From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pg1-f200.google.com (mail-pg1-f200.google.com [209.85.215.200]) by kanga.kvack.org (Postfix) with ESMTP id 041596B7119 for ; Tue, 4 Dec 2018 17:48:45 -0500 (EST) Received: by mail-pg1-f200.google.com with SMTP id g188so9866407pgc.22 for ; Tue, 04 Dec 2018 14:48:44 -0800 (PST) Received: from mail-sor-f65.google.com (mail-sor-f65.google.com. [209.85.220.65]) by mx.google.com with SMTPS id o19sor24439096pll.44.2018.12.04.14.48.43 for (Google Transport Security); Tue, 04 Dec 2018 14:48:43 -0800 (PST) Content-Type: text/plain; charset=utf-8 Mime-Version: 1.0 (Mac OS X Mail 12.1 \(3445.101.1\)) Subject: Re: [PATCH 1/2] vmalloc: New flag for flush before releasing pages From: Nadav Amit In-Reply-To: Date: Tue, 4 Dec 2018 14:48:40 -0800 Content-Transfer-Encoding: quoted-printable Message-Id: <20CC2F71-308D-42E2-8C54-F64D7CC3863F@gmail.com> References: <20181128000754.18056-1-rick.p.edgecombe@intel.com> <20181128000754.18056-2-rick.p.edgecombe@intel.com> <4883FED1-D0EC-41B0-A90F-1A697756D41D@gmail.com> <08141F66-F3E6-4CC5-AF91-1ED5F101A54C@gmail.com> Sender: owner-linux-mm@kvack.org List-ID: To: Andy Lutomirski Cc: Rick Edgecombe , Andrew Morton , Will Deacon , Linux-MM , LKML , Kernel Hardening , "Naveen N . Rao" , Anil S Keshavamurthy , "David S. Miller" , Masami Hiramatsu , Steven Rostedt , Ingo Molnar , Alexei Starovoitov , Daniel Borkmann , jeyu@kernel.org, Network Development , Ard Biesheuvel , Jann Horn , Kristen Carlson Accardi , Dave Hansen , "Dock, Deneen T" , Peter Zijlstra > On Dec 4, 2018, at 11:48 AM, Andy Lutomirski wrote: >=20 > On Tue, Dec 4, 2018 at 11:45 AM Nadav Amit = wrote: >>> On Dec 4, 2018, at 10:56 AM, Andy Lutomirski = wrote: >>>=20 >>> On Mon, Dec 3, 2018 at 5:43 PM Nadav Amit = wrote: >>>>> On Nov 27, 2018, at 4:07 PM, Rick Edgecombe = wrote: >>>>>=20 >>>>> Since vfree will lazily flush the TLB, but not lazily free the = underlying pages, >>>>> it often leaves stale TLB entries to freed pages that could get = re-used. This is >>>>> undesirable for cases where the memory being freed has special = permissions such >>>>> as executable. >>>>=20 >>>> So I am trying to finish my patch-set for preventing transient W+X = mappings >>>> from taking space, by handling kprobes & ftrace that I missed = (thanks again for >>>> pointing it out). >>>>=20 >>>> But all of the sudden, I don=E2=80=99t understand why we have the = problem that this >>>> (your) patch-set deals with at all. We already change the mappings = to make >>>> the memory writable before freeing the memory, so why can=E2=80=99t = we make it >>>> non-executable at the same time? Actually, why do we make the = module memory, >>>> including its data executable before freeing it??? >>>=20 >>> All the code you're looking at is IMO a very awkward and possibly >>> incorrect of doing what's actually necessary: putting the direct map >>> the way it wants to be. >>>=20 >>> Can't we shove this entirely mess into vunmap? Have a flag (as part >>> of vmalloc like in Rick's patch or as a flag passed to a vfree = variant >>> directly) that makes the vunmap code that frees the underlying pages >>> also reset their permissions? >>>=20 >>> Right now, we muck with set_memory_rw() and set_memory_nx(), which >>> both have very awkward (and inconsistent with each other!) semantics >>> when called on vmalloc memory. And they have their own flushes, = which >>> is inefficient. Maybe the right solution is for vunmap to remove = the >>> vmap area PTEs, call into a function like set_memory_rw() that = resets >>> the direct maps to their default permissions *without* flushing, and >>> then to do a single flush for everything. Or, even better, to cause >>> the change_page_attr code to do the flush and also to flush the vmap >>> area all at once so that very small free operations can flush single >>> pages instead of flushing globally. >>=20 >> Thanks for the explanation. I read it just after I realized that = indeed the >> whole purpose of this code is to get cpa_process_alias() >> update the corresponding direct mapping. >>=20 >> This thing (pageattr.c) indeed seems over-engineered and very = unintuitive. >> Right now I have a list of patch-sets that I owe, so I don=E2=80=99t = have the time >> to deal with it. >>=20 >> But, I still think that disable_ro_nx() should not call = set_memory_x(). >> IIUC, this breaks W+X of the direct-mapping which correspond with the = module >> memory. Does it ever stop being W+X?? I=E2=80=99ll have another look. >=20 > Dunno. I did once chase down a bug where some memory got freed while > it was still read-only, and the results were hilarious and hard to > debug, since the explosion happened long after the buggy code > finished. This piece of code causes me pain and misery. So, it turns out that the direct map is *not* changed if you just change the NX-bit. See change_page_attr_set_clr(): /* No alias checking for _NX bit modifications */ checkalias =3D (pgprot_val(mask_set) | pgprot_val(mask_clr)) !=3D = _PAGE_NX; How many levels of abstraction are broken in the way? What would happen if somebody tries to change the NX-bit and some other bit in the PTE? Luckily, I don=E2=80=99t think someone does=E2=80=A6 at least for now. So, again, I think the change I proposed makes sense. nios2 does not = have set_memory_x() and it will not be affected. [ I can add a comment, although I don=E2=80=99t have know if nios2 has = an NX bit, and I don=E2=80=99t find any code that defines PTEs. Actually where is = pte_present() of nios2 being defined? Whatever. ]