[PATCH v3 00/11] sysctl: treewide: constify ctl_table argument of sysctl handlers

[Joel Granados]

On Fri, May 03, 2024 at 04:09:40PM +0200, Thomas Weißschuh wrote:
> Hey Joel,
> 
...
> > # Motivation
> > As I read it, the motivation for these constification efforts are:
> > 1. It provides increased safety: Having things in .rodata section reduces the
> >    attack surface. This is especially relevant for structures that have function
> >    pointers (like ctl_table); having these in .rodata means that these pointers
> >    always point to the "intended" function and cannot be changed.
> > 2. Compiler optimizations: This was just a comment in the patchsets that I have
> >    mentioned ([3,4,5]). Do you know what optimizations specifically? Does it
> >    have to do with enhancing locality for the data in .rodata? Do you have other
> >    specific optimizations in mind?
> 
> I don't know about anything that would make it faster.
> It's more about safety and transmission of intent to API users,
> especially callback implementers.
Noted.

...
> > # Show the move
> > I created [8] because there is no easy way to validate which objects made it
> > into .rodata. I ran [8] for your Dec 2nd patcheset [7] and there are less in
> > .rodata than I expected (the results are in [9]) Why is that? Is it something
> > that has not been posted to the lists yet? 
> 
> Constifying the APIs only *allows* the actual table to be constified
> themselves.
> Then each table definition will have to be touched and "const" added.
That is what I thought. thx for clarifying.

> 
> See patches 17 and 18 in [7] for two examples.
> 
> Some tables in net/ are already "const" as the static definitions are
> never registered themselves but only their copies are.
> 
...

best

-- 

Joel Granados
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-riscv/attachments/20240508/ad04f2b9/attachment.sig>