From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: [PATCH] rhashtable: don't attempt to grow when at max_size Date: Thu, 23 Apr 2015 11:59:19 -0400 (EDT) Message-ID: <20150423.115919.1353583175267783165.davem@davemloft.net> References: <1429799923-28122-1-git-send-email-johannes@sipsolutions.net> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, kaber@trash.net, tgraf@suug.ch, johannes.berg@intel.com, herbert@gondor.apana.org.au To: johannes@sipsolutions.net Return-path: Received: from shards.monkeyblade.net ([149.20.54.216]:42190 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S966289AbbDWP7W (ORCPT ); Thu, 23 Apr 2015 11:59:22 -0400 In-Reply-To: <1429799923-28122-1-git-send-email-johannes@sipsolutions.net> Sender: netdev-owner@vger.kernel.org List-ID: From: Johannes Berg Date: Thu, 23 Apr 2015 16:38:43 +0200 > From: Johannes Berg > > The conversion of mac80211's station table to rhashtable had a bug > that I found by accident in code review, that hadn't been found as > rhashtable apparently managed to have a maximum hash chain length > of one (!) in all our testing. > > In order to test the bug and verify the fix I set my rhashtable's > max_size very low (4) in order to force getting hash collisions. > > At that point, rhashtable WARNed in rhashtable_insert_rehash() but > didn't actually reject the hash table insertion. This caused it to > lose insertions - my master list of stations would have 9 entries, > but the rhashtable only had 5. This may warrant a deeper look, but > that WARN_ON() just shouldn't happen. > > Fix this by not returning true from rht_grow_above_100() when the > rhashtable's max_size has been reached - in this case the user is > explicitly configuring it to be at most that big, so even if it's > now above 100% it shouldn't attempt to resize. > > This fixes the "lost insertion" issue and consequently allows my > code to display its error (and verify my fix for it.) > > Signed-off-by: Johannes Berg It looks fine to me, but I'll let Herbert and Thomas review this.