Re: KMSAN: uninit-value in usb_autopm_put_interface

[Alan Stern]

On Tue, 17 Sep 2019, Dmitry Vyukov wrote:

> On Mon, Sep 16, 2019 at 10:31 PM Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Mon, 16 Sep 2019, syzbot wrote:
> >
> > > Hello,
> > >
> > > syzbot found the following crash on:
> > >
> > > HEAD commit:    014077b5 DO-NOT-SUBMIT: usb-fuzzer: main usb gadget fuzzer..
> > > git tree:       https://github.com/google/kmsan.git master
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=16a7dde1600000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=f03c659d0830ab8d
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=e1d1a6e595adbd2458f1
> > > compiler:       clang version 9.0.0 (/home/glider/llvm/clang
> > > 80fee25776c2fb61e74c1ecb1a523375c2500b69)
> > > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=176303e1600000
> > > C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=10e8f23e600000

> > This is probably the same problem that was fixed in the Logitech driver
> > earlier.  The fix still appears to be in linux-next (commit
> > 5f9242775bb6).
> >
> > Shouldn't syzbot wait until after the merge window before running tests
> > like this?
> 
> 
> Merge window is a weak notion and may be not enough either (all trees
> do not necessary update at that point and syzbot does not necessary
> rebuild all of them successfully). syzbot uses another criteria: if
> you say a bug is fixed by commit X, it will wait until commit X
> reaches all of tested trees and will report the same crash signature
> again only after that. This procedure was specifically designed to not
> produce duplicate reports about the same bug.
> So either the bug wasn't really fixed, or this is another bug, or
> syzbot was given a wrong commit.

Hmmm.  Which are the "tested trees"?

This bug (e1d1a6e595adbd2458f1) is marked as a duplicate of 
3cbe5cd105d2ad56a1df.  The dashboard link says that bug was fixed by 
commit "HID: logitech: Fix general protection fault caused by Logitech 
driver" -- which is correct, as far as I know.

That commit is present in linux-next, as mentioned above.  As of 10:44 
EDT today, it is not present in Linus's tree, according to

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/log/drivers/hid/hid-lg.c

(in fact, no commits affecting drivers/hid/hid-lg.c in that tree are 
dated after 2019-07-10).

Furthermore, according to

https://github.com/google/kmsan/blob/master/drivers/hid/hid-lg.c?h=014077b5

the source code actually used by syzbot for this test doesn't have that 
commit either.  (BTW, is there any way to get a git log out of github?  
It would be nice not to have to download the whole source file -- and 
I'm not certain that this URL really does point to the version of the 
file that syzbot used.)

So what's really going on?

Alan Stern