Re: WARNING in apparmor_cred_free

[Casey Schaufler]

On 1/11/2019 2:43 PM, Casey Schaufler wrote:
> On 1/11/2019 2:30 PM, John Johansen wrote:
>> On 1/11/19 2:11 PM, Casey Schaufler wrote:
>>> On 1/11/2019 1:43 AM, syzbot wrote:
>>>> Hello,
>>>>
>>>> syzbot found the following crash on:
>>>>
>>>> HEAD commit:ÂÂÂ b808822a75a3 Add linux-next specific files for 20190111
>>>> git tree:ÂÂÂÂÂÂ linux-next
>>>> console output: https://syzkaller.appspot.com/x/log.txt?x=179c22f7400000
>>>> kernel config:Â https://syzkaller.appspot.com/x/.config?x=c052ead0aed5001b
>>>> dashboard link: https://syzkaller.appspot.com/bug?extid=69ca07954461f189e808
>>>> compiler:ÂÂÂÂÂÂ gcc (GCC) 9.0.0 20181231 (experimental)
>>>> syz repro:ÂÂÂÂÂ https://syzkaller.appspot.com/x/repro.syz?x=162d947f400000
>>>> C reproducer:ÂÂ https://syzkaller.appspot.com/x/repro.c?x=139f6c37400000
>>>>
>>>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>>>> Reported-by: syzbot+69ca07954461f189e808@xxxxxxxxxxxxxxxxxxxxxxxxx
>>>>
>>>> ------------[ cut here ]------------
>>>> AppArmor WARN cred_label: ((!blob)):
>>>> WARNING: CPU: 0 PID: 0 at security/apparmor/include/cred.h:30 cred_label security/apparmor/include/cred.h:30 [inline]
>>>> WARNING: CPU: 0 PID: 0 at security/apparmor/include/cred.h:30 apparmor_cred_free+0x12f/0x1a0 security/apparmor/lsm.c:62
>>>> Kernel panic - not syncing: panic_on_warn set ...
>>>> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.0.0-rc1-next-20190111 #10
>>>> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
>>>> Call Trace:
>>>> Â<IRQ>
>>>> Â__dump_stack lib/dump_stack.c:77 [inline]
>>>> Âdump_stack+0x1db/0x2d0 lib/dump_stack.c:113
>>>> Âpanic+0x2cb/0x65c kernel/panic.c:214
>>>> Â__warn.cold+0x20/0x48 kernel/panic.c:571
>>>> Âreport_bug+0x263/0x2b0 lib/bug.c:186
>>>> Âfixup_bug arch/x86/kernel/traps.c:178 [inline]
>>>> Âfixup_bug arch/x86/kernel/traps.c:173 [inline]
>>>> Âdo_error_trap+0x11b/0x200 arch/x86/kernel/traps.c:271
>>>> Âdo_invalid_op+0x37/0x50 arch/x86/kernel/traps.c:290
>>>> Âinvalid_op+0x14/0x20 arch/x86/entry/entry_64.S:973
>>>> RIP: 0010:cred_label security/apparmor/include/cred.h:30 [inline]
>>>> RIP: 0010:apparmor_cred_free+0x12f/0x1a0 security/apparmor/lsm.c:62
>>>> Code: 7c 88 48 c7 c7 00 d0 7c 88 e8 fd 70 f2 fd 0f 0b eb a9 e8 54 3f 29 fe 48 c7 c6 c0 df 7c 88 48 c7 c7 00 d0 7c 88 e8 e1 70 f2 fd <0f> 0b 48 b8 00 00 00 00 00 fc ff df 80 38 00 75 4a 4c 8b 2c 25 00
>>>> RSP: 0018:ffff8880ae6079f8 EFLAGS: 00010286
>>>> RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
>>>> RDX: 0000000000000100 RSI: ffffffff81687fa6 RDI: 0000000000000006
>>>> RBP: ffff8880ae607a18 R08: ffffffff8987dec0 R09: 0000000000000000
>>>> R10: 0000000000000000 R11: 0000000000000000 R12: ffff8880a86b3100
>>>> R13: ffff8880a86b3100 R14: ffff8880a86b3188 R15: dffffc0000000000
>>>> Âsecurity_cred_free+0x4b/0xf0 security/security.c:1490
>>> The obvious thing to do is put a check in security_cred_free
>>> for a NULL cred->security, in which case the LSM hooks
>>> wouldn't get called.
>> Right, but the question is should we? To my thinking we shouldn't
>> ever have a cred without cred->security, unless the cred was
>> allocated but a later step in its construction, say allocating
>> ->security failed.
> If allocating ->security fails in security_cred_alloc_blank()
> or security_prepare_creds() you don't have to do anything but
> fail because the LSM hooks are not called before the allocation.
>
>> In which case I'd rather see the cred directly freed and not
>> call into security_cred_free() as I like being able to detect
>> corrupt creds.
> I think we need to look for some bit of code that's setting
> cred->security to NULL inappropriately.

If security_cred_alloc_blank() fails for lack of memory
in cred_alloc_blank() abort_creds() will be called. This
in turn calls put_cred() and put_cred_rcu(), which will
call security_cred_free() with ->security set to NULL.

put_cred_rcu() is the only caller of security_cred_free().
The ->security == NULL check can be in either put_cred_rcu()
or in security_cred_free(). I suggest the latter as the
cleanest option.