From: Ted Bullock <tbullock@comlore.com>
Subject: Re: const-ify tables
To: Christian Weisgerber <naddy@mips.inka.de>
Cc: gameoftrees@openbsd.org
Date: Tue, 15 Feb 2022 14:29:17 -0700



On 2022-02-15 2:12 p.m., Christian Weisgerber wrote:
> Ted Bullock:
> 
>> Interesting question to me, if you force writing to one of these const
>> variables (ignoring compiler warnings) what happens to the program? Are
>> there any security benefits to moving data to supposedly read only memory?
> 
> They are actually placed in a memory section that is mapped read-only.
> 
> $ cat t.c
> const int i = 1;
> 
> int
> main(void)
> {
>         int *j;
> 
>         j = &i;
>         *j = 2;
> 
>         return 0;
> }
> $ cc -o t t.c
> t.c:8:4: warning: assigning to 'int *' from 'const int *' discards qualifiers [-Wincompatible-pointer-types-discards-qualifiers]
>         j = &i;
>           ^ ~~
> 1 warning generated.
> $ objdump -t t
> ...
> 00000000000004c8 g     O .rodata        0000000000000004 i
> ...
> $ ./t
> Segmentation fault (core dumped) 

Definitely strong improvement then, I like that.

> 
> The tables in got.c etc. contain pointers, so they are placed in
> the .data.rel.ro section, where the ld.so(1) dynamic linker first
> updates the pointers to their final values and then maps the region
> read-only before calling the main program.
> 




-- 
Ted Bullock <tbullock@comlore.com>