Re: [syzbot] [bpf?] KCSAN: data-race in bpf_obj_memcpy / bpf_obj_memcpy

[Mykyta Yatsenko]

On 4/20/26 3:13 PM, syzbot wrote:
> Hello,
>
> syzbot found the following issue on:
>
> HEAD commit:    c1f49dea2b8f Merge tag 'mm-hotfixes-stable-2026-04-19-00-1..
> git tree:       upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=10ec34ce580000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=d3740f7f69b18f59
> dashboard link: https://syzkaller.appspot.com/bug?extid=44044637ef892e79ca2b
> compiler:       Debian clang version 21.1.8 (++20251221033036+2078da43e25a-1~exp1~20251221153213.50), Debian LLD 21.1.8
>
> Unfortunately, I don't have any reproducer for this issue yet.
>
> Downloadable assets:
> disk image: https://storage.googleapis.com/syzbot-assets/4ed91de40e47/disk-c1f49dea.raw.xz
> vmlinux: https://storage.googleapis.com/syzbot-assets/7353bf53627b/vmlinux-c1f49dea.xz
> kernel image: https://storage.googleapis.com/syzbot-assets/ab6db1fcd59d/bzImage-c1f49dea.xz
>
> IMPORTANT: if you fix the issue, please add the following tag to the commit:
> Reported-by: syzbot+44044637ef892e79ca2b@xxxxxxxxxxxxxxxxxxxxxxxxx
>
> netlink: 676 bytes leftover after parsing attributes in process `syz.4.735'.
> ==================================================================
> BUG: KCSAN: data-race in bpf_obj_memcpy / bpf_obj_memcpy
>
> write to 0xffffe8ffffa24c00 of 1404 bytes by task 6603 on cpu 0:
>   bpf_obj_memcpy+0x13c/0x1a0 include/linux/bpf.h:-1
>   copy_map_value include/linux/bpf.h:557 [inline]
>   bpf_percpu_array_update+0x1e1/0x2d0 kernel/bpf/arraymap.c:443
>   bpf_map_update_value+0x260/0x570 kernel/bpf/syscall.c:275
>   generic_map_update_batch+0x52d/0x680 kernel/bpf/syscall.c:2025
>   bpf_map_do_batch+0x25c/0x380 kernel/bpf/syscall.c:5689
>   __sys_bpf+0x6a2/0x7e0 kernel/bpf/syscall.c:-1
>   __do_sys_bpf kernel/bpf/syscall.c:6361 [inline]
>   __se_sys_bpf kernel/bpf/syscall.c:6359 [inline]
>   __x64_sys_bpf+0x41/0x50 kernel/bpf/syscall.c:6359
>   x64_sys_call+0x10cb/0x3020 arch/x86/include/generated/asm/syscalls_64.h:322
>   do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
>   do_syscall_64+0x12c/0x3b0 arch/x86/entry/syscall_64.c:94
>   entry_SYSCALL_64_after_hwframe+0x77/0x7f
>
> write to 0xffffe8ffffa24c00 of 1404 bytes by task 6604 on cpu 1:
>   bpf_obj_memcpy+0x13c/0x1a0 include/linux/bpf.h:-1
>   copy_map_value include/linux/bpf.h:557 [inline]
>   bpf_percpu_array_update+0x1e1/0x2d0 kernel/bpf/arraymap.c:443
>   bpf_map_update_value+0x260/0x570 kernel/bpf/syscall.c:275
>   generic_map_update_batch+0x52d/0x680 kernel/bpf/syscall.c:2025
>   bpf_map_do_batch+0x25c/0x380 kernel/bpf/syscall.c:5689
>   __sys_bpf+0x6a2/0x7e0 kernel/bpf/syscall.c:-1
>   __do_sys_bpf kernel/bpf/syscall.c:6361 [inline]
>   __se_sys_bpf kernel/bpf/syscall.c:6359 [inline]
>   __x64_sys_bpf+0x41/0x50 kernel/bpf/syscall.c:6359
>   x64_sys_call+0x10cb/0x3020 arch/x86/include/generated/asm/syscalls_64.h:322
>   do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
>   do_syscall_64+0x12c/0x3b0 arch/x86/entry/syscall_64.c:94
>   entry_SYSCALL_64_after_hwframe+0x77/0x7f

This looks like a design choice - no explicit synchronization for percpu 
data updates, for performance reasons.
From the syscall side it's possible to use external lock. From BPF in 
NMI context torn writes risk is acceptable.