posix_fallocate fails on ZFS+FreeBSD

[P-E-Meunier]

This error never shows up on Linux because glibc tries to emulate fallocate. However, it seems this call could be emulated with truncate on FreeBSD.

[ryanavella]

@P-E-Meunier by "emulate," do you mean a call that avoids unnecessary truncating? i.e. something like this?

#[cfg(target_os = "freebsd")]
pub fn allocate(file: &File, len: u64) -> Result<()> {
    let fd = file.as_raw_fd();
    let len = len.try_into().unwrap();

    let ret = unsafe {
        let mut stat: libc::stat = mem::zeroed();
        let len_new = if libc::fstat(fd, std::ptr::addr_of_mut!(stat)) == 0 {
            stat.st_size.max(len)
        } else {
            len
        };
        libc::ftruncate(file.as_raw_fd(), len_new)
    };

    if ret == 0 { Ok(()) } else { Err(Error::last_os_error()) }
}

[P-E-Meunier]

I believe so, but it was quite some time ago, I don't remember exactly what the issue was. It was also reported to me indirectly.
My use of fs2 , when running on ZFS, is also hard to understand, since I'm doing copy-on-write (in the sanakirja crate) on top of a copy-on-write filesystem.

[rtyler]

I have a current version of FreeBSD and ZFS running with this crate. posix_fallocate will fail with EINVAL.

diff --git a/src/unix.rs b/src/unix.rs
index 5c2f4f2..79f7595 100644
--- a/src/unix.rs
+++ b/src/unix.rs
@@ -96,7 +96,6 @@ pub fn allocated_size(file: &File) -> Result<u64> {
 }

 #[cfg(any(target_os = "linux",
-          target_os = "freebsd",
           target_os = "android",
           target_os = "emscripten",
           target_os = "nacl"))]
@@ -138,6 +137,7 @@ pub fn allocate(file: &File, len: u64) -> Result<()> {

 #[cfg(any(target_os = "openbsd",
           target_os = "netbsd",
+          target_os = "freebsd",
           target_os = "dragonfly",
           target_os = "solaris",
           target_os = "haiku"))]

A suitable workaround seems to be to make FreeBSD use the same allocate implementation as the other BSDs, but I'm not yet sure if this is a Good Idea ™️

I tested another approach which would use ftruncate(2) on FreeBSD. Either really works for the higher level application I am using here which relies on this functionality.

I am not sure whether @danburkert has any preference to the approach taken here.

[ryanavella]

@rtyler I'm not sure if fs2 is maintained any more, the last commit was >5 years ago and several important PR's have been left open for nearly as long.

You may consider fs4 as an alternative. It is a fork of fs2 that is now actively maintained, and they accepted my PR to fix the FreeBSD+ZFS issue you are describing.

[rtyler]

@ryanavella fs4 does bubble up the EINVAL but still does not provider a suitable allocate implementaiton, but I will take that up with that maintainer instead 😄

[ryanavella]

I think the reason FreeBSD returns EINVAL with posix_fallocate is that fallocate doesn't really make much sense on a copy-on-write filesystem.

My workaround is to match on fs4::allocate and ignore the returned error if it is EINVAL, instead of blindly bubbling it up with the ? operator. In a lot of cases fs4::allocate will be paired with a ftruncate anyways, so there isn't much to do if EINVAL is returned other than continue.