Re: ntdll: Randomize security cookie when available (try 4)

[Alexandre Julliard]

André Hentschel  writes:
> @@ -1278,6 +1281,26 @@ static NTSTATUS map_image( HANDLE hmapping, int fd, char *base, SIZE_T total_siz
>          }
>      }
> -
> - /* randomize security cookie */
> -
> - if (IMAGE_DIRECTORY_ENTRY_LOAD_CONFIG < nt->OptionalHeader.NumberOfRvaAndSizes &&
> -
>    (pos = nt->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_LOAD_CONFIG].VirtualAddress))
> - {
> -
>    IMAGE_LOAD_CONFIG_DIRECTORY *loadcfg = (IMAGE_LOAD_CONFIG_DIRECTORY *)(ptr + pos);
> -
>    ULONG_PTR *cookie = (ULONG_PTR *)loadcfg->SecurityCookie;
> -
>    struct timeval tv;
> -
> -
>    gettimeofday( &tv, NULL );
> -
>    srand( tv.tv_usec * getpid() );
> -
>    *cookie = rand();

That's not much better. It doesn't make sense to reset the seed on every
call.

ntdll: Randomize security cookie when available (try 4)

[André Hentschel]

Am 29.06.2015 um 14:16 schrieb Alexandre Julliard:
> André Hentschel  writes:
> > @@ -1278,6 +1281,26 @@ static NTSTATUS map_image( HANDLE hmapping, int fd, char *base, SIZE_T total_siz
> >          }
> >      }
> > -
> > - /* randomize security cookie */
> > -
> > - if (IMAGE_DIRECTORY_ENTRY_LOAD_CONFIG < nt->OptionalHeader.NumberOfRvaAndSizes &&
> > -
> >    (pos = nt->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_LOAD_CONFIG].VirtualAddress))
> > - {
> > -
> >    IMAGE_LOAD_CONFIG_DIRECTORY *loadcfg = (IMAGE_LOAD_CONFIG_DIRECTORY *)(ptr + pos);
> > -
> >    ULONG_PTR *cookie = (ULONG_PTR *)loadcfg->SecurityCookie;
> > -
> >    struct timeval tv;
> > -
> > -
> >    gettimeofday( &tv, NULL );
> > -
> >    srand( tv.tv_usec * getpid() );
> > -
> >    *cookie = rand();
>
> That's not much better. It doesn't make sense to reset the seed on every
> call.

Where would you see the best point for srand()?
__wine_process_init()?
Should i do it with a "once" variable?
Or would you prefer something like in SystemFunction036 using /dev/urandom (suggested by Bruno)?

ntdll: Randomize security cookie when available (try 4)

[Alexandre Julliard]

André Hentschel  writes:
> Where would you see the best point for srand()?
> __wine_process_init()?
> Should i do it with a "once" variable?
> Or would you prefer something like in SystemFunction036 using /dev/urandom (suggested by Bruno)?

Using srand() is not a good idea since it's process-global. Something
like RtlRandom is probably better.