Re: [PATCH] ld.so: Enable preloading of new symbol versions [BZ #24974]

[Carlos O'Donell <carlos at redhat dot com>]

On 10/2/19 1:00 PM, Florian Weimer wrote:
> * Carlos O'Donell:
> 
>> On 10/2/19 7:36 AM, Florian Weimer wrote:
>>> Can we please move the discussion forward?
>>
>> I think we've settled on:
>>
>> * No env var.
>> * Add a DT_ tag to disable the check.
>> * Users building the special preload DSOs need to set the
>>   new DT_ tag.
>> * Everything else is OK.
> 
> I have no recollection of that decision, or anyone else requesting that
> we follow this approach.  I expect that this will require a more
> invasive change to the dynamic linker.

Sorry, I thought that was what we were talking about when hashing out
the details to avoid a new env var. Always good to practice 3 way communication :-)
 
> Alan Modra pointed out how to avoid binutils changes for this:
> 
>   <https://sourceware.org/ml/binutils/2019-09/msg00287.html>
> 
> I don't know if we can get a tag reservation for GNU gABI for this
> feature.  I couldn't get one the last time, for .got.plt isolation (aka
> safe lazy binding), so that project is now essentially dead.

Previous rejection doesn't mean you will get rejected again?

I know it's a platitude, but "if at first you don't succeed..." keep
trying until you understand who the key decision makers are and their
particular requirements.

> I'm slightly concerned that what started out as a simple change is now
> something that needs coordination between glibc, binutils (at least for
> readelf), elfutils, and the GNU generic ABI.  Is this really the
> consensus on this list?

Right now it's just me and you talking about this feature.

*I* was hoping to avoid an env var that the user has to set to magically
disable the check. With the DT_ tag you could disable the check just
for one DSO, and keep the check for all others?

Your suggestion of a DT_ tag seemed like a really nice idea. Do you dislike
the idea for some reason? Only because it's more work and might fail?

-- 
Cheers,
Carlos.