enhancement: better alpine compatibility

[semarie]

The following PR fixes few build errors while running make and make test on alpine linux.

It mostly adds compatibility declaration on C stuff (unixpwd and forkexecd compoments).

The more important changes is regarding getpwent_r and getspent_r usage in unixpwd.
As it, these functions doesn't exists in musl-libc. So I switched back to more standard variants (getpwent and getspent). There are explicitly not reentrants, but getpwent_r and getspent_r weren't fully reentrants too (due to FILE sharing between threads), so I assume it doesn't introduce regression from this point of vue.

With the PR, I am able to run ./configure && make && make test without errors (tested on alpine edge).

[freddy77]

If you don't add some locking amongst getpwent calls you could end up with one thread giving result from a function called in another thread. This can be a security concern.

[last-genius]

Some of these changes are small and reasonable (portability is always good!), but I'm not sure how much these are going to help you work with xapi on alpine. xs-opam will require quite a few packages that might not be available on alpine, and I'm sure there's more work to get some of our dependencies to build with musl libc too. Approximating the build environment might be the easiest option - most of us use Fedora or similar (I'm personally using a Fedora distrobox on Arch Linux).

UPD: I'd love to be proven wrong - if these are the only blockers for you currently, that'd be great

[andyhhp]

Commits 1, 4 and 5 are trivial and fine. The use of non-recursive functions needs more careful review

[semarie]

@freddy77 I agree that concurrent threads could lead to problems. but the current code isn't different: concurrent calls to getpwent_r will result in similar problems.

The getpwent_rman page documents it as:

The function getpwent_r() is not really reentrant since it shares
the reading position in the stream with all other threads.

getpwent_r only avoids using static memory for the returned result, but the source (the passwd file opened) is still shared between threads.

so if reentrance is mandated the current code is already wrong.

@last-genius the sole problem I had with dependencies is dlm. I had to patch xs-opam to avoid the explicit reject of alpine. But the library built fine from source (but only the library, tools like dlm_controld needs additional stuff)

[freddy77]

Why not using fopen/fclose and fgetpwent_r/fgetspent_r where available instead? There are also lckpwdf/ulckpwdf although musl implementation for these is empty...

[freddy77]

This can be dangerous. Potentially for an error we are truncating /etc/passwd file instead of returning error.

[semarie]

yes, fgetpwent_r and fgetspent_r seems to be the way for reentrance (but not for musl).

For now, I will drop the changes for getpwent and getspent usage, so only uncontroversal changes will be in the PR. Next I will look for doing another PR for fgetpwent_r and fgetspent_r usage. And last go back to musl compat case to see what could be done.

[last-genius]

I believe Andy used to have the maintainer status and his approvals were green, did anything change?

[andyhhp]

I've never been a maintainer in Xapi.