linux-kernel - Re: [PATCH 3/4] KVM: selftests: Use asserting kvm_ioctl() macros when getting ARM page sizes

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 4 Aug 2023 08:23:54 -0700
From:   Sean Christopherson <seanjc@...gle.com>
To:     Michal Luczaj <mhal@...x.co>
Cc:     Marc Zyngier <maz@...nel.org>,
        Oliver Upton <oliver.upton@...ux.dev>,
        linux-arm-kernel@...ts.infradead.org, kvmarm@...ts.linux.dev,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/4] KVM: selftests: Use asserting kvm_ioctl() macros when
 getting ARM page sizes

On Fri, Aug 04, 2023, Michal Luczaj wrote:
> On 8/4/23 02:42, Sean Christopherson wrote:
> > Use kvm_ioctl() instead of open coding equivalent ioctl()+TEST_ASSERT()
> > calls when getting the support page sizes on ARM.  The macro usage is a
> > little funky since the "kvm_fd" parameter implies an actual /dev/kvm fd,
> > but on the other hand the code is invoking KVM ioctl()s.
> > 
> > Alternatively, the core utilities could expose a vm_open()+vm_close()
> > pair so that the ARM code could create a dummy, on-stack VM+vCPU pair and
> > use {vm,vcpu}_ioctl() as appropriate.  But the odds of something breaking
> > due to oddball, partial usage of kvm_vm and kvm_vcpu structures is much
> > higher than realizing meaningful benefit from using {vm,vcpu}_ioctl().
> 
> Since you're doing the cleanup, does mmio_warning_test qualify for the
> same (funky usage ahead)?

Hmm, I'm heavily leaning towards deleting that test entirely.  It's almost
literally a copy+paste of the most basic syzkaller test, e.g. spawn a vCPU with
no backing memory and watch it die a horrible death.  Unless I'm missing something,
the test is complete overkill too, e.g. I highly doubt the original KVM bug required
userspace to fork() and whatnot, but syzkaller spawns threads for everything and
so that got copied into the selftest.

And this stuff is just silly:

	TEST_REQUIRE(host_cpu_is_intel);

	TEST_REQUIRE(!vm_is_unrestricted_guest(NULL));

because crashing the VM doesn't require Intel, nor does it require !URG, those
just happened to be the conditions for the bug.

As much as I like having explicit testcases, adding a new selftest for every bug
that syzkaller finds is neither realistic nor productive.  In other words, I think
we should treat syzkaller as being part of KVM's test infrastructure.

I'll send a patch to nuke the test.

> -       kvm = open("/dev/kvm", O_RDWR);
> -       TEST_ASSERT(kvm != -1, "failed to open /dev/kvm");
> -       kvmvm = __kvm_ioctl(kvm, KVM_CREATE_VM, NULL);
> -       TEST_ASSERT(kvmvm > 0, KVM_IOCTL_ERROR(KVM_CREATE_VM, kvmvm));
> -       kvmcpu = ioctl(kvmvm, KVM_CREATE_VCPU, 0);
> -       TEST_ASSERT(kvmcpu != -1, KVM_IOCTL_ERROR(KVM_CREATE_VCPU, kvmcpu));
> +       kvm = open_path_or_exit(KVM_DEV_PATH, O_RDWR);
> +       kvmvm = kvm_fd_ioctl(kvm, KVM_CREATE_VM, NULL);
> +       kvmcpu = kvm_fd_ioctl(kvmvm, KVM_CREATE_VCPU, NULL);
> 
> Side note, just in case this wasn't your intention: no kvm@ in cc.

Wasn't intentional, I was moving too fast at the end of the day and missed that
KVM wasn't Cc'd.  Grr.