Return-Path: MIME-Version: 1.0 In-Reply-To: <20160623064137.GA8958@gondor.apana.org.au> References: <20160622004214.GA27134@gondor.apana.org.au> <20160623034825.GB7819@gondor.apana.org.au> <20160623064137.GA8958@gondor.apana.org.au> From: Andy Lutomirski Date: Thu, 23 Jun 2016 15:11:00 -0700 Message-ID: Subject: Re: Doing crypto in small stack buffers (bluetooth vs vmalloc-stack crash, etc) To: Herbert Xu Cc: Andy Lutomirski , linux-bluetooth@vger.kernel.org, Johan Hedberg , Gustavo Padovan , Marcel Holtmann , linux-crypto@vger.kernel.org, "David S. Miller" , "linux-kernel@vger.kernel.org" , Linus Torvalds , Network Development Content-Type: text/plain; charset=UTF-8 List-ID: On Wed, Jun 22, 2016 at 11:41 PM, Herbert Xu wrote: > On Thu, Jun 23, 2016 at 11:48:25AM +0800, Herbert Xu wrote: >> >> No we never had such an API in the kernel. However, I see that >> rxkad does some pretty silly things and we should be able to avoid >> using the stack in pretty much all cases. Let me try to come up with >> something. > > Here it is: > > ---8<--- > Subject: rxrpc: Avoid using stack memory in SG lists in rxkad Looks reasonable to me. Unless anyone tells me otherwise, my plan is to queue it in my virtually-mapped stack series and to ask Ingo to apply it via -tip. If it went in via the networking tree, that would work as well, but it would introduce a bisectability problem. Thanks! --Andy