tracing-journald fix for #1698 fails to build on glibc < 2.27

[pkhuong]

Bug Report

Building tracing-journald 0.2.2 on Ubuntu Xenial (supported until 2024), Debian Stretch (supported until June 2022) or RHEL 7 (supported until 2024) fails with in function `tracing_journald::memfd::create': ... github.com-1ecc6299db9ec823/tracing-journald-0.2.2/src/memfd.rs:11: undefined reference to `memfd_create'.

This syscall was introduced in Linux 3.17 (in 2014, safely out of LTS), but only exposed by glibc in 2.27 (released in Feb 2018). Unfortunately, libc (used by #1744) just calls out to glibc for this Linux extension, instead of directly invoking a syscall(2) stub, which would always link and return ENOSYS on very old kernels.

Version

$ cargo tree | grep tracing
│   │   │   │   └── tracing v0.1.29
│   │   │   │       ├── tracing-attributes v0.1.18 (proc-macro)
│   │   │   │       └── tracing-core v0.1.21
│   │   │   │   ├── tracing v0.1.29 (*)
│   ├── tracing v0.1.29 (*)
│   ├── tracing-subscriber v0.3.7
│   │   ├── tracing v0.1.29 (*)
│   │   ├── tracing-core v0.1.21 (*)
│   │   └── tracing-log v0.1.2
│   │       └── tracing-core v0.1.21 (*)
│   └── tracing v0.1.29 (*)
├── tracing v0.1.29 (*)
├── tracing-journald v0.2.2
│   ├── tracing-core v0.1.21 (*)
│   └── tracing-subscriber v0.3.7 (*)
├── tracing-subscriber v0.3.7 (*)
    ├── tracing v0.1.29 (*)

Platform

Ubuntu Xenial, Debian Stretch, RHEL 7, on x86-64.

Crates

tracing-journald >= 0.2.1

Description

I try to build a crate that depends on tracing-journald. Linking fails because no one defines memfd_create. Pinning to 0.2.0 restores the build.

[hawkw]

cc @lunaryorn, @Ralith

i'm not sure what the best way to handle compatibility with earlier glibc versions is in Rust. possibly a build script to detect the glibc version and set a cfg flag to enable/disable the memfd code?

[hawkw]

alternatively, i suppose we could just make the system call ourselves rather than depending on it from libc...

[swsnr]

I'm sorry that my changes broke things, but I would not like contribute to fixing this, to be honest.

The OP mentions RHEL 7, Ubuntu Xenial, and Debian Stretch. Stretch is end of live, Ubuntu Xenial is in ESM (i.e. you've got to pay for support), and RHEL is flat out enterprise. Whoever uses those distributions works in some kind of commercial enterprise environment, and should definitely have funding and resources to fix things on their own or contract it out for good money, instead of trying to get the community to fix it for free.

The OP can just pin to 0.2.0 which is as good as trying to cfg-away the memfd_create call 🙂

Edit: I'm willing to review any merge requests to tracing-journald of course 🙂

[Ralith]

I agree that pinning is a reasonable solution to this, though I'd be willing to review a PR to invoke syscall ourselves.

[9999years]

I'm from one of those enterprise environments interested in patching this to invoke syscall but the docs are not exactly elucidating -- where should I start?

(Pinning 0.2.0 doesn't work because I need to set the syslog identifier, introduced in 0.2.1.)

EDIT: Wait, I think I found the syscall table, nvm.