[Bug 264208] make install[kernel|world] from read only /usr/obj throws permission denied warnings.

From: <bugzilla-noreply_at_freebsd.org>
Date: Sat, 23 May 2026 20:19:48 UTC
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=264208

--- Comment #5 from Bruce M Simpson <bms@FreeBSD.org> ---
I ran into something like this last night across both NFSv3 and NFSv4. Just a
reminder that MAKE_OBJDIR_CHECK_WRITABLE is a make(1) environment variable, not
a share/mk variable, though it arguably belongs in /etc/src-env.conf.

Does FreeBSD bmake have a problem when attempting to read Makefiles or included
rule fragments, and not logging EPERM to stderr appropriately? I had to track
that down manually when rules & targets were mysteriously missing from build.

TL;DR Root Cause Analysis: Doing stupid stuff in Git, i.e. core.filemode=false,
to workaround the Win11 9P IFS driver being stupid about POSIX file mode bits.
So, indirectly, in my case, it was Microsoft. The reason I probably never saw
the problem until now is because it was only enabled on my main branch
checkout, NOT my worktrees, which is where I was normally building and testing
from.
This change was made to appease SourceTree across 9P. core.sharedrepository=all
 is probably needed for me from now on as it also affected CGit.

pin:/usr/src % git config set --local core.filemode true

We come full circle back to the length description in this thread:
https://stackoverflow.com/questions/69339954/what-are-the-consequences-of-git-config-core-filemode-false

At /usr/src, git ls-files -s | less confirmed the modes were sane.
But because of core.filemode=false they were previously not used.

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