[PATCH] PT_GNU_STACK

[Andrew Cagney]

> On Sun, Jun 01, 2003 at 05:34:00PM -0400, Andrew Cagney wrote:
> 
>> >Hi Andrew,
>> >
>> >
> 
>> >>Wouldn't two note sections work better:
>> >>
>> >>	if (.note.GNU-stack.rwx)
>> >>	  stack read-write-executable
>> >>	else if (.note.GNU-stack.rw)
>> >>	  stack read-write
> 
>> 
>> Having empty sections would let binutils ignore the problem entirely. 
>> Instead it could just pass the sections through.
> 
> 
> That's not true. If ld doesn't handle them specially
> during ld -r and final link, then there is no way to differentiate between
> all input files .note.GNU-stack.rw and some input files .note.GNU-stack.rw
> and some input files without any note section (in the former case
> non-exec stack can be used, in the latter case most probably shouldn't,
> as no assumptions can be made on whether the remaining objects need
> or don't need exec stack).

That's true.  But is it reasonable to assume that a non-branded .o file 
should be rwx?  It's going to make it too easy to accidently create 
programs with an executible stack.

> Another thing is efficiency. If there is more than one PT_ segment for this
> stack or if kernel/ld.so has to read some segment and analyze it to find
> out whether exec stack is needed or not, it means more runtime overhead.
> And IMHO the default should be clear from what the ABI allows or doesn't
> allow. If say i?86-*-linux* so far could assume executable stack, then the
> default for old .o files must be executable stack.

The intent was names, not contents, could be used to make the decision 
and that would avoid the double read.

Andrew