Re: libpq OpenSSL and multithreading

From: Peter Eisentraut <peter(at)eisentraut(dot)org>
To: Daniel Gustafsson <daniel(at)yesql(dot)se>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: libpq OpenSSL and multithreading
Date: 2025-07-02 11:44:55
Message-ID: 27dd8d8c-174c-4f66-b404-1e5d83c3e075@eisentraut.org
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 27.06.25 19:24, Daniel Gustafsson wrote:
> The OpenSSL code in libpq have two issues for multithreading: the verify_cb
> callback use a global variable to pass back error detail state and there is one
> use of strerror().

Slightly misleading title: This is actually about the *backend* libpq code.

> The attached fixes both, with no functional change, in order to pave the way
> for multithreading:
>
> * Rather than using a global variable the callback use a new member in the
> Port struct for passing the string, and the Port struct is in turn passed as
> private data in the SSL object

Couldn't this also be done by making that global variable thread-local?
But getting rid of it is even nicer.

> * The strerror call is replaced with a strerror_r call using the already
> existing errorbuffer

This one was already discussed some time ago at
<https://www.postgresql.org/message-id/flat/daa87d79-c044-46c4-8458-8d77241ed7b0%40eisentraut.org>:

> But the bigger issue is that the use of a static buffer makes
> this not thread-safe, so having it use strerror_r to fill that
> buffer is just putting lipstick on a pig.

It looks like your patch doesn't address that?

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Ashutosh Bapat 2025-07-02 12:20:18 Re: Using failover slots for PG-non_PG logical replication
Previous Message Ashutosh Bapat 2025-07-02 11:37:28 Re: Adding basic NUMA awareness