netdev - Re: [RFC 1/2] net/tls: don't clear socket error if strparser aborted

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] [day] [month] [year] [list]

Message-Id: <20191007.145817.1363255118392494597.davem@davemloft.net>
Date:   Mon, 07 Oct 2019 14:58:17 +0200 (CEST)
From:   David Miller <davem@...emloft.net>
To:     jakub.kicinski@...ronome.com
Cc:     netdev@...r.kernel.org, edumazet@...gle.com, willemb@...gle.com,
        oss-drivers@...ronome.com, davejwatson@...com, borisp@...lanox.com,
        aviadye@...lanox.com, john.fastabend@...il.com,
        daniel@...earbox.net, simon.horman@...ronome.com
Subject: Re: [RFC 1/2] net/tls: don't clear socket error if strparser
 aborted

From: Jakub Kicinski <jakub.kicinski@...ronome.com>
Date: Sun,  6 Oct 2019 20:53:22 -0700

> If strparser encounters an error it reports it on the socket and
> stops any further processing. TLS RX will currently pick up that
> error code with sock_error(), and report it to user space once.
> Subsequent read calls will block indefinitely.
> 
> Since the error condition is not cleared and processing is not
> restarted it seems more correct to keep returning the error
> rather than sleeping.
> 
> Signed-off-by: Jakub Kicinski <jakub.kicinski@...ronome.com>
> Reviewed-by: Simon Horman <simon.horman@...ronome.com>

So I guess this is all about what the usual socket error code
semantics work, which is report-and-clear.

States which should signal errors are always checked in the
various code socket call paths and if necessary the socket
error is resignalled.

>From what I'm seeing here, the issue is that the strparser stopped
state isn't resampled at the appropriate spot.  So I would rather
see that fixed rather than having socket errors become level
triggered in this special case.