On 24/08/2024 11:42, Peter Eisentraut wrote:
Here is a patch to replace a getpwuid() call in the backend, for
thread-safety.
This is AFAICT the only call in the getpw*() family that needs to be
dealt with.
(There is also a getgrnam() call, but that is called very early in the
postmaster, before multiprocessing, so we can leave that as is.)
The solution here is actually quite attractive: We can replace the
getpwuid() call by the existing wrapper function pg_get_user_name(),
which internally uses getpwuid_r(). This also makes the code a bit
simpler. The same function is already used in libpq for a purpose that
mirrors the backend use, so it's also nicer to use the same function for
consistency.
Makes sense.
The temporary buffers are a bit funky. pg_get_user_name() internally
uses a BUFSIZE-sized area to hold the result of getpwuid_r(). If the
pg_get_user_name() caller passes a buffer smaller than BUFSIZE, the user
id might get truncated. I don't think that's a concern on any real-world
system, and the callers do pass a large-enough buffer so truncation
can't happen. At a minimum, it would be good to add a comment to
pg_get_user_name() along the lines of "if 'buflen' is smaller than
BUFSIZE, the result might be truncated".
Come to think of it, the pg_get_user_name() function is just a thin
wrapper around getpwuid_r(). It doesn't provide a lot of value. Perhaps
we should remove pg_get_user_name() and pg_get_user_home_dir()
altogether and call getpwuid_r() directly.
But no objection to committing this as it is, either.
--
Heikki Linnakangas
Neon (https://neon.tech)
