->range < cur->range)
2584 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2585 * since both cannot be used for packet access and safe(old)
2586 * pointer has smaller off that could be used for further
2587 * 'if (ptr > data_end)' check
2589 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2590 * that we cannot access the packet.
2591 * The safe range is:
2592 * [ptr, ptr + range - off)
2593 * so whenever off >=range, it means no safe bytes from this pointer.
2594 * When comparing old->off <= cur->off, it means that older code
2595 * went with smaller offset and that offset was later
2596 * used to figure out the safe range after 'if (ptr > data_end)' check
2597 * Say, 'old' state was explored like:
2598 * ... R3(off=0, r=0)
2600 * ... now R4(off=20,r=0) <-- here
2601 * if (R4 > data_end)
2602 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2603 * ... the code further went all the way to bpf_exit.
2604 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2605 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2606 * goes further, such cur_R4 will give larger safe packet range after
2607 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2608 * so they will be good with r=30 and we can prune the search.
2610 if (old->off <= cur->off &&
2611 old->off >= old->range && cur->off >= cur->range)
2617 /* compare two verifier states
2619 * all states stored in state_list are known to be valid, since
2620 * verifier reached 'bpf_exit' instruction through them
2622 * this function is called when verifier exploring different branches of
2623 * execution popped from the state stack. If it sees an old state that has
2624 * more strict register state and more strict stack state then this execution
2625 * branch doesn't need to be explored further, since verifier already
2626 * concluded that more strict state leads to valid finish.
2628 * Therefore two states are equivalent if register state is more conservative
2629 * and explored stack state is more conservative than the current one.
2632 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2633 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2635 * In other words if current stack state (one being explored) has more
2636 * valid slots than old one that already passed validation, it means
2637 * the verifier can stop exploring and conclude that current state is valid too
2639 * Similarly with registers. If explored state has register type as invalid
2640 * whereas register type in current state is meaningful, it means that
2641 * the current state will reach 'bpf_exit' instruction safely
2643 static bool states_equal(struct bpf_verifier_env *env,
2644 struct bpf_verifier_state *old,
2645 struct bpf_verifier_state *cur)
2647 bool varlen_map_access = env->varlen_map_value_access;
2648 struct bpf_reg_state *rold, *rcur;
2651 for (i = 0; i < MAX_BPF_REG; i++) {
2652 rold = &old->regs[i];
2653 rcur = &cur->regs[i];
2655 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2658 /* If the ranges were not the same, but everything else was and
2659 * we didn't do a variable access into a map then we are a-ok.
2661 if (!varlen_map_access &&
2662 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2665 /* If we didn't map access then again we don't care about the
2666 * mismatched range values and it's ok if our old type was
2667 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2669 if (rold->type == NOT_INIT ||
2670 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2671 rcur->type != NOT_INIT))
2674 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2675 compare_ptrs_to_packet(rold, rcur))
2681 for (i = 0; i < MAX_BPF_STACK; i++) {
2682 if (old->stack_slot_type[i] == STACK_INVALID)
2684 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2685 /* Ex: old explored (safe) state has STACK_SPILL in
2686 * this stack slot, but current has has STACK_MISC ->
2687 * this verifier states are not equivalent,
2688 * return false to continue verification of this path
2691 if (i % BPF_REG_SIZE)
2693 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2694 &cur->spilled_regs[i / BPF_REG_SIZE],
2695 sizeof(old->spilled_regs[0])))
2696 /* when explored and current stack slot types are
2697 * the same, check that stored pointers types
2698 * are the same as well.
2699 * Ex: explored safe path could have stored
2700 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2701 * but current path has stored:
2702 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2703 * such verifier states are not equivalent.
2704 * return false to continue verification of this path
2713 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2715 struct bpf_verifier_state_list *new_sl;
2716 struct bpf_verifier_state_list *sl;
2718 sl = env->explored_states[insn_idx];
2720 /* this 'insn_idx' instruction wasn't marked, so we will not
2721 * be doing state search here
2725 while (sl != STATE_LIST_MARK) {
2726 if (states_equal(env, &sl->state, &env->cur_state))
2727 /* reached equivalent register/stack state,
2734 /* there were no equivalent states, remember current one.
2735 * technically the current state is not proven to be safe yet,
2736 * but it will either reach bpf_exit (which means it's safe) or
2737 * it will be rejected. Since there are no loops, we won't be
2738 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2740 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2744 /* add new state to the head of linked list */
2745 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2746 new_sl->next = env->explored_states[insn_idx];
2747 env->explored_states[insn_idx] = new_sl;
2751 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2752 int insn_idx, int prev_insn_idx)
2754 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2757 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2760 static int do_check(struct bpf_verifier_env *env)
2762 struct bpf_verifier_state *state = &env->cur_state;
2763 struct bpf_insn *insns = env->prog->insnsi;
2764 struct bpf_reg_state *regs = state->regs;
2765 int insn_cnt = env->prog->len;
2766 int insn_idx, prev_insn_idx = 0;
2767 int insn_processed = 0;
2768 bool do_print_state = false;
2770 init_reg_state(regs);
2772 env->varlen_map_value_access = false;
2774 struct bpf_insn *insn;
2778 if (insn_idx >= insn_cnt) {
2779 verbose("invalid insn idx %d insn_cnt %d\n",
2780 insn_idx, insn_cnt);
2784 insn = &insns[insn_idx];
2785 class = BPF_CLASS(insn->code);
2787 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2788 verbose("BPF program is too large. Processed %d insn\n",
2793 err = is_state_visited(env, insn_idx);
2797 /* found equivalent state, can prune the search */
2800 verbose("\nfrom %d to %d: safe\n",
2801 prev_insn_idx, insn_idx);
2803 verbose("%d: safe\n", insn_idx);
2805 goto process_bpf_exit;
2808 if (log_level && do_print_state) {
2809 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2810 print_verifier_state(&env->cur_state);
2811 do_print_state = false;
2815 verbose("%d: ", insn_idx);
2816 print_bpf_insn(insn);
2819 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2823 if (class == BPF_ALU || class == BPF_ALU64) {
2824 err = check_alu_op(env, insn);
2828 } else if (class == BPF_LDX) {
2829 enum bpf_reg_type *prev_src_type, src_reg_type;
2831 /* check for reserved fields is already done */
2833 /* check src operand */
2834 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2838 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2842 src_reg_type = regs[insn->src_reg].type;
2844 /* check that memory (src_reg + off) is readable,
2845 * the state of dst_reg will be updated by this func
2847 err = check_mem_access(env, insn->src_reg, insn->off,
2848 BPF_SIZE(insn->code), BPF_READ,
2853 if (BPF_SIZE(insn->code) != BPF_W &&
2854 BPF_SIZE(insn->code) != BPF_DW) {
2859 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2861 if (*prev_src_type == NOT_INIT) {
2863 * dst_reg = *(u32 *)(src_reg + off)
2864 * save type to validate intersecting paths
2866 *prev_src_type = src_reg_type;
2868 } else if (src_reg_type != *prev_src_type &&
2869 (src_reg_type == PTR_TO_CTX ||
2870 *prev_src_type == PTR_TO_CTX)) {
2871 /* ABuser program is trying to use the same insn
2872 * dst_reg = *(u32*) (src_reg + off)
2873 * with different pointer types:
2874 * src_reg == ctx in one branch and
2875 * src_reg == stack|map in some other branch.
2878 verbose("same insn cannot be used with different pointers\n");
2882 } else if (class == BPF_STX) {
2883 enum bpf_reg_type *prev_dst_type, dst_reg_type;
2885 if (BPF_MODE(insn->code) == BPF_XADD) {
2886 err = check_xadd(env, insn);
2893 /* check src1 operand */
2894 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2897 /* check src2 operand */
2898 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2902 dst_reg_type = regs[insn->dst_reg].type;
2904 /* check that memory (dst_reg + off) is writeable */
2905 err = check_mem_access(env, insn->dst_reg, insn->off,
2906 BPF_SIZE(insn->code), BPF_WRITE,
2911 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2913 if (*prev_dst_type == NOT_INIT) {
2914 *prev_dst_type = dst_reg_type;
2915 } else if (dst_reg_type != *prev_dst_type &&
2916 (dst_reg_type == PTR_TO_CTX ||
2917 *prev_dst_type == PTR_TO_CTX)) {
2918 verbose("same insn cannot be used with different pointers\n");
2922 } else if (class == BPF_ST) {
2923 if (BPF_MODE(insn->code) != BPF_MEM ||
2924 insn->src_reg != BPF_REG_0) {
2925 verbose("BPF_ST uses reserved fields\n");
2928 /* check src operand */
2929 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2933 /* check that memory (dst_reg + off) is writeable */
2934 err = check_mem_access(env, insn->dst_reg, insn->off,
2935 BPF_SIZE(insn->code), BPF_WRITE,
2940 } else if (class == BPF_JMP) {
2941 u8 opcode = BPF_OP(insn->code);
2943 if (opcode == BPF_CALL) {
2944 if (BPF_SRC(insn->code) != BPF_K ||
2946 insn->src_reg != BPF_REG_0 ||
2947 insn->dst_reg != BPF_REG_0) {
2948 verbose("BPF_CALL uses reserved fields\n");
2952 err = check_call(env, insn->imm, insn_idx);
2956 } else if (opcode == BPF_JA) {
2957 if (BPF_SRC(insn->code) != BPF_K ||
2959 insn->src_reg != BPF_REG_0 ||
2960 insn->dst_reg != BPF_REG_0) {
2961 verbose("BPF_JA uses reserved fields\n");
2965 insn_idx += insn->off + 1;
2968 } else if (opcode == BPF_EXIT) {
2969 if (BPF_SRC(insn->code) != BPF_K ||
2971 insn->src_reg != BPF_REG_0 ||
2972 insn->dst_reg != BPF_REG_0) {
2973 verbose("BPF_EXIT uses reserved fields\n");
2977 /* eBPF calling convetion is such that R0 is used
2978 * to return the value from eBPF program.
2979 * Make sure that it's readable at this time
2980 * of bpf_exit, which means that program wrote
2981 * something into it earlier
2983 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2987 if (is_pointer_value(env, BPF_REG_0)) {
2988 verbose("R0 leaks addr as return value\n");
2993 insn_idx = pop_stack(env, &prev_insn_idx);
2997 do_print_state = true;
3001 err = check_cond_jmp_op(env, insn, &insn_idx);
3005 } else if (class == BPF_LD) {
3006 u8 mode = BPF_MODE(insn->code);
3008 if (mode == BPF_ABS || mode == BPF_IND) {
3009 err = check_ld_abs(env, insn);
3013 } else if (mode == BPF_IMM) {
3014 err = check_ld_imm(env, insn);
3020 verbose("invalid BPF_LD mode\n");
3023 reset_reg_range_values(regs, insn->dst_reg);
3025 verbose("unknown insn class %d\n", class);
3032 verbose("processed %d insns\n", insn_processed);
3036 static int check_map_prog_compatibility(struct bpf_map *map,
3037 struct bpf_prog *prog)
3040 if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
3041 (map->map_type == BPF_MAP_TYPE_HASH ||
3042 map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
3043 (map->map_flags & BPF_F_NO_PREALLOC)) {
3044 verbose("perf_event programs can only use preallocated hash map\n");
3050 /* look for pseudo eBPF instructions that access map FDs and
3051 * replace them with actual map pointers
3053 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3055 struct bpf_insn *insn = env->prog->insnsi;
3056 int insn_cnt = env->prog->len;
3059 err = bpf_prog_calc_tag(env->prog);
3063 for (i = 0; i < insn_cnt; i++, insn++) {
3064 if (BPF_CLASS(insn->code) == BPF_LDX &&
3065 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3066 verbose("BPF_LDX uses reserved fields\n");
3070 if (BPF_CLASS(insn->code) == BPF_STX &&
3071 ((BPF_MODE(insn->code) != BPF_MEM &&
3072 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3073 verbose("BPF_STX uses reserved fields\n");
3077 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3078 struct bpf_map *map;
3081 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3082 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3084 verbose("invalid bpf_ld_imm64 insn\n");
3088 if (insn->src_reg == 0)
3089 /* valid generic load 64-bit imm */
3092 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3093 verbose("unrecognized bpf_ld_imm64 insn\n");
3097 f = fdget(insn->imm);
3098 map = __bpf_map_get(f);
3100 verbose("fd %d is not pointing to valid bpf_map\n",
3102 return PTR_ERR(map);
3105 err = check_map_prog_compatibility(map, env->prog);
3111 /* store map pointer inside BPF_LD_IMM64 instruction */
3112 insn[0].imm = (u32) (unsigned long) map;
3113 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3115 /* check whether we recorded this map already */
3116 for (j = 0; j < env->used_map_cnt; j++)
3117 if (env->used_maps[j] == map) {
3122 if (env->used_map_cnt >= MAX_USED_MAPS) {
3127 /* hold the map. If the program is rejected by verifier,
3128 * the map will be released by release_maps() or it
3129 * will be used by the valid program until it's unloaded
3130 * and all maps are released in free_bpf_prog_info()
3132 map = bpf_map_inc(map, false);
3135 return PTR_ERR(map);
3137 env->used_maps[env->used_map_cnt++] = map;
3146 /* now all pseudo BPF_LD_IMM64 instructions load valid
3147 * 'struct bpf_map *' into a register instead of user map_fd.
3148 * These pointers will be used later by verifier to validate map access.
3153 /* drop refcnt of maps used by the rejected program */
3154 static void release_maps(struct bpf_verifier_env *env)
3158 for (i = 0; i < env->used_map_cnt; i++)
3159 bpf_map_put(env->used_maps[i]);
3162 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3163 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3165 struct bpf_insn *insn = env->prog->insnsi;
3166 int insn_cnt = env->prog->len;
3169 for (i = 0; i < insn_cnt; i++, insn++)
3170 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3174 /* single env->prog->insni[off] instruction was replaced with the range
3175 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3176 * [0, off) and [off, end) to new locations, so the patched range stays zero
3178 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3181 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3185 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3188 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3189 memcpy(new_data + off + cnt - 1, old_data + off,
3190 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3191 env->insn_aux_data = new_data;
3196 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3197 const struct bpf_insn *patch, u32 len)
3199 struct bpf_prog *new_prog;
3201 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3204 if (adjust_insn_aux_data(env, new_prog->len, off, len))
3209 /* convert load instructions that access fields of 'struct __sk_buff'
3210 * into sequence of instructions that access fields of 'struct sk_buff'
3212 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3214 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3215 const int insn_cnt = env->prog->len;
3216 struct bpf_insn insn_buf[16], *insn;
3217 struct bpf_prog *new_prog;
3218 enum bpf_access_type type;
3219 int i, cnt, delta = 0;
3221 if (ops->gen_prologue) {
3222 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3224 if (cnt >= ARRAY_SIZE(insn_buf)) {
3225 verbose("bpf verifier is misconfigured\n");
3228 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3232 env->prog = new_prog;
3237 if (!ops->convert_ctx_access)
3240 insn = env->prog->insnsi + delta;
3242 for (i = 0; i < insn_cnt; i++, insn++) {
3243 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3244 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3245 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3246 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3248 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3249 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3250 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3251 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3256 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3259 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3260 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3261 verbose("bpf verifier is misconfigured\n");
3265 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3271 /* keep walking new program and skip insns we just inserted */
3272 env->prog = new_prog;
3273 insn = new_prog->insnsi + i + delta;
3279 /* fixup insn->imm field of bpf_call instructions
3280 * and inline eligible helpers as explicit sequence of BPF instructions
3282 * this function is called after eBPF program passed verification
3284 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3286 struct bpf_prog *prog = env->prog;
3287 struct bpf_insn *insn = prog->insnsi;
3288 const struct bpf_func_proto *fn;
3289 const int insn_cnt = prog->len;
3290 struct bpf_insn insn_buf[16];
3291 struct bpf_prog *new_prog;
3292 struct bpf_map *map_ptr;
3293 int i, cnt, delta = 0;
3295 for (i = 0; i < insn_cnt; i++, insn++) {
3296 if (insn->code != (BPF_JMP | BPF_CALL))
3299 if (insn->imm == BPF_FUNC_get_route_realm)
3300 prog->dst_needed = 1;
3301 if (insn->imm == BPF_FUNC_get_prandom_u32)
3302 bpf_user_rnd_init_once();
3303 if (insn->imm == BPF_FUNC_xdp_adjust_head)
3304 prog->xdp_adjust_head = 1;
3305 if (insn->imm == BPF_FUNC_tail_call) {
3306 /* mark bpf_tail_call as different opcode to avoid
3307 * conditional branch in the interpeter for every normal
3308 * call and to prevent accidental JITing by JIT compiler
3309 * that doesn't support bpf_tail_call yet
3312 insn->code |= BPF_X;
3316 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3317 map_ptr = env->insn_aux_data[i + delta].map_ptr;
3318 if (map_ptr == BPF_MAP_PTR_POISON ||
3319 !map_ptr->ops->map_gen_lookup)
3320 goto patch_call_imm;
3322 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3323 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3324 verbose("bpf verifier is misconfigured\n");
3328 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3335 /* keep walking new program and skip insns we just inserted */
3336 env->prog = prog = new_prog;
3337 insn = new_prog->insnsi + i + delta;
3342 fn = prog->aux->ops->get_func_proto(insn->imm);
3343 /* all functions that have prototype and verifier allowed
3344 * programs to call them, must be real in-kernel functions
3347 verbose("kernel subsystem misconfigured func %s#%d\n",
3348 func_id_name(insn->imm), insn->imm);
3351 insn->imm = fn->func - __bpf_call_base;
3357 static void free_states(struct bpf_verifier_env *env)
3359 struct bpf_verifier_state_list *sl, *sln;
3362 if (!env->explored_states)
3365 for (i = 0; i < env->prog->len; i++) {
3366 sl = env->explored_states[i];
3369 while (sl != STATE_LIST_MARK) {
3376 kfree(env->explored_states);
3379 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3381 char __user *log_ubuf = NULL;
3382 struct bpf_verifier_env *env;
3385 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3386 * allocate/free it every time bpf_check() is called
3388 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3392 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3395 if (!env->insn_aux_data)
3399 /* grab the mutex to protect few globals used by verifier */
3400 mutex_lock(&bpf_verifier_lock);
3402 if (attr->log_level || attr->log_buf || attr->log_size) {
3403 /* user requested verbose verifier output
3404 * and supplied buffer to store the verification trace
3406 log_level = attr->log_level;
3407 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3408 log_size = attr->log_size;
3412 /* log_* values have to be sane */
3413 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3414 log_level == 0 || log_ubuf == NULL)
3418 log_buf = vmalloc(log_size);
3425 ret = replace_map_fd_with_map_ptr(env);
3427 goto skip_full_check;
3429 env->explored_states = kcalloc(env->prog->len,
3430 sizeof(struct bpf_verifier_state_list *),
3433 if (!env->explored_states)
3434 goto skip_full_check;
3436 ret = check_cfg(env);
3438 goto skip_full_check;
3440 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3442 ret = do_check(env);
3445 while (pop_stack(env, NULL) >= 0);
3449 /* program is valid, convert *(u32*)(ctx + off) accesses */
3450 ret = convert_ctx_accesses(env);
3453 ret = fixup_bpf_calls(env);
3455 if (log_level && log_len >= log_size - 1) {
3456 BUG_ON(log_len >= log_size);
3457 /* verifier log exceeded user supplied buffer */
3459 /* fall through to return what was recorded */
3462 /* copy verifier log back to user space including trailing zero */
3463 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3468 if (ret == 0 && env->used_map_cnt) {
3469 /* if program passed verifier, update used_maps in bpf_prog_info */
3470 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3471 sizeof(env->used_maps[0]),
3474 if (!env->prog->aux->used_maps) {
3479 memcpy(env->prog->aux->used_maps, env->used_maps,
3480 sizeof(env->used_maps[0]) * env->used_map_cnt);
3481 env->prog->aux->used_map_cnt = env->used_map_cnt;
3483 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3484 * bpf_ld_imm64 instructions
3486 convert_pseudo_ld_imm64(env);
3492 if (!env->prog->aux->used_maps)
3493 /* if we didn't copy map pointers into bpf_prog_info, release
3494 * them now. Otherwise free_bpf_prog_info() will release them.
3499 mutex_unlock(&bpf_verifier_lock);
3500 vfree(env->insn_aux_data);
3506 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3509 struct bpf_verifier_env *env;
3512 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3516 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3519 if (!env->insn_aux_data)
3522 env->analyzer_ops = ops;
3523 env->analyzer_priv = priv;
3525 /* grab the mutex to protect few globals used by verifier */
3526 mutex_lock(&bpf_verifier_lock);
3530 env->explored_states = kcalloc(env->prog->len,
3531 sizeof(struct bpf_verifier_state_list *),
3534 if (!env->explored_states)
3535 goto skip_full_check;
3537 ret = check_cfg(env);
3539 goto skip_full_check;
3541 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3543 ret = do_check(env);
3546 while (pop_stack(env, NULL) >= 0);
3549 mutex_unlock(&bpf_verifier_lock);
3550 vfree(env->insn_aux_data);
3555 EXPORT_SYMBOL_GPL(bpf_analyzer);