Re: [dpdk-dev] [PATCH] eal: add madvise to avoid dump memory

[Feng Li <lifeng1519@gmail.com>]

Hi Anatoly,

This program could run like this:
gcc test_map.c
./a.out 2 &
gcore `pidof a.out`

You will get a core dump sized to 2GiB.

Thanks,
Feng Li

Burakov, Anatoly <anatoly.burakov@intel.com> 于2020年4月21日周二 下午8:19写道：
>
> On 21-Apr-20 12:06 PM, Feng Li wrote:
> > #include <sys/mman.h>
> > #include <time.h>
> > #include <stdint.h>
> > #include <unistd.h>
> > #include <stdlib.h>
> > #include <stdio.h>
> >
> > int main(int argc, char** argv) {
> >      // FIXME(fengli): XXXXX
> >      uint64_t gb = atoi(argv[1]);
> >      void* ptr = mmap(0, gb << 30, PROT_NONE, MAP_PRIVATE |
> > MAP_ANONYMOUS, -1, 0);
> >      if (ptr == (void*)-1) {
> >          perror("[-] mmap failed with MAP_PRIVATE | MAP_ANONYMOUS");
> >          exit(1);
> >      }
> >      while(1)
> >          sleep(1);
> >      return 0;
> > }
>
> DONTDUMP is available since Linux 3.4. I presume our minimum kernel
> version is higher than that.
>
> I have little idea of how dumping works, but reading the manpage for
> madvise, DONTDUMP should be the way to go here. Also, reading up on
> PROT_NONE, i can't find any references to this memory necessarily being
> excluded from core dumps.
>
> That said, I've run the program above, and i got a core dump sized
> ~100K. Do i need any special configuration to trigger core dump that
> would include that anonymous memory?
>
> --
> Thanks,
> Anatoly