Make memfd_create syscall directly
#1912
Conversation
There was a problem hiding this comment.
this looks good to me!
AFAIK, the memfd_create function just makes the syscall exactly the way this code does (or at least, the musl version does; the glibc source tree is harder to navigate so i gave up), so i don't think there's anything we need to worry about by doing it ourselves. looks good to me!
also, i believe the relevant syscall has existed in Linux since approximately forever ago, so we probably don't need to worry about it not being available. |
|
CI failure is due to a dependency breaking MSRV and is unrelated, I'll resolve that separately --- you don't need to worry about it. |
|
updated with the latest master, which should fix the MSRV checks |
There was a problem hiding this comment.
I have a small nitpick, but otherwise 👍
As for the question, yes, if the syscall doesn't exist the call will fail with ENOSYS and the message is lost.
I'd just take this risk; in the ticket the OP said it exists since some 3.x version, and I don't think anyone still uses kernels that old 🙂
But if you'd like to make it safe I guess we could take a look at the systemd code from before the call was added to glibc to see what it used instead back then, and use that as a fallback on ENOSYS.
But personally I'd just document it and bother no more 🙂
| /// RHEL 7, etc. | ||
| /// | ||
| /// See: https://github.com/tokio-rs/tracing/issues/1879 | ||
| fn memfd_create_syscall(flags: c_uint) -> i64 { |
There was a problem hiding this comment.
This is broken in 32-bit environments where c_long is, of course, 32 bit
There was a problem hiding this comment.
ah, good catch, that should return a c_long
# 0.2.3 (February 7, 2022) ### Fixed - Fixed missing `memfd_create` with `glibc` versions < 2.25 ([tokio-rs#1912]) ### Changed - Updated minimum supported Rust version to 1.49.0 ([tokio-rs#1913]) Thanks to @9999years for contributing to this release! [tokio-rs#1912]: tokio-rs#1912 [tokio-rs#1913]: tokio-rs#1913
Fixes #1879
Motivation
journald-tracing>=0.2.1doesn't build with old glibc.Solution
Make the
memfd_createsyscall ourselves.Open questions
memfd_createisn't supported? Do large payloads just get trashed?cc @lunaryorn @Ralith