gai_strerror() is not thread-safe on Windows

Thomas Munro <thomas.munro@gmail.com>

From: Thomas Munro <thomas.munro@gmail.com>
To: pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2023-12-04T03:21:24Z
Lists: pgsql-hackers

Attachments

Hi,

Commit 5579388d removed a bunch of dead code, formerly needed for old
systems that lacked getaddrinfo() in the early days of IPv6.  We
already used the system getaddrinfo() via either configure-time tests
(Unix) or runtime tests (Windows using attempt-to-find-with-dlsym that
always succeeded on modern systems), so no modern system needed the
fallback code, except for one small detail:

getaddrinfo() has a companion function to spit out human readable
error messages, and although Windows has that too, it's not thread
safe[1].  libpq shouldn't call it, or else an unlucky multi-threaded
program might see an error message messed up by another thread.

Here's a patch to put that bit back.  It's simpler than before: the
original replacement had a bunch of #ifdefs for various historical
reasons, but now we can just handle the 8 documented EAI errors on
Windows.

Noticed while wondering why the list of symbols reported in bug #18219
didn't include gai_strerrorA.  That turned out to be because it is
static inline in ws2tcpip.h, and its definition set alarm bells
ringing.  Avoid.

[1] https://learn.microsoft.com/en-us/windows/win32/api/ws2tcpip/nf-ws2tcpip-getaddrinfo

Commits

  1. Fix gai_strerror() thread-safety on Windows.

  2. Update comment, generation mem contexts have a "keeper" block

  3. Remove replacement code for getaddrinfo.