netdev - Re: [PATCH] net: netpoll: ensure skb_pool list is always initialized

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: <20250103-passionate-mighty-seagull-29cebe@leitao>
Date: Fri, 3 Jan 2025 05:24:22 -0800
From: Breno Leitao <leitao@...ian.org>
To: Jakub Kicinski <kuba@...nel.org>, jsperbeck@...gle.com
Cc: John Sperbeck <jsperbeck@...gle.com>,
	"David S . Miller" <davem@...emloft.net>,
	Eric Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>,
	netdev@...r.kernel.org, Simon Horman <horms@...nel.org>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] net: netpoll: ensure skb_pool list is always initialized

On Mon, Dec 30, 2024 at 05:57:07PM -0800, Jakub Kicinski wrote:
> On Sat, 21 Dec 2024 17:23:34 -0800 John Sperbeck wrote:
> > Move the skb_pool list initialization into __netpoll_setup().  Also,
> > have netpoll_setup() call this before allocating its initial pool of
> > packets.
> > 
> > Fixes: 6c59f16f1770 ("net: netpoll: flush skb pool during cleanup")
> 
> The fixes tag seems to be off by one? Wasn't the problem was introduced
> by commit 221a9c1df790 ("net: netpoll: Individualize the skb pool") ?

You are correct. The regression was caused by 221a9c1df790 ("net:
netpoll: Individualize the skb pool"), when I mistakenly moved
skb_queue_head_init() out of netpoll_init() into netpoll_setup().

> Since __netpoll_setup() can be called by other drivers, shouldn't 
> we move refill in there? Since the pool is per np?

I'd say so.

It is not a big deal to have it in netpoll_setup(), since find_skb()
will refill the SKBs in the very first message being transmitted
(which will undesirably delay the very first netconsole TX).

On the other side, having it in refill_skbs() in the __netpoll_setup(),
as Jakub suggested, will avoid this extra delay in the very first TX,
and make the workflow cohesive.

Anyway, thanks John for spotting this regression and working in the fix,
--breno