[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20220815084313.GA5059@katalix.com>
Date: Mon, 15 Aug 2022 09:43:14 +0100
From: Tom Parkin <tparkin@...alix.com>
To: Jakub Sitnicki <jakub@...udflare.com>
Cc: Jakub Kicinski <kuba@...nel.org>, netdev@...r.kernel.org,
kernel-team@...udflare.com, van fantasy <g1042620637@...il.com>
Subject: Re: [PATCH net] l2tp: Serialize access to sk_user_data with sock lock
On Fri, Aug 12, 2022 at 11:54:43 +0200, Jakub Sitnicki wrote:
> On Thu, Aug 11, 2022 at 10:23 AM -07, Jakub Kicinski wrote:
> > On Wed, 10 Aug 2022 12:28:48 +0200 Jakub Sitnicki wrote:
> >> Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
> >
> > That tag immediately sets off red flags. Please find the commit where
> > to code originates, not where it was last moved.
>
> The code move happened in v2.6.35. There's no point in digging further, IMHO.
At the time of fd558d186df2, sk_user_data was checked/set by the newly
added function l2tp_tunnel_create. The only callpath for
l2tp_tunnel_create was via. pppol2tp_connect which called
l2tp_tunnel_create with lock_sock held (and indeed still does).
I think the addition of the netlink API (which added a new callpath
for l2tp_tunnel_create via. l2tp_nl_cmd_tunnel_create which was *not*
lock_sock-protected) is perhaps the right commit to point to?
309795f4bec2 ("l2tp: Add netlink control API for L2TP")
Download attachment "signature.asc" of type "application/pgp-signature" (489 bytes)
Powered by blists - more mailing lists