[Bug libc/30306] ld.so.conf could easily / should support binary specific LD_PATH directory search path settings ?

["adhemerval.zanella at linaro dot org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=30306

Adhemerval Zanella <adhemerval.zanella at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |adhemerval.zanella at linaro dot o
                   |                            |rg

--- Comment #4 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
(In reply to Jason Vas Dias from comment #3)
> RE: Comment #2: 
> > Why can't you just add /usr/local/lib64 unconditionally to /etc/ld.so.conf?
> 
> Thanks, Andreas - 
> 
> Because there may be many other libraries in /usr/local/lib64 - in fact,
> there is a libgcc-shared.so and a libstdc++.so.6 in /usr/local/lib64,
> which I'd rather not pick up at all - I want to be able to move
> libbpf's libraries to some /usr/local/libbpf-v1.1.0/lib64 directory ,
> and specify that they apply ONLY to links needed for /usr/bin/perf,
> I do NOT want these updated libbpf* links to update the system
> libbpf at all, which MUST remain at v0.7.0, else everything that
> uses libbpf (except perf) will break .
> 
> It just seems to me that having some very simple extra config
> in ld.so.conf, to apply a LD_PATH only to certain executables,
> would prevent a whole host of error prone disasters involved 
> with binary patching in of a DT_RUNPATH ELF note .

We already have RPATH, which has precedence over the loader cache and it is not
subject to DF_1_NODEFLIB.  RPATH also accepts $ORIGIN, which does not tie where
the binary is installed; and it is also simpler to use since do not need admin
credential (to update ld.so.config and run ldconfig).

I don't have a strong opinion, but this seems to just replicate a subset of
RPATH and it will add an extra complexity on loader cache (which will need to
keep track of whether the entry has a path associated, and add a new type to
handle the directory suggestion).  

It also adds some corner cases: 

  - should it just support ET_EXEC? 
  - For ET_DYN, should it tie to the filesystem name or with its SONAME? 
  - The LD_LIBRARY_PATH is handled earlier than the cache and it is not subject
to DF_1_NODEFLIB, however your suggestion ties with LD_LIBRARY_PATH.

So I am not sure if this will be really an improvement.

-- 
You are receiving this mail because:
You are on the CC list for the bug.