[lkml]   [2020]   [Nov]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectSolving the portable binaries problem in the elf loader
Hi Guys,

I remember watching this YouTube video years ago and something Linus
said really stuck with me:

It was around the whole binaries, package management problem, because
it's been something that has been troublesome for me, building separate
binaries for separate Linux platforms.

I've examined the various efforts to fix this: flatpack, snap,
AppImage, FatELF, docker etc. The kindof solve the problem but not
100%, some aspects of their approaches are nice. I think this
problem can be addressed in the ELF loader. By the way I don't claim to
be an expert on ELF, I know just enough to do my job.

So ELF tends to depend on a filename with a version embedded in the
filename to determine if the shared library is the correct one to load,
/bin/true example:

ldd /bin/true (0x00007ffe0218a000) => /lib/x86_64-linux-gnu/ (0x00007f47086c9000)
/lib64/ (0x00007f47088e5000)

What if instead of depending on a filename on determining if the
library is the correct version, we used a checksum (like a SHA or
something even cheaper) taking a little influence from how git
determines if patches are valid, this is a relatively cheap operation:

time sha1sum /lib/x86_64-linux-gnu/
4fcd76645607f38d91f65654ddd6b9770b5ea54a /lib/x86_64-linux-gnu/

real 0m0.008s
user 0m0.008s
sys 0m0.001s

You could store a volatile cache of checksummed libraries in memory to
ensure you don't do this more than once for each inode that contains a

package managers generally add their own versioning system, because
kernelspace doesn't really give you one. With this ELF loading feature
you now have that. It also opens up the possibility of distributing
binaries and libraries in a more P2P fashion as a checksum is core to
the ELF loading process.

If this feature was added I think it would percolate down to the
various distros, package managers in time.

The end product would be more cross-platform binaries, without doing
tricks like fat binaries. In code you could maintain backwards
compatibility like:

if (legacy_elf) {
return 0;


I wouldn't mind hearing thoughts from Linus on this as it fixes many of
the problems he mentions like:

"you don't make binaries for Linux you, make binaries for Fedora 19 for
Oh 20, maybe there's even like rl5 from ten year ago"

with the checksummed library approach you fix this problem. When you
execute the binary and if it doesn't have all the checksummed libraries
available you pull a library with the correct checksum.

"using shared libraries is not an option when the libraries are
experimental and the libraries are used by two people and one of them
is crazy so every other day some ABI breaks right so"

Fixes this problem because you embed checksums for the libraries you
want your binary to load, of course you must update that checksum in
the main binary if you want to upgrade. That actually doesn't seem
like a huge problem as checksums are fixed width.

"they break binary compatibility left and right they update glibc and
everything breaks you a you can"

again, the checksums fix this.

"can't have application writers do 15 billion different versions"

I think if you centralize how versioning is done by using checksums in
the elf loader you relieve this problem.

"I also guarantee you that every single desktop distribution will care
about Valve binaries so the problem is valve will build everything
statically linked and create huge binaries"

You are still using dynamic linking using this model, so this problem
is solved.

I could keep going but I think it solves most of the problems he
talks about (and I tend to have the same problems), I think it could be
solved in the ELF loader. Anybody see potential here? I'd greatly
appreciate feedback.

As a side effect you also gain more security around loading libraries,
although that's not the problem I was trying to solve here, it's the
many builds, portable binaries problem I'm trying to poke.

Might even save the environment a little, many more builds of binaries
around the world than needed, the checksum when dynamically
linking/loading seems like a fair tradeoff.

 \ /
  Last update: 2020-11-02 17:10    [W:0.033 / U:0.264 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site