Re: [syzbot] [mm?] kernel BUG in __khugepaged_enter

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

+cc Thomas in case the commit it's sat at is indicative, there does seem to be
some weirdness with MMF_MULTIPROCESS processes (i.e. CLONE_VM but !CLONE_THREAD)
resulting in possible memory corruption?

We kinda need a repro to be sure though I think...

On Mon, Feb 16, 2026 at 02:43:17PM +0000, Lorenzo Stoakes wrote:
> On Mon, Feb 16, 2026 at 03:40:21PM +0100, David Hildenbrand (Arm) wrote:
> > On 2/14/26 17:40, syzbot wrote:
> > > Hello,
> > >
> > > syzbot found the following issue on:
> > >
> > > HEAD commit:    1e83ccd5921a sched/mmcid: Don't assume CID is CPU owned on..
> > > git tree:       upstream
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=1169dae6580000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=54ae71b284dd0e13
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=6b554d491efbe066b701
> > > compiler:       gcc (Debian 14.2.0-19) 14.2.0, GNU ld (GNU Binutils for Debian) 2.44
> > >
> > > Unfortunately, I don't have any reproducer for this issue yet.

We're going to need one I fear :)

> > >
> > > Downloadable assets:
> > > disk image: https://storage.googleapis.com/syzbot-assets/ed43f42e3ea1/disk-1e83ccd5.raw.xz
> > > vmlinux: https://storage.googleapis.com/syzbot-assets/d8af54a32588/vmlinux-1e83ccd5.xz
> > > kernel image: https://storage.googleapis.com/syzbot-assets/34e6a8cc1037/bzImage-1e83ccd5.xz
> > >
> > > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > > Reported-by: syzbot+6b554d491efbe066b701@syzkaller.appspotmail.com
> > >
> > > ------------[ cut here ]------------
> > > kernel BUG at mm/khugepaged.c:438!
> > > Oops: invalid opcode: 0000 [#1] SMP KASAN PTI
> > > CPU: 0 UID: 0 PID: 16472 Comm: syz.3.2372 Tainted: G     U  W    L XTNJ syzkaller #0 PREEMPT(full)
> > > Tainted: [U]=USER, [W]=WARN, [L]=SOFTLOCKUP, [X]=AUX, [T]=RANDSTRUCT, [N]=TEST, [J]=FWCTL
> > > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/24/2026
> > > RIP: 0010:__khugepaged_enter+0x30a/0x380 mm/khugepaged.c:438
> > > Code: 64 7e 8e e8 a8 dc 66 ff e8 93 e6 8d ff 5b 5d 41 5c 41 5d 41 5e 41 5f e9 04 6c 04 09 e8 7f e6 8d ff 48 89 df e8 17 33 d9 ff 90 <0f> 0b 48 89 ef e8 dc 51 f8 ff e9 3b fd ff ff e8 f2 52 f8 ff e9 e1
> > > RSP: 0018:ffffc9000e98fba8 EFLAGS: 00010292
> > > RAX: 000000000000031f RBX: ffff888079b24980 RCX: 0000000000000000
> > > RDX: 000000000000031f RSI: ffffffff81e5b2c9 RDI: fffff52001d31f1c
> > > RBP: 0000000000000000 R08: 0000000000000005 R09: 0000000000000000
> > > R10: 0000000080000000 R11: 0000000000000001 R12: 0000000008100177
> > > R13: ffff88804adf9510 R14: 0000000000000000 R15: 0000000000000000
> > > FS:  00007f06093436c0(0000) GS:ffff8881245b1000(0000) knlGS:0000000000000000
> > > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > CR2: 00007fff341d3f52 CR3: 00000000319b0000 CR4: 00000000003526f0
> > > Call Trace:
> > >   <TASK>
> > >   khugepaged_enter_vma mm/khugepaged.c:467 [inline]
> > >   khugepaged_enter_vma+0x137/0x2c0 mm/khugepaged.c:461
> > >   do_huge_pmd_anonymous_page+0x1c8/0x1c00 mm/huge_memory.c:1469
> > >   create_huge_pmd mm/memory.c:6102 [inline]
> > >   __handle_mm_fault+0x1e96/0x2b50 mm/memory.c:6376
> > >   handle_mm_fault+0x36d/0xa20 mm/memory.c:6583
> > >   do_user_addr_fault+0x5a3/0x12f0 arch/x86/mm/fault.c:1334

	vma = lock_vma_under_rcu(mm, address);
	if (!vma)
		goto lock_mmap; <--- didn't jump there, so is a VMA lock.

	if (unlikely(access_error(error_code, vma))) {
		bad_area_access_error(regs, error_code, address, NULL, vma);
		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
		return;
	}
	fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs); <-- here

> > >   handle_page_fault arch/x86/mm/fault.c:1474 [inline]
> > >   exc_page_fault+0x6f/0xd0 arch/x86/mm/fault.c:1527
> > >   asm_exc_page_fault+0x26/0x30 arch/x86/include/asm/idtentry.h:618
> >
> > This is the VM_BUG_ON_MM(hpage_collapse_test_exit(mm), mm), which checks
> >
> > 	atomic_read(&mm->mm_users) == 0;

Yeah, and that just shouldn't be possible, so maybe memory corruption?

The crash log indicates the system is tainted by softlock
https://syzkaller.appspot.com/text?tag=CrashLog&x=1169dae6580000 so something's
gone horribly wrong there...

(from crash log)
[  696.104336][T16472] pgd ffff8880319b0000 mm_users 0 mm_count 2 pgtables_bytes 155648 map_count 32

VMA's still there so exit_mmap() hasn't run yet...

But hmm we injected a fault :)

[  696.293779][T16475] FAULT_INJECTION: forcing a failure.
[  696.293779][T16475] name fail_page_alloc, interval 1, probability 0, space 0, times 0

[  696.332139][T16475]  dump_stack_lvl+0x100/0x190
[  696.332164][T16475]  should_fail_ex.cold+0x5/0xa
[  696.332178][T16475]  ? prepare_alloc_pages+0x16d/0x5f0
[  696.332200][T16475]  should_fail_alloc_page+0xeb/0x140
[  696.332219][T16475]  prepare_alloc_pages+0x1f0/0x5f0
[  696.332241][T16475]  __alloc_frozen_pages_noprof+0x193/0x2410
[  696.332258][T16475]  ? stack_trace_save+0x8e/0xc0
[  696.332277][T16475]  ? __pfx_stack_trace_save+0x10/0x10
[  696.332297][T16475]  ? stack_depot_save_flags+0x27/0x9d0
[  696.332315][T16475]  ? __lock_acquire+0x4a5/0x2630
[  696.332331][T16475]  ? kasan_save_stack+0x3f/0x50
[  696.332346][T16475]  ? __pfx___alloc_frozen_pages_noprof+0x10/0x10
[  696.332360][T16475]  ? copy_time_ns+0xf6/0x800
[  696.332379][T16475]  ? unshare_nsproxy_namespaces+0xc3/0x1f0
[  696.332408][T16475]  ? __x64_sys_unshare+0x31/0x40
[  696.332423][T16475]  ? do_syscall_64+0x106/0xf80
[  696.332437][T16475]  ? entry_SYSCALL_64_after_hwframe+0x77/0x7f
[  696.332460][T16475]  ? __sanitizer_cov_trace_switch+0x54/0x90
[  696.332480][T16475]  ? policy_nodemask+0xed/0x4f0
[  696.332500][T16475]  alloc_pages_mpol+0x1fb/0x550
[  696.332519][T16475]  ? __pfx_alloc_pages_mpol+0x10/0x10
[  696.332542][T16475]  alloc_pages_noprof+0x131/0x390
[  696.332560][T16475]  copy_time_ns+0x11a/0x800

So:

static struct nsproxy *create_new_namespaces(u64 flags,
	struct task_struct *tsk, struct user_namespace *user_ns,
	struct fs_struct *new_fs)
{
	...
	new_nsp->time_ns_for_children = copy_time_ns(flags, user_ns,
				tsk->nsproxy->time_ns_for_children); <- -ENOMEM
	if (IS_ERR(new_nsp->time_ns_for_children)) {
		err = PTR_ERR(new_nsp->time_ns_for_children);
		goto out_time;
	}

	...

out_time:
	put_net(new_nsp->net_ns);
out_net:
	put_cgroup_ns(new_nsp->cgroup_ns);
out_cgroup:
	put_pid_ns(new_nsp->pid_ns_for_children);
out_pid:
	put_ipc_ns(new_nsp->ipc_ns);
out_ipc:
	put_uts_ns(new_nsp->uts_ns);
out_uts:
	put_mnt_ns(new_nsp->mnt_ns);
out_ns:
	kmem_cache_free(nsproxy_cachep, new_nsp);
	return ERR_PTR(err);
}

So we're putting the world... maybe some of this is buggy?

[  696.332578][T16475]  ? copy_cgroup_ns+0x71/0x970
[  696.332601][T16475]  create_new_namespaces+0x48a/0xac0

So:

int unshare_nsproxy_namespaces(unsigned long unshare_flags,
	struct nsproxy **new_nsp, struct cred *new_cred, struct fs_struct *new_fs)
{
	...
	*new_nsp = create_new_namespaces(unshare_flags, current, user_
					 new_fs ? new_fs : current->fs);
	if (IS_ERR(*new_nsp)) {
		err = PTR_ERR(*new_nsp);
		goto out;
	}
	...
out:
	return err;
}


[  696.332626][T16475]  unshare_nsproxy_namespaces+0xc3/0x1f0
[  696.332648][T16475]  ksys_unshare+0x455/0xab0

So:

int ksys_unshare(unsigned long unshare_flags)
{
	...
	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
					 new_cred, new_fs);
	if (err)
		goto bad_unshare_cleanup_cred;
	...
bad_unshare_cleanup_cred:
	if (new_cred)
		put_cred(new_cred);
bad_unshare_cleanup_fd:
	if (new_fd)
		put_files_struct(new_fd);

bad_unshare_cleanup_fs:
	if (new_fs)
		free_fs_struct(new_fs);

bad_unshare_out:
	return err;
}

And again we're putting all the things... maybe something buggy here?

Perhaps this unshare is racing with something else?

OTOH, we _already_ had mm_users = 0 at this point (as per mm dump) so. Probably
something before got us into this state?

[  696.332664][T16475]  ? __pfx_ksys_unshare+0x10/0x10
[  696.332679][T16475]  ? xfd_validate_state+0x129/0x190
[  696.332702][T16475]  __x64_sys_unshare+0x31/0x40
[  696.332717][T16475]  do_syscall_64+0x106/0xf80
[  696.332730][T16475]  ? clear_bhb_loop+0x40/0x90
[  696.332747][T16475]  entry_SYSCALL_64_after_hwframe+0x77/0x7f

Also from mm dump:

flags: 00000000,840007fd

MMF_TOPDOWN | MMF_MULTIPROCESS | (core dump flags)

No MMF_VM_HUGEPAGE...

MMF_MULTIPROCESS marks this as shared between processes, as set in
copy_process() -> copy_oom_score_adj() which has a guard:

	/* Skip if spawning a thread or using vfork */
	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
		return;

Which grabs the mm in __set_oom_adj() which as per commit 44a70adec910 ("mm,
oom_adj: make sure processes sharing mm have same view of oom_score_adj")
suggests processes were cloned with CLONE_VM but not CLONE_SIGHAND (which
presumably implies !CLONE_THREAD).

Anyway it's hard to know with a repro.

Cheers, Lorenzo