Re: kernel BUG at net/rxrpc/local_object.c:LINE!

[Dmitry Vyukov]

On Fri, Jul 5, 2019 at 2:12 PM Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> > syzbot <syzbot+1e0edc4b8b7494c28450@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > I *think* the reproducer boils down to the attached, but I can't get syzkaller
> > to work and the attached sample does not cause the oops to occur.  Can you try
> > it in your environment?
> >
> > > The bug was bisected to:
> > >
> > > commit 46894a13599a977ac35411b536fb3e0b2feefa95
> > > Author: David Howells <dhowells@xxxxxxxxxx>
> > > Date:   Thu Oct 4 08:32:28 2018 +0000
> > >
> > >     rxrpc: Use IPv4 addresses throught the IPv6
> >
> > This might not be the correct bisection point.  If you look at the attached
> > sample, you're mixing AF_INET and AF_INET6.  If you try AF_INET throughout,
> > that might get a different point.  On the other hand, since you've bound the
> > socket, the AF_INET6 passed to socket() should be ignored.
> >
> > David
> > ---
> > #include <stdio.h>
> > #include <stdlib.h>
> > #include <string.h>
> > #include <unistd.h>
> > #include <sys/socket.h>
> > #include <arpa/inet.h>
> > #include <linux/rxrpc.h>
> >
> > static const unsigned char inet4_addr[4] = {
> >         0xe0, 0x00, 0x00, 0x01
> > };
> >
> > int main(void)
> > {
> >         struct sockaddr_rxrpc srx;
> >         int fd;
> >
> >         memset(&srx, 0, sizeof(srx));
> >         srx.srx_family                  = AF_RXRPC;
> >         srx.srx_service                 = 0;
> >         srx.transport_type              = AF_INET;
> >         srx.transport_len               = sizeof(srx.transport.sin);
> >         srx.transport.sin.sin_family    = AF_INET;
> >         srx.transport.sin.sin_port      = htons(0x4e21);
> >         memcpy(&srx.transport.sin.sin_addr, inet4_addr, 4);
> >
> >         fd = socket(AF_RXRPC, SOCK_DGRAM, AF_INET6);
> >         if (fd == -1) {
> >                 perror("socket");
> >                 exit(1);
> >         }
> >
> >         if (bind(fd, (struct sockaddr *)&srx, sizeof(srx)) == -1) {
> >                 perror("bind");
> >                 exit(1);
> >         }
> >
> >         sleep(20);
> >
> >         // Whilst sleeping, hit with:
> >         // echo -e '\0\0\0\0\0\0\0\0' | ncat -4u --send-only 224.0.0.1 20001
> >
> >         return 0;
> > }
>
> Hi David,
>
> I can't re-reproduce it locally in qemu either. Though, syzbot managed
> to re-reproduce it reliably during bisection (maybe there is some
> difference in hardware and as the result the injected ethernet packet
> would need some different values). Let's try to ask it again to make
> sure:
> #syz test: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> master
>
> Re bisection, I don't know if there are some more subtle things as
> play (you are in the better position to judge that), but bisection log
> looks good, it tracked the target crash throughout and wasn't
> distracted by any unrelated bugs, etc. So I don't see any obvious
> reasons to not trust it.

FWIW here is a more complete translation of the syzkaller repro to C using:

$ syz-prog2c -prog /tmp/prog -threaded -collide -repeat=0 -procs=6
-sandbox=namespace -enable=tun,net_dev,net_reset,cgroups,close_fds
-tmpdir -segv

https://gist.githubusercontent.com/dvyukov/f712ca7c3a0d355ce63823d7882c2934/raw/7a72635b99e5a85599a6bcf9b7901fa9d8c494d4/repro.c

However, both syzbot and me won't able to repro with this C program,
so it is expected that it does not reproduce the crash for some
reason.