Issue with `write_all_at` on GitHub Action

[allevo]

write_all_at doesn't write the buffer at the given index. The content is written at the end of the file.
I created a repo on my github https://github.com/allevo/rust-fs-write-all-bug to show the error.

Locally on Mac M2, the test on that repo passes, but on CI it doesn't.

NB:

• write_all_at and read_exact_at doesn't change the cursor position
• flush or/and sync_all doesn't fix

I tried this code:

use std::{io::Write, os::unix::fs::FileExt, path::PathBuf};

pub fn run(file_path: PathBuf) {
    let mut page_file = std::fs::File::options()
        .create(true)
        .append(true)
        .read(true)
        .open(&file_path)
        .unwrap();

    let n: u32 = 10;
    let buf: [u8; 4] = n.to_be_bytes();
    page_file
        .write_all_at(&buf, 0).unwrap();

    page_file.flush().unwrap();
    page_file
        .sync_all().unwrap();

    let mut buf: [u8; 4] = [0; 4];
    page_file
        .read_exact_at(&mut buf, 0).unwrap();
    let n = u32::from_be_bytes(buf);

    page_file.seek(SeekFrom::Start(0)).unwrap();

    let buf: [u8; 4] = (n + 1).to_be_bytes();
    page_file
        .write_all_at(&buf, 0).unwrap();
}

I expected to see this happen: The file content will be [0, 0, 0, 11]

Instead, this happened: [0, 0, 0, 10, 0, 0, 0, 11]

Meta

rustup --version && rustc --version --verbose:

info: This is the version for the rustup toolchain manager, not the rustc compiler.
rustup 1.2[8](https://github.com/allevo/rust-fs-write-all-bug/actions/runs/14934973609/job/41960164587#step:3:9).1 (f9edccde0 2025-03-05)
info: The currently active `rustc` version is `rustc 1.86.0 (05f[9](https://github.com/allevo/rust-fs-write-all-bug/actions/runs/14934973609/job/41960164587#step:3:10)846f8 2025-03-31)`
rustc 1.86.0 (05f9846f8 2025-03-31)
binary: rustc
commit-hash: 05f9846f893b09a1be1fc8560e33fc3c815cfecb
commit-date: 2025-03-31
host: x86_64-unknown-linux-gnu
release: 1.86.0
LLVM version: 19.1.7

Backtrace

In this case it is not important, the assertion fails

[paolobarbolini]

POSIX requires that opening a file with the O_APPEND flag should have no affect on the location at which pwrite() writes data. However, on Linux, if a file is opened with O_APPEND, pwrite() appends data to the end of the file, regardless of the value of offset.

This seems to be a know bug. See the Bugs section https://linux.die.net/man/2/pwrite64

[the8472]

Btw, die.net hasn't been updated in a long time, use man7.org or distro documentation for manpages.

Linux 6.9 introduced RWF_NOAPPEND as a flag to say "no really, do a positioned write even on O_APPEND".

https://man7.org/linux/man-pages/man2/pwritev2.2.html

[the8472]

The flags aren't in libc yet, rust-lang/libc#4452