Date: Wed, 9 Feb 2022 15:11:29 -0800
From: Andres Freund <andres@anarazel.de>
To: Justin Pryzby <pryzby@telsasoft.com>
Cc: pgsql-hackers@postgresql.org, Robert Haas <robertmhaas@gmail.com>,
	Thomas Munro <thomas.munro@gmail.com>
Subject: Re: wrong fds used for refilenodes after pg_upgrade relfilenode
 changes
Message-ID: <20220209231129.g2hbb624nn33md3k@alap3.anarazel.de>
References: <20220209220004.kb3dgtn2x2k2gtdm@alap3.anarazel.de>
 <20220209224230.GS31460@telsasoft.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220209224230.GS31460@telsasoft.com>
Archived-At: <https://www.postgresql.org/message-id/20220209231129.g2hbb624nn33md3k%40alap3.anarazel.de>
Precedence: bulk

Hi,

On 2022-02-09 16:42:30 -0600, Justin Pryzby wrote:
> On Wed, Feb 09, 2022 at 02:00:04PM -0800, Andres Freund wrote:
> > On linux we can do so by a) checking if readlink(/proc/self/fd/$fd) points to
> > a filename ending in " (deleted)", b) doing fstat(fd) and checking if st_nlink
> > == 0.
> 
> You could also stat() the file in proc/self/fd/N and compare st_ino.  It
> "looks" like a symlink (in which case that wouldn't work) but it's actually a
> Very Special File.  You can even recover deleted, still-opened files that way..

Yea, the readlink() thing above relies on it being a /proc/self/fd/$fd being a
"Very Special File".

In most places we'd not have convenient access to a inode / filename to
compare it to. I don't think there's any additional information we could gain
anyway, compared to looking at st_nlink == 0 and then doing a readlink() to
get the filename?


> PS. I didn't know pg_upgrade knew about Reply-To ;)

Ugh, formatting fail...

Greetings,

Andres Freund