[syzbot] [bpf?] KMSAN: uninit-value in ieee802154_subif_start_xmit

[syzbot]

Hello,

syzbot found the following issue on:

HEAD commit: 25ad10658dc1 riscv, bpf: Adapt bpf trampoline to optimized..
git tree: bpf-next
console+strace: https://syzkaller.appspot.com/x/log.txt?x=147cbb29a80000
kernel config: https://syzkaller.appspot.com/x/.config?x=8acaeb93ad7c6aaa
dashboard link: https://syzkaller.appspot.com/bug?extid=d61b595e9205573133b3
compiler: gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro: https://syzkaller.appspot.com/x/repro.syz?x=14d73ccea80000
C reproducer: https://syzkaller.appspot.com/x/repro.c?x=1276aedea80000

Downloadable assets:
disk image: https://storage.googleapis.com/syzbot-assets/3d378cc13d42/disk-25ad1065.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/44580fd5d1af/vmlinux-25ad1065.xz
kernel image: https://storage.googleapis.com/syzbot-assets/840587618b41/bzImage-25ad1065.xz

The issue was bisected to:

commit 8100928c881482a73ed8bd499d602bab0fe55608
Author: Yonghong Song <yongho...@linux.dev>
Date: Fri Jul 28 01:12:02 2023 +0000

bpf: Support new sign-extension mov insns

bisection log: https://syzkaller.appspot.com/x/bisect.txt?x=17970c5da80000
final oops: https://syzkaller.appspot.com/x/report.txt?x=14570c5da80000
console output: https://syzkaller.appspot.com/x/log.txt?x=10570c5da80000

IMPORTANT: if you fix the issue, please add the following tag to the commit:
Reported-by: syzbot+d61b59...@syzkaller.appspotmail.com
Fixes: 8100928c8814 ("bpf: Support new sign-extension mov insns")

general protection fault, probably for non-canonical address 0xdffffc0000000f4f: 0000 [#1] PREEMPT SMP KASAN
KASAN: probably user-memory-access in range [0x0000000000007a78-0x0000000000007a7f]
CPU: 1 PID: 0 Comm: swapper/1 Not tainted 6.5.0-rc2-syzkaller-00619-g25ad10658dc1 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 07/12/2023
RIP: 0010:strnchr+0x25/0x80 lib/string.c:403
Code: 00 00 00 00 90 f3 0f 1e fa 53 48 01 fe 48 bb 00 00 00 00 00 fc ff df 48 83 ec 18 eb 28 48 89 f8 48 89 f9 48 c1 e8 03 83 e1 07 <0f> b6 04 18 38 c8 7f 04 84 c0 75 25 0f b6 07 38 d0 74 15 48 83 c7
RSP: 0018:ffffc90000177848 EFLAGS: 00010046
RAX: 0000000000000f4f RBX: dffffc0000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000007a7b RDI: 0000000000007a78
RBP: 0000000000000001 R08: 0000000000000005 R09: 0000000000000000
R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000007a78
R13: ffffc900001779b0 R14: 0000000000000000 R15: 0000000000000003
FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00005611db5094b8 CR3: 0000000028ef0000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
bpf_bprintf_prepare+0x127/0x1490 kernel/bpf/helpers.c:823
____bpf_trace_printk kernel/trace/bpf_trace.c:385 [inline]
bpf_trace_printk+0xdb/0x180 kernel/trace/bpf_trace.c:375
bpf_prog_ebeed182d92b487f+0x38/0x3c
bpf_dispatcher_nop_func include/linux/bpf.h:1180 [inline]
__bpf_prog_run include/linux/filter.h:609 [inline]
bpf_prog_run include/linux/filter.h:616 [inline]
__bpf_trace_run kernel/trace/bpf_trace.c:2269 [inline]
bpf_trace_run1+0x148/0x400 kernel/trace/bpf_trace.c:2307
__bpf_trace_rcu_utilization+0x8e/0xc0 include/trace/events/rcu.h:27
trace_rcu_utilization+0xcd/0x120 include/trace/events/rcu.h:27
rcu_note_context_switch+0x6c/0x1ac0 kernel/rcu/tree_plugin.h:318
__schedule+0x293/0x59f0 kernel/sched/core.c:6610
schedule_idle+0x5b/0x80 kernel/sched/core.c:6814
do_idle+0x288/0x3f0 kernel/sched/idle.c:310
cpu_startup_entry+0x18/0x20 kernel/sched/idle.c:379
start_secondary+0x200/0x290 arch/x86/kernel/smpboot.c:326
secondary_startup_64_no_verify+0x167/0x16b
</TASK>
Modules linked in:
---[ end trace 0000000000000000 ]---
RIP: 0010:strnchr+0x25/0x80 lib/string.c:403
Code: 00 00 00 00 90 f3 0f 1e fa 53 48 01 fe 48 bb 00 00 00 00 00 fc ff df 48 83 ec 18 eb 28 48 89 f8 48 89 f9 48 c1 e8 03 83 e1 07 <0f> b6 04 18 38 c8 7f 04 84 c0 75 25 0f b6 07 38 d0 74 15 48 83 c7
RSP: 0018:ffffc90000177848 EFLAGS: 00010046

RAX: 0000000000000f4f RBX: dffffc0000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000007a7b RDI: 0000000000007a78
RBP: 0000000000000001 R08: 0000000000000005 R09: 0000000000000000
R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000007a78
R13: ffffc900001779b0 R14: 0000000000000000 R15: 0000000000000003
FS: 0000000000000000(0000) GS:ffff8880b9900000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00005611db5094b8 CR3: 0000000028ef0000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
----------------
Code disassembly (best guess):
0: 00 00 add %al,(%rax)
2: 00 00 add %al,(%rax)
4: 90 nop
5: f3 0f 1e fa endbr64
9: 53 push %rbx
a: 48 01 fe add %rdi,%rsi
d: 48 bb 00 00 00 00 00 movabs $0xdffffc0000000000,%rbx
14: fc ff df
17: 48 83 ec 18 sub $0x18,%rsp
1b: eb 28 jmp 0x45
1d: 48 89 f8 mov %rdi,%rax
20: 48 89 f9 mov %rdi,%rcx
23: 48 c1 e8 03 shr $0x3,%rax
27: 83 e1 07 and $0x7,%ecx
* 2a: 0f b6 04 18 movzbl (%rax,%rbx,1),%eax <-- trapping instruction
2e: 38 c8 cmp %cl,%al
30: 7f 04 jg 0x36
32: 84 c0 test %al,%al
34: 75 25 jne 0x5b
36: 0f b6 07 movzbl (%rdi),%eax
39: 38 d0 cmp %dl,%al
3b: 74 15 je 0x52
3d: 48 rex.W
3e: 83 .byte 0x83
3f: c7 .byte 0xc7


---
This report is generated by a bot. It may contain errors.
See https://goo.gl/tpsmEJ for more information about syzbot.
syzbot engineers can be reached at syzk...@googlegroups.com.

syzbot will keep track of this issue. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot.
For information about bisection process see: https://goo.gl/tpsmEJ#bisection

If the bug is already fixed, let syzbot know by replying with:
#syz fix: exact-commit-title

If you want syzbot to run the reproducer, reply with:
#syz test: git://repo/address.git branch-or-commit-hash
If you attach or paste a git patch, syzbot will apply it before testing.

If you want to change bug's subsystems, reply with:
#syz set subsystems: new-subsystem
(See the list of subsystem names on the web dashboard)

If the bug is a duplicate of another bug, reply with:
#syz dup: exact-subject-of-another-report

If you want to undo deduplication, reply with:
#syz undup

[Yonghong Song]

On 8/6/23 4:23 PM, syzbot wrote:
> Hello,
>
> syzbot found the following issue on:
>
> HEAD commit: 25ad10658dc1 riscv, bpf: Adapt bpf trampoline to optimized..
> git tree: bpf-next
> console+strace: https://syzkaller.appspot.com/x/log.txt?x=147cbb29a80000
> kernel config: https://syzkaller.appspot.com/x/.config?x=8acaeb93ad7c6aaa
> dashboard link: https://syzkaller.appspot.com/bug?extid=d61b595e9205573133b3
> compiler: gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
> syz repro: https://syzkaller.appspot.com/x/repro.syz?x=14d73ccea80000
> C reproducer: https://syzkaller.appspot.com/x/repro.c?x=1276aedea80000
>
> Downloadable assets:
> disk image: https://storage.googleapis.com/syzbot-assets/3d378cc13d42/disk-25ad1065.raw.xz
> vmlinux: https://storage.googleapis.com/syzbot-assets/44580fd5d1af/vmlinux-25ad1065.xz
> kernel image: https://storage.googleapis.com/syzbot-assets/840587618b41/bzImage-25ad1065.xz
>
> The issue was bisected to:
>
> commit 8100928c881482a73ed8bd499d602bab0fe55608
> Author: Yonghong Song <yongho...@linux.dev>
> Date: Fri Jul 28 01:12:02 2023 +0000
>
> bpf: Support new sign-extension mov insns

Thanks for reporting. I will look into this ASAP.

[...]

[Eduard Zingerman]

On Sun, 2023-08-06 at 23:40 -0700, Yonghong Song wrote:
>
> On 8/6/23 4:23 PM, syzbot wrote:
> > Hello,
> >
> > syzbot found the following issue on:
> >
> > HEAD commit: 25ad10658dc1 riscv, bpf: Adapt bpf trampoline to optimized..
> > git tree: bpf-next
> > console+strace: https://syzkaller.appspot.com/x/log.txt?x=147cbb29a80000
> > kernel config: https://syzkaller.appspot.com/x/.config?x=8acaeb93ad7c6aaa
> > dashboard link: https://syzkaller.appspot.com/bug?extid=d61b595e9205573133b3
> > compiler: gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
> > syz repro: https://syzkaller.appspot.com/x/repro.syz?x=14d73ccea80000
> > C reproducer: https://syzkaller.appspot.com/x/repro.c?x=1276aedea80000
> >
> > Downloadable assets:
> > disk image: https://storage.googleapis.com/syzbot-assets/3d378cc13d42/disk-25ad1065.raw.xz
> > vmlinux: https://storage.googleapis.com/syzbot-assets/44580fd5d1af/vmlinux-25ad1065.xz
> > kernel image: https://storage.googleapis.com/syzbot-assets/840587618b41/bzImage-25ad1065.xz
> >
> > The issue was bisected to:
> >
> > commit 8100928c881482a73ed8bd499d602bab0fe55608
> > Author: Yonghong Song <yongho...@linux.dev>
> > Date: Fri Jul 28 01:12:02 2023 +0000
> >
> > bpf: Support new sign-extension mov insns
>
> Thanks for reporting. I will look into this ASAP.

Hi Yonghong,

I guess it's your night and my morning, so I did some initial assessment.
The BPF program being loaded is:

0 : (62) *(u32 *)(r10 -8) = 553656332
1 : (bf) r1 = (s16)r10
2 : (07) r1 += -8
3 : (b7) r2 = 3
4 : (bd) if r2 <= r1 goto pc+0
5 : (85) call bpf_trace_printk#6
6 : (b7) r0 = 0
7 : (95) exit

(Note: when using bpftool (prog dump xlated id <some-id>) the disassembly
of the instruction #1 is incorrectly printed as "1: (bf) r1 = r10")

The error occurs when instruction #5 (call to printk) is executed.
An incorrect address for the format string is passed to printk.
Disassembly of the jited program looks as follows:

$ bpftool prog dump jited id <some-id>
bpf_prog_ebeed182d92b487f:
0: nopl (%rax,%rax)
5: nop
7: pushq %rbp
8: movq %rsp, %rbp
b: subq $8, %rsp
12: movl $553656332, -8(%rbp)
19: movswq %bp, %rdi ; <---- Note movswq %bp !
1d: addq $-8, %rdi
21: movl $3, %esi
26: cmpq %rdi, %rsi
29: jbe 0x2b
2b: callq 0xffffffffe11c484c
30: xorl %eax, %eax
32: leave
33: retq

Note jit instruction #19 corresponding to BPF instruction #1, which
loads truncated and sign-extended value of %rbp's first byte as an
address of format string.

Here is how verifier log looks for (slightly modified) program:

func#0 @0
0: R1=ctx(off=0,imm=0) R10=fp0
; asm volatile (" \n\
0: (b7) r1 = 553656332 ; R1_w=553656332
1: (63) *(u32 *)(r10 -8) = r1 ; R1_w=553656332 R10=fp0 fp-8=553656332
2: (bf) r1 = (s16)r10 ; R1_w=fp0 R10=fp0
3: (07) r1 += -8 ; R1_w=fp-8
4: (b7) r2 = 3 ; R2_w=3
5: (bd) if r2 <= r1 goto pc+0 ; R1_w=fp-8 R2_w=3
6: (85) call bpf_trace_printk#6
mark_precise: frame0: last_idx 6 first_idx 0 subseq_idx -1
...
mark_precise: frame0: falling back to forcing all scalars precise
7: R0=scalar()
7: (b7) r0 = 0 ; R0_w=0
8: (95) exit

from 5 to 6: R1_w=fp-8 R2_w=3 R10=fp0 fp-8=553656332
6: (85) call bpf_trace_printk#6
mark_precise: frame0: last_idx 6 first_idx 0 subseq_idx -1
...
mark_precise: frame0: falling back to forcing all scalars precise
7: safe

Note the following line:

2: (bf) r1 = (s16)r10 ; R1_w=fp0 R10=fp0

Verifier incorrectly marked r1 as fp0, hence not noticing the problem
with address passed to printk.

Thanks,
Eduard.

[Eduard Zingerman]

On Mon, 2023-08-07 at 16:11 +0300, Eduard Zingerman wrote:
[...]

> $ bpftool prog dump jited id <some-id>
> bpf_prog_ebeed182d92b487f:
> 0: nopl (%rax,%rax)
> 5: nop
> 7: pushq %rbp
> 8: movq %rsp, %rbp
> b: subq $8, %rsp
> 12: movl $553656332, -8(%rbp)
> 19: movswq %bp, %rdi ; <---- Note movswq %bp !
> 1d: addq $-8, %rdi
> 21: movl $3, %esi
> 26: cmpq %rdi, %rsi
> 29: jbe 0x2b
> 2b: callq 0xffffffffe11c484c
> 30: xorl %eax, %eax
> 32: leave
> 33: retq
>
> Note jit instruction #19 corresponding to BPF instruction #1, which
> loads truncated and sign-extended value of %rbp's first byte as an
> address of format string.

Correction: sign-extended value of %rbp's first *two* bytes,
disassembly with opcodes:

19: movswq %bp, %rdi
48 0f bf fd

[...]

[Yonghong Song]

Thanks, Eduard. Right. I am also able to dump xlated code like
below:

0: (62) *(u32 *)(r10 -8) = 553656332
1: (bf) r1 = (s16)r10
2: (07) r1 += -8
3: (b7) r2 = 3
4: (bd) if r2 <= r1 goto pc+0

5: (85) call bpf_trace_printk#-138320

6: (b7) r0 = 0
7: (95) exit

Something like below can fix the problem,

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 132f25dab931..db72619551b2 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -13171,6 +13171,7 @@ static int check_alu_op(struct bpf_verifier_env
*env, struct bpf_insn *insn)
if (no_sext && need_id)
src_reg->id =
++env->id_gen;
copy_register_state(dst_reg,
src_reg);
+ dst_reg->type = SCALAR_VALUE;
if (!no_sext)
dst_reg->id = 0;
coerce_reg_to_size_sx(dst_reg,
insn->off >> 3);

After insn 1, we need change r1 type to SCALAR_VALUE. Will add
the the test to selftest and submit the patch to fix the problem
today.

>
> Thanks,
> Eduard.
>
>>>
>>> bisection log: https://syzkaller.appspot.com/x/bisect.txt?x=17970c5da80000
>>> final oops: https://syzkaller.appspot.com/x/report.txt?x=14570c5da80000
>>> console output: https://syzkaller.appspot.com/x/log.txt?x=10570c5da80000
>>>

[...]

[Eduard Zingerman]

Should this be an error?
Like in the same function but slightly below, when u32 moves are
processed:

/* R1 = (u32) R2 */
if (is_pointer_value(env, insn->src_reg)) {
verbose(env,
"R%d partial copy of pointer\n",
insn->src_reg);
return -EACCES;
} else { ...

[Yonghong Song]

Right, this is indeed better for unprivileged prog run.
I have submitted a patch to fix the issue reported by syzbot.
Please help review. Thanks!