[Bug 272120] touch(1) -r modifies file's birthtime

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=272120

--- Comment #5 from Alexey Dokuchaev <[email protected]> ---
(In reply to Simon Tettmar from comment #4)
> [VFS] deliberately sets va_birthtime to the requested mtime when the
> requested mtime is earlier than the current birthtime.
I've been thinking about this since when Mark had made his analysis in comment
#2, and could not come to peace with the rationale behind existing behavior.

> I cannot convince myself that having mtimes prior to birthtimes can ever
> be a good thing and so if no one objects I will look at updating touch(1)
> documentation to explain this behaviour.
I don't quite see why these timestamps should be interconnected at all.  Inode
birth time should be constant throughout the inode's life, due to the mere
definition of birth.  Non-constant birth time is semantically wrong. 
Modification time applies to file contents which has nothing to do with its
container's inode.

I guess I see the value of better describing current (bogus) behavior to reduce
user confusion, but I'll probably start patching my kernels to stop messing
with btimes after they were assigned.

-- 
You are receiving this mail because:
You are the assignee for the bug.