Re: general protection fault in skb_put

[Dmitry Vyukov]

On Mon, Mar 11, 2019 at 7:10 PM James Smart <jsmart2021@xxxxxxxxx> wrote:
>
> On 3/11/2019 9:40 AM, Dmitry Vyukov wrote:
> > On Mon, Mar 11, 2019 at 5:20 PM 'James Smart' via syzkaller-bugs
> > <syzkaller-bugs@xxxxxxxxxxxxxxxx> wrote:
> >>
> >> On 3/11/2019 6:20 AM, syzbot wrote:
> >>> syzbot has bisected this bug to:
> >>>
> >>> commit 97faec531460c949d7120672b8c77e2f41f8d6d7
> >>> Author: James Smart <jsmart2021@xxxxxxxxx>
> >>> Date:   Thu Sep 13 23:17:38 2018 +0000
> >>>
> >>>      nvme_fc: add 'nvme_discovery' sysfs attribute to fc transport device
> >>>
> >>> bisection log:
> >>> https://syzkaller.appspot.com/x/bisect.txt?x=121f55db200000
> >>> start commit:   97faec53 nvme_fc: add 'nvme_discovery' sysfs attribute
> >>> to ..
> >>> git tree:       linux-next
> >>> final crash: https://syzkaller.appspot.com/x/report.txt?x=111f55db200000
> >>> console output: https://syzkaller.appspot.com/x/log.txt?x=161f55db200000
> >>> kernel config: https://syzkaller.appspot.com/x/.config?x=59aefae07c771af6
> >>> dashboard link:
> >>> https://syzkaller.appspot.com/bug?extid=65788f9af9d54844389e
> >>> userspace arch: amd64
> >>> syz repro: https://syzkaller.appspot.com/x/repro.syz?x=178e0798c00000
> >>> C reproducer: https://syzkaller.appspot.com/x/repro.c?x=11b4f0b0c00000
> >>>
> >>> Reported-by: syzbot+65788f9af9d54844389e@xxxxxxxxxxxxxxxxxxxxxxxxx
> >>> Fixes: 97faec53 ("nvme_fc: add 'nvme_discovery' sysfs attribute to fc
> >>> transport device")
> >>
> >> can someone contact me as to what this thing is doing and how to
> >> interpret all the logs.  nvme_fc isn't remotely in any of the logs and
> >> doesn't use skb's unless the underlying udev_uevents are using them.
> >
> > Hi James,
> >
> > What exactly is unclear/needs interpretation? syzbot did what is
> > commonly known as kernel/git bisection process. This is a new feature
> > so there can be some rough edges. Hopefully we can improve the
> > representation together.
> >
> > Thanks
> >
> Everything is unclear. You're telling me that an error occurred and that
> you reduced it to the git submit where the error starts appearing.
>
> Usually there would be something in the base crash, which I'm looking at
> in https://syzkaller.appspot.com/x/report.txt?x=111f55db200000 which
> would point back at something in the patch or related to it. There are
> no relationships.  I can't quite figure out what the base test actually
> did that generated the failure to see if there's any possible relationship.
>
> Everything in the base crash stacktrace points to an issue in the
> bluetooth uart driver doing all the logging - not the patch called out.

Everything up to this point is perfectly correct. So lots of things
seem to be clear to you ;)
The base test case is provided in under the "syz/C repro" links in the
original report and in the bisection results report.

> So this looks like a failure of your infrastructure.

I agree that the result seems to be unrelated to the original crash.
What is the root cause is a good question. You can see the exact
history of how bisection progressed any why it ended up at the commit
it ended up over the "bisection log" link.
Kernel is unfortunately (or fortunately) is not a single-threaded
deterministic user-space parser library without global state where
everything can be bisected precisely. There is a very long tail of
other problems as well. E.g. the same reproducer triggering multiple
bugs at once, of different bugs at different commit ranges. At the
same time lots of people asked for bisection of bugs. So this is where
we are.
I've started collecting all cases with incorrect bisection results, so
that we can draw broader conclusions later and bucket common root
causes:
https://github.com/google/syzkaller/issues/1051
Added this case too.