Re: Detecting the availability of VSYSCALL

* Andy Lutomirski:

On Tue, Jun 25, 2019 at 1:47 PM Florian Weimer [off-list ref] wrote:

quoted

* Andy Lutomirski:

quoted

quoted

We want binaries that run fast on VSYSCALL kernels, but can fall back to
full system calls on kernels that do not have them (instead of
crashing).

Define "VSYSCALL kernels."  On any remotely recent kernel (*all* new
kernels and all kernels for the last several years that haven't
specifically requested vsyscall=native), using vsyscalls is much, much
slower than just doing syscalls.  I know a way you can tell whether
vsyscalls are fast, but it's unreliable, and I'm disinclined to
suggest it.  There are also at least two pending patch series that
will interfere.

The fast path is for the benefit of the 2.6.32-based kernel in Red Hat
Enterprise Linux 6.  It doesn't have the vsyscall emulation code yet, I
think.

My hope is to produce (statically linked) binaries that run as fast on
that kernel as they run today, but can gracefully fall back to something
else on kernels without vsyscall support.

quoted

quoted

We could parse the vDSO and prefer the functions found there, but this
is for the statically linked case.  We currently do not have a (minimal)
dynamic loader there in that version of the code base, so that doesn't
really work for us.

Is anything preventing you from adding a vDSO parser?  I wrote one
just for this type of use:

$ wc -l tools/testing/selftests/vDSO/parse_vdso.c
269 tools/testing/selftests/vDSO/parse_vdso.c

(289 lines includes quite a bit of comment.)

I'm worried that if I use a custom parser and the binaries start
crashing again because something changed in the kernel (within the scope
permitted by the ELF specification), the kernel won't be fixed.

That is, we'd be in exactly the same situation as today.

With my maintainer hat on, the kernel won't do that.  Obviously a
review of my parser would be appreciated, but I consider it to be
fully supported, just like glibc and musl's parsers are fully
supported.  Sadly, I *also* consider the version Go forked for a while
(now fixed) to be supported.  Sigh.

We've been burnt once, otherwise we wouldn't be having this
conversation.  It's not just what the kernel does by default; if it's
configurable, it will be disabled by some, and if it's label as
“security hardening”, the userspace ABI promise is suddenly forgotten
and it's all userspace's fault for not supporting the new way.

It looks like parsing the vDSO is the only way forward, and we have to
move in that direction if we move at all.

It's tempting to read the machine code on the vsyscall page and analyze
that, but vsyscall=none behavior changed at one point, and you no longer
any mapping there at all.  So that doesn't work, either.

I do hope the next userspace ABI break will have an option to undo it on
a per-container basis.  Or at least a flag to detect it.

Thanks,
Florian