[Libguestfs] alternatives for hooking dlopen() without LD_LIBRARY_PATH or LD_AUDIT?

[Eric Blake]

On 2/21/20 9:00 AM, Florian Weimer wrote:
> * Richard W. M. Jones:
> 
>> On Fri, Feb 21, 2020 at 01:19:34PM +0100, Florian Weimer wrote:
>>> I think what confuses me is that keep talking about a single binary, but
>>> clearly there is this separate vddk DSO, and there is talk of plugins.
>>> So it seems to me that multiple files are involved already?
>>
>> nbdkit is a standalone binary that happens to be able to load plugins
>> from a well-known path, eg nbdkit-vddk-plugin.so.  nbdkit knows the
>> path for plugins, and there's a wrapper allowing it to get local
>> plugins even when it's still in the build directory.  Adding another
>> file would mean another path (or overloading the meaning of the plugin
>> path) and just makes the whole thing more fragile and complex.
>>
>> Having said all that, what would also solve this is either an API for
>> updating LD_LIBRARY_PATH after the program has started; or making
>> setenv ("LD_LIBRARY_PATH",...) DTRT*; or some kind of dlopen() variant
>> which takes a library path as an extra parameter.
> 
> Have you tried adding DT_RUNPATH or DT_RPATH to nbdkit-vddk-plugin.so

Post-processing an existing closed-source .so shipped from an external 
vendor might have negative consequences - while it may be possible to 
modify the ELF image to add a DT_RUNPATH entry or modify the DT_NEEDED 
entries to use anchored names based on $ORIGIN rather than bare names, 
there's no telling if such modification would be in violation of 
agreements or even cause failure to load if the proprietary code is 
using shenanigans like validating a checksum of shipped binaries to 
detect tampering.

> Or does the path have to be chosen dynamically?

So, since we cannot fix the existing product, and have no idea if/when 
the vendor will release an updated version that has saner libraries, a 
dynamic search path is the only option to getting their product to load 
(but whether that is done by LD_LIBRARY_PATH, LD_AUDIT, pre-loading 
libraries, or something else, is where we have a bit more control).

> 
> If you merely want to prevent loading of libstdc++.so or libcrypto.so by
> vddk, it may be possible to explicitly dlopen DSOs of that name before
> loading vddk.  But there is an existing bug where we do not duplicate
> properly on soname alone, so we may have to fix that first.

The problem then becomes "given an arbitrary libvixDiskLib.so, how do we 
determine the dependency of bare libraries it would want to load so that 
we can pre-emptively load those libraries by direct path name first". 
My re-exec solution was nice - it works for all versions of VDDK. 
Whereas with scraping the binary to see what DT_NEEDED entries it has is 
a bit more image-specific: for example, VDDK 5.5.5 loads 
"libcrypto.so.1.0.0", VDDK-6.5 loads "libcrypto.so.1.0.2".  I don't want 
to hard-code any of those library names into nbdkit-vddk-plugin.so, 
because that would needlessly tie the open source program into a 
specific VDDK release, instead of working with all of them.  However, if 
it is easy enough to compute our own topological sort of DT_NEEDED for a 
given VDDK .so, and then pre-load in leaf-first order, so that by the 
time we finally get around to dlopen()ing libvixDiskLib.so and calling 
its init function, then all of the subsequent dlopen()s performed by 
VDDK code will succeed right off the bat because the library has already 
been loaded in memory.  Or at least, that's what I'm hoping your 
suggesting.  And if we do that, it would avoid the need to re-exec with 
LD_LIBRARY_PATH set.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org