Re: [RFC 3/5] elf: Add support to memory sealing

[Mike Hommey <mh@glandium.org>]

On Fri, Jun 28, 2024 at 08:06:46AM +0200, Florian Weimer wrote:
> * Mike Hommey:
> 
> > On Fri, Jun 28, 2024 at 07:51:05AM +0200, Florian Weimer wrote:
> >> * Mike Hommey:
> >> 
> >> > I just realized this can't work. For some reason I had the impression
> >> > the mseal was applied to the RELRO segment, but it's over the entire
> >> > library, which makes sense, in hindsight. The problem is that if we
> >> > remove RELRO, then... we can't even reapply it afterwards because of the
> >> > mseal, leaving us with a writable data section.
> >> > But if we disable mseal, we only get to disable it for everything, not
> >> > only our libs! (and only if we re-exec with GLIBC_TUNABLES set?)
> >> 
> >> We can introduce a flag in a dynamic tag at the same time we implement
> >> mseal.  The flag would isntruct the dynamic linker to skip mseal.  It's
> >> going to be some time until link editors know about the flag, but that
> >> doesn't matter in your case because you have a custom linker anyway,
> >> more or less.
> >
> > That would be the most useful, thank you. Are you thinking about some
> > DT_FLAGS/DT_FLAGS_1, or some other (new) tag?
> 
> I think we'd add this to the GNU generic ABI or the Linux ABI, so we
> can't use the existing flag tags.

https://sourceware.org/gnu-gabi/program-loading-and-dynamic-linking.txt
lists DT_GNU_FLAGS_1, although it's not in glibc.

Mike