[BUG] bpf: verifier: False warning for helpers in speculative branches

1 view
Skip to first unread message

Yinhao Hu

unread,
Dec 23, 2025, 6:04:08 AM (2 days ago) Dec 23
to bpf, dz...@hust.edu.cn, M2024...@hust.edu.cn, a...@kernel.org, dan...@iogearbox.net, john.fa...@gmail.com, and...@kernel.org, marti...@linux.dev, edd...@gmail.com, so...@kernel.org, yongho...@linux.dev, kps...@kernel.org, s...@fomichev.me, hao...@google.com, jo...@kernel.org, hust-os-ker...@googlegroups.com
Our fuzzer discovered a verifier bug in the BPF subsystem. The warning
triggers when Spectre mitigation is enabled and a write-performing
helper call is placed in a speculatively-executed branch.

The BPF verifier assumes `insn_aux->nospec_result` is only set for
direct memory writes (e.g., `*(u32*)(r1+off) = r2`). However, it fails
to account for helper calls (e.g., `bpf_skb_load_bytes_relative`) that
perform writes to stack memory.

The problem: `BPF_CALL` instructions have `BPF_CLASS(insn->code) ==
BPF_JMP`, which triggers the warning check. The code comment states:

```c
/* "This can currently never happen because nospec_result is only
* used for the write-ops `*(size*)(dst_reg+off)=src_reg|imm32`
* which must never skip the following insn."
*/
```

However, helper calls break this assumption:
- Helpers like `bpf_skb_load_bytes_relative` write to stack memory
- `check_helper_call()` loops through `meta.access_size`, calling
`check_mem_access(..., BPF_WRITE)`
- `check_stack_write()` sets `insn_aux->nospec_result = 1`
- Since `BPF_CALL` is encoded as `BPF_JMP | BPF_CALL`, the warning fires

Reported-by: Yinhao Hu <ddd...@hust.edu.cn>
Reported-by: Kaiyan Mei <M2024...@hust.edu.cn>
Reviewed-by: Dongliang Mu <dz...@hust.edu.cn>

### Trigger Condition

The warning occurs when both flags are set:
1. `state->speculative = 1` — Verifier processes a branch that won't
execute (marked during `check_cond_jmp_op`)
2. `insn_aux->nospec_result = 1` — A helper performs stack writes (set
during `check_helper_call`)

### Execution Flow

```
1. Drop capabilities → Enable Spectre mitigation
2. Load BPF program
└─> do_check()
├─> check_cond_jmp_op() → Marks dead branch as speculative
│ └─> push_stack(..., speculative=true)
├─> pop_stack() → state->speculative = 1
├─> check_helper_call() → Processes helper in dead branch
│ └─> check_mem_access(..., BPF_WRITE)
│ └─> insn_aux->nospec_result = 1
└─> Checks: state->speculative && insn_aux->nospec_result
└─> BPF_CLASS(insn->code) == BPF_JMP → WARNING
```

### Warning

```yaml
------------[ cut here ]------------
verifier bug: speculation barrier after jump instruction may not have
the desired effect (BPF_CLASS(insn->code) == BPF_JMP ||
BPF_CLASS(insn->code) == BPF_JMP32)
WARNING: CPU: 0 PID: 9956 at kernel/bpf/verifier.c:20536 do_check
kernel/bpf/verifier.c:20536 [inline]
WARNING: CPU: 0 PID: 9956 at kernel/bpf/verifier.c:20536
do_check_common+0xac7b/0xb200 kernel/bpf/verifier.c:23784
Modules linked in:
CPU: 0 UID: 0 PID: 9956 Comm: syz-executor206 Not tainted
6.18.0-rc4-g93ce3bee311d #3 PREEMPT(full)
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1
04/01/2014
RIP: 0010:do_check kernel/bpf/verifier.c:20536 [inline]
RIP: 0010:do_check_common+0xac7b/0xb200 kernel/bpf/verifier.c:23784
Code: 00 e9 2b 84 ff ff e8 f4 ea 4c 00 e9 31 83 ff ff e8 6a 47 e0 ff c6
05 b3 8d 6c 0f 01 90 48 c7 c7 c0 ab 76 8b e8 a6 64 9f ff 90 <0f> 0b 90
90 e9 96 83 ff ff e8 c7 ea 4c 00 e9 29 89 ff ff e8 1d eb
RSP: 0018:ffa00000080df5e0 EFLAGS: 00010282
RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffffffff817acafe
RDX: ff11000108f0ca00 RSI: ffffffff817acb0b RDI: 0000000000000001
RBP: 0000000000000017 R08: 0000000000000001 R09: ffe21c00142c4841
R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
R13: 0000000000000000 R14: ff11000024320000 R15: dffffc0000000000
FS: 000055558abb53c0(0000) GS:ff1100010ccd0000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000200000000040 CR3: 0000000028fc5000 CR4: 0000000000753ef0
PKRU: 55555554
Call Trace:
<TASK>
do_check_main kernel/bpf/verifier.c:23867 [inline]
bpf_check+0x9382/0xb930 kernel/bpf/verifier.c:25174
bpf_prog_load+0x17a6/0x2960 kernel/bpf/syscall.c:3095
__sys_bpf+0x1971/0x5390 kernel/bpf/syscall.c:6171
__do_sys_bpf kernel/bpf/syscall.c:6281 [inline]
__se_sys_bpf kernel/bpf/syscall.c:6279 [inline]
__x64_sys_bpf+0x7d/0xc0 kernel/bpf/syscall.c:6279
do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
do_syscall_64+0xcb/0xfa0 arch/x86/entry/syscall_64.c:94
entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7f13824ac64d
Code: 28 c3 e8 46 1e 00 00 66 0f 1f 44 00 00 f3 0f 1e fa 48 89 f8 48 89
f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01
f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007ffc6d73d488 EFLAGS: 00000246 ORIG_RAX: 0000000000000141
RAX: ffffffffffffffda RBX: 00007ffc6d73d698 RCX: 00007f13824ac64d
RDX: 0000000000000094 RSI: 0000200000000a00 RDI: 0000000000000005
RBP: 0000000000000001 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000002 R11: 0000000000000246 R12: 0000000000000001
R13: 00007ffc6d73d688 R14: 00007f1382529530 R15: 0000000000000001
</TASK>
```

### Proof of Concept

Tested on:
- Linux next 6.19.0-rc1-next-20251219 (commit
cc3aa43b44bdb43dfbac0fcb51c56594a11338a8)
- bpf next (commit ac1c5bc7c4c7e20e2070e6eaa673fc3e11619dbb)

```c
#define _GNU_SOURCE
#include <linux/bpf.h>
#include <linux/filter.h>
#include <stdio.h>
#include <string.h>
#include <sys/syscall.h>
#include <unistd.h>
#include <stdint.h>

int main(void)
{
/* Setup memory for capset (optional for most systems) */
syscall(__NR_mmap, 0x200000000000ul, 0x1000000ul, 7, 0x32, -1, 0);

/* Drop capabilities to enable Spectre mitigation */
*(uint32_t*)0x200000000040 = 0x20080522; /*
_LINUX_CAPABILITY_VERSION_3 */
*(uint32_t*)0x200000000044 = 0;
memset((void*)0x200000000080, 0, 24);
syscall(__NR_capset, 0x200000000040ul, 0x200000000080ul);

/* BPF program: write-performing helper in dead branch */
struct bpf_insn prog[] = {
/* r0 = 0 */
{ .code = BPF_ALU64 | BPF_MOV | BPF_K, .dst_reg = BPF_REG_0,
.imm = 0,},
/* if r0 != 1 goto +6 */
{.code = BPF_JMP | BPF_JNE | BPF_K, .dst_reg = BPF_REG_0, .imm =
1, .off = 6,},
/* R2 = offset */
{.code = BPF_ALU64 | BPF_MOV | BPF_K, .dst_reg = BPF_REG_2, .imm
= 0,},
/* R3 = R10 - 16 */
{.code = BPF_ALU64 | BPF_MOV | BPF_X, .dst_reg = BPF_REG_3,
.src_reg = BPF_REG_10,},
{.code = BPF_ALU64 | BPF_ADD | BPF_K, .dst_reg = BPF_REG_3, .imm
= -16,},
/* R4 = 4 */
{.code = BPF_ALU64 | BPF_MOV | BPF_K, .dst_reg = BPF_REG_4, .imm
= 4,},
/* R5 = flags */
{.code = BPF_ALU64 | BPF_MOV | BPF_K, .dst_reg = BPF_REG_5, .imm
= 0,},
/* call helper 68 */
{.code = BPF_JMP | BPF_CALL, .imm =
BPF_FUNC_skb_load_bytes_relative,},
/* exit */
{.code = BPF_JMP | BPF_EXIT,},
};

char log_buf[65536] = {0};
union bpf_attr attr = {
.prog_type = BPF_PROG_TYPE_SOCKET_FILTER,
.insns = (uint64_t)prog,
.insn_cnt = sizeof(prog) / sizeof(prog[0]),
.license = (uint64_t)"GPL",
.log_buf = (uint64_t)log_buf,
.log_size = sizeof(log_buf),
.log_level = 2,
};

int fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr));
if (fd < 0) {
perror("bpf");
fprintf(stderr, "\nVerifier log:\n%s\n", log_buf);
return 1;
}

printf("Loaded (fd=%d) — Check dmesg for WARNING\n", fd);
close(fd);
return 0;
}
```
config-linux-next
Reply all
Reply to author
Forward
0 new messages