Re: lost connection to test machine (3)

From: Dmitry Vyukov
Date: Thu Dec 28 2017 - 03:50:43 EST


On Wed, Dec 27, 2017 at 10:36 PM, Florian Westphal <fw@xxxxxxxxx> wrote:
> Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> On Wed, Dec 27, 2017 at 7:18 PM, syzbot
>> <syzbot+4396883fa8c4f64e0175@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
>> > Hello,
>> >
>> > syzkaller hit the following crash on
>> > beacbc68ac3e23821a681adb30b45dc55b17488d
>> > git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/master
>> > compiler: gcc (GCC) 7.1.1 20170620
>> > .config is attached
>> > Raw console output is attached.
>> > C reproducer is attached
>> > syzkaller reproducer is attached. See https://goo.gl/kgGztJ
>> > for information about syzkaller reproducers
>> >
>> >
>> > IMPORTANT: if you fix the bug, please add the following tag to the commit:
>> > Reported-by: <syzbot+4396883fa8c4f64e0175@xxxxxxxxxxxxxxxxxxxxxxxxx>
>> > It will help syzbot understand when the bug is fixed. See footer for
>> > details.
>> > If you forward the report, please keep this part and the footer.
>>
>> +netfilter maintainers
>>
>> Here is cleaned reproducer:
>>
>> // autogenerated by syzkaller (http://github.com/google/syzkaller)
>> #include <sys/types.h>
>> #include <sys/socket.h>
>> #include <netinet/in.h>
>> #include <netinet/tcp.h>
>> #include <linux/if.h>
>> #include <linux/netfilter_ipv4/ip_tables.h>
>>
>> int main()
>> {
>> int fd;
>>
>> fd = socket(AF_INET, SOCK_STREAM, IPPROTO_IP);
>> struct ipt_replace opt = {};
>> opt.num_counters = 1;
>> opt.size = -1;
>> setsockopt(fd, SOL_IP, 0x40, &opt, 0x4);
>> return 0;
>> }
>>
>>
>> What happens there is that here:
>>
>> struct xt_table_info *xt_alloc_table_info(unsigned int size)
>> {
>> ...
>> if ((SMP_ALIGN(size) >> PAGE_SHIFT) + 2 > totalram_pages)
>> return NULL;
>>
>> size = -1 and SMP_ALIGN(size) = 0, so this still tries to allocate
>> 4GB+delta bytes.
>>
>> I don't understand why this uses SMP_ALIGN since we add 2 pages on
>> top, it seems that we could just drop SMP_ALIGN and local SMP_ALIGN
>> definition altogether.
>
> Looking at history.git this seems to be a left over from back when
> iptables allocated size * num_cpus() (and used an SMP_ALIGN based offset
> for each cpu).
>
> So yes, I think we can just toss/drop this.


Thanks.
I've mailed "netfilter: fix int overflow in xt_alloc_table_info()" to fix this.