linux-kernel - [PATCH v2 0/3] KVM: VMX: Fix for kexec VMCLEAR and VMXON cleanup

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 for Android: free password hash cracker in your pocket

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

Message-Id: <20200321193751.24985-1-sean.j.christopherson@intel.com>
Date:   Sat, 21 Mar 2020 12:37:48 -0700
From:   Sean Christopherson <sean.j.christopherson@...el.com>
To:     Paolo Bonzini <pbonzini@...hat.com>
Cc:     Sean Christopherson <sean.j.christopherson@...el.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Jim Mattson <jmattson@...gle.com>,
        Joerg Roedel <joro@...tes.org>, kvm@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: [PATCH v2 0/3] KVM: VMX: Fix for kexec VMCLEAR and VMXON cleanup

Patch 1 fixes a a theoretical bug where a crashdump NMI that arrives
while KVM is messing with the percpu VMCS list would result in one or more
VMCSes not being cleared, potentially causing memory corruption in the new
kexec'd kernel.

Patch 2 is cleanup that's made possible by patch 1.

Patch 3 isn't directly related, but it conflicts with the crash cleanup
changes, both from a code and a semantics perspective.  Without the crash
cleanup, IMO hardware_enable() should do crash_disable_local_vmclear()
if VMXON fails, i.e. clean up after itself.  But hardware_disable()
doesn't even do crash_disable_local_vmclear() (which is what got me
looking at that code in the first place).  Basing the VMXON change on top
of the crash cleanup avoids the debate entirely.

v2:
  - Inverted the code flow, i.e. move code from loaded_vmcs_init() to
    __loaded_vmcs_clear().  Trying to share loaded_vmcs_init() with
    alloc_loaded_vmcs() was taking more code than it saved. [Paolo]


Gory details on the crashdump bug:

I verified my analysis of the NMI bug by simulating what would happen if
an NMI arrived in the middle of list_add() and list_del().  The below
output matches expectations, e.g. nothing hangs, the entry being added
doesn't show up, and the entry being deleted _does_ show up.

[    8.205898] KVM: testing NMI in list_add()
[    8.205898] KVM: testing NMI in list_del()
[    8.205899] KVM: found e3
[    8.205899] KVM: found e2
[    8.205899] KVM: found e1
[    8.205900] KVM: found e3
[    8.205900] KVM: found e1

static void vmx_test_list(struct list_head *list, struct list_head *e1,
                          struct list_head *e2, struct list_head *e3)
{
        struct list_head *tmp;

        list_for_each(tmp, list) {
                if (tmp == e1)
                        pr_warn("KVM: found e1\n");
                else if (tmp == e2)
                        pr_warn("KVM: found e2\n");
                else if (tmp == e3)
                        pr_warn("KVM: found e3\n");
                else
                        pr_warn("KVM: kaboom\n");
        }
}

static int __init vmx_init(void)
{
        LIST_HEAD(list);
        LIST_HEAD(e1);
        LIST_HEAD(e2);
        LIST_HEAD(e3);

        pr_warn("KVM: testing NMI in list_add()\n");

        list.next->prev = &e1;
        vmx_test_list(&list, &e1, &e2, &e3);

        e1.next = list.next;
        vmx_test_list(&list, &e1, &e2, &e3);

        e1.prev = &list;
        vmx_test_list(&list, &e1, &e2, &e3);

        INIT_LIST_HEAD(&list);
        INIT_LIST_HEAD(&e1);

        list_add(&e1, &list);
        list_add(&e2, &list);
        list_add(&e3, &list);

        pr_warn("KVM: testing NMI in list_del()\n");

        e3.prev = &e1;
        vmx_test_list(&list, &e1, &e2, &e3);

        list_del(&e2);
        list.prev = &e1;
        vmx_test_list(&list, &e1, &e2, &e3);
}

Sean Christopherson (3):
  KVM: VMX: Always VMCLEAR in-use VMCSes during crash with kexec support
  KVM: VMX: Fold loaded_vmcs_init() into alloc_loaded_vmcs()
  KVM: VMX: Gracefully handle faults on VMXON

 arch/x86/kvm/vmx/vmx.c | 103 ++++++++++++++++-------------------------
 arch/x86/kvm/vmx/vmx.h |   1 -
 2 files changed, 40 insertions(+), 64 deletions(-)

-- 
2.24.1