Re: [PATCH] Fix glibc 2.34 ABI omission (missing GLIBC_2.34 in dynamic loader)

[Florian Weimer <fweimer@redhat.com>]

* Andreas Schwab:

> On Dez 08 2021, Florian Weimer via Libc-alpha wrote:
>
>> The glibc 2.34 release really should have added a GLIBC_2.34
>> symbol to the dynamic loader.
>
> Well, the ship has sailed.

Has it?  It's just software, we can change it.

It's different from other ABI changes in that we wouldn't be using the
symbol version node immediately, and the actual symbol (if we start to
use it) would already be there in libc.so.6.

>> With it, we could move functions such as dlopen or pthread_key_create
>> that work on process-global state into the dynamic loader (once we
>> have fixed a longstanding issue with static linking).
>
> Why does this depend on the frozen release?

Say we move dlopen into ld.so in glibc 2.36.  It will no longer be
present in libc.so.6.  Applications linked against glibc 2.36 will
continue referencing dlopen@GLIBC_2.34, but with a soname baked into the
version reference that mentions the dynamic loader (say, ld64.so.1).  On
the current glibc 2.34 release, the dynamic loader sees that soname,
checks if the object has a version node for GLIBC_2.34, and refuses
loading the application because ld64.so.1 does not have this version
node.

Thanks,
Florian