netdev - Re: [REGRESSION] 5.15.88 and onwards no longer return EADDRINUSE from bind

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20230213164455.36911-1-kuniyu@amazon.com>
Date:   Mon, 13 Feb 2023 08:44:55 -0800
From:   Kuniyuki Iwashima <kuniyu@...zon.com>
To:     <w@....eu>
CC:     <gregkh@...uxfoundation.org>, <netdev@...r.kernel.org>,
        <regressions@...ts.linux.dev>, <stable@...r.kernel.org>,
        <winter@...ter.cafe>, <kuniyu@...zon.com>
Subject: Re: [REGRESSION] 5.15.88 and onwards no longer return EADDRINUSE from bind

From:   Willy Tarreau <w@....eu>
Date:   Mon, 13 Feb 2023 08:52:34 +0100
> Hi Greg,
> 
> On Mon, Feb 13, 2023 at 08:25:34AM +0100, Greg KH wrote:
> > On Mon, Feb 13, 2023 at 05:27:03AM +0100, Willy Tarreau wrote:
> > > Hi,
> > > 
> > > [CCed netdev]
> > > 
> > > On Sun, Feb 12, 2023 at 10:38:40PM -0500, Winter wrote:
> > > > Hi all,
> > > > 
> > > > I'm facing the same issue as
> > > > https://lore.kernel.org/stable/CAFsF8vL4CGFzWMb38_XviiEgxoKX0GYup=JiUFXUOmagdk9CRg@mail.gmail.com/,
> > > > but on 5.15. I've bisected it across releases to 5.15.88, and can reproduce
> > > > on 5.15.93.
> > > > 
> > > > However, I cannot seem to find the identified problematic commit in the 5.15
> > > > branch, so I'm unsure if this is a different issue or not.
> > > > 
> > > > There's a few ways to reproduce this issue, but the one I've been using is
> > > > running libuv's (https://github.com/libuv/libuv) tests, specifically tests
> > > > 271 and 277.
> > > 
> > > >From the linked patch:
> > > 
> > >   https://lore.kernel.org/stable/20221228144337.512799851@linuxfoundation.org/
> > 
> > But that commit only ended up in 6.0.y, not 5.15, so how is this an
> > issue in 5.15.y?
> 
> Hmmm I plead -ENOCOFFEE on my side, I hadn't notice the "can't find the
> problematic commit", you're right indeed.
> 
> However if the issue happened in 5.15.88, the only part touching the
> network listening area is this one which may introduce an EINVAL on
> one listening path, but that seems unrelated to me given that it's
> only for ULP that libuv doesn't seem to be using:
> 
>   dadd0dcaa67d ("net/ulp: prevent ULP without clone op from entering the LISTEN status")

This commit accidentally backports a part of 7a7160edf1bf ("net: Return
errno in sk->sk_prot->get_port().") and removed err = -EADDRINUSE in
inet_csk_listen_start().  Then, listen() will return 0 even if ->get_port()
actually fails and returns 1.

I can send a small revert or a whole backport, but which is preferable ?
The original patch is not for stable, but it will make future backports
easy.

Thanks,
Kuniyuki