Re: [RFC PATCH v2 1/2] mm/mremap: Fix uprobe anon page be overwritten when expanding vma during mremap

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Tue, May 27, 2025 at 05:30:08PM +0200, Oleg Nesterov wrote:
> Not that this is really important, but the test-case looks broken,
>
> On 05/27, Pu Lehui wrote:
> >
> > #define _GNU_SOURCE
> > #include <fcntl.h>
> > #include <unistd.h>
> > #include <syscall.h>
> > #include <sys/mman.h>
> > #include <linux/perf_event.h>
> >
> > int main(int argc, char *argv[])
> > {
> >     int fd = open(FNAME, O_RDWR|O_CREAT, 0600);
>
> FNAME is not defined
>
> >     struct perf_event_attr attr = {
> >         .type = 9,
>
> Cough ;) Yes I too used perf_event_attr.type == 9 when I wrote another
> test-case. Because I am lazy and this is what I see in
> /sys/bus/event_source/devices/uprobe/type on my machine.
>
> But me should not assume that perf_pmu_register(&perf_uprobe) -> idr_alloc()
> will return 9.
>
> >     write(fd, "x", 1);
>
> looks unnecessary.
>
> Oleg.
>

While I agree we should probably try to do this nicely, in defence of Pu I think
this is adapted from the syzkaller horror show :P and that code does tend to
just insert random integers etc.

It would be good to refine this into something more robust if possible and
ideally add as a self-test, however!