Incorrect errno used with %m for backend code

Hi all,

I have been reviewing the use of errno in the backend code when %m is
used in the logs, and found more places than when I looked at improving
the error messages for read() calls which bloat the errno value because
of intermediate system calls. As the problem is separate and should be
back-patched, I have preferred beginning a new thread.

A couple of places use CloseTransientFile without bothering much that
this can overwrite errno. I was tempted in keeping errno saved and kept
if set to a non-zero value, but refrained from doing so as some callers
may rely on the existing behavior, and the attached shows better the
intention.

Attached is a patch with everything I have spotted. Any opinions or
thoughts in getting this back-patched?

Thanks,
--
Michael