Re: mm: GPF in find_get_pages_tag

[Dmitry Vyukov]

On Fri, Jul 15, 2016 at 10:21 PM, Ross Zwisler
<ross.zwisler@xxxxxxxxxxxxxxx> wrote:
> On Fri, Jul 15, 2016 at 09:07:48PM +0200, Dmitry Vyukov wrote:
>> On Fri, Jul 15, 2016 at 9:03 PM, Ross Zwisler
>> <ross.zwisler@xxxxxxxxxxxxxxx> wrote:
>> >> // autogenerated by syzkaller (http://github.com/google/syzkaller)
>> >> #include <pthread.h>
>> >> #include <stdint.h>
>> >> #include <string.h>
>> >> #include <stdio.h>
>> >> #include <sys/syscall.h>
>> >> #include <unistd.h>
>> >>
>> >> int fd;
>> >> char buf[8192];
>> >> char filename[256];
>> >>
>> >> void* thr(void* arg)
>> >> {
>> >>   switch ((long)arg) {
>> >>   case 0:
>> >>     write(fd, buf, 0x1001ul);
>> >>     break;
>> >>   case 1:
>> >>     fdatasync(fd);
>> >>     break;
>> >>   case 2:
>> >>     ftruncate(fd, 2);
>> >>     break;
>> >>   case 3:
>> >>     write(fd, buf, 0x20ul);
>> >>     break;
>> >>   case 5:
>> >>     fd = open(filename, 0x50042ul, 0x41ul);
>> >>     break;
>> >
>> > This open() code is unreachable because the thread argument will only be 0-4,
>> > right?  Should this be "case 4"?
>>
>> I am not sure. I think it I just copy-pasted the program that
>> triggered the crash for me. Andrey should have a valid reproducer, in
>> the other thread he said that he can reproduce it. Andrey, did you
>> change 5 to 4?
>
> Ah, sorry if I wasn't clear.  I don't think you need the open() call to have a
> valid reproducer.  In mine, in fact, I only use the first three - the error
> happens with a combination of write(), fdatasync() and ftruncate().
>
> I just wanted to note that the test program (which was autogenerated?) had an
> unreachable case in the switch() statement. :)
>
> Thanks for this testing, by the way!


Ah, OK. I modified the program by hand to make it trigger the bug more
frequently. So I think the bug was introduced by me. Generator should
not generate dead cases.