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>
26 static const struct bpf_verifier_ops * const bpf_verifier_ops[] = {
27 #define BPF_PROG_TYPE(_id, _name) \
28 [_id] = & _name ## _verifier_ops,
29 #define BPF_MAP_TYPE(_id, _ops)
30 #include <linux/bpf_types.h>
35 /* bpf_check() is a static code analyzer that walks eBPF program
36 * instruction by instruction and updates register/stack state.
37 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
39 * The first pass is depth-first-search to check that the program is a DAG.
40 * It rejects the following programs:
41 * - larger than BPF_MAXINSNS insns
42 * - if loop is present (detected via back-edge)
43 * - unreachable insns exist (shouldn't be a forest. program = one function)
44 * - out of bounds or malformed jumps
45 * The second pass is all possible path descent from the 1st insn.
46 * Since it's analyzing all pathes through the program, the length of the
47 * analysis is limited to 64k insn, which may be hit even if total number of
48 * insn is less then 4K, but there are too many branches that change stack/regs.
49 * Number of 'branches to be analyzed' is limited to 1k
51 * On entry to each instruction, each register has a type, and the instruction
52 * changes the types of the registers depending on instruction semantics.
53 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
56 * All registers are 64-bit.
57 * R0 - return register
58 * R1-R5 argument passing registers
59 * R6-R9 callee saved registers
60 * R10 - frame pointer read-only
62 * At the start of BPF program the register R1 contains a pointer to bpf_context
63 * and has type PTR_TO_CTX.
65 * Verifier tracks arithmetic operations on pointers in case:
66 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
67 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
68 * 1st insn copies R10 (which has FRAME_PTR) type into R1
69 * and 2nd arithmetic instruction is pattern matched to recognize
70 * that it wants to construct a pointer to some element within stack.
71 * So after 2nd insn, the register R1 has type PTR_TO_STACK
72 * (and -20 constant is saved for further stack bounds checking).
73 * Meaning that this reg is a pointer to stack plus known immediate constant.
75 * Most of the time the registers have SCALAR_VALUE type, which
76 * means the register has some value, but it's not a valid pointer.
77 * (like pointer plus pointer becomes SCALAR_VALUE type)
79 * When verifier sees load or store instructions the type of base register
80 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer
81 * types recognized by check_mem_access() function.
83 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
84 * and the range of [ptr, ptr + map's value_size) is accessible.
86 * registers used to pass values to function calls are checked against
87 * function argument constraints.
89 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
90 * It means that the register type passed to this function must be
91 * PTR_TO_STACK and it will be used inside the function as
92 * 'pointer to map element key'
94 * For example the argument constraints for bpf_map_lookup_elem():
95 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
96 * .arg1_type = ARG_CONST_MAP_PTR,
97 * .arg2_type = ARG_PTR_TO_MAP_KEY,
99 * ret_type says that this function returns 'pointer to map elem value or null'
100 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
101 * 2nd argument should be a pointer to stack, which will be used inside
102 * the helper function as a pointer to map element key.
104 * On the kernel side the helper function looks like:
105 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
107 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
108 * void *key = (void *) (unsigned long) r2;
111 * here kernel can access 'key' and 'map' pointers safely, knowing that
112 * [key, key + map->key_size) bytes are valid and were initialized on
113 * the stack of eBPF program.
116 * Corresponding eBPF program may look like:
117 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
118 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
119 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
120 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
121 * here verifier looks at prototype of map_lookup_elem() and sees:
122 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
123 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
125 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
126 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
127 * and were initialized prior to this call.
128 * If it's ok, then verifier allows this BPF_CALL insn and looks at
129 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
130 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
131 * returns ether pointer to map value or NULL.
133 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
134 * insn, the register holding that pointer in the true branch changes state to
135 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
136 * branch. See check_cond_jmp_op().
138 * After the call R0 is set to return type of the function and registers R1-R5
139 * are set to NOT_INIT to indicate that they are no longer readable.
142 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
143 struct bpf_verifier_stack_elem {
144 /* verifer state is 'st'
145 * before processing instruction 'insn_idx'
146 * and after processing instruction 'prev_insn_idx'
148 struct bpf_verifier_state st;
151 struct bpf_verifier_stack_elem *next;
154 #define BPF_COMPLEXITY_LIMIT_INSNS 131072
155 #define BPF_COMPLEXITY_LIMIT_STACK 1024
157 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
159 struct bpf_call_arg_meta {
160 struct bpf_map *map_ptr;
167 static DEFINE_MUTEX(bpf_verifier_lock);
169 /* log_level controls verbosity level of eBPF verifier.
170 * verbose() is used to dump the verification trace to the log, so the user
171 * can figure out what's wrong with the program
173 static __printf(2, 3) void verbose(struct bpf_verifier_env *env,
174 const char *fmt, ...)
176 struct bpf_verifer_log *log = &env->log;
180 if (!log->level || !log->ubuf || bpf_verifier_log_full(log))
184 n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args);
187 WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1,
188 "verifier log line truncated - local buffer too short\n");
190 n = min(log->len_total - log->len_used - 1, n);
193 if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1))
199 static bool type_is_pkt_pointer(enum bpf_reg_type type)
201 return type == PTR_TO_PACKET ||
202 type == PTR_TO_PACKET_META;
205 /* string representation of 'enum bpf_reg_type' */
206 static const char * const reg_type_str[] = {
208 [SCALAR_VALUE] = "inv",
209 [PTR_TO_CTX] = "ctx",
210 [CONST_PTR_TO_MAP] = "map_ptr",
211 [PTR_TO_MAP_VALUE] = "map_value",
212 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
213 [PTR_TO_STACK] = "fp",
214 [PTR_TO_PACKET] = "pkt",
215 [PTR_TO_PACKET_META] = "pkt_meta",
216 [PTR_TO_PACKET_END] = "pkt_end",
219 static void print_verifier_state(struct bpf_verifier_env *env,
220 struct bpf_verifier_state *state)
222 struct bpf_reg_state *reg;
226 for (i = 0; i < MAX_BPF_REG; i++) {
227 reg = &state->regs[i];
231 verbose(env, " R%d=%s", i, reg_type_str[t]);
232 if ((t == SCALAR_VALUE || t == PTR_TO_STACK) &&
233 tnum_is_const(reg->var_off)) {
234 /* reg->off should be 0 for SCALAR_VALUE */
235 verbose(env, "%lld", reg->var_off.value + reg->off);
237 verbose(env, "(id=%d", reg->id);
238 if (t != SCALAR_VALUE)
239 verbose(env, ",off=%d", reg->off);
240 if (type_is_pkt_pointer(t))
241 verbose(env, ",r=%d", reg->range);
242 else if (t == CONST_PTR_TO_MAP ||
243 t == PTR_TO_MAP_VALUE ||
244 t == PTR_TO_MAP_VALUE_OR_NULL)
245 verbose(env, ",ks=%d,vs=%d",
246 reg->map_ptr->key_size,
247 reg->map_ptr->value_size);
248 if (tnum_is_const(reg->var_off)) {
249 /* Typically an immediate SCALAR_VALUE, but
250 * could be a pointer whose offset is too big
253 verbose(env, ",imm=%llx", reg->var_off.value);
255 if (reg->smin_value != reg->umin_value &&
256 reg->smin_value != S64_MIN)
257 verbose(env, ",smin_value=%lld",
258 (long long)reg->smin_value);
259 if (reg->smax_value != reg->umax_value &&
260 reg->smax_value != S64_MAX)
261 verbose(env, ",smax_value=%lld",
262 (long long)reg->smax_value);
263 if (reg->umin_value != 0)
264 verbose(env, ",umin_value=%llu",
265 (unsigned long long)reg->umin_value);
266 if (reg->umax_value != U64_MAX)
267 verbose(env, ",umax_value=%llu",
268 (unsigned long long)reg->umax_value);
269 if (!tnum_is_unknown(reg->var_off)) {
272 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
273 verbose(env, ",var_off=%s", tn_buf);
279 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
280 if (state->stack[i].slot_type[0] == STACK_SPILL)
281 verbose(env, " fp%d=%s",
282 -MAX_BPF_STACK + i * BPF_REG_SIZE,
283 reg_type_str[state->stack[i].spilled_ptr.type]);
288 static int copy_stack_state(struct bpf_verifier_state *dst,
289 const struct bpf_verifier_state *src)
293 if (WARN_ON_ONCE(dst->allocated_stack < src->allocated_stack)) {
294 /* internal bug, make state invalid to reject the program */
295 memset(dst, 0, sizeof(*dst));
298 memcpy(dst->stack, src->stack,
299 sizeof(*src->stack) * (src->allocated_stack / BPF_REG_SIZE));
303 /* do_check() starts with zero-sized stack in struct bpf_verifier_state to
304 * make it consume minimal amount of memory. check_stack_write() access from
305 * the program calls into realloc_verifier_state() to grow the stack size.
306 * Note there is a non-zero 'parent' pointer inside bpf_verifier_state
307 * which this function copies over. It points to previous bpf_verifier_state
308 * which is never reallocated
310 static int realloc_verifier_state(struct bpf_verifier_state *state, int size,
313 u32 old_size = state->allocated_stack;
314 struct bpf_stack_state *new_stack;
315 int slot = size / BPF_REG_SIZE;
317 if (size <= old_size || !size) {
320 state->allocated_stack = slot * BPF_REG_SIZE;
321 if (!size && old_size) {
327 new_stack = kmalloc_array(slot, sizeof(struct bpf_stack_state),
333 memcpy(new_stack, state->stack,
334 sizeof(*new_stack) * (old_size / BPF_REG_SIZE));
335 memset(new_stack + old_size / BPF_REG_SIZE, 0,
336 sizeof(*new_stack) * (size - old_size) / BPF_REG_SIZE);
338 state->allocated_stack = slot * BPF_REG_SIZE;
340 state->stack = new_stack;
344 static void free_verifier_state(struct bpf_verifier_state *state,
352 /* copy verifier state from src to dst growing dst stack space
353 * when necessary to accommodate larger src stack
355 static int copy_verifier_state(struct bpf_verifier_state *dst,
356 const struct bpf_verifier_state *src)
360 err = realloc_verifier_state(dst, src->allocated_stack, false);
363 memcpy(dst, src, offsetof(struct bpf_verifier_state, allocated_stack));
364 return copy_stack_state(dst, src);
367 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx,
370 struct bpf_verifier_state *cur = env->cur_state;
371 struct bpf_verifier_stack_elem *elem, *head = env->head;
374 if (env->head == NULL)
378 err = copy_verifier_state(cur, &head->st);
383 *insn_idx = head->insn_idx;
385 *prev_insn_idx = head->prev_insn_idx;
387 free_verifier_state(&head->st, false);
394 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
395 int insn_idx, int prev_insn_idx)
397 struct bpf_verifier_state *cur = env->cur_state;
398 struct bpf_verifier_stack_elem *elem;
401 elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
405 elem->insn_idx = insn_idx;
406 elem->prev_insn_idx = prev_insn_idx;
407 elem->next = env->head;
410 err = copy_verifier_state(&elem->st, cur);
413 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
414 verbose(env, "BPF program is too complex\n");
419 /* pop all elements and return */
420 while (!pop_stack(env, NULL, NULL));
424 #define CALLER_SAVED_REGS 6
425 static const int caller_saved[CALLER_SAVED_REGS] = {
426 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
429 static void __mark_reg_not_init(struct bpf_reg_state *reg);
431 /* Mark the unknown part of a register (variable offset or scalar value) as
432 * known to have the value @imm.
434 static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm)
437 reg->var_off = tnum_const(imm);
438 reg->smin_value = (s64)imm;
439 reg->smax_value = (s64)imm;
440 reg->umin_value = imm;
441 reg->umax_value = imm;
444 /* Mark the 'variable offset' part of a register as zero. This should be
445 * used only on registers holding a pointer type.
447 static void __mark_reg_known_zero(struct bpf_reg_state *reg)
449 __mark_reg_known(reg, 0);
452 static void mark_reg_known_zero(struct bpf_verifier_env *env,
453 struct bpf_reg_state *regs, u32 regno)
455 if (WARN_ON(regno >= MAX_BPF_REG)) {
456 verbose(env, "mark_reg_known_zero(regs, %u)\n", regno);
457 /* Something bad happened, let's kill all regs */
458 for (regno = 0; regno < MAX_BPF_REG; regno++)
459 __mark_reg_not_init(regs + regno);
462 __mark_reg_known_zero(regs + regno);
465 static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg)
467 return type_is_pkt_pointer(reg->type);
470 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg)
472 return reg_is_pkt_pointer(reg) ||
473 reg->type == PTR_TO_PACKET_END;
476 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
477 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg,
478 enum bpf_reg_type which)
480 /* The register can already have a range from prior markings.
481 * This is fine as long as it hasn't been advanced from its
484 return reg->type == which &&
487 tnum_equals_const(reg->var_off, 0);
490 /* Attempts to improve min/max values based on var_off information */
491 static void __update_reg_bounds(struct bpf_reg_state *reg)
493 /* min signed is max(sign bit) | min(other bits) */
494 reg->smin_value = max_t(s64, reg->smin_value,
495 reg->var_off.value | (reg->var_off.mask & S64_MIN));
496 /* max signed is min(sign bit) | max(other bits) */
497 reg->smax_value = min_t(s64, reg->smax_value,
498 reg->var_off.value | (reg->var_off.mask & S64_MAX));
499 reg->umin_value = max(reg->umin_value, reg->var_off.value);
500 reg->umax_value = min(reg->umax_value,
501 reg->var_off.value | reg->var_off.mask);
504 /* Uses signed min/max values to inform unsigned, and vice-versa */
505 static void __reg_deduce_bounds(struct bpf_reg_state *reg)
507 /* Learn sign from signed bounds.
508 * If we cannot cross the sign boundary, then signed and unsigned bounds
509 * are the same, so combine. This works even in the negative case, e.g.
510 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
512 if (reg->smin_value >= 0 || reg->smax_value < 0) {
513 reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
515 reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
519 /* Learn sign from unsigned bounds. Signed bounds cross the sign
520 * boundary, so we must be careful.
522 if ((s64)reg->umax_value >= 0) {
523 /* Positive. We can't learn anything from the smin, but smax
524 * is positive, hence safe.
526 reg->smin_value = reg->umin_value;
527 reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
529 } else if ((s64)reg->umin_value < 0) {
530 /* Negative. We can't learn anything from the smax, but smin
531 * is negative, hence safe.
533 reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value,
535 reg->smax_value = reg->umax_value;
539 /* Attempts to improve var_off based on unsigned min/max information */
540 static void __reg_bound_offset(struct bpf_reg_state *reg)
542 reg->var_off = tnum_intersect(reg->var_off,
543 tnum_range(reg->umin_value,
547 /* Reset the min/max bounds of a register */
548 static void __mark_reg_unbounded(struct bpf_reg_state *reg)
550 reg->smin_value = S64_MIN;
551 reg->smax_value = S64_MAX;
553 reg->umax_value = U64_MAX;
556 /* Mark a register as having a completely unknown (scalar) value. */
557 static void __mark_reg_unknown(struct bpf_reg_state *reg)
559 reg->type = SCALAR_VALUE;
562 reg->var_off = tnum_unknown;
563 __mark_reg_unbounded(reg);
566 static void mark_reg_unknown(struct bpf_verifier_env *env,
567 struct bpf_reg_state *regs, u32 regno)
569 if (WARN_ON(regno >= MAX_BPF_REG)) {
570 verbose(env, "mark_reg_unknown(regs, %u)\n", regno);
571 /* Something bad happened, let's kill all regs */
572 for (regno = 0; regno < MAX_BPF_REG; regno++)
573 __mark_reg_not_init(regs + regno);
576 __mark_reg_unknown(regs + regno);
579 static void __mark_reg_not_init(struct bpf_reg_state *reg)
581 __mark_reg_unknown(reg);
582 reg->type = NOT_INIT;
585 static void mark_reg_not_init(struct bpf_verifier_env *env,
586 struct bpf_reg_state *regs, u32 regno)
588 if (WARN_ON(regno >= MAX_BPF_REG)) {
589 verbose(env, "mark_reg_not_init(regs, %u)\n", regno);
590 /* Something bad happened, let's kill all regs */
591 for (regno = 0; regno < MAX_BPF_REG; regno++)
592 __mark_reg_not_init(regs + regno);
595 __mark_reg_not_init(regs + regno);
598 static void init_reg_state(struct bpf_verifier_env *env,
599 struct bpf_reg_state *regs)
603 for (i = 0; i < MAX_BPF_REG; i++) {
604 mark_reg_not_init(env, regs, i);
605 regs[i].live = REG_LIVE_NONE;
609 regs[BPF_REG_FP].type = PTR_TO_STACK;
610 mark_reg_known_zero(env, regs, BPF_REG_FP);
612 /* 1st arg to a function */
613 regs[BPF_REG_1].type = PTR_TO_CTX;
614 mark_reg_known_zero(env, regs, BPF_REG_1);
618 SRC_OP, /* register is used as source operand */
619 DST_OP, /* register is used as destination operand */
620 DST_OP_NO_MARK /* same as above, check only, don't mark */
623 static void mark_reg_read(const struct bpf_verifier_state *state, u32 regno)
625 struct bpf_verifier_state *parent = state->parent;
627 if (regno == BPF_REG_FP)
628 /* We don't need to worry about FP liveness because it's read-only */
632 /* if read wasn't screened by an earlier write ... */
633 if (state->regs[regno].live & REG_LIVE_WRITTEN)
635 /* ... then we depend on parent's value */
636 parent->regs[regno].live |= REG_LIVE_READ;
638 parent = state->parent;
642 static int check_reg_arg(struct bpf_verifier_env *env, u32 regno,
645 struct bpf_reg_state *regs = env->cur_state->regs;
647 if (regno >= MAX_BPF_REG) {
648 verbose(env, "R%d is invalid\n", regno);
653 /* check whether register used as source operand can be read */
654 if (regs[regno].type == NOT_INIT) {
655 verbose(env, "R%d !read_ok\n", regno);
658 mark_reg_read(env->cur_state, regno);
660 /* check whether register used as dest operand can be written to */
661 if (regno == BPF_REG_FP) {
662 verbose(env, "frame pointer is read only\n");
665 regs[regno].live |= REG_LIVE_WRITTEN;
667 mark_reg_unknown(env, regs, regno);
672 static bool is_spillable_regtype(enum bpf_reg_type type)
675 case PTR_TO_MAP_VALUE:
676 case PTR_TO_MAP_VALUE_OR_NULL:
680 case PTR_TO_PACKET_META:
681 case PTR_TO_PACKET_END:
682 case CONST_PTR_TO_MAP:
689 /* check_stack_read/write functions track spill/fill of registers,
690 * stack boundary and alignment are checked in check_mem_access()
692 static int check_stack_write(struct bpf_verifier_env *env,
693 struct bpf_verifier_state *state, int off,
694 int size, int value_regno)
696 int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err;
698 err = realloc_verifier_state(state, round_up(slot + 1, BPF_REG_SIZE),
702 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
703 * so it's aligned access and [off, off + size) are within stack limits
705 if (!env->allow_ptr_leaks &&
706 state->stack[spi].slot_type[0] == STACK_SPILL &&
707 size != BPF_REG_SIZE) {
708 verbose(env, "attempt to corrupt spilled pointer on stack\n");
712 if (value_regno >= 0 &&
713 is_spillable_regtype(state->regs[value_regno].type)) {
715 /* register containing pointer is being spilled into stack */
716 if (size != BPF_REG_SIZE) {
717 verbose(env, "invalid size of register spill\n");
721 /* save register state */
722 state->stack[spi].spilled_ptr = state->regs[value_regno];
723 state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN;
725 for (i = 0; i < BPF_REG_SIZE; i++)
726 state->stack[spi].slot_type[i] = STACK_SPILL;
728 /* regular write of data into stack */
729 state->stack[spi].spilled_ptr = (struct bpf_reg_state) {};
731 for (i = 0; i < size; i++)
732 state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] =
738 static void mark_stack_slot_read(const struct bpf_verifier_state *state, int slot)
740 struct bpf_verifier_state *parent = state->parent;
743 /* if read wasn't screened by an earlier write ... */
744 if (state->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN)
746 /* ... then we depend on parent's value */
747 parent->stack[slot].spilled_ptr.live |= REG_LIVE_READ;
749 parent = state->parent;
753 static int check_stack_read(struct bpf_verifier_env *env,
754 struct bpf_verifier_state *state, int off, int size,
757 int i, slot = -off - 1, spi = slot / BPF_REG_SIZE;
760 if (state->allocated_stack <= slot) {
761 verbose(env, "invalid read from stack off %d+0 size %d\n",
765 stype = state->stack[spi].slot_type;
767 if (stype[0] == STACK_SPILL) {
768 if (size != BPF_REG_SIZE) {
769 verbose(env, "invalid size of register spill\n");
772 for (i = 1; i < BPF_REG_SIZE; i++) {
773 if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) {
774 verbose(env, "corrupted spill memory\n");
779 if (value_regno >= 0) {
780 /* restore register state from stack */
781 state->regs[value_regno] = state->stack[spi].spilled_ptr;
782 mark_stack_slot_read(state, spi);
786 for (i = 0; i < size; i++) {
787 if (stype[(slot - i) % BPF_REG_SIZE] != STACK_MISC) {
788 verbose(env, "invalid read from stack off %d+%d size %d\n",
793 if (value_regno >= 0)
794 /* have read misc data from the stack */
795 mark_reg_unknown(env, state->regs, value_regno);
800 /* check read/write into map element returned by bpf_map_lookup_elem() */
801 static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
804 struct bpf_reg_state *regs = cur_regs(env);
805 struct bpf_map *map = regs[regno].map_ptr;
807 if (off < 0 || size <= 0 || off + size > map->value_size) {
808 verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n",
809 map->value_size, off, size);
815 /* check read/write into a map element with possible variable offset */
816 static int check_map_access(struct bpf_verifier_env *env, u32 regno,
819 struct bpf_verifier_state *state = env->cur_state;
820 struct bpf_reg_state *reg = &state->regs[regno];
823 /* We may have adjusted the register to this map value, so we
824 * need to try adding each of min_value and max_value to off
825 * to make sure our theoretical access will be safe.
828 print_verifier_state(env, state);
829 /* The minimum value is only important with signed
830 * comparisons where we can't assume the floor of a
831 * value is 0. If we are using signed variables for our
832 * index'es we need to make sure that whatever we use
833 * will have a set floor within our range.
835 if (reg->smin_value < 0) {
836 verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
840 err = __check_map_access(env, regno, reg->smin_value + off, size);
842 verbose(env, "R%d min value is outside of the array range\n",
847 /* If we haven't set a max value then we need to bail since we can't be
848 * sure we won't do bad things.
849 * If reg->umax_value + off could overflow, treat that as unbounded too.
851 if (reg->umax_value >= BPF_MAX_VAR_OFF) {
852 verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n",
856 err = __check_map_access(env, regno, reg->umax_value + off, size);
858 verbose(env, "R%d max value is outside of the array range\n",
863 #define MAX_PACKET_OFF 0xffff
865 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
866 const struct bpf_call_arg_meta *meta,
867 enum bpf_access_type t)
869 switch (env->prog->type) {
870 case BPF_PROG_TYPE_LWT_IN:
871 case BPF_PROG_TYPE_LWT_OUT:
872 /* dst_input() and dst_output() can't write for now */
876 case BPF_PROG_TYPE_SCHED_CLS:
877 case BPF_PROG_TYPE_SCHED_ACT:
878 case BPF_PROG_TYPE_XDP:
879 case BPF_PROG_TYPE_LWT_XMIT:
880 case BPF_PROG_TYPE_SK_SKB:
882 return meta->pkt_access;
884 env->seen_direct_write = true;
891 static int __check_packet_access(struct bpf_verifier_env *env, u32 regno,
894 struct bpf_reg_state *regs = cur_regs(env);
895 struct bpf_reg_state *reg = ®s[regno];
897 if (off < 0 || size <= 0 || (u64)off + size > reg->range) {
898 verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
899 off, size, regno, reg->id, reg->off, reg->range);
905 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
908 struct bpf_reg_state *regs = cur_regs(env);
909 struct bpf_reg_state *reg = ®s[regno];
912 /* We may have added a variable offset to the packet pointer; but any
913 * reg->range we have comes after that. We are only checking the fixed
917 /* We don't allow negative numbers, because we aren't tracking enough
918 * detail to prove they're safe.
920 if (reg->smin_value < 0) {
921 verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
925 err = __check_packet_access(env, regno, off, size);
927 verbose(env, "R%d offset is outside of the packet\n", regno);
933 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
934 static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
935 enum bpf_access_type t, enum bpf_reg_type *reg_type)
937 struct bpf_insn_access_aux info = {
938 .reg_type = *reg_type,
941 if (env->ops->is_valid_access &&
942 env->ops->is_valid_access(off, size, t, &info)) {
943 /* A non zero info.ctx_field_size indicates that this field is a
944 * candidate for later verifier transformation to load the whole
945 * field and then apply a mask when accessed with a narrower
946 * access than actual ctx access size. A zero info.ctx_field_size
947 * will only allow for whole field access and rejects any other
948 * type of narrower access.
950 *reg_type = info.reg_type;
952 if (env->analyzer_ops)
955 env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size;
956 /* remember the offset of last byte accessed in ctx */
957 if (env->prog->aux->max_ctx_offset < off + size)
958 env->prog->aux->max_ctx_offset = off + size;
962 verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size);
966 static bool __is_pointer_value(bool allow_ptr_leaks,
967 const struct bpf_reg_state *reg)
972 return reg->type != SCALAR_VALUE;
975 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
977 return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno);
980 static int check_pkt_ptr_alignment(struct bpf_verifier_env *env,
981 const struct bpf_reg_state *reg,
982 int off, int size, bool strict)
987 /* Byte size accesses are always allowed. */
988 if (!strict || size == 1)
991 /* For platforms that do not have a Kconfig enabling
992 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
993 * NET_IP_ALIGN is universally set to '2'. And on platforms
994 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
995 * to this code only in strict mode where we want to emulate
996 * the NET_IP_ALIGN==2 checking. Therefore use an
997 * unconditional IP align value of '2'.
1001 reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off));
1002 if (!tnum_is_aligned(reg_off, size)) {
1005 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1007 "misaligned packet access off %d+%s+%d+%d size %d\n",
1008 ip_align, tn_buf, reg->off, off, size);
1015 static int check_generic_ptr_alignment(struct bpf_verifier_env *env,
1016 const struct bpf_reg_state *reg,
1017 const char *pointer_desc,
1018 int off, int size, bool strict)
1020 struct tnum reg_off;
1022 /* Byte size accesses are always allowed. */
1023 if (!strict || size == 1)
1026 reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off));
1027 if (!tnum_is_aligned(reg_off, size)) {
1030 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1031 verbose(env, "misaligned %saccess off %s+%d+%d size %d\n",
1032 pointer_desc, tn_buf, reg->off, off, size);
1039 static int check_ptr_alignment(struct bpf_verifier_env *env,
1040 const struct bpf_reg_state *reg,
1043 bool strict = env->strict_alignment;
1044 const char *pointer_desc = "";
1046 switch (reg->type) {
1048 case PTR_TO_PACKET_META:
1049 /* Special case, because of NET_IP_ALIGN. Given metadata sits
1050 * right in front, treat it the very same way.
1052 return check_pkt_ptr_alignment(env, reg, off, size, strict);
1053 case PTR_TO_MAP_VALUE:
1054 pointer_desc = "value ";
1057 pointer_desc = "context ";
1060 pointer_desc = "stack ";
1065 return check_generic_ptr_alignment(env, reg, pointer_desc, off, size,
1069 /* check whether memory at (regno + off) is accessible for t = (read | write)
1070 * if t==write, value_regno is a register which value is stored into memory
1071 * if t==read, value_regno is a register which will receive the value from memory
1072 * if t==write && value_regno==-1, some unknown value is stored into memory
1073 * if t==read && value_regno==-1, don't care what we read from memory
1075 static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off,
1076 int bpf_size, enum bpf_access_type t,
1079 struct bpf_verifier_state *state = env->cur_state;
1080 struct bpf_reg_state *regs = cur_regs(env);
1081 struct bpf_reg_state *reg = regs + regno;
1084 size = bpf_size_to_bytes(bpf_size);
1088 /* alignment checks will add in reg->off themselves */
1089 err = check_ptr_alignment(env, reg, off, size);
1093 /* for access checks, reg->off is just part of off */
1096 if (reg->type == PTR_TO_MAP_VALUE) {
1097 if (t == BPF_WRITE && value_regno >= 0 &&
1098 is_pointer_value(env, value_regno)) {
1099 verbose(env, "R%d leaks addr into map\n", value_regno);
1103 err = check_map_access(env, regno, off, size);
1104 if (!err && t == BPF_READ && value_regno >= 0)
1105 mark_reg_unknown(env, regs, value_regno);
1107 } else if (reg->type == PTR_TO_CTX) {
1108 enum bpf_reg_type reg_type = SCALAR_VALUE;
1110 if (t == BPF_WRITE && value_regno >= 0 &&
1111 is_pointer_value(env, value_regno)) {
1112 verbose(env, "R%d leaks addr into ctx\n", value_regno);
1115 /* ctx accesses must be at a fixed offset, so that we can
1116 * determine what type of data were returned.
1120 "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n",
1121 regno, reg->off, off - reg->off);
1124 if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
1127 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1129 "variable ctx access var_off=%s off=%d size=%d",
1133 err = check_ctx_access(env, insn_idx, off, size, t, ®_type);
1134 if (!err && t == BPF_READ && value_regno >= 0) {
1135 /* ctx access returns either a scalar, or a
1136 * PTR_TO_PACKET[_META,_END]. In the latter
1137 * case, we know the offset is zero.
1139 if (reg_type == SCALAR_VALUE)
1140 mark_reg_unknown(env, regs, value_regno);
1142 mark_reg_known_zero(env, regs,
1144 regs[value_regno].id = 0;
1145 regs[value_regno].off = 0;
1146 regs[value_regno].range = 0;
1147 regs[value_regno].type = reg_type;
1150 } else if (reg->type == PTR_TO_STACK) {
1151 /* stack accesses must be at a fixed offset, so that we can
1152 * determine what type of data were returned.
1153 * See check_stack_read().
1155 if (!tnum_is_const(reg->var_off)) {
1158 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
1159 verbose(env, "variable stack access var_off=%s off=%d size=%d",
1163 off += reg->var_off.value;
1164 if (off >= 0 || off < -MAX_BPF_STACK) {
1165 verbose(env, "invalid stack off=%d size=%d\n", off,
1170 if (env->prog->aux->stack_depth < -off)
1171 env->prog->aux->stack_depth = -off;
1174 err = check_stack_write(env, state, off, size,
1177 err = check_stack_read(env, state, off, size,
1179 } else if (reg_is_pkt_pointer(reg)) {
1180 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
1181 verbose(env, "cannot write into packet\n");
1184 if (t == BPF_WRITE && value_regno >= 0 &&
1185 is_pointer_value(env, value_regno)) {
1186 verbose(env, "R%d leaks addr into packet\n",
1190 err = check_packet_access(env, regno, off, size);
1191 if (!err && t == BPF_READ && value_regno >= 0)
1192 mark_reg_unknown(env, regs, value_regno);
1194 verbose(env, "R%d invalid mem access '%s'\n", regno,
1195 reg_type_str[reg->type]);
1199 if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ &&
1200 regs[value_regno].type == SCALAR_VALUE) {
1201 /* b/h/w load zero-extends, mark upper bits as known 0 */
1202 regs[value_regno].var_off =
1203 tnum_cast(regs[value_regno].var_off, size);
1204 __update_reg_bounds(®s[value_regno]);
1209 static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn)
1213 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
1215 verbose(env, "BPF_XADD uses reserved fields\n");
1219 /* check src1 operand */
1220 err = check_reg_arg(env, insn->src_reg, SRC_OP);
1224 /* check src2 operand */
1225 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
1229 if (is_pointer_value(env, insn->src_reg)) {
1230 verbose(env, "R%d leaks addr into mem\n", insn->src_reg);
1234 /* check whether atomic_add can read the memory */
1235 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1236 BPF_SIZE(insn->code), BPF_READ, -1);
1240 /* check whether atomic_add can write into the same memory */
1241 return check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
1242 BPF_SIZE(insn->code), BPF_WRITE, -1);
1245 /* Does this register contain a constant zero? */
1246 static bool register_is_null(struct bpf_reg_state reg)
1248 return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0);
1251 /* when register 'regno' is passed into function that will read 'access_size'
1252 * bytes from that pointer, make sure that it's within stack boundary
1253 * and all elements of stack are initialized.
1254 * Unlike most pointer bounds-checking functions, this one doesn't take an
1255 * 'off' argument, so it has to add in reg->off itself.
1257 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
1258 int access_size, bool zero_size_allowed,
1259 struct bpf_call_arg_meta *meta)
1261 struct bpf_verifier_state *state = env->cur_state;
1262 struct bpf_reg_state *regs = state->regs;
1263 int off, i, slot, spi;
1265 if (regs[regno].type != PTR_TO_STACK) {
1266 /* Allow zero-byte read from NULL, regardless of pointer type */
1267 if (zero_size_allowed && access_size == 0 &&
1268 register_is_null(regs[regno]))
1271 verbose(env, "R%d type=%s expected=%s\n", regno,
1272 reg_type_str[regs[regno].type],
1273 reg_type_str[PTR_TO_STACK]);
1277 /* Only allow fixed-offset stack reads */
1278 if (!tnum_is_const(regs[regno].var_off)) {
1281 tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off);
1282 verbose(env, "invalid variable stack read R%d var_off=%s\n",
1285 off = regs[regno].off + regs[regno].var_off.value;
1286 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
1288 verbose(env, "invalid stack type R%d off=%d access_size=%d\n",
1289 regno, off, access_size);
1293 if (env->prog->aux->stack_depth < -off)
1294 env->prog->aux->stack_depth = -off;
1296 if (meta && meta->raw_mode) {
1297 meta->access_size = access_size;
1298 meta->regno = regno;
1302 for (i = 0; i < access_size; i++) {
1303 slot = -(off + i) - 1;
1304 spi = slot / BPF_REG_SIZE;
1305 if (state->allocated_stack <= slot ||
1306 state->stack[spi].slot_type[slot % BPF_REG_SIZE] !=
1308 verbose(env, "invalid indirect read from stack off %d+%d size %d\n",
1309 off, i, access_size);
1316 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1317 int access_size, bool zero_size_allowed,
1318 struct bpf_call_arg_meta *meta)
1320 struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno];
1322 switch (reg->type) {
1324 case PTR_TO_PACKET_META:
1325 return check_packet_access(env, regno, reg->off, access_size);
1326 case PTR_TO_MAP_VALUE:
1327 return check_map_access(env, regno, reg->off, access_size);
1328 default: /* scalar_value|ptr_to_stack or invalid ptr */
1329 return check_stack_boundary(env, regno, access_size,
1330 zero_size_allowed, meta);
1334 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1335 enum bpf_arg_type arg_type,
1336 struct bpf_call_arg_meta *meta)
1338 struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno];
1339 enum bpf_reg_type expected_type, type = reg->type;
1342 if (arg_type == ARG_DONTCARE)
1345 err = check_reg_arg(env, regno, SRC_OP);
1349 if (arg_type == ARG_ANYTHING) {
1350 if (is_pointer_value(env, regno)) {
1351 verbose(env, "R%d leaks addr into helper function\n",
1358 if (type_is_pkt_pointer(type) &&
1359 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1360 verbose(env, "helper access to the packet is not allowed\n");
1364 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1365 arg_type == ARG_PTR_TO_MAP_VALUE) {
1366 expected_type = PTR_TO_STACK;
1367 if (!type_is_pkt_pointer(type) &&
1368 type != expected_type)
1370 } else if (arg_type == ARG_CONST_SIZE ||
1371 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1372 expected_type = SCALAR_VALUE;
1373 if (type != expected_type)
1375 } else if (arg_type == ARG_CONST_MAP_PTR) {
1376 expected_type = CONST_PTR_TO_MAP;
1377 if (type != expected_type)
1379 } else if (arg_type == ARG_PTR_TO_CTX) {
1380 expected_type = PTR_TO_CTX;
1381 if (type != expected_type)
1383 } else if (arg_type == ARG_PTR_TO_MEM ||
1384 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1385 expected_type = PTR_TO_STACK;
1386 /* One exception here. In case function allows for NULL to be
1387 * passed in as argument, it's a SCALAR_VALUE type. Final test
1388 * happens during stack boundary checking.
1390 if (register_is_null(*reg))
1391 /* final test in check_stack_boundary() */;
1392 else if (!type_is_pkt_pointer(type) &&
1393 type != PTR_TO_MAP_VALUE &&
1394 type != expected_type)
1396 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1398 verbose(env, "unsupported arg_type %d\n", arg_type);
1402 if (arg_type == ARG_CONST_MAP_PTR) {
1403 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1404 meta->map_ptr = reg->map_ptr;
1405 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1406 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1407 * check that [key, key + map->key_size) are within
1408 * stack limits and initialized
1410 if (!meta->map_ptr) {
1411 /* in function declaration map_ptr must come before
1412 * map_key, so that it's verified and known before
1413 * we have to check map_key here. Otherwise it means
1414 * that kernel subsystem misconfigured verifier
1416 verbose(env, "invalid map_ptr to access map->key\n");
1419 if (type_is_pkt_pointer(type))
1420 err = check_packet_access(env, regno, reg->off,
1421 meta->map_ptr->key_size);
1423 err = check_stack_boundary(env, regno,
1424 meta->map_ptr->key_size,
1426 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1427 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1428 * check [value, value + map->value_size) validity
1430 if (!meta->map_ptr) {
1431 /* kernel subsystem misconfigured verifier */
1432 verbose(env, "invalid map_ptr to access map->value\n");
1435 if (type_is_pkt_pointer(type))
1436 err = check_packet_access(env, regno, reg->off,
1437 meta->map_ptr->value_size);
1439 err = check_stack_boundary(env, regno,
1440 meta->map_ptr->value_size,
1442 } else if (arg_type == ARG_CONST_SIZE ||
1443 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1444 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1446 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1447 * from stack pointer 'buf'. Check it
1448 * note: regno == len, regno - 1 == buf
1451 /* kernel subsystem misconfigured verifier */
1453 "ARG_CONST_SIZE cannot be first argument\n");
1457 /* The register is SCALAR_VALUE; the access check
1458 * happens using its boundaries.
1461 if (!tnum_is_const(reg->var_off))
1462 /* For unprivileged variable accesses, disable raw
1463 * mode so that the program is required to
1464 * initialize all the memory that the helper could
1465 * just partially fill up.
1469 if (reg->smin_value < 0) {
1470 verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n",
1475 if (reg->umin_value == 0) {
1476 err = check_helper_mem_access(env, regno - 1, 0,
1483 if (reg->umax_value >= BPF_MAX_VAR_SIZ) {
1484 verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1488 err = check_helper_mem_access(env, regno - 1,
1490 zero_size_allowed, meta);
1495 verbose(env, "R%d type=%s expected=%s\n", regno,
1496 reg_type_str[type], reg_type_str[expected_type]);
1500 static int check_map_func_compatibility(struct bpf_verifier_env *env,
1501 struct bpf_map *map, int func_id)
1506 /* We need a two way check, first is from map perspective ... */
1507 switch (map->map_type) {
1508 case BPF_MAP_TYPE_PROG_ARRAY:
1509 if (func_id != BPF_FUNC_tail_call)
1512 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1513 if (func_id != BPF_FUNC_perf_event_read &&
1514 func_id != BPF_FUNC_perf_event_output &&
1515 func_id != BPF_FUNC_perf_event_read_value)
1518 case BPF_MAP_TYPE_STACK_TRACE:
1519 if (func_id != BPF_FUNC_get_stackid)
1522 case BPF_MAP_TYPE_CGROUP_ARRAY:
1523 if (func_id != BPF_FUNC_skb_under_cgroup &&
1524 func_id != BPF_FUNC_current_task_under_cgroup)
1527 /* devmap returns a pointer to a live net_device ifindex that we cannot
1528 * allow to be modified from bpf side. So do not allow lookup elements
1531 case BPF_MAP_TYPE_DEVMAP:
1532 if (func_id != BPF_FUNC_redirect_map)
1535 /* Restrict bpf side of cpumap, open when use-cases appear */
1536 case BPF_MAP_TYPE_CPUMAP:
1537 if (func_id != BPF_FUNC_redirect_map)
1540 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1541 case BPF_MAP_TYPE_HASH_OF_MAPS:
1542 if (func_id != BPF_FUNC_map_lookup_elem)
1545 case BPF_MAP_TYPE_SOCKMAP:
1546 if (func_id != BPF_FUNC_sk_redirect_map &&
1547 func_id != BPF_FUNC_sock_map_update &&
1548 func_id != BPF_FUNC_map_delete_elem)
1555 /* ... and second from the function itself. */
1557 case BPF_FUNC_tail_call:
1558 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1561 case BPF_FUNC_perf_event_read:
1562 case BPF_FUNC_perf_event_output:
1563 case BPF_FUNC_perf_event_read_value:
1564 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1567 case BPF_FUNC_get_stackid:
1568 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1571 case BPF_FUNC_current_task_under_cgroup:
1572 case BPF_FUNC_skb_under_cgroup:
1573 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1576 case BPF_FUNC_redirect_map:
1577 if (map->map_type != BPF_MAP_TYPE_DEVMAP &&
1578 map->map_type != BPF_MAP_TYPE_CPUMAP)
1581 case BPF_FUNC_sk_redirect_map:
1582 if (map->map_type != BPF_MAP_TYPE_SOCKMAP)
1585 case BPF_FUNC_sock_map_update:
1586 if (map->map_type != BPF_MAP_TYPE_SOCKMAP)
1595 verbose(env, "cannot pass map_type %d into func %s#%d\n",
1596 map->map_type, func_id_name(func_id), func_id);
1600 static int check_raw_mode(const struct bpf_func_proto *fn)
1604 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1606 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1608 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1610 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1612 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1615 return count > 1 ? -EINVAL : 0;
1618 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
1619 * are now invalid, so turn them into unknown SCALAR_VALUE.
1621 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1623 struct bpf_verifier_state *state = env->cur_state;
1624 struct bpf_reg_state *regs = state->regs, *reg;
1627 for (i = 0; i < MAX_BPF_REG; i++)
1628 if (reg_is_pkt_pointer_any(®s[i]))
1629 mark_reg_unknown(env, regs, i);
1631 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
1632 if (state->stack[i].slot_type[0] != STACK_SPILL)
1634 reg = &state->stack[i].spilled_ptr;
1635 if (reg_is_pkt_pointer_any(reg))
1636 __mark_reg_unknown(reg);
1640 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1642 const struct bpf_func_proto *fn = NULL;
1643 struct bpf_reg_state *regs;
1644 struct bpf_call_arg_meta meta;
1648 /* find function prototype */
1649 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1650 verbose(env, "invalid func %s#%d\n", func_id_name(func_id),
1655 if (env->ops->get_func_proto)
1656 fn = env->ops->get_func_proto(func_id);
1659 verbose(env, "unknown func %s#%d\n", func_id_name(func_id),
1664 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1665 if (!env->prog->gpl_compatible && fn->gpl_only) {
1666 verbose(env, "cannot call GPL only function from proprietary program\n");
1670 changes_data = bpf_helper_changes_pkt_data(fn->func);
1672 memset(&meta, 0, sizeof(meta));
1673 meta.pkt_access = fn->pkt_access;
1675 /* We only support one arg being in raw mode at the moment, which
1676 * is sufficient for the helper functions we have right now.
1678 err = check_raw_mode(fn);
1680 verbose(env, "kernel subsystem misconfigured func %s#%d\n",
1681 func_id_name(func_id), func_id);
1686 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1689 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1692 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1695 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1698 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1702 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1703 * is inferred from register state.
1705 for (i = 0; i < meta.access_size; i++) {
1706 err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1);
1711 regs = cur_regs(env);
1712 /* reset caller saved regs */
1713 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1714 mark_reg_not_init(env, regs, caller_saved[i]);
1715 check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
1718 /* update return register (already marked as written above) */
1719 if (fn->ret_type == RET_INTEGER) {
1720 /* sets type to SCALAR_VALUE */
1721 mark_reg_unknown(env, regs, BPF_REG_0);
1722 } else if (fn->ret_type == RET_VOID) {
1723 regs[BPF_REG_0].type = NOT_INIT;
1724 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1725 struct bpf_insn_aux_data *insn_aux;
1727 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1728 /* There is no offset yet applied, variable or fixed */
1729 mark_reg_known_zero(env, regs, BPF_REG_0);
1730 regs[BPF_REG_0].off = 0;
1731 /* remember map_ptr, so that check_map_access()
1732 * can check 'value_size' boundary of memory access
1733 * to map element returned from bpf_map_lookup_elem()
1735 if (meta.map_ptr == NULL) {
1737 "kernel subsystem misconfigured verifier\n");
1740 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1741 regs[BPF_REG_0].id = ++env->id_gen;
1742 insn_aux = &env->insn_aux_data[insn_idx];
1743 if (!insn_aux->map_ptr)
1744 insn_aux->map_ptr = meta.map_ptr;
1745 else if (insn_aux->map_ptr != meta.map_ptr)
1746 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1748 verbose(env, "unknown return type %d of func %s#%d\n",
1749 fn->ret_type, func_id_name(func_id), func_id);
1753 err = check_map_func_compatibility(env, meta.map_ptr, func_id);
1758 clear_all_pkt_pointers(env);
1762 static void coerce_reg_to_32(struct bpf_reg_state *reg)
1764 /* clear high 32 bits */
1765 reg->var_off = tnum_cast(reg->var_off, 4);
1767 __update_reg_bounds(reg);
1770 static bool signed_add_overflows(s64 a, s64 b)
1772 /* Do the add in u64, where overflow is well-defined */
1773 s64 res = (s64)((u64)a + (u64)b);
1780 static bool signed_sub_overflows(s64 a, s64 b)
1782 /* Do the sub in u64, where overflow is well-defined */
1783 s64 res = (s64)((u64)a - (u64)b);
1790 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
1791 * Caller should also handle BPF_MOV case separately.
1792 * If we return -EACCES, caller may want to try again treating pointer as a
1793 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
1795 static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
1796 struct bpf_insn *insn,
1797 const struct bpf_reg_state *ptr_reg,
1798 const struct bpf_reg_state *off_reg)
1800 struct bpf_reg_state *regs = cur_regs(env), *dst_reg;
1801 bool known = tnum_is_const(off_reg->var_off);
1802 s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value,
1803 smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value;
1804 u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value,
1805 umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value;
1806 u8 opcode = BPF_OP(insn->code);
1807 u32 dst = insn->dst_reg;
1809 dst_reg = ®s[dst];
1811 if (WARN_ON_ONCE(known && (smin_val != smax_val))) {
1812 print_verifier_state(env, env->cur_state);
1814 "verifier internal error: known but bad sbounds\n");
1817 if (WARN_ON_ONCE(known && (umin_val != umax_val))) {
1818 print_verifier_state(env, env->cur_state);
1820 "verifier internal error: known but bad ubounds\n");
1824 if (BPF_CLASS(insn->code) != BPF_ALU64) {
1825 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
1826 if (!env->allow_ptr_leaks)
1828 "R%d 32-bit pointer arithmetic prohibited\n",
1833 if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
1834 if (!env->allow_ptr_leaks)
1835 verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n",
1839 if (ptr_reg->type == CONST_PTR_TO_MAP) {
1840 if (!env->allow_ptr_leaks)
1841 verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n",
1845 if (ptr_reg->type == PTR_TO_PACKET_END) {
1846 if (!env->allow_ptr_leaks)
1847 verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n",
1852 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
1853 * The id may be overwritten later if we create a new variable offset.
1855 dst_reg->type = ptr_reg->type;
1856 dst_reg->id = ptr_reg->id;
1860 /* We can take a fixed offset as long as it doesn't overflow
1861 * the s32 'off' field
1863 if (known && (ptr_reg->off + smin_val ==
1864 (s64)(s32)(ptr_reg->off + smin_val))) {
1865 /* pointer += K. Accumulate it into fixed offset */
1866 dst_reg->smin_value = smin_ptr;
1867 dst_reg->smax_value = smax_ptr;
1868 dst_reg->umin_value = umin_ptr;
1869 dst_reg->umax_value = umax_ptr;
1870 dst_reg->var_off = ptr_reg->var_off;
1871 dst_reg->off = ptr_reg->off + smin_val;
1872 dst_reg->range = ptr_reg->range;
1875 /* A new variable offset is created. Note that off_reg->off
1876 * == 0, since it's a scalar.
1877 * dst_reg gets the pointer type and since some positive
1878 * integer value was added to the pointer, give it a new 'id'
1879 * if it's a PTR_TO_PACKET.
1880 * this creates a new 'base' pointer, off_reg (variable) gets
1881 * added into the variable offset, and we copy the fixed offset
1884 if (signed_add_overflows(smin_ptr, smin_val) ||
1885 signed_add_overflows(smax_ptr, smax_val)) {
1886 dst_reg->smin_value = S64_MIN;
1887 dst_reg->smax_value = S64_MAX;
1889 dst_reg->smin_value = smin_ptr + smin_val;
1890 dst_reg->smax_value = smax_ptr + smax_val;
1892 if (umin_ptr + umin_val < umin_ptr ||
1893 umax_ptr + umax_val < umax_ptr) {
1894 dst_reg->umin_value = 0;
1895 dst_reg->umax_value = U64_MAX;
1897 dst_reg->umin_value = umin_ptr + umin_val;
1898 dst_reg->umax_value = umax_ptr + umax_val;
1900 dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off);
1901 dst_reg->off = ptr_reg->off;
1902 if (reg_is_pkt_pointer(ptr_reg)) {
1903 dst_reg->id = ++env->id_gen;
1904 /* something was added to pkt_ptr, set range to zero */
1909 if (dst_reg == off_reg) {
1910 /* scalar -= pointer. Creates an unknown scalar */
1911 if (!env->allow_ptr_leaks)
1912 verbose(env, "R%d tried to subtract pointer from scalar\n",
1916 /* We don't allow subtraction from FP, because (according to
1917 * test_verifier.c test "invalid fp arithmetic", JITs might not
1918 * be able to deal with it.
1920 if (ptr_reg->type == PTR_TO_STACK) {
1921 if (!env->allow_ptr_leaks)
1922 verbose(env, "R%d subtraction from stack pointer prohibited\n",
1926 if (known && (ptr_reg->off - smin_val ==
1927 (s64)(s32)(ptr_reg->off - smin_val))) {
1928 /* pointer -= K. Subtract it from fixed offset */
1929 dst_reg->smin_value = smin_ptr;
1930 dst_reg->smax_value = smax_ptr;
1931 dst_reg->umin_value = umin_ptr;
1932 dst_reg->umax_value = umax_ptr;
1933 dst_reg->var_off = ptr_reg->var_off;
1934 dst_reg->id = ptr_reg->id;
1935 dst_reg->off = ptr_reg->off - smin_val;
1936 dst_reg->range = ptr_reg->range;
1939 /* A new variable offset is created. If the subtrahend is known
1940 * nonnegative, then any reg->range we had before is still good.
1942 if (signed_sub_overflows(smin_ptr, smax_val) ||
1943 signed_sub_overflows(smax_ptr, smin_val)) {
1944 /* Overflow possible, we know nothing */
1945 dst_reg->smin_value = S64_MIN;
1946 dst_reg->smax_value = S64_MAX;
1948 dst_reg->smin_value = smin_ptr - smax_val;
1949 dst_reg->smax_value = smax_ptr - smin_val;
1951 if (umin_ptr < umax_val) {
1952 /* Overflow possible, we know nothing */
1953 dst_reg->umin_value = 0;
1954 dst_reg->umax_value = U64_MAX;
1956 /* Cannot overflow (as long as bounds are consistent) */
1957 dst_reg->umin_value = umin_ptr - umax_val;
1958 dst_reg->umax_value = umax_ptr - umin_val;
1960 dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off);
1961 dst_reg->off = ptr_reg->off;
1962 if (reg_is_pkt_pointer(ptr_reg)) {
1963 dst_reg->id = ++env->id_gen;
1964 /* something was added to pkt_ptr, set range to zero */
1972 /* bitwise ops on pointers are troublesome, prohibit for now.
1973 * (However, in principle we could allow some cases, e.g.
1974 * ptr &= ~3 which would reduce min_value by 3.)
1976 if (!env->allow_ptr_leaks)
1977 verbose(env, "R%d bitwise operator %s on pointer prohibited\n",
1978 dst, bpf_alu_string[opcode >> 4]);
1981 /* other operators (e.g. MUL,LSH) produce non-pointer results */
1982 if (!env->allow_ptr_leaks)
1983 verbose(env, "R%d pointer arithmetic with %s operator prohibited\n",
1984 dst, bpf_alu_string[opcode >> 4]);
1988 __update_reg_bounds(dst_reg);
1989 __reg_deduce_bounds(dst_reg);
1990 __reg_bound_offset(dst_reg);
1994 static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env,
1995 struct bpf_insn *insn,
1996 struct bpf_reg_state *dst_reg,
1997 struct bpf_reg_state src_reg)
1999 struct bpf_reg_state *regs = cur_regs(env);
2000 u8 opcode = BPF_OP(insn->code);
2001 bool src_known, dst_known;
2002 s64 smin_val, smax_val;
2003 u64 umin_val, umax_val;
2005 if (BPF_CLASS(insn->code) != BPF_ALU64) {
2006 /* 32-bit ALU ops are (32,32)->64 */
2007 coerce_reg_to_32(dst_reg);
2008 coerce_reg_to_32(&src_reg);
2010 smin_val = src_reg.smin_value;
2011 smax_val = src_reg.smax_value;
2012 umin_val = src_reg.umin_value;
2013 umax_val = src_reg.umax_value;
2014 src_known = tnum_is_const(src_reg.var_off);
2015 dst_known = tnum_is_const(dst_reg->var_off);
2019 if (signed_add_overflows(dst_reg->smin_value, smin_val) ||
2020 signed_add_overflows(dst_reg->smax_value, smax_val)) {
2021 dst_reg->smin_value = S64_MIN;
2022 dst_reg->smax_value = S64_MAX;
2024 dst_reg->smin_value += smin_val;
2025 dst_reg->smax_value += smax_val;
2027 if (dst_reg->umin_value + umin_val < umin_val ||
2028 dst_reg->umax_value + umax_val < umax_val) {
2029 dst_reg->umin_value = 0;
2030 dst_reg->umax_value = U64_MAX;
2032 dst_reg->umin_value += umin_val;
2033 dst_reg->umax_value += umax_val;
2035 dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off);
2038 if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
2039 signed_sub_overflows(dst_reg->smax_value, smin_val)) {
2040 /* Overflow possible, we know nothing */
2041 dst_reg->smin_value = S64_MIN;
2042 dst_reg->smax_value = S64_MAX;
2044 dst_reg->smin_value -= smax_val;
2045 dst_reg->smax_value -= smin_val;
2047 if (dst_reg->umin_value < umax_val) {
2048 /* Overflow possible, we know nothing */
2049 dst_reg->umin_value = 0;
2050 dst_reg->umax_value = U64_MAX;
2052 /* Cannot overflow (as long as bounds are consistent) */
2053 dst_reg->umin_value -= umax_val;
2054 dst_reg->umax_value -= umin_val;
2056 dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off);
2059 dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off);
2060 if (smin_val < 0 || dst_reg->smin_value < 0) {
2061 /* Ain't nobody got time to multiply that sign */
2062 __mark_reg_unbounded(dst_reg);
2063 __update_reg_bounds(dst_reg);
2066 /* Both values are positive, so we can work with unsigned and
2067 * copy the result to signed (unless it exceeds S64_MAX).
2069 if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) {
2070 /* Potential overflow, we know nothing */
2071 __mark_reg_unbounded(dst_reg);
2072 /* (except what we can learn from the var_off) */
2073 __update_reg_bounds(dst_reg);
2076 dst_reg->umin_value *= umin_val;
2077 dst_reg->umax_value *= umax_val;
2078 if (dst_reg->umax_value > S64_MAX) {
2079 /* Overflow possible, we know nothing */
2080 dst_reg->smin_value = S64_MIN;
2081 dst_reg->smax_value = S64_MAX;
2083 dst_reg->smin_value = dst_reg->umin_value;
2084 dst_reg->smax_value = dst_reg->umax_value;
2088 if (src_known && dst_known) {
2089 __mark_reg_known(dst_reg, dst_reg->var_off.value &
2090 src_reg.var_off.value);
2093 /* We get our minimum from the var_off, since that's inherently
2094 * bitwise. Our maximum is the minimum of the operands' maxima.
2096 dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off);
2097 dst_reg->umin_value = dst_reg->var_off.value;
2098 dst_reg->umax_value = min(dst_reg->umax_value, umax_val);
2099 if (dst_reg->smin_value < 0 || smin_val < 0) {
2100 /* Lose signed bounds when ANDing negative numbers,
2101 * ain't nobody got time for that.
2103 dst_reg->smin_value = S64_MIN;
2104 dst_reg->smax_value = S64_MAX;
2106 /* ANDing two positives gives a positive, so safe to
2107 * cast result into s64.
2109 dst_reg->smin_value = dst_reg->umin_value;
2110 dst_reg->smax_value = dst_reg->umax_value;
2112 /* We may learn something more from the var_off */
2113 __update_reg_bounds(dst_reg);
2116 if (src_known && dst_known) {
2117 __mark_reg_known(dst_reg, dst_reg->var_off.value |
2118 src_reg.var_off.value);
2121 /* We get our maximum from the var_off, and our minimum is the
2122 * maximum of the operands' minima
2124 dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off);
2125 dst_reg->umin_value = max(dst_reg->umin_value, umin_val);
2126 dst_reg->umax_value = dst_reg->var_off.value |
2127 dst_reg->var_off.mask;
2128 if (dst_reg->smin_value < 0 || smin_val < 0) {
2129 /* Lose signed bounds when ORing negative numbers,
2130 * ain't nobody got time for that.
2132 dst_reg->smin_value = S64_MIN;
2133 dst_reg->smax_value = S64_MAX;
2135 /* ORing two positives gives a positive, so safe to
2136 * cast result into s64.
2138 dst_reg->smin_value = dst_reg->umin_value;
2139 dst_reg->smax_value = dst_reg->umax_value;
2141 /* We may learn something more from the var_off */
2142 __update_reg_bounds(dst_reg);
2145 if (umax_val > 63) {
2146 /* Shifts greater than 63 are undefined. This includes
2147 * shifts by a negative number.
2149 mark_reg_unknown(env, regs, insn->dst_reg);
2152 /* We lose all sign bit information (except what we can pick
2155 dst_reg->smin_value = S64_MIN;
2156 dst_reg->smax_value = S64_MAX;
2157 /* If we might shift our top bit out, then we know nothing */
2158 if (dst_reg->umax_value > 1ULL << (63 - umax_val)) {
2159 dst_reg->umin_value = 0;
2160 dst_reg->umax_value = U64_MAX;
2162 dst_reg->umin_value <<= umin_val;
2163 dst_reg->umax_value <<= umax_val;
2166 dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val);
2168 dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val);
2169 /* We may learn something more from the var_off */
2170 __update_reg_bounds(dst_reg);
2173 if (umax_val > 63) {
2174 /* Shifts greater than 63 are undefined. This includes
2175 * shifts by a negative number.
2177 mark_reg_unknown(env, regs, insn->dst_reg);
2180 /* BPF_RSH is an unsigned shift, so make the appropriate casts */
2181 if (dst_reg->smin_value < 0) {
2183 /* Sign bit will be cleared */
2184 dst_reg->smin_value = 0;
2186 /* Lost sign bit information */
2187 dst_reg->smin_value = S64_MIN;
2188 dst_reg->smax_value = S64_MAX;
2191 dst_reg->smin_value =
2192 (u64)(dst_reg->smin_value) >> umax_val;
2195 dst_reg->var_off = tnum_rshift(dst_reg->var_off,
2198 dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val);
2199 dst_reg->umin_value >>= umax_val;
2200 dst_reg->umax_value >>= umin_val;
2201 /* We may learn something more from the var_off */
2202 __update_reg_bounds(dst_reg);
2205 mark_reg_unknown(env, regs, insn->dst_reg);
2209 __reg_deduce_bounds(dst_reg);
2210 __reg_bound_offset(dst_reg);
2214 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
2217 static int adjust_reg_min_max_vals(struct bpf_verifier_env *env,
2218 struct bpf_insn *insn)
2220 struct bpf_reg_state *regs = cur_regs(env), *dst_reg, *src_reg;
2221 struct bpf_reg_state *ptr_reg = NULL, off_reg = {0};
2222 u8 opcode = BPF_OP(insn->code);
2225 dst_reg = ®s[insn->dst_reg];
2227 if (dst_reg->type != SCALAR_VALUE)
2229 if (BPF_SRC(insn->code) == BPF_X) {
2230 src_reg = ®s[insn->src_reg];
2231 if (src_reg->type != SCALAR_VALUE) {
2232 if (dst_reg->type != SCALAR_VALUE) {
2233 /* Combining two pointers by any ALU op yields
2234 * an arbitrary scalar.
2236 if (!env->allow_ptr_leaks) {
2237 verbose(env, "R%d pointer %s pointer prohibited\n",
2239 bpf_alu_string[opcode >> 4]);
2242 mark_reg_unknown(env, regs, insn->dst_reg);
2245 /* scalar += pointer
2246 * This is legal, but we have to reverse our
2247 * src/dest handling in computing the range
2249 rc = adjust_ptr_min_max_vals(env, insn,
2251 if (rc == -EACCES && env->allow_ptr_leaks) {
2252 /* scalar += unknown scalar */
2253 __mark_reg_unknown(&off_reg);
2254 return adjust_scalar_min_max_vals(
2260 } else if (ptr_reg) {
2261 /* pointer += scalar */
2262 rc = adjust_ptr_min_max_vals(env, insn,
2264 if (rc == -EACCES && env->allow_ptr_leaks) {
2265 /* unknown scalar += scalar */
2266 __mark_reg_unknown(dst_reg);
2267 return adjust_scalar_min_max_vals(
2268 env, insn, dst_reg, *src_reg);
2273 /* Pretend the src is a reg with a known value, since we only
2274 * need to be able to read from this state.
2276 off_reg.type = SCALAR_VALUE;
2277 __mark_reg_known(&off_reg, insn->imm);
2279 if (ptr_reg) { /* pointer += K */
2280 rc = adjust_ptr_min_max_vals(env, insn,
2282 if (rc == -EACCES && env->allow_ptr_leaks) {
2283 /* unknown scalar += K */
2284 __mark_reg_unknown(dst_reg);
2285 return adjust_scalar_min_max_vals(
2286 env, insn, dst_reg, off_reg);
2292 /* Got here implies adding two SCALAR_VALUEs */
2293 if (WARN_ON_ONCE(ptr_reg)) {
2294 print_verifier_state(env, env->cur_state);
2295 verbose(env, "verifier internal error: unexpected ptr_reg\n");
2298 if (WARN_ON(!src_reg)) {
2299 print_verifier_state(env, env->cur_state);
2300 verbose(env, "verifier internal error: no src_reg\n");
2303 return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg);
2306 /* check validity of 32-bit and 64-bit arithmetic operations */
2307 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
2309 struct bpf_reg_state *regs = cur_regs(env);
2310 u8 opcode = BPF_OP(insn->code);
2313 if (opcode == BPF_END || opcode == BPF_NEG) {
2314 if (opcode == BPF_NEG) {
2315 if (BPF_SRC(insn->code) != 0 ||
2316 insn->src_reg != BPF_REG_0 ||
2317 insn->off != 0 || insn->imm != 0) {
2318 verbose(env, "BPF_NEG uses reserved fields\n");
2322 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
2323 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) ||
2324 BPF_CLASS(insn->code) == BPF_ALU64) {
2325 verbose(env, "BPF_END uses reserved fields\n");
2330 /* check src operand */
2331 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
2335 if (is_pointer_value(env, insn->dst_reg)) {
2336 verbose(env, "R%d pointer arithmetic prohibited\n",
2341 /* check dest operand */
2342 err = check_reg_arg(env, insn->dst_reg, DST_OP);
2346 } else if (opcode == BPF_MOV) {
2348 if (BPF_SRC(insn->code) == BPF_X) {
2349 if (insn->imm != 0 || insn->off != 0) {
2350 verbose(env, "BPF_MOV uses reserved fields\n");
2354 /* check src operand */
2355 err = check_reg_arg(env, insn->src_reg, SRC_OP);
2359 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
2360 verbose(env, "BPF_MOV uses reserved fields\n");
2365 /* check dest operand */
2366 err = check_reg_arg(env, insn->dst_reg, DST_OP);
2370 if (BPF_SRC(insn->code) == BPF_X) {
2371 if (BPF_CLASS(insn->code) == BPF_ALU64) {
2373 * copy register state to dest reg
2375 regs[insn->dst_reg] = regs[insn->src_reg];
2376 regs[insn->dst_reg].live |= REG_LIVE_WRITTEN;
2379 if (is_pointer_value(env, insn->src_reg)) {
2381 "R%d partial copy of pointer\n",
2385 mark_reg_unknown(env, regs, insn->dst_reg);
2386 /* high 32 bits are known zero. */
2387 regs[insn->dst_reg].var_off = tnum_cast(
2388 regs[insn->dst_reg].var_off, 4);
2389 __update_reg_bounds(®s[insn->dst_reg]);
2393 * remember the value we stored into this reg
2395 regs[insn->dst_reg].type = SCALAR_VALUE;
2396 __mark_reg_known(regs + insn->dst_reg, insn->imm);
2399 } else if (opcode > BPF_END) {
2400 verbose(env, "invalid BPF_ALU opcode %x\n", opcode);
2403 } else { /* all other ALU ops: and, sub, xor, add, ... */
2405 if (BPF_SRC(insn->code) == BPF_X) {
2406 if (insn->imm != 0 || insn->off != 0) {
2407 verbose(env, "BPF_ALU uses reserved fields\n");
2410 /* check src1 operand */
2411 err = check_reg_arg(env, insn->src_reg, SRC_OP);
2415 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
2416 verbose(env, "BPF_ALU uses reserved fields\n");
2421 /* check src2 operand */
2422 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
2426 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
2427 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
2428 verbose(env, "div by zero\n");
2432 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
2433 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
2434 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
2436 if (insn->imm < 0 || insn->imm >= size) {
2437 verbose(env, "invalid shift %d\n", insn->imm);
2442 /* check dest operand */
2443 err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
2447 return adjust_reg_min_max_vals(env, insn);
2453 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
2454 struct bpf_reg_state *dst_reg,
2455 enum bpf_reg_type type,
2456 bool range_right_open)
2458 struct bpf_reg_state *regs = state->regs, *reg;
2462 if (dst_reg->off < 0 ||
2463 (dst_reg->off == 0 && range_right_open))
2464 /* This doesn't give us any range */
2467 if (dst_reg->umax_value > MAX_PACKET_OFF ||
2468 dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF)
2469 /* Risk of overflow. For instance, ptr + (1<<63) may be less
2470 * than pkt_end, but that's because it's also less than pkt.
2474 new_range = dst_reg->off;
2475 if (range_right_open)
2478 /* Examples for register markings:
2480 * pkt_data in dst register:
2484 * if (r2 > pkt_end) goto <handle exception>
2489 * if (r2 < pkt_end) goto <access okay>
2490 * <handle exception>
2493 * r2 == dst_reg, pkt_end == src_reg
2494 * r2=pkt(id=n,off=8,r=0)
2495 * r3=pkt(id=n,off=0,r=0)
2497 * pkt_data in src register:
2501 * if (pkt_end >= r2) goto <access okay>
2502 * <handle exception>
2506 * if (pkt_end <= r2) goto <handle exception>
2510 * pkt_end == dst_reg, r2 == src_reg
2511 * r2=pkt(id=n,off=8,r=0)
2512 * r3=pkt(id=n,off=0,r=0)
2514 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2515 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
2516 * and [r3, r3 + 8-1) respectively is safe to access depending on
2520 /* If our ids match, then we must have the same max_value. And we
2521 * don't care about the other reg's fixed offset, since if it's too big
2522 * the range won't allow anything.
2523 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
2525 for (i = 0; i < MAX_BPF_REG; i++)
2526 if (regs[i].type == type && regs[i].id == dst_reg->id)
2527 /* keep the maximum range already checked */
2528 regs[i].range = max(regs[i].range, new_range);
2530 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
2531 if (state->stack[i].slot_type[0] != STACK_SPILL)
2533 reg = &state->stack[i].spilled_ptr;
2534 if (reg->type == type && reg->id == dst_reg->id)
2535 reg->range = max_t(u16, reg->range, new_range);
2539 /* Adjusts the register min/max values in the case that the dst_reg is the
2540 * variable register that we are working on, and src_reg is a constant or we're
2541 * simply doing a BPF_K check.
2542 * In JEQ/JNE cases we also adjust the var_off values.
2544 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2545 struct bpf_reg_state *false_reg, u64 val,
2548 /* If the dst_reg is a pointer, we can't learn anything about its
2549 * variable offset from the compare (unless src_reg were a pointer into
2550 * the same object, but we don't bother with that.
2551 * Since false_reg and true_reg have the same type by construction, we
2552 * only need to check one of them for pointerness.
2554 if (__is_pointer_value(false, false_reg))
2559 /* If this is false then we know nothing Jon Snow, but if it is
2560 * true then we know for sure.
2562 __mark_reg_known(true_reg, val);
2565 /* If this is true we know nothing Jon Snow, but if it is false
2566 * we know the value for sure;
2568 __mark_reg_known(false_reg, val);
2571 false_reg->umax_value = min(false_reg->umax_value, val);
2572 true_reg->umin_value = max(true_reg->umin_value, val + 1);
2575 false_reg->smax_value = min_t(s64, false_reg->smax_value, val);
2576 true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1);
2579 false_reg->umin_value = max(false_reg->umin_value, val);
2580 true_reg->umax_value = min(true_reg->umax_value, val - 1);
2583 false_reg->smin_value = max_t(s64, false_reg->smin_value, val);
2584 true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1);
2587 false_reg->umax_value = min(false_reg->umax_value, val - 1);
2588 true_reg->umin_value = max(true_reg->umin_value, val);
2591 false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1);
2592 true_reg->smin_value = max_t(s64, true_reg->smin_value, val);
2595 false_reg->umin_value = max(false_reg->umin_value, val + 1);
2596 true_reg->umax_value = min(true_reg->umax_value, val);
2599 false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1);
2600 true_reg->smax_value = min_t(s64, true_reg->smax_value, val);
2606 __reg_deduce_bounds(false_reg);
2607 __reg_deduce_bounds(true_reg);
2608 /* We might have learned some bits from the bounds. */
2609 __reg_bound_offset(false_reg);
2610 __reg_bound_offset(true_reg);
2611 /* Intersecting with the old var_off might have improved our bounds
2612 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2613 * then new var_off is (0; 0x7f...fc) which improves our umax.
2615 __update_reg_bounds(false_reg);
2616 __update_reg_bounds(true_reg);
2619 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
2622 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2623 struct bpf_reg_state *false_reg, u64 val,
2626 if (__is_pointer_value(false, false_reg))
2631 /* If this is false then we know nothing Jon Snow, but if it is
2632 * true then we know for sure.
2634 __mark_reg_known(true_reg, val);
2637 /* If this is true we know nothing Jon Snow, but if it is false
2638 * we know the value for sure;
2640 __mark_reg_known(false_reg, val);
2643 true_reg->umax_value = min(true_reg->umax_value, val - 1);
2644 false_reg->umin_value = max(false_reg->umin_value, val);
2647 true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1);
2648 false_reg->smin_value = max_t(s64, false_reg->smin_value, val);
2651 true_reg->umin_value = max(true_reg->umin_value, val + 1);
2652 false_reg->umax_value = min(false_reg->umax_value, val);
2655 true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1);
2656 false_reg->smax_value = min_t(s64, false_reg->smax_value, val);
2659 true_reg->umax_value = min(true_reg->umax_value, val);
2660 false_reg->umin_value = max(false_reg->umin_value, val + 1);
2663 true_reg->smax_value = min_t(s64, true_reg->smax_value, val);
2664 false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1);
2667 true_reg->umin_value = max(true_reg->umin_value, val);
2668 false_reg->umax_value = min(false_reg->umax_value, val - 1);
2671 true_reg->smin_value = max_t(s64, true_reg->smin_value, val);
2672 false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1);
2678 __reg_deduce_bounds(false_reg);
2679 __reg_deduce_bounds(true_reg);
2680 /* We might have learned some bits from the bounds. */
2681 __reg_bound_offset(false_reg);
2682 __reg_bound_offset(true_reg);
2683 /* Intersecting with the old var_off might have improved our bounds
2684 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2685 * then new var_off is (0; 0x7f...fc) which improves our umax.
2687 __update_reg_bounds(false_reg);
2688 __update_reg_bounds(true_reg);
2691 /* Regs are known to be equal, so intersect their min/max/var_off */
2692 static void __reg_combine_min_max(struct bpf_reg_state *src_reg,
2693 struct bpf_reg_state *dst_reg)
2695 src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value,
2696 dst_reg->umin_value);
2697 src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value,
2698 dst_reg->umax_value);
2699 src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value,
2700 dst_reg->smin_value);
2701 src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value,
2702 dst_reg->smax_value);
2703 src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off,
2705 /* We might have learned new bounds from the var_off. */
2706 __update_reg_bounds(src_reg);
2707 __update_reg_bounds(dst_reg);
2708 /* We might have learned something about the sign bit. */
2709 __reg_deduce_bounds(src_reg);
2710 __reg_deduce_bounds(dst_reg);
2711 /* We might have learned some bits from the bounds. */
2712 __reg_bound_offset(src_reg);
2713 __reg_bound_offset(dst_reg);
2714 /* Intersecting with the old var_off might have improved our bounds
2715 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2716 * then new var_off is (0; 0x7f...fc) which improves our umax.
2718 __update_reg_bounds(src_reg);
2719 __update_reg_bounds(dst_reg);
2722 static void reg_combine_min_max(struct bpf_reg_state *true_src,
2723 struct bpf_reg_state *true_dst,
2724 struct bpf_reg_state *false_src,
2725 struct bpf_reg_state *false_dst,
2730 __reg_combine_min_max(true_src, true_dst);
2733 __reg_combine_min_max(false_src, false_dst);
2738 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2741 struct bpf_reg_state *reg = ®s[regno];
2743 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2744 /* Old offset (both fixed and variable parts) should
2745 * have been known-zero, because we don't allow pointer
2746 * arithmetic on pointers that might be NULL.
2748 if (WARN_ON_ONCE(reg->smin_value || reg->smax_value ||
2749 !tnum_equals_const(reg->var_off, 0) ||
2751 __mark_reg_known_zero(reg);
2755 reg->type = SCALAR_VALUE;
2756 } else if (reg->map_ptr->inner_map_meta) {
2757 reg->type = CONST_PTR_TO_MAP;
2758 reg->map_ptr = reg->map_ptr->inner_map_meta;
2760 reg->type = PTR_TO_MAP_VALUE;
2762 /* We don't need id from this point onwards anymore, thus we
2763 * should better reset it, so that state pruning has chances
2770 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2771 * be folded together at some point.
2773 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2776 struct bpf_reg_state *regs = state->regs;
2777 u32 id = regs[regno].id;
2780 for (i = 0; i < MAX_BPF_REG; i++)
2781 mark_map_reg(regs, i, id, is_null);
2783 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) {
2784 if (state->stack[i].slot_type[0] != STACK_SPILL)
2786 mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null);
2790 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2791 struct bpf_insn *insn, int *insn_idx)
2793 struct bpf_verifier_state *other_branch, *this_branch = env->cur_state;
2794 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2795 u8 opcode = BPF_OP(insn->code);
2798 if (opcode > BPF_JSLE) {
2799 verbose(env, "invalid BPF_JMP opcode %x\n", opcode);
2803 if (BPF_SRC(insn->code) == BPF_X) {
2804 if (insn->imm != 0) {
2805 verbose(env, "BPF_JMP uses reserved fields\n");
2809 /* check src1 operand */
2810 err = check_reg_arg(env, insn->src_reg, SRC_OP);
2814 if (is_pointer_value(env, insn->src_reg)) {
2815 verbose(env, "R%d pointer comparison prohibited\n",
2820 if (insn->src_reg != BPF_REG_0) {
2821 verbose(env, "BPF_JMP uses reserved fields\n");
2826 /* check src2 operand */
2827 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
2831 dst_reg = ®s[insn->dst_reg];
2833 /* detect if R == 0 where R was initialized to zero earlier */
2834 if (BPF_SRC(insn->code) == BPF_K &&
2835 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2836 dst_reg->type == SCALAR_VALUE &&
2837 tnum_equals_const(dst_reg->var_off, insn->imm)) {
2838 if (opcode == BPF_JEQ) {
2839 /* if (imm == imm) goto pc+off;
2840 * only follow the goto, ignore fall-through
2842 *insn_idx += insn->off;
2845 /* if (imm != imm) goto pc+off;
2846 * only follow fall-through branch, since
2847 * that's where the program will go
2853 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2857 /* detect if we are comparing against a constant value so we can adjust
2858 * our min/max values for our dst register.
2859 * this is only legit if both are scalars (or pointers to the same
2860 * object, I suppose, but we don't support that right now), because
2861 * otherwise the different base pointers mean the offsets aren't
2864 if (BPF_SRC(insn->code) == BPF_X) {
2865 if (dst_reg->type == SCALAR_VALUE &&
2866 regs[insn->src_reg].type == SCALAR_VALUE) {
2867 if (tnum_is_const(regs[insn->src_reg].var_off))
2868 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2869 dst_reg, regs[insn->src_reg].var_off.value,
2871 else if (tnum_is_const(dst_reg->var_off))
2872 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2873 ®s[insn->src_reg],
2874 dst_reg->var_off.value, opcode);
2875 else if (opcode == BPF_JEQ || opcode == BPF_JNE)
2876 /* Comparing for equality, we can combine knowledge */
2877 reg_combine_min_max(&other_branch->regs[insn->src_reg],
2878 &other_branch->regs[insn->dst_reg],
2879 ®s[insn->src_reg],
2880 ®s[insn->dst_reg], opcode);
2882 } else if (dst_reg->type == SCALAR_VALUE) {
2883 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2884 dst_reg, insn->imm, opcode);
2887 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2888 if (BPF_SRC(insn->code) == BPF_K &&
2889 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2890 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2891 /* Mark all identical map registers in each branch as either
2892 * safe or unknown depending R == 0 or R != 0 conditional.
2894 mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE);
2895 mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ);
2896 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2897 dst_reg->type == PTR_TO_PACKET &&
2898 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2899 /* pkt_data' > pkt_end */
2900 find_good_pkt_pointers(this_branch, dst_reg,
2901 PTR_TO_PACKET, false);
2902 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2903 dst_reg->type == PTR_TO_PACKET_END &&
2904 regs[insn->src_reg].type == PTR_TO_PACKET) {
2905 /* pkt_end > pkt_data' */
2906 find_good_pkt_pointers(other_branch, ®s[insn->src_reg],
2907 PTR_TO_PACKET, true);
2908 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLT &&
2909 dst_reg->type == PTR_TO_PACKET &&
2910 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2911 /* pkt_data' < pkt_end */
2912 find_good_pkt_pointers(other_branch, dst_reg, PTR_TO_PACKET,
2914 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLT &&
2915 dst_reg->type == PTR_TO_PACKET_END &&
2916 regs[insn->src_reg].type == PTR_TO_PACKET) {
2917 /* pkt_end < pkt_data' */
2918 find_good_pkt_pointers(this_branch, ®s[insn->src_reg],
2919 PTR_TO_PACKET, false);
2920 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2921 dst_reg->type == PTR_TO_PACKET &&
2922 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2923 /* pkt_data' >= pkt_end */
2924 find_good_pkt_pointers(this_branch, dst_reg,
2925 PTR_TO_PACKET, true);
2926 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2927 dst_reg->type == PTR_TO_PACKET_END &&
2928 regs[insn->src_reg].type == PTR_TO_PACKET) {
2929 /* pkt_end >= pkt_data' */
2930 find_good_pkt_pointers(other_branch, ®s[insn->src_reg],
2931 PTR_TO_PACKET, false);
2932 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLE &&
2933 dst_reg->type == PTR_TO_PACKET &&
2934 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2935 /* pkt_data' <= pkt_end */
2936 find_good_pkt_pointers(other_branch, dst_reg,
2937 PTR_TO_PACKET, false);
2938 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLE &&
2939 dst_reg->type == PTR_TO_PACKET_END &&
2940 regs[insn->src_reg].type == PTR_TO_PACKET) {
2941 /* pkt_end <= pkt_data' */
2942 find_good_pkt_pointers(this_branch, ®s[insn->src_reg],
2943 PTR_TO_PACKET, true);
2944 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2945 dst_reg->type == PTR_TO_PACKET_META &&
2946 reg_is_init_pkt_pointer(®s[insn->src_reg], PTR_TO_PACKET)) {
2947 find_good_pkt_pointers(this_branch, dst_reg,
2948 PTR_TO_PACKET_META, false);
2949 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLT &&
2950 dst_reg->type == PTR_TO_PACKET_META &&
2951 reg_is_init_pkt_pointer(®s[insn->src_reg], PTR_TO_PACKET)) {
2952 find_good_pkt_pointers(other_branch, dst_reg,
2953 PTR_TO_PACKET_META, false);
2954 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2955 reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
2956 regs[insn->src_reg].type == PTR_TO_PACKET_META) {
2957 find_good_pkt_pointers(other_branch, ®s[insn->src_reg],
2958 PTR_TO_PACKET_META, false);
2959 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JLE &&
2960 reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) &&
2961 regs[insn->src_reg].type == PTR_TO_PACKET_META) {
2962 find_good_pkt_pointers(this_branch, ®s[insn->src_reg],
2963 PTR_TO_PACKET_META, false);
2964 } else if (is_pointer_value(env, insn->dst_reg)) {
2965 verbose(env, "R%d pointer comparison prohibited\n",
2970 print_verifier_state(env, this_branch);
2974 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2975 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2977 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2979 return (struct bpf_map *) (unsigned long) imm64;
2982 /* verify BPF_LD_IMM64 instruction */
2983 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2985 struct bpf_reg_state *regs = cur_regs(env);
2988 if (BPF_SIZE(insn->code) != BPF_DW) {
2989 verbose(env, "invalid BPF_LD_IMM insn\n");
2992 if (insn->off != 0) {
2993 verbose(env, "BPF_LD_IMM64 uses reserved fields\n");
2997 err = check_reg_arg(env, insn->dst_reg, DST_OP);
3001 if (insn->src_reg == 0) {
3002 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
3004 regs[insn->dst_reg].type = SCALAR_VALUE;
3005 __mark_reg_known(®s[insn->dst_reg], imm);
3009 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
3010 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
3012 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
3013 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
3017 static bool may_access_skb(enum bpf_prog_type type)
3020 case BPF_PROG_TYPE_SOCKET_FILTER:
3021 case BPF_PROG_TYPE_SCHED_CLS:
3022 case BPF_PROG_TYPE_SCHED_ACT:
3029 /* verify safety of LD_ABS|LD_IND instructions:
3030 * - they can only appear in the programs where ctx == skb
3031 * - since they are wrappers of function calls, they scratch R1-R5 registers,
3032 * preserve R6-R9, and store return value into R0
3035 * ctx == skb == R6 == CTX
3038 * SRC == any register
3039 * IMM == 32-bit immediate
3042 * R0 - 8/16/32-bit skb data converted to cpu endianness
3044 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
3046 struct bpf_reg_state *regs = cur_regs(env);
3047 u8 mode = BPF_MODE(insn->code);
3050 if (!may_access_skb(env->prog->type)) {
3051 verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
3055 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
3056 BPF_SIZE(insn->code) == BPF_DW ||
3057 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
3058 verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n");
3062 /* check whether implicit source operand (register R6) is readable */
3063 err = check_reg_arg(env, BPF_REG_6, SRC_OP);
3067 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
3069 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
3073 if (mode == BPF_IND) {
3074 /* check explicit source operand */
3075 err = check_reg_arg(env, insn->src_reg, SRC_OP);
3080 /* reset caller saved regs to unreadable */
3081 for (i = 0; i < CALLER_SAVED_REGS; i++) {
3082 mark_reg_not_init(env, regs, caller_saved[i]);
3083 check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK);
3086 /* mark destination R0 register as readable, since it contains
3087 * the value fetched from the packet.
3088 * Already marked as written above.
3090 mark_reg_unknown(env, regs, BPF_REG_0);
3094 static int check_return_code(struct bpf_verifier_env *env)
3096 struct bpf_reg_state *reg;
3097 struct tnum range = tnum_range(0, 1);
3099 switch (env->prog->type) {
3100 case BPF_PROG_TYPE_CGROUP_SKB:
3101 case BPF_PROG_TYPE_CGROUP_SOCK:
3102 case BPF_PROG_TYPE_SOCK_OPS:
3108 reg = cur_regs(env) + BPF_REG_0;
3109 if (reg->type != SCALAR_VALUE) {
3110 verbose(env, "At program exit the register R0 is not a known value (%s)\n",
3111 reg_type_str[reg->type]);
3115 if (!tnum_in(range, reg->var_off)) {
3116 verbose(env, "At program exit the register R0 ");
3117 if (!tnum_is_unknown(reg->var_off)) {
3120 tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off);
3121 verbose(env, "has value %s", tn_buf);
3123 verbose(env, "has unknown scalar value");
3125 verbose(env, " should have been 0 or 1\n");
3131 /* non-recursive DFS pseudo code
3132 * 1 procedure DFS-iterative(G,v):
3133 * 2 label v as discovered
3134 * 3 let S be a stack
3136 * 5 while S is not empty
3138 * 7 if t is what we're looking for:
3140 * 9 for all edges e in G.adjacentEdges(t) do
3141 * 10 if edge e is already labelled
3142 * 11 continue with the next edge
3143 * 12 w <- G.adjacentVertex(t,e)
3144 * 13 if vertex w is not discovered and not explored
3145 * 14 label e as tree-edge
3146 * 15 label w as discovered
3149 * 18 else if vertex w is discovered
3150 * 19 label e as back-edge
3152 * 21 // vertex w is explored
3153 * 22 label e as forward- or cross-edge
3154 * 23 label t as explored
3159 * 0x11 - discovered and fall-through edge labelled
3160 * 0x12 - discovered and fall-through and branch edges labelled
3171 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
3173 static int *insn_stack; /* stack of insns to process */
3174 static int cur_stack; /* current stack index */
3175 static int *insn_state;
3177 /* t, w, e - match pseudo-code above:
3178 * t - index of current instruction
3179 * w - next instruction
3182 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
3184 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
3187 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
3190 if (w < 0 || w >= env->prog->len) {
3191 verbose(env, "jump out of range from insn %d to %d\n", t, w);
3196 /* mark branch target for state pruning */
3197 env->explored_states[w] = STATE_LIST_MARK;
3199 if (insn_state[w] == 0) {
3201 insn_state[t] = DISCOVERED | e;
3202 insn_state[w] = DISCOVERED;
3203 if (cur_stack >= env->prog->len)
3205 insn_stack[cur_stack++] = w;
3207 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
3208 verbose(env, "back-edge from insn %d to %d\n", t, w);
3210 } else if (insn_state[w] == EXPLORED) {
3211 /* forward- or cross-edge */
3212 insn_state[t] = DISCOVERED | e;
3214 verbose(env, "insn state internal bug\n");
3220 /* non-recursive depth-first-search to detect loops in BPF program
3221 * loop == back-edge in directed graph
3223 static int check_cfg(struct bpf_verifier_env *env)
3225 struct bpf_insn *insns = env->prog->insnsi;
3226 int insn_cnt = env->prog->len;
3230 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
3234 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
3240 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
3241 insn_stack[0] = 0; /* 0 is the first instruction */
3247 t = insn_stack[cur_stack - 1];
3249 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
3250 u8 opcode = BPF_OP(insns[t].code);
3252 if (opcode == BPF_EXIT) {
3254 } else if (opcode == BPF_CALL) {
3255 ret = push_insn(t, t + 1, FALLTHROUGH, env);
3260 if (t + 1 < insn_cnt)
3261 env->explored_states[t + 1] = STATE_LIST_MARK;
3262 } else if (opcode == BPF_JA) {
3263 if (BPF_SRC(insns[t].code) != BPF_K) {
3267 /* unconditional jump with single edge */
3268 ret = push_insn(t, t + insns[t].off + 1,
3274 /* tell verifier to check for equivalent states
3275 * after every call and jump
3277 if (t + 1 < insn_cnt)
3278 env->explored_states[t + 1] = STATE_LIST_MARK;
3280 /* conditional jump with two edges */
3281 env->explored_states[t] = STATE_LIST_MARK;
3282 ret = push_insn(t, t + 1, FALLTHROUGH, env);
3288 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
3295 /* all other non-branch instructions with single
3298 ret = push_insn(t, t + 1, FALLTHROUGH, env);
3306 insn_state[t] = EXPLORED;
3307 if (cur_stack-- <= 0) {
3308 verbose(env, "pop stack internal bug\n");
3315 for (i = 0; i < insn_cnt; i++) {
3316 if (insn_state[i] != EXPLORED) {
3317 verbose(env, "unreachable insn %d\n", i);
3322 ret = 0; /* cfg looks good */
3330 /* check %cur's range satisfies %old's */
3331 static bool range_within(struct bpf_reg_state *old,
3332 struct bpf_reg_state *cur)
3334 return old->umin_value <= cur->umin_value &&
3335 old->umax_value >= cur->umax_value &&
3336 old->smin_value <= cur->smin_value &&
3337 old->smax_value >= cur->smax_value;
3340 /* Maximum number of register states that can exist at once */
3341 #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
3347 /* If in the old state two registers had the same id, then they need to have
3348 * the same id in the new state as well. But that id could be different from
3349 * the old state, so we need to track the mapping from old to new ids.
3350 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
3351 * regs with old id 5 must also have new id 9 for the new state to be safe. But
3352 * regs with a different old id could still have new id 9, we don't care about
3354 * So we look through our idmap to see if this old id has been seen before. If
3355 * so, we require the new id to match; otherwise, we add the id pair to the map.
3357 static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap)
3361 for (i = 0; i < ID_MAP_SIZE; i++) {
3362 if (!idmap[i].old) {
3363 /* Reached an empty slot; haven't seen this id before */
3364 idmap[i].old = old_id;
3365 idmap[i].cur = cur_id;
3368 if (idmap[i].old == old_id)
3369 return idmap[i].cur == cur_id;
3371 /* We ran out of idmap slots, which should be impossible */
3376 /* Returns true if (rold safe implies rcur safe) */
3377 static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur,
3378 struct idpair *idmap)
3380 if (!(rold->live & REG_LIVE_READ))
3381 /* explored state didn't use this */
3384 if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, live)) == 0)
3387 if (rold->type == NOT_INIT)
3388 /* explored state can't have used this */
3390 if (rcur->type == NOT_INIT)
3392 switch (rold->type) {
3394 if (rcur->type == SCALAR_VALUE) {
3395 /* new val must satisfy old val knowledge */
3396 return range_within(rold, rcur) &&
3397 tnum_in(rold->var_off, rcur->var_off);
3399 /* if we knew anything about the old value, we're not
3400 * equal, because we can't know anything about the
3401 * scalar value of the pointer in the new value.
3403 return rold->umin_value == 0 &&
3404 rold->umax_value == U64_MAX &&
3405 rold->smin_value == S64_MIN &&
3406 rold->smax_value == S64_MAX &&
3407 tnum_is_unknown(rold->var_off);
3409 case PTR_TO_MAP_VALUE:
3410 /* If the new min/max/var_off satisfy the old ones and
3411 * everything else matches, we are OK.
3412 * We don't care about the 'id' value, because nothing
3413 * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL)
3415 return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 &&
3416 range_within(rold, rcur) &&
3417 tnum_in(rold->var_off, rcur->var_off);
3418 case PTR_TO_MAP_VALUE_OR_NULL:
3419 /* a PTR_TO_MAP_VALUE could be safe to use as a
3420 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
3421 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
3422 * checked, doing so could have affected others with the same
3423 * id, and we can't check for that because we lost the id when
3424 * we converted to a PTR_TO_MAP_VALUE.
3426 if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL)
3428 if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)))
3430 /* Check our ids match any regs they're supposed to */
3431 return check_ids(rold->id, rcur->id, idmap);
3432 case PTR_TO_PACKET_META:
3434 if (rcur->type != rold->type)
3436 /* We must have at least as much range as the old ptr
3437 * did, so that any accesses which were safe before are
3438 * still safe. This is true even if old range < old off,
3439 * since someone could have accessed through (ptr - k), or
3440 * even done ptr -= k in a register, to get a safe access.
3442 if (rold->range > rcur->range)
3444 /* If the offsets don't match, we can't trust our alignment;
3445 * nor can we be sure that we won't fall out of range.
3447 if (rold->off != rcur->off)
3449 /* id relations must be preserved */
3450 if (rold->id && !check_ids(rold->id, rcur->id, idmap))
3452 /* new val must satisfy old val knowledge */
3453 return range_within(rold, rcur) &&
3454 tnum_in(rold->var_off, rcur->var_off);
3456 case CONST_PTR_TO_MAP:
3458 case PTR_TO_PACKET_END:
3459 /* Only valid matches are exact, which memcmp() above
3460 * would have accepted
3463 /* Don't know what's going on, just say it's not safe */
3467 /* Shouldn't get here; if we do, say it's not safe */
3472 static bool stacksafe(struct bpf_verifier_state *old,
3473 struct bpf_verifier_state *cur,
3474 struct idpair *idmap)
3478 /* if explored stack has more populated slots than current stack
3479 * such stacks are not equivalent
3481 if (old->allocated_stack > cur->allocated_stack)
3484 /* walk slots of the explored stack and ignore any additional
3485 * slots in the current stack, since explored(safe) state
3488 for (i = 0; i < old->allocated_stack; i++) {
3489 spi = i / BPF_REG_SIZE;
3491 if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID)
3493 if (old->stack[spi].slot_type[i % BPF_REG_SIZE] !=
3494 cur->stack[spi].slot_type[i % BPF_REG_SIZE])
3495 /* Ex: old explored (safe) state has STACK_SPILL in
3496 * this stack slot, but current has has STACK_MISC ->
3497 * this verifier states are not equivalent,
3498 * return false to continue verification of this path
3501 if (i % BPF_REG_SIZE)
3503 if (old->stack[spi].slot_type[0] != STACK_SPILL)
3505 if (!regsafe(&old->stack[spi].spilled_ptr,
3506 &cur->stack[spi].spilled_ptr,
3508 /* when explored and current stack slot are both storing
3509 * spilled registers, check that stored pointers types
3510 * are the same as well.
3511 * Ex: explored safe path could have stored
3512 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
3513 * but current path has stored:
3514 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
3515 * such verifier states are not equivalent.
3516 * return false to continue verification of this path
3523 /* compare two verifier states
3525 * all states stored in state_list are known to be valid, since
3526 * verifier reached 'bpf_exit' instruction through them
3528 * this function is called when verifier exploring different branches of
3529 * execution popped from the state stack. If it sees an old state that has
3530 * more strict register state and more strict stack state then this execution
3531 * branch doesn't need to be explored further, since verifier already
3532 * concluded that more strict state leads to valid finish.
3534 * Therefore two states are equivalent if register state is more conservative
3535 * and explored stack state is more conservative than the current one.
3538 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
3539 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
3541 * In other words if current stack state (one being explored) has more
3542 * valid slots than old one that already passed validation, it means
3543 * the verifier can stop exploring and conclude that current state is valid too
3545 * Similarly with registers. If explored state has register type as invalid
3546 * whereas register type in current state is meaningful, it means that
3547 * the current state will reach 'bpf_exit' instruction safely
3549 static bool states_equal(struct bpf_verifier_env *env,
3550 struct bpf_verifier_state *old,
3551 struct bpf_verifier_state *cur)
3553 struct idpair *idmap;
3557 idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL);
3558 /* If we failed to allocate the idmap, just say it's not safe */
3562 for (i = 0; i < MAX_BPF_REG; i++) {
3563 if (!regsafe(&old->regs[i], &cur->regs[i], idmap))
3567 if (!stacksafe(old, cur, idmap))
3575 /* A write screens off any subsequent reads; but write marks come from the
3576 * straight-line code between a state and its parent. When we arrive at a
3577 * jump target (in the first iteration of the propagate_liveness() loop),
3578 * we didn't arrive by the straight-line code, so read marks in state must
3579 * propagate to parent regardless of state's write marks.
3581 static bool do_propagate_liveness(const struct bpf_verifier_state *state,
3582 struct bpf_verifier_state *parent)
3584 bool writes = parent == state->parent; /* Observe write marks */
3585 bool touched = false; /* any changes made? */
3590 /* Propagate read liveness of registers... */
3591 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
3592 /* We don't need to worry about FP liveness because it's read-only */
3593 for (i = 0; i < BPF_REG_FP; i++) {
3594 if (parent->regs[i].live & REG_LIVE_READ)
3596 if (writes && (state->regs[i].live & REG_LIVE_WRITTEN))
3598 if (state->regs[i].live & REG_LIVE_READ) {
3599 parent->regs[i].live |= REG_LIVE_READ;
3603 /* ... and stack slots */
3604 for (i = 0; i < state->allocated_stack / BPF_REG_SIZE &&
3605 i < parent->allocated_stack / BPF_REG_SIZE; i++) {
3606 if (parent->stack[i].slot_type[0] != STACK_SPILL)
3608 if (state->stack[i].slot_type[0] != STACK_SPILL)
3610 if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ)
3613 (state->stack[i].spilled_ptr.live & REG_LIVE_WRITTEN))
3615 if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) {
3616 parent->stack[i].spilled_ptr.live |= REG_LIVE_READ;
3623 /* "parent" is "a state from which we reach the current state", but initially
3624 * it is not the state->parent (i.e. "the state whose straight-line code leads
3625 * to the current state"), instead it is the state that happened to arrive at
3626 * a (prunable) equivalent of the current state. See comment above
3627 * do_propagate_liveness() for consequences of this.
3628 * This function is just a more efficient way of calling mark_reg_read() or
3629 * mark_stack_slot_read() on each reg in "parent" that is read in "state",
3630 * though it requires that parent != state->parent in the call arguments.
3632 static void propagate_liveness(const struct bpf_verifier_state *state,
3633 struct bpf_verifier_state *parent)
3635 while (do_propagate_liveness(state, parent)) {
3636 /* Something changed, so we need to feed those changes onward */
3638 parent = state->parent;
3642 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
3644 struct bpf_verifier_state_list *new_sl;
3645 struct bpf_verifier_state_list *sl;
3646 struct bpf_verifier_state *cur = env->cur_state;
3649 sl = env->explored_states[insn_idx];
3651 /* this 'insn_idx' instruction wasn't marked, so we will not
3652 * be doing state search here
3656 while (sl != STATE_LIST_MARK) {
3657 if (states_equal(env, &sl->state, cur)) {
3658 /* reached equivalent register/stack state,
3660 * Registers read by the continuation are read by us.
3661 * If we have any write marks in env->cur_state, they
3662 * will prevent corresponding reads in the continuation
3663 * from reaching our parent (an explored_state). Our
3664 * own state will get the read marks recorded, but
3665 * they'll be immediately forgotten as we're pruning
3666 * this state and will pop a new one.
3668 propagate_liveness(&sl->state, cur);
3674 /* there were no equivalent states, remember current one.
3675 * technically the current state is not proven to be safe yet,
3676 * but it will either reach bpf_exit (which means it's safe) or
3677 * it will be rejected. Since there are no loops, we won't be
3678 * seeing this 'insn_idx' instruction again on the way to bpf_exit
3680 new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL);
3684 /* add new state to the head of linked list */
3685 err = copy_verifier_state(&new_sl->state, cur);
3687 free_verifier_state(&new_sl->state, false);
3691 new_sl->next = env->explored_states[insn_idx];
3692 env->explored_states[insn_idx] = new_sl;
3693 /* connect new state to parentage chain */
3694 cur->parent = &new_sl->state;
3695 /* clear write marks in current state: the writes we did are not writes
3696 * our child did, so they don't screen off its reads from us.
3697 * (There are no read marks in current state, because reads always mark
3698 * their parent and current state never has children yet. Only
3699 * explored_states can get read marks.)
3701 for (i = 0; i < BPF_REG_FP; i++)
3702 cur->regs[i].live = REG_LIVE_NONE;
3703 for (i = 0; i < cur->allocated_stack / BPF_REG_SIZE; i++)
3704 if (cur->stack[i].slot_type[0] == STACK_SPILL)
3705 cur->stack[i].spilled_ptr.live = REG_LIVE_NONE;
3709 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
3710 int insn_idx, int prev_insn_idx)
3712 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
3715 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
3718 static int do_check(struct bpf_verifier_env *env)
3720 struct bpf_verifier_state *state;
3721 struct bpf_insn *insns = env->prog->insnsi;
3722 struct bpf_reg_state *regs;
3723 int insn_cnt = env->prog->len;
3724 int insn_idx, prev_insn_idx = 0;
3725 int insn_processed = 0;
3726 bool do_print_state = false;
3728 state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL);
3731 env->cur_state = state;
3732 init_reg_state(env, state->regs);
3733 state->parent = NULL;
3736 struct bpf_insn *insn;
3740 if (insn_idx >= insn_cnt) {
3741 verbose(env, "invalid insn idx %d insn_cnt %d\n",
3742 insn_idx, insn_cnt);
3746 insn = &insns[insn_idx];
3747 class = BPF_CLASS(insn->code);
3749 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
3751 "BPF program is too large. Processed %d insn\n",
3756 err = is_state_visited(env, insn_idx);
3760 /* found equivalent state, can prune the search */
3761 if (env->log.level) {
3763 verbose(env, "\nfrom %d to %d: safe\n",
3764 prev_insn_idx, insn_idx);
3766 verbose(env, "%d: safe\n", insn_idx);
3768 goto process_bpf_exit;
3774 if (env->log.level > 1 || (env->log.level && do_print_state)) {
3775 if (env->log.level > 1)
3776 verbose(env, "%d:", insn_idx);
3778 verbose(env, "\nfrom %d to %d:",
3779 prev_insn_idx, insn_idx);
3780 print_verifier_state(env, state);
3781 do_print_state = false;
3784 if (env->log.level) {
3785 verbose(env, "%d: ", insn_idx);
3786 print_bpf_insn(verbose, env, insn,
3787 env->allow_ptr_leaks);
3790 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
3794 regs = cur_regs(env);
3795 if (class == BPF_ALU || class == BPF_ALU64) {
3796 err = check_alu_op(env, insn);
3800 } else if (class == BPF_LDX) {
3801 enum bpf_reg_type *prev_src_type, src_reg_type;
3803 /* check for reserved fields is already done */
3805 /* check src operand */
3806 err = check_reg_arg(env, insn->src_reg, SRC_OP);
3810 err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK);
3814 src_reg_type = regs[insn->src_reg].type;
3816 /* check that memory (src_reg + off) is readable,
3817 * the state of dst_reg will be updated by this func
3819 err = check_mem_access(env, insn_idx, insn->src_reg, insn->off,
3820 BPF_SIZE(insn->code), BPF_READ,
3825 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
3827 if (*prev_src_type == NOT_INIT) {
3829 * dst_reg = *(u32 *)(src_reg + off)
3830 * save type to validate intersecting paths
3832 *prev_src_type = src_reg_type;
3834 } else if (src_reg_type != *prev_src_type &&
3835 (src_reg_type == PTR_TO_CTX ||
3836 *prev_src_type == PTR_TO_CTX)) {
3837 /* ABuser program is trying to use the same insn
3838 * dst_reg = *(u32*) (src_reg + off)
3839 * with different pointer types:
3840 * src_reg == ctx in one branch and
3841 * src_reg == stack|map in some other branch.
3844 verbose(env, "same insn cannot be used with different pointers\n");
3848 } else if (class == BPF_STX) {
3849 enum bpf_reg_type *prev_dst_type, dst_reg_type;
3851 if (BPF_MODE(insn->code) == BPF_XADD) {
3852 err = check_xadd(env, insn_idx, insn);
3859 /* check src1 operand */
3860 err = check_reg_arg(env, insn->src_reg, SRC_OP);
3863 /* check src2 operand */
3864 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
3868 dst_reg_type = regs[insn->dst_reg].type;
3870 /* check that memory (dst_reg + off) is writeable */
3871 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
3872 BPF_SIZE(insn->code), BPF_WRITE,
3877 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
3879 if (*prev_dst_type == NOT_INIT) {
3880 *prev_dst_type = dst_reg_type;
3881 } else if (dst_reg_type != *prev_dst_type &&
3882 (dst_reg_type == PTR_TO_CTX ||
3883 *prev_dst_type == PTR_TO_CTX)) {
3884 verbose(env, "same insn cannot be used with different pointers\n");
3888 } else if (class == BPF_ST) {
3889 if (BPF_MODE(insn->code) != BPF_MEM ||
3890 insn->src_reg != BPF_REG_0) {
3891 verbose(env, "BPF_ST uses reserved fields\n");
3894 /* check src operand */
3895 err = check_reg_arg(env, insn->dst_reg, SRC_OP);
3899 /* check that memory (dst_reg + off) is writeable */
3900 err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off,
3901 BPF_SIZE(insn->code), BPF_WRITE,
3906 } else if (class == BPF_JMP) {
3907 u8 opcode = BPF_OP(insn->code);
3909 if (opcode == BPF_CALL) {
3910 if (BPF_SRC(insn->code) != BPF_K ||
3912 insn->src_reg != BPF_REG_0 ||
3913 insn->dst_reg != BPF_REG_0) {
3914 verbose(env, "BPF_CALL uses reserved fields\n");
3918 err = check_call(env, insn->imm, insn_idx);
3922 } else if (opcode == BPF_JA) {
3923 if (BPF_SRC(insn->code) != BPF_K ||
3925 insn->src_reg != BPF_REG_0 ||
3926 insn->dst_reg != BPF_REG_0) {
3927 verbose(env, "BPF_JA uses reserved fields\n");
3931 insn_idx += insn->off + 1;
3934 } else if (opcode == BPF_EXIT) {
3935 if (BPF_SRC(insn->code) != BPF_K ||
3937 insn->src_reg != BPF_REG_0 ||
3938 insn->dst_reg != BPF_REG_0) {
3939 verbose(env, "BPF_EXIT uses reserved fields\n");
3943 /* eBPF calling convetion is such that R0 is used
3944 * to return the value from eBPF program.
3945 * Make sure that it's readable at this time
3946 * of bpf_exit, which means that program wrote
3947 * something into it earlier
3949 err = check_reg_arg(env, BPF_REG_0, SRC_OP);
3953 if (is_pointer_value(env, BPF_REG_0)) {
3954 verbose(env, "R0 leaks addr as return value\n");
3958 err = check_return_code(env);
3962 err = pop_stack(env, &prev_insn_idx, &insn_idx);
3968 do_print_state = true;
3972 err = check_cond_jmp_op(env, insn, &insn_idx);
3976 } else if (class == BPF_LD) {
3977 u8 mode = BPF_MODE(insn->code);
3979 if (mode == BPF_ABS || mode == BPF_IND) {
3980 err = check_ld_abs(env, insn);
3984 } else if (mode == BPF_IMM) {
3985 err = check_ld_imm(env, insn);
3991 verbose(env, "invalid BPF_LD mode\n");
3995 verbose(env, "unknown insn class %d\n", class);
4002 verbose(env, "processed %d insns, stack depth %d\n", insn_processed,
4003 env->prog->aux->stack_depth);
4007 static int check_map_prealloc(struct bpf_map *map)
4009 return (map->map_type != BPF_MAP_TYPE_HASH &&
4010 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
4011 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
4012 !(map->map_flags & BPF_F_NO_PREALLOC);
4015 static int check_map_prog_compatibility(struct bpf_verifier_env *env,
4016 struct bpf_map *map,
4017 struct bpf_prog *prog)
4020 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
4021 * preallocated hash maps, since doing memory allocation
4022 * in overflow_handler can crash depending on where nmi got
4025 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
4026 if (!check_map_prealloc(map)) {
4027 verbose(env, "perf_event programs can only use preallocated hash map\n");
4030 if (map->inner_map_meta &&
4031 !check_map_prealloc(map->inner_map_meta)) {
4032 verbose(env, "perf_event programs can only use preallocated inner hash map\n");
4039 /* look for pseudo eBPF instructions that access map FDs and
4040 * replace them with actual map pointers
4042 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
4044 struct bpf_insn *insn = env->prog->insnsi;
4045 int insn_cnt = env->prog->len;
4048 err = bpf_prog_calc_tag(env->prog);
4052 for (i = 0; i < insn_cnt; i++, insn++) {
4053 if (BPF_CLASS(insn->code) == BPF_LDX &&
4054 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
4055 verbose(env, "BPF_LDX uses reserved fields\n");
4059 if (BPF_CLASS(insn->code) == BPF_STX &&
4060 ((BPF_MODE(insn->code) != BPF_MEM &&
4061 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
4062 verbose(env, "BPF_STX uses reserved fields\n");
4066 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
4067 struct bpf_map *map;
4070 if (i == insn_cnt - 1 || insn[1].code != 0 ||
4071 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
4073 verbose(env, "invalid bpf_ld_imm64 insn\n");
4077 if (insn->src_reg == 0)
4078 /* valid generic load 64-bit imm */
4081 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
4083 "unrecognized bpf_ld_imm64 insn\n");
4087 f = fdget(insn->imm);
4088 map = __bpf_map_get(f);
4090 verbose(env, "fd %d is not pointing to valid bpf_map\n",
4092 return PTR_ERR(map);
4095 err = check_map_prog_compatibility(env, map, env->prog);
4101 /* store map pointer inside BPF_LD_IMM64 instruction */
4102 insn[0].imm = (u32) (unsigned long) map;
4103 insn[1].imm = ((u64) (unsigned long) map) >> 32;
4105 /* check whether we recorded this map already */
4106 for (j = 0; j < env->used_map_cnt; j++)
4107 if (env->used_maps[j] == map) {
4112 if (env->used_map_cnt >= MAX_USED_MAPS) {
4117 /* hold the map. If the program is rejected by verifier,
4118 * the map will be released by release_maps() or it
4119 * will be used by the valid program until it's unloaded
4120 * and all maps are released in free_bpf_prog_info()
4122 map = bpf_map_inc(map, false);
4125 return PTR_ERR(map);
4127 env->used_maps[env->used_map_cnt++] = map;
4136 /* now all pseudo BPF_LD_IMM64 instructions load valid
4137 * 'struct bpf_map *' into a register instead of user map_fd.
4138 * These pointers will be used later by verifier to validate map access.
4143 /* drop refcnt of maps used by the rejected program */
4144 static void release_maps(struct bpf_verifier_env *env)
4148 for (i = 0; i < env->used_map_cnt; i++)
4149 bpf_map_put(env->used_maps[i]);
4152 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
4153 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
4155 struct bpf_insn *insn = env->prog->insnsi;
4156 int insn_cnt = env->prog->len;
4159 for (i = 0; i < insn_cnt; i++, insn++)
4160 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
4164 /* single env->prog->insni[off] instruction was replaced with the range
4165 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
4166 * [0, off) and [off, end) to new locations, so the patched range stays zero
4168 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
4171 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
4175 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
4178 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
4179 memcpy(new_data + off + cnt - 1, old_data + off,
4180 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
4181 env->insn_aux_data = new_data;
4186 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
4187 const struct bpf_insn *patch, u32 len)
4189 struct bpf_prog *new_prog;
4191 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
4194 if (adjust_insn_aux_data(env, new_prog->len, off, len))
4199 /* convert load instructions that access fields of 'struct __sk_buff'
4200 * into sequence of instructions that access fields of 'struct sk_buff'
4202 static int convert_ctx_accesses(struct bpf_verifier_env *env)
4204 const struct bpf_verifier_ops *ops = env->ops;
4205 int i, cnt, size, ctx_field_size, delta = 0;
4206 const int insn_cnt = env->prog->len;
4207 struct bpf_insn insn_buf[16], *insn;
4208 struct bpf_prog *new_prog;
4209 enum bpf_access_type type;
4210 bool is_narrower_load;
4213 if (ops->gen_prologue) {
4214 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
4216 if (cnt >= ARRAY_SIZE(insn_buf)) {
4217 verbose(env, "bpf verifier is misconfigured\n");
4220 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
4224 env->prog = new_prog;
4229 if (!ops->convert_ctx_access)
4232 insn = env->prog->insnsi + delta;
4234 for (i = 0; i < insn_cnt; i++, insn++) {
4235 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
4236 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
4237 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
4238 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
4240 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
4241 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
4242 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
4243 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
4248 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
4251 ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size;
4252 size = BPF_LDST_BYTES(insn);
4254 /* If the read access is a narrower load of the field,
4255 * convert to a 4/8-byte load, to minimum program type specific
4256 * convert_ctx_access changes. If conversion is successful,
4257 * we will apply proper mask to the result.
4259 is_narrower_load = size < ctx_field_size;
4260 if (is_narrower_load) {
4261 u32 off = insn->off;
4264 if (type == BPF_WRITE) {
4265 verbose(env, "bpf verifier narrow ctx access misconfigured\n");
4270 if (ctx_field_size == 4)
4272 else if (ctx_field_size == 8)
4275 insn->off = off & ~(ctx_field_size - 1);
4276 insn->code = BPF_LDX | BPF_MEM | size_code;
4280 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog,
4282 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) ||
4283 (ctx_field_size && !target_size)) {
4284 verbose(env, "bpf verifier is misconfigured\n");
4288 if (is_narrower_load && size < target_size) {
4289 if (ctx_field_size <= 4)
4290 insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg,
4291 (1 << size * 8) - 1);
4293 insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg,
4294 (1 << size * 8) - 1);
4297 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
4303 /* keep walking new program and skip insns we just inserted */
4304 env->prog = new_prog;
4305 insn = new_prog->insnsi + i + delta;
4311 /* fixup insn->imm field of bpf_call instructions
4312 * and inline eligible helpers as explicit sequence of BPF instructions
4314 * this function is called after eBPF program passed verification
4316 static int fixup_bpf_calls(struct bpf_verifier_env *env)
4318 struct bpf_prog *prog = env->prog;
4319 struct bpf_insn *insn = prog->insnsi;
4320 const struct bpf_func_proto *fn;
4321 const int insn_cnt = prog->len;
4322 struct bpf_insn insn_buf[16];
4323 struct bpf_prog *new_prog;
4324 struct bpf_map *map_ptr;
4325 int i, cnt, delta = 0;
4327 for (i = 0; i < insn_cnt; i++, insn++) {
4328 if (insn->code != (BPF_JMP | BPF_CALL))
4331 if (insn->imm == BPF_FUNC_get_route_realm)
4332 prog->dst_needed = 1;
4333 if (insn->imm == BPF_FUNC_get_prandom_u32)
4334 bpf_user_rnd_init_once();
4335 if (insn->imm == BPF_FUNC_tail_call) {
4336 /* If we tail call into other programs, we
4337 * cannot make any assumptions since they can
4338 * be replaced dynamically during runtime in
4339 * the program array.
4341 prog->cb_access = 1;
4342 env->prog->aux->stack_depth = MAX_BPF_STACK;
4344 /* mark bpf_tail_call as different opcode to avoid
4345 * conditional branch in the interpeter for every normal
4346 * call and to prevent accidental JITing by JIT compiler
4347 * that doesn't support bpf_tail_call yet
4350 insn->code = BPF_JMP | BPF_TAIL_CALL;
4354 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
4355 * handlers are currently limited to 64 bit only.
4357 if (ebpf_jit_enabled() && BITS_PER_LONG == 64 &&
4358 insn->imm == BPF_FUNC_map_lookup_elem) {
4359 map_ptr = env->insn_aux_data[i + delta].map_ptr;
4360 if (map_ptr == BPF_MAP_PTR_POISON ||
4361 !map_ptr->ops->map_gen_lookup)
4362 goto patch_call_imm;
4364 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
4365 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
4366 verbose(env, "bpf verifier is misconfigured\n");
4370 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
4377 /* keep walking new program and skip insns we just inserted */
4378 env->prog = prog = new_prog;
4379 insn = new_prog->insnsi + i + delta;
4383 if (insn->imm == BPF_FUNC_redirect_map) {
4384 /* Note, we cannot use prog directly as imm as subsequent
4385 * rewrites would still change the prog pointer. The only
4386 * stable address we can use is aux, which also works with
4387 * prog clones during blinding.
4389 u64 addr = (unsigned long)prog->aux;
4390 struct bpf_insn r4_ld[] = {
4391 BPF_LD_IMM64(BPF_REG_4, addr),
4394 cnt = ARRAY_SIZE(r4_ld);
4396 new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt);
4401 env->prog = prog = new_prog;
4402 insn = new_prog->insnsi + i + delta;
4405 fn = env->ops->get_func_proto(insn->imm);
4406 /* all functions that have prototype and verifier allowed
4407 * programs to call them, must be real in-kernel functions
4411 "kernel subsystem misconfigured func %s#%d\n",
4412 func_id_name(insn->imm), insn->imm);
4415 insn->imm = fn->func - __bpf_call_base;
4421 static void free_states(struct bpf_verifier_env *env)
4423 struct bpf_verifier_state_list *sl, *sln;
4426 if (!env->explored_states)
4429 for (i = 0; i < env->prog->len; i++) {
4430 sl = env->explored_states[i];
4433 while (sl != STATE_LIST_MARK) {
4435 free_verifier_state(&sl->state, false);
4441 kfree(env->explored_states);
4444 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
4446 struct bpf_verifier_env *env;
4447 struct bpf_verifer_log *log;
4450 /* 'struct bpf_verifier_env' can be global, but since it's not small,
4451 * allocate/free it every time bpf_check() is called
4453 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
4458 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
4461 if (!env->insn_aux_data)
4464 env->ops = bpf_verifier_ops[env->prog->type];
4466 /* grab the mutex to protect few globals used by verifier */
4467 mutex_lock(&bpf_verifier_lock);
4469 if (attr->log_level || attr->log_buf || attr->log_size) {
4470 /* user requested verbose verifier output
4471 * and supplied buffer to store the verification trace
4473 log->level = attr->log_level;
4474 log->ubuf = (char __user *) (unsigned long) attr->log_buf;
4475 log->len_total = attr->log_size;
4478 /* log attributes have to be sane */
4479 if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 ||
4480 !log->level || !log->ubuf)
4484 env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT);
4485 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
4486 env->strict_alignment = true;
4488 ret = replace_map_fd_with_map_ptr(env);
4490 goto skip_full_check;
4492 env->explored_states = kcalloc(env->prog->len,
4493 sizeof(struct bpf_verifier_state_list *),
4496 if (!env->explored_states)
4497 goto skip_full_check;
4499 ret = check_cfg(env);
4501 goto skip_full_check;
4503 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
4505 ret = do_check(env);
4506 free_verifier_state(env->cur_state, true);
4507 env->cur_state = NULL;
4510 while (!pop_stack(env, NULL, NULL));
4514 /* program is valid, convert *(u32*)(ctx + off) accesses */
4515 ret = convert_ctx_accesses(env);
4518 ret = fixup_bpf_calls(env);
4520 if (log->level && bpf_verifier_log_full(log))
4522 if (log->level && !log->ubuf) {
4524 goto err_release_maps;
4527 if (ret == 0 && env->used_map_cnt) {
4528 /* if program passed verifier, update used_maps in bpf_prog_info */
4529 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
4530 sizeof(env->used_maps[0]),
4533 if (!env->prog->aux->used_maps) {
4535 goto err_release_maps;
4538 memcpy(env->prog->aux->used_maps, env->used_maps,
4539 sizeof(env->used_maps[0]) * env->used_map_cnt);
4540 env->prog->aux->used_map_cnt = env->used_map_cnt;
4542 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
4543 * bpf_ld_imm64 instructions
4545 convert_pseudo_ld_imm64(env);
4549 if (!env->prog->aux->used_maps)
4550 /* if we didn't copy map pointers into bpf_prog_info, release
4551 * them now. Otherwise free_bpf_prog_info() will release them.
4556 mutex_unlock(&bpf_verifier_lock);
4557 vfree(env->insn_aux_data);
4563 static const struct bpf_verifier_ops * const bpf_analyzer_ops[] = {
4564 [BPF_PROG_TYPE_XDP] = &xdp_analyzer_ops,
4565 [BPF_PROG_TYPE_SCHED_CLS] = &tc_cls_act_analyzer_ops,
4568 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
4571 struct bpf_verifier_env *env;
4574 if (prog->type >= ARRAY_SIZE(bpf_analyzer_ops) ||
4575 !bpf_analyzer_ops[prog->type])
4578 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
4582 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
4585 if (!env->insn_aux_data)
4588 env->ops = bpf_analyzer_ops[env->prog->type];
4589 env->analyzer_ops = ops;
4590 env->analyzer_priv = priv;
4592 /* grab the mutex to protect few globals used by verifier */
4593 mutex_lock(&bpf_verifier_lock);
4595 env->strict_alignment = false;
4596 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
4597 env->strict_alignment = true;
4599 env->explored_states = kcalloc(env->prog->len,
4600 sizeof(struct bpf_verifier_state_list *),
4603 if (!env->explored_states)
4604 goto skip_full_check;
4606 ret = check_cfg(env);
4608 goto skip_full_check;
4610 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
4612 ret = do_check(env);
4613 free_verifier_state(env->cur_state, true);
4614 env->cur_state = NULL;
4617 while (!pop_stack(env, NULL, NULL));
4620 mutex_unlock(&bpf_verifier_lock);
4621 vfree(env->insn_aux_data);
4626 EXPORT_SYMBOL_GPL(bpf_analyzer);