Re: WARNING in get_unlocked_entry

[Matthew Wilcox]

On Tue, Oct 30, 2018 at 08:00:03AM -0700, syzbot wrote:
> syzbot found the following crash on:
> 
> HEAD commit:    4b42745211af Merge tag 'armsoc-soc' of git://git.kernel.or..
> git tree:       upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=1187d06d400000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=93932074d01b4a5
> dashboard link: https://syzkaller.appspot.com/bug?extid=4fd0c066d82852499145
> compiler:       gcc (GCC) 8.0.1 20180413 (experimental)
> 
> Unfortunately, I don't have any reproducer for this crash yet.

Hmmpf.  Would have been nice if syzbot had caught this during its tests
of the -next tree ...

> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> Reported-by: syzbot+4fd0c066d82852499145@xxxxxxxxxxxxxxxxxxxxxxxxx
> 
> EXT4-fs (sda1): Cannot specify journal on remount
> EXT4-fs (sda1): DAX enabled. Warning: EXPERIMENTAL, use at your own risk
> EXT4-fs (sda1): warning: refusing change of dax flag with busy inodes while
> remounting
> WARNING: CPU: 0 PID: 12870 at fs/dax.c:227 get_unlocked_entry+0x3ac/0x4d0
> fs/dax.c:227
> Kernel panic - not syncing: panic_on_warn set ...
> 
> CPU: 0 PID: 12870 Comm: syz-executor3 Not tainted 4.19.0+ #87
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
> Google 01/01/2011
> Call Trace:
>  __dump_stack lib/dump_stack.c:77 [inline]
>  dump_stack+0x244/0x39d lib/dump_stack.c:113
>  panic+0x238/0x4e7 kernel/panic.c:184
>  __warn.cold.8+0x20/0x4a kernel/panic.c:536
>  report_bug+0x254/0x2d0 lib/bug.c:186
>  fixup_bug arch/x86/kernel/traps.c:178 [inline]
>  do_error_trap+0x11b/0x200 arch/x86/kernel/traps.c:271
>  do_invalid_op+0x36/0x40 arch/x86/kernel/traps.c:290
>  invalid_op+0x14/0x20 arch/x86/entry/entry_64.S:966
> RIP: 0010:get_unlocked_entry+0x3ac/0x4d0 fs/dax.c:227
> Code: e8 31 ff 83 e0 01 48 89 c6 48 89 85 10 ff ff ff e8 09 18 96 ff 48 8b
> 85 10 ff ff ff 48 85 c0 0f 85 34 fe ff ff e8 c4 16 96 ff <0f> 0b e8 bd 16 96
> ff 48 b8 00 00 00 00 00 fc ff df 48 03 85 f8 fe
> RSP: 0018:ffff88018025e7b8 EFLAGS: 00010012
> RAX: 0000000000040000 RBX: ffff88018025ead8 RCX: ffffc90008983000
> RDX: 0000000000000341 RSI: ffffffff81e94d3c RDI: 0000000000000007
> RBP: ffff88018025e8c8 R08: ffff88017f862280 R09: ffffed003004bd08
> R10: ffffed003004bd08 R11: 0000000000000003 R12: dffffc0000000000
> R13: ffffea0005e01040 R14: ffff88018025e800 R15: ffff88018025e8a0
>  grab_mapping_entry fs/dax.c:447 [inline]

OK, so I'd like to know what value actually got loaded into 'entry'.
Maybe my x86-fu is weak, but I don't understand how this register dump
correlates with the instructions I'm seeing.

Here's the C code:

                entry = xas_load(xas);
                if (!entry || xa_is_internal(entry) ||
                                WARN_ON_ONCE(!xa_is_value(entry)) ||
                                !dax_is_locked(entry))
                        return entry;


I downloaded the .config that syzbot was using and built fs/dax.o.
Disassembling it, I get:

                                WARN_ON_ONCE(!xa_is_value(entry)) ||
    5e53:       31 ff                   xor    %edi,%edi
    5e55:       83 e0 01                and    $0x1,%eax
    5e58:       48 89 c6                mov    %rax,%rsi
    5e5b:       48 89 85 10 ff ff ff    mov    %rax,-0xf0(%rbp)
    5e62:       e8 00 00 00 00          callq  5e67 <get_unlocked_entry+0x397>
                        5e63: R_X86_64_PLT32    __sanitizer_cov_trace_const_cmp8
-0x4
    5e67:       48 8b 85 10 ff ff ff    mov    -0xf0(%rbp),%rax
    5e6e:       48 85 c0                test   %rax,%rax
    5e71:       0f 85 34 fe ff ff       jne    5cab <get_unlocked_entry+0x1db>
    5e77:       e8 00 00 00 00          callq  5e7c <get_unlocked_entry+0x3ac>
                        5e78: R_X86_64_PLT32    __sanitizer_cov_trace_pc-0x4
    5e7c:       0f 0b                   ud2    

That ud2 is the WARN_ON trigger.  It's at offset 0x3ac from the start of
get_unlocked_entry() so it matches the dump.

The key is 'test %rax,%rax' at 5e6e.  If that passes, we jump to 5cab
where we do xa_to_value() in order to implement dax_is_locked().  So is
'entry' actually 0x0000000000040000 like the dump says?

That's where I get confused.  The previous insn (5e67) loads rax from what
I presume is the stack, having previously stored it there at insn 535b.
But insn 5e51 should have reduced %rax to just 0 or 1 in the low bit of
the register.  So clearly I have my understanding of x86 insns confused.
If someone could help me out here, I'd be grateful.