GIMP help shouldn't need execstack, should it?

[John Reiser]

On 01/08/2011 04:50 AM, Göran Uddeborg wrote:
> It turns out the help-browser does not have any execstack flag set.
> And none of the libraries involved either.
> 
> I used strace to see what actually was done, and found out that this
> happens when the process tries to mmap() some anonymous memory:
> 
>     17020 gettimeofday( <unfinished ...>
>     17020 <... gettimeofday resumed> {1294488756, 742289}, NULL) = 0
>     17020 mmap(NULL, 2147483648, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0 <unfinished ...>
>     17020 <... mmap resumed> )              = -1 EACCES (Permission denied)
>     17020 --- SIGSEGV (Segmentation fault) @ 0 (0) ---
> 
> Looking a bit more (and repeating how to debug applications that fork
> with gdb:-) I found that this is in the
> /usr/lib64/libwebkitgtk-1.0.so.0, with the stack trace included below.
> 
> I don't know anything about webkitgtk, but strings such as "jit" in
> that trace make me suspect that maybe this could be intentional.  Now
> I'm not sure what to do.  Should I bug report webkitgtk.

Yes, file a bug against webkitgtk.  Include the output from strace.

>   And if so,
> is there any "right" way to do just-in-time compilation, if that is
> indeed what happens?  Any good pointers I could add to such a bug
> report?

This works for me: allocate space first as PROT_NONE, then MAP_FIXED the
allocated space to the desired protection:

    addr = mmap(0,    length, PROT_NONE, MAP_PRIVATE_MAP_ANONYMOUS, -1, 0);
    if (MAP_FAILED==addr) ...
    addr = mmap(addr, length, PROT_READ|PROT_WRITE|PROT_EXEC,
        MAP_FIXED|MAP_PRIVATE_MAP_ANONYMOUS, -1, 0);
    if (MAP_FAILED==addr) ...

--