From mboxrd@z Thu Jan 1 00:00:00 1970 From: Radim =?utf-8?B?S3LEjW3DocWZ?= Subject: Re: [PATCH v4 2/2] KVM: move vcpu id checking to archs Date: Thu, 21 Apr 2016 19:36:11 +0200 Message-ID: <20160421173611.GB30356@potion> References: <146124809455.32509.15232948272580716135.stgit@bahia.huguette.org> <146124811255.32509.17679765789502091772.stgit@bahia.huguette.org> <20160421160018.GA31953@potion> <20160421184500.6cb5fd8a@bahia.huguette.org> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Paolo Bonzini , james.hogan@imgtec.com, mingo@redhat.com, linux-mips@linux-mips.org, kvm@vger.kernel.org, linux-kernel@vger.kernel.org, David Hildenbrand , qemu-ppc@nongnu.org, Cornelia Huck , Paul Mackerras , David Gibson To: Greg Kurz Return-path: Received: from mx1.redhat.com ([209.132.183.28]:35181 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751851AbcDURgR (ORCPT ); Thu, 21 Apr 2016 13:36:17 -0400 Content-Disposition: inline In-Reply-To: <20160421184500.6cb5fd8a@bahia.huguette.org> Sender: kvm-owner@vger.kernel.org List-ID: 2016-04-21 18:45+0200, Greg Kurz: > On Thu, 21 Apr 2016 18:00:19 +0200 > Radim Kr=C4=8Dm=C3=A1=C5=99 wrote: >> 2016-04-21 16:20+0200, Greg Kurz: >> > Commit 338c7dbadd26 ("KVM: Improve create VCPU parameter (CVE-2013= -4587)") >> > introduced a check to prevent potential kernel memory corruption i= n case >> > the vcpu id is too great. >> >=20 >> > Unfortunately this check assumes vcpu ids grow in sequence with a = common >> > difference of 1, which is wrong: archs are free to use vcpu id as = they fit. >> > For example, QEMU originated vcpu ids for PowerPC cpus running in = boot3s_hv >> > mode, can grow with a common difference of 2, 4 or 8: if KVM_MAX_V= CPUS is >> > 1024, guests may be limited down to 128 vcpus on POWER8. >> >=20 >> > This means the check does not belong here and should be moved to s= ome arch >> > specific function: kvm_arch_vcpu_create() looks like a good candid= ate. >> >=20 >> > ARM and s390 already have such a check. >> >=20 >> > I could not spot any path in the PowerPC or common KVM code where = a vcpu >> > id is used as described in the above commit: I believe PowerPC can= live >> > without this check. >> >=20 >> > In the end, this patch simply moves the check to MIPS and x86. Whi= le here, >> > we also update the documentation to dissociate vcpu ids from the m= aximum >> > number of vcpus per virtual machine. >> >=20 >> > Acked-by: James Hogan >> > Acked-by: Cornelia Huck >> > Signed-off-by: Greg Kurz >> > --- >> > v4: - updated subject for more clarity on what the patch does >> > - added James's and Connie's A-b tags >> > - updated documentation >> >=20 >> > Documentation/virtual/kvm/api.txt | 7 +++---- >> > arch/mips/kvm/mips.c | 7 ++++++- >> > arch/x86/kvm/x86.c | 3 +++ >> > virt/kvm/kvm_main.c | 3 --- >> > 4 files changed, 12 insertions(+), 8 deletions(-) >> >=20 >> > diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/vir= tual/kvm/api.txt >> > index 4d0542c5206b..486a1d783b82 100644 >> > --- a/Documentation/virtual/kvm/api.txt >> > +++ b/Documentation/virtual/kvm/api.txt >> > @@ -199,11 +199,10 @@ Type: vm ioctl >> > Parameters: vcpu id (apic id on x86) >> > Returns: vcpu fd on success, -1 on error >> > =20 >> > -This API adds a vcpu to a virtual machine. The vcpu id is a smal= l integer >> > -in the range [0, max_vcpus). >> > +This API adds a vcpu to a virtual machine. The vcpu id is a posi= tive integer. =20 >>=20 >> Userspace won't be able to tell if KVM_CREATE_VCPU failed because it >> provided too high vcpu_id to an old KVM or because new KVM failed in >> other areas. Not a huge problem (because I expect that userspace wi= ll >> die on both), but a new KVM_CAP would be able to disambiguate it. >>=20 >> Toggleable capability doesn't seem necessary and only PowerPC change= s, >> so the capability could be arch specific ... I think that a generic = one >> makes more sense, though. >> >=20 > I'm not sure userspace can disambiguate all the cases where KVM_CREAT= E_VCPU > returns EINVAL already... and, FWIW, QEMU simply exits if it gets an = error. Yes, userspace cannot disambiguate, but would have the option of not doing something that is destined to fail, like with KVM_CAP_MAX_VCPU. > So I understand your concern but would we have a user for this ? I think so, new userspace on pre-patch KVM is the most likely one. Userspace cannot tell that KVM doesn't support the extension and behaving like on patched KVM would result in a failure with cryptic error message, because KVM only returns EINVAL. Btw. PowerPC QEMU tries vcpu_id >=3D KVM_MAX_VCPUS and fails, instead o= f recognizing that the user wanted too much? >> Userspace also doesn't know the vcpu id limit anymore, and it might >> care. What do you think about returning the arch-specific limit (or= the >> highest positive integer) as int in KVM_CAP_MAX_VCPU_ID? >>=20 >=20 > This is partly true: only arch agnostic code would be lost. >=20 > Moreover this is a problem for powerpc only at the moment and userspa= ce code > can compute the vcpu_id limit out of KVM_CAP_MAX_VCPUS and KVM_CAP_PP= C_SMT. How would that work on KVM without this patch? > For other architectures, it is simply KVM_MAX_VCPUS. (Other architectures would not implement the capability.) >> I think this would also clarify the connection between VCPU limit an= d >> VCPU_ID limit. Or is a boolean cap better? >>=20 >=20 > Well, I'm not fan of adding a generic API to handle a corner case... I don't like it either, but I think that introducing the capability is worth avoided problems. > = maybe later > if we have other scenarios where vcpu ids need to cross the limit ? x86 is going to have that soon too -- vcpu_id will be able to range fro= m 0 to 2^32-1 (or 2^31), but MAX_CPUS related data structures probably won't be improved to actually scale, so MAX_CPUS will remain lower.