libbsnmp: fix local client socket creation

[glebius]

Summary

Revert "bsnmp: Use mkstemp when creating clients local socket"

The use of mktemp() was intentional here. We don't want a regular file to
be created. If created, it guarantees that following bind(2) would always
fail with EADDRINUSE and thus open_client_local() always fails.

This reverts commit 81e0e7b9e36d6a25b3af6482811318e085537d2f.

[glebius]

I am really puzzled how D34550 was created and how did it pass your review :)

[des]

I assume the original concern here was that mktemp() is subject to TOCTTOU races. Thus, it would be better to use mkdtemp() to create a unique temporary directory, and create our socket inside that directory.

[glebius]

In D51031#1164827, @des wrote:

I assume the original concern here was that mktemp() is subject to TOCTTOU races. Thus, it would be better to use mkdtemp() to create a unique temporary directory, and create our socket inside that directory.

Let's just revert this to original working state. I have upcoming review on this function that skips binding at all. This binding is not needed. The patch would still allow to use a bound socket for SNMP client, but that would be optional. The path name would be either specified by caller or no path name at all. This makes mktemp/mkstemp point moot.

[rew]

In D51031#1164660, @glebius wrote:

I am really puzzled how D34550 was created and how did it pass your review :)

same way as this one, LGTM :)

side note, I don't recall the issue NetApp was having and AFAIK that information is not public.

but I trust your vision and if this change becomes an issue for NetApp, they can adapt.

[glebius]

Note: 4 more reviews stacked on top of this one, if anyone interested.

[harti]

Find for me.