Received-SPF: pass (google.com: best guess record for domain of linux-wireless-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <00161d6069cda67dbd8b918dd987e01dc1a3dab3.camel@sipsolutions.net>
Subject: Re: [RFCv2 1/4] nl80211: Fix broken non-split wiphy dumps
From:   Johannes Berg <johannes@sipsolutions.net>
To:     Denis Kenzior <denkenz@gmail.com>, linux-wireless@vger.kernel.org
Date:   Fri, 30 Aug 2019 11:10:13 +0200
In-Reply-To: <f7c98da178677cbb0cad3568f4ea4ab85171edd8.camel@sipsolutions.net> (sfid-20190830_110356_499003_5F22B3F6)
References: <20190816192703.12445-1-denkenz@gmail.com>
         (sfid-20190816_212729_636741_39C4CEB6) <f7c98da178677cbb0cad3568f4ea4ab85171edd8.camel@sipsolutions.net>
         (sfid-20190830_110356_499003_5F22B3F6)
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.30.5 (3.30.5-1.fc29) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-wireless-owner@vger.kernel.org
Precedence: bulk

On Fri, 2019-08-30 at 11:03 +0200, Johannes Berg wrote:
> On Fri, 2019-08-16 at 14:27 -0500, Denis Kenzior wrote:
> > If a (legacy) client requested a wiphy dump but did not provide the
> > NL80211_ATTR_SPLIT_WIPHY_DUMP attribute, the dump was supposed to be
> > composed of purely non-split NEW_WIPHY messages, with 1 wiphy per
> > message.  At least this was the intent after commit:
> > 3713b4e364ef ("nl80211: allow splitting wiphy information in dumps")
> > 
> > However, in reality the non-split dumps were broken very shortly after.
> > Perhaps around commit:
> > fe1abafd942f ("nl80211: re-add channel width and extended capa advertising")
> 
> Fun. I guess we updated all userspace quickly enough to not actually
> have any issues there. As far as I remember, nobody ever complained, so
> I guess people just updated their userspace.

Actually, going back in time to the code there (e.g. iw and hostap), it
seems that it quite possibly never was a userspace issue, just an issue
with netlink allocating a 4k SKB by default for dumps.

Even then, libnl would've defaulted to a 16k recvmsg() buffer size, and
we didn't override that anywhere.

So more likely, this was all fixed by kernel 9063e21fb026 ("netlink:
autosize skb lengthes") about a year after we ran into the problem.

johannes