[syzbot] possible deadlock in hugetlb_fault

[syzbot]

Hello,

syzbot found the following issue on:

HEAD commit: f2f32f8af2b0 Merge tag 'for-6.1-rc3-tag' of git://git.kern..
git tree: upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=137d52ca880000
kernel config: https://syzkaller.appspot.com/x/.config?x=d080a4bd239918dd
dashboard link: https://syzkaller.appspot.com/bug?extid=ca56f14c500045350f93
compiler: gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.2
userspace arch: i386

Unfortunately, I don't have any reproducer for this issue yet.

Downloadable assets:
disk image: https://storage.googleapis.com/syzbot-assets/b4f72e7a4c11/disk-f2f32f8a.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/3f88997ad7c9/vmlinux-f2f32f8a.xz
kernel image: https://storage.googleapis.com/syzbot-assets/b4b5b3963e2d/bzImage-f2f32f8a.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:
Reported-by: syzbot+ca56f1...@syzkaller.appspotmail.com

======================================================
WARNING: possible circular locking dependency detected
6.1.0-rc3-syzkaller-00152-gf2f32f8af2b0 #0 Not tainted
------------------------------------------------------
syz-executor.2/5665 is trying to acquire lock:
ffff88801c74c298 (&mm->mmap_lock#2){++++}-{3:3}, at: __might_fault+0xa1/0x170 mm/memory.c:5645

but task is already holding lock:
ffff88801c4f3078 (&vma_lock->rw_sema){++++}-{3:3}, at: hugetlb_vma_lock_read mm/hugetlb.c:6816 [inline]
ffff88801c4f3078 (&vma_lock->rw_sema){++++}-{3:3}, at: hugetlb_fault+0x40a/0x2060 mm/hugetlb.c:5859

which lock already depends on the new lock.


the existing dependency chain (in reverse order) is:

-> #1 (&vma_lock->rw_sema){++++}-{3:3}:
down_write+0x90/0x220 kernel/locking/rwsem.c:1562
hugetlb_vma_lock_write mm/hugetlb.c:6834 [inline]
__unmap_hugepage_range_final+0x97/0x340 mm/hugetlb.c:5202
unmap_single_vma+0x23d/0x2a0 mm/memory.c:1690
unmap_vmas+0x21e/0x370 mm/memory.c:1733
exit_mmap+0x189/0x7a0 mm/mmap.c:3090
__mmput+0x128/0x4c0 kernel/fork.c:1185
mmput+0x5c/0x70 kernel/fork.c:1207
exit_mm kernel/exit.c:516 [inline]
do_exit+0xa39/0x2a20 kernel/exit.c:807
do_group_exit+0xd0/0x2a0 kernel/exit.c:950
get_signal+0x21a1/0x2430 kernel/signal.c:2858
arch_do_signal_or_restart+0x82/0x2300 arch/x86/kernel/signal.c:869
exit_to_user_mode_loop kernel/entry/common.c:168 [inline]
exit_to_user_mode_prepare+0x15f/0x250 kernel/entry/common.c:203
__syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
syscall_exit_to_user_mode+0x19/0x50 kernel/entry/common.c:296
__do_fast_syscall_32+0x72/0xf0 arch/x86/entry/common.c:181
do_fast_syscall_32+0x2f/0x70 arch/x86/entry/common.c:203
entry_SYSENTER_compat_after_hwframe+0x70/0x82

-> #0 (&mm->mmap_lock#2){++++}-{3:3}:
check_prev_add kernel/locking/lockdep.c:3097 [inline]
check_prevs_add kernel/locking/lockdep.c:3216 [inline]
validate_chain kernel/locking/lockdep.c:3831 [inline]
__lock_acquire+0x2a43/0x56d0 kernel/locking/lockdep.c:5055
lock_acquire kernel/locking/lockdep.c:5668 [inline]
lock_acquire+0x1df/0x630 kernel/locking/lockdep.c:5633
__might_fault mm/memory.c:5646 [inline]
__might_fault+0x104/0x170 mm/memory.c:5639
_copy_from_user+0x25/0x170 lib/usercopy.c:13
copy_from_user include/linux/uaccess.h:161 [inline]
snd_rawmidi_kernel_write1+0x366/0x880 sound/core/rawmidi.c:1549
snd_rawmidi_write+0x273/0xbb0 sound/core/rawmidi.c:1618
vfs_write+0x2d7/0xdd0 fs/read_write.c:582
ksys_write+0x1e8/0x250 fs/read_write.c:637
do_syscall_32_irqs_on arch/x86/entry/common.c:112 [inline]
__do_fast_syscall_32+0x65/0xf0 arch/x86/entry/common.c:178
do_fast_syscall_32+0x2f/0x70 arch/x86/entry/common.c:203
entry_SYSENTER_compat_after_hwframe+0x70/0x82

other info that might help us debug this:

Possible unsafe locking scenario:

CPU0 CPU1
---- ----
lock(&vma_lock->rw_sema);
lock(&mm->mmap_lock#2);
lock(&vma_lock->rw_sema);
lock(&mm->mmap_lock#2);

*** DEADLOCK ***

1 lock held by syz-executor.2/5665:
#0: ffff88801c4f3078 (&vma_lock->rw_sema){++++}-{3:3}, at: hugetlb_vma_lock_read mm/hugetlb.c:6816 [inline]
#0: ffff88801c4f3078 (&vma_lock->rw_sema){++++}-{3:3}, at: hugetlb_fault+0x40a/0x2060 mm/hugetlb.c:5859

stack backtrace:
CPU: 1 PID: 5665 Comm: syz-executor.2 Not tainted 6.1.0-rc3-syzkaller-00152-gf2f32f8af2b0 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/11/2022
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
check_noncircular+0x25f/0x2e0 kernel/locking/lockdep.c:2177
check_prev_add kernel/locking/lockdep.c:3097 [inline]
check_prevs_add kernel/locking/lockdep.c:3216 [inline]
validate_chain kernel/locking/lockdep.c:3831 [inline]
__lock_acquire+0x2a43/0x56d0 kernel/locking/lockdep.c:5055
lock_acquire kernel/locking/lockdep.c:5668 [inline]
lock_acquire+0x1df/0x630 kernel/locking/lockdep.c:5633
__might_fault mm/memory.c:5646 [inline]
__might_fault+0x104/0x170 mm/memory.c:5639
_copy_from_user+0x25/0x170 lib/usercopy.c:13
copy_from_user include/linux/uaccess.h:161 [inline]
snd_rawmidi_kernel_write1+0x366/0x880 sound/core/rawmidi.c:1549
snd_rawmidi_write+0x273/0xbb0 sound/core/rawmidi.c:1618
vfs_write+0x2d7/0xdd0 fs/read_write.c:582
ksys_write+0x1e8/0x250 fs/read_write.c:637
do_syscall_32_irqs_on arch/x86/entry/common.c:112 [inline]
__do_fast_syscall_32+0x65/0xf0 arch/x86/entry/common.c:178
do_fast_syscall_32+0x2f/0x70 arch/x86/entry/common.c:203
entry_SYSENTER_compat_after_hwframe+0x70/0x82
RIP: 0023:0xf7fad549
Code: 03 74 c0 01 10 05 03 74 b8 01 10 06 03 74 b4 01 10 07 03 74 b0 01 10 08 03 74 d8 01 00 00 00 00 00 51 52 55 89 e5 0f 34 cd 80 <5d> 5a 59 c3 90 90 90 90 8d b4 26 00 00 00 00 8d b4 26 00 00 00 00
RSP: 002b:00000000f7f875cc EFLAGS: 00000296 ORIG_RAX: 0000000000000004
RAX: ffffffffffffffda RBX: 0000000000000004 RCX: 0000000020000000
RDX: 00000000c86ade39 RSI: 0000000000000000 RDI: 0000000000000000
RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000000
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
</TASK>
----------------
Code disassembly (best guess):
0: 03 74 c0 01 add 0x1(%rax,%rax,8),%esi
4: 10 05 03 74 b8 01 adc %al,0x1b87403(%rip) # 0x1b8740d
a: 10 06 adc %al,(%rsi)
c: 03 74 b4 01 add 0x1(%rsp,%rsi,4),%esi
10: 10 07 adc %al,(%rdi)
12: 03 74 b0 01 add 0x1(%rax,%rsi,4),%esi
16: 10 08 adc %cl,(%rax)
18: 03 74 d8 01 add 0x1(%rax,%rbx,8),%esi
1c: 00 00 add %al,(%rax)
1e: 00 00 add %al,(%rax)
20: 00 51 52 add %dl,0x52(%rcx)
23: 55 push %rbp
24: 89 e5 mov %esp,%ebp
26: 0f 34 sysenter
28: cd 80 int $0x80
* 2a: 5d pop %rbp <-- trapping instruction
2b: 5a pop %rdx
2c: 59 pop %rcx
2d: c3 retq
2e: 90 nop
2f: 90 nop
30: 90 nop
31: 90 nop
32: 8d b4 26 00 00 00 00 lea 0x0(%rsi,%riz,1),%esi
39: 8d b4 26 00 00 00 00 lea 0x0(%rsi,%riz,1),%esi


---
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.

[Mike Kravetz]

I may not be reading the report correctly, but I can not see how we acquire the
hugetlb vma_lock before trying to acquire mmap_lock in stack 0. We would not
acquire the vma_lock until we enter hugetlb fault processing (not in the stack).

Adding Miaohe Lin on Cc due to previous help with vma_lock potential deadlock
situations. Miaohe, does this make sense to you?
--
Mike Kravetz

[Miaohe Lin]

Hi Mike,
This doesn't make sense for me too. Stack #1 shows that syz-executor is releasing
its address space while stack #0 shows another thread is serving the write syscall.
In this case, mm->mm_users is 0 and all threads in this process should be serving
do_exit()? But I could be easily wrong. Also I can't see how vma_lock is locked before
trying to acquire mmap_lock in above stacks. Might this be a false positive?

Thanks,
Miaohe Lin

[Dmitry Vyukov]

The unlock of vma_lock is conditional:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/hugetlb.c?id=f2f32f8af2b0ca9d619e5183eae3eed431793baf#n6840

and the condition is:

static bool __vma_shareable_flags_pmd(struct vm_area_struct *vma)
{
return vma->vm_flags & (VM_MAYSHARE | VM_SHARED) &&
vma->vm_private_data;
}

Is it possible that the condition has changed between vma lock and unlock?
What mutexes protect vma->vm_flags/vm_private_data?

That would make the report perfectly sensible.

FWIW the test case that was running is this, that's the syscalls that
were running concurrently:

07:56:56 executing program 2:
r0 = syz_open_dev$sndmidi(&(0x7f0000000040), 0x2, 0x141101)
r1 = dup(r0)
setsockopt$inet_sctp_SCTP_I_WANT_MAPPED_V4_ADDR(r1, 0x84, 0xc,
&(0x7f0000000080), 0x4) (async)
write$6lowpan_enable(r1, &(0x7f0000000000)='0', 0xc86ade39) (async)
mmap(&(0x7f0000000000/0xb36000)=nil, 0xb36000, 0x3, 0x68831,
0xffffffffffffffff, 0x0) (async)
madvise(&(0x7f0000000000/0x600000)=nil, 0x600003, 0x4) (async, rerun: 32)
mremap(&(0x7f00007a0000/0x3000)=nil, 0x3000, 0x2000, 0x7,
&(0x7f0000835000/0x2000)=nil) (rerun: 32)

[Dmitry Vyukov]

This new bug report seems to confirm the hypothesis:

WARNING: locking bug in hugetlb_no_page
https://syzkaller.appspot.com/bug?extid=d07c65298d2c15eafcb0

[Miaohe Lin]

Thanks Dmitry!

I believe Mike found the root cause and his proposed patches in next-20221111 will fix the problem:
https://lore.kernel.org/linux-mm/20221111232628.290...@oracle.com/

Let's keep looking. :) Thanks for your work!

Thanks,
Miaohe Lin

>
>
>>>> Adding Miaohe Lin on Cc due to previous help with vma_lock potential deadlock
>>>> situations. Miaohe, does this make sense to you?
>>>>
>>>
>>> Hi Mike,
>>> This doesn't make sense for me too. Stack #1 shows that syz-executor is releasing
>>> its address space while stack #0 shows another thread is serving the write syscall.
>>> In this case, mm->mm_users is 0 and all threads in this process should be serving
>>> do_exit()? But I could be easily wrong. Also I can't see how vma_lock is locked before
>>> trying to acquire mmap_lock in above stacks. Might this be a false positive?

> .
>

[Dmitry Vyukov]

On Mon, 14 Nov 2022 at 03:35, 'Miaohe Lin' via syzkaller-bugs

Let's tell syzbot about the fix so that it reports similar issues in future:

#syz fix:
hugetlb: don't delete vma_lock in hugetlb MADV_DONTNEED processing

> Thanks,
> Miaohe Lin
>
>
> >
> >
> >>>> Adding Miaohe Lin on Cc due to previous help with vma_lock potential deadlock
> >>>> situations. Miaohe, does this make sense to you?
> >>>>
> >>>
> >>> Hi Mike,
> >>> This doesn't make sense for me too. Stack #1 shows that syz-executor is releasing
> >>> its address space while stack #0 shows another thread is serving the write syscall.
> >>> In this case, mm->mm_users is 0 and all threads in this process should be serving
> >>> do_exit()? But I could be easily wrong. Also I can't see how vma_lock is locked before
> >>> trying to acquire mmap_lock in above stacks. Might this be a false positive?
> > .
> >
>

> --
> You received this message because you are subscribed to the Google Groups "syzkaller-bugs" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to syzkaller-bug...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/syzkaller-bugs/4c5fa6b4-6278-ce2b-73c2-883eb734bdc7%40huawei.com.