10:27 a.m.
---
I sent this as a patch to get a review on the contents itself, but I believe it
should live as an article on our wiki page afterwards.
docs/launch_security_sev.html.in | 536 +++++++++++++++++++++++++++++++
1 file changed, 536 insertions(+)
create mode 100644 docs/launch_security_sev.html.in
diff --git a/docs/launch_security_sev.html.in b/docs/launch_security_sev.html.in
new file mode 100644
index 0000000000..42db10b33a
--- /dev/null
+++ b/docs/launch_security_sev.html.in
@@ -0,0 +1,536 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <meta charset="UTF-8"/>
+ <meta name="viewport" content="width=device-width,
initial-scale=1"/>
+ <link rel="stylesheet" type="text/css"
href="main.css"/>
+ <link rel="apple-touch-icon" sizes="180x180"
href="/apple-touch-icon.png"/>
+ <link rel="icon" type="image/png" sizes="32x32"
href="/favicon-32x32.png"/>
+ <link rel="icon" type="image/png" sizes="16x16"
href="/favicon-16x16.png"/>
+ <link rel="manifest" href="/manifest.json"/>
+ <meta name="theme-color" content="#ffffff"/>
+ <title>libvirt: Launch security with AMD SEV</title>
+ <meta name="description" content="libvirt, virtualization,
virtualization API"/>
+ <script type="text/javascript" src="js/main.js">
+ <!--// forces non-empty element-->
+ </script>
+ </head>
+ <body onload="pageload()">
+ <div id="body">
+ <div id="content">
+ <h1>Launch security with AMD SEV</h1>
+ <ul>
+ <li>
+ <a href="#Host">Enabling SEV on the host</a>
+ </li>
+ <li>
+ <a href="#Virt">Checking SEV support in the virt
stack</a>
+ </li>
+ <li>
+ <a href="#Configuration">VM Configuration</a>
+ </li>
+ <li>
+ <a href="#Limitations">Limitations</a>
+ </li>
+ <li>
+ <a href="#Examples">Full domain XML examples</a>
+ </li>
+ </ul>
+ </div>
+ <p>
+ Storage encryption in modern public cloud computing is a common practice.
+ However, from the point of view of a user of these cloud workloads, a
+ significant amount of trust needs to be put in the cloud platform security as
+ well as integrity (was the hypervisor tampered?). For this reason there's
ever
+ rising demand for securing data in use, i.e. memory encryption.
+ One of the solutions addressing this matter is AMD SEV.
+ </p>
+ <h2>AMD SEV</h2>
+ <p>
+ SEV (Secure Encrypted Virtualization) is a feature extension of AMD's SME
(Secure
+ Memory Encryption) intended for KVM virtual machines which is supported
+ primarily on AMD's EPYC CPU line. In contrast to SME, SEV uses a unique
memory encryption
+ key for each VM. The whole encryption of memory pages is completely
transparent
+ to the hypervisor and happens in the AMD firmware.
+ For more details about the technology itself, you can visit
+ <a href="https://developer.amd.com/sev/">AMD's developer
portal</a>.
+ </p>
+ <h3>
+ <a id="Host">Enabling SEV on the host</a>
+ <a class="headerlink" href="#Host" title="Permalink
to this headline">¶</a>
+ </h3>
+ <p>
+ Before VMs can make use of the SEV feature you need to make sure your
+ AMD CPU does support SEV. You can check whether SEV is among the CPU
+ flags with:
+ </p>
+
+ <pre>
+$ cat /proc/cpuinfo | grep sev
+...
+sme ssbd sev ibpb</pre>
+
+ <p>
+ Next step is to enable SEV in the kernel, because it is disabled by default.
+ This is done by putting the following onto the kernel command line:
+ </p>
+ <pre>
+mem_encrypt=on kvm_amd.sev=1
+ </pre>
+
+ <p>
+ To make the changes persistent, append the above to the variable holding
+ parameters of the kernel command line in
+ <code>/etc/default/grub</code> to preserve SEV settings across
reboots
+ </p>
+
+ <pre>
+$ cat /etc/default/grub
+...
+GRUB_CMDLINE_LINUX="... mem_encrypt=on kvm_amd.sev=1"
+$ grub2-mkconfig -o /boot/efi/EFI/<distro>/grub.cfg</pre>
+
+ <p>
+ <code>mem_encrypt=on</code> turns on the SME memory encryption
feature on
+ the host which is required for SEV to work. The
<code>kvm_amd.sev</code>
+ parameter actually enables SEV in the kvm module. It can be set on the
+ command line alongside <code>mem_encrypt</code> like shown above, or
it
+ can be put into a module config under <code>/etc/modprobe.d/</code>
+ </p>
+ <pre>
+$ cat /etc/modprobe.d/sev.conf
+options kvm_amd sev=1
+ </pre>
+ <p>
+ After rebooting the host, you should see SEV being enabled in the kernel:
+ </p>
+ <pre>
+$ cat /sys/module/kvm_amd/parameters/sev
+1
+ </pre>
+
+ <h2>
+ <a id="Virt">Checking SEV support in the virt stack</a>
+ <a class="headerlink" href="#Virt" title="Permalink
to this headline">¶</a>
+ </h2>
+ <p>
+ <b>Note: All of the commands bellow need to be run with root
privileges.</b>
+ </p>
+ <p>
+ First make sure you have the following packages in the specified versions:
+ </p>
+ <ul>
+ <li>
+ libvirt >= 4.5.0 (>5.1.0 recommended due to additional SEV bugfixes)
+ </li>
+ <li>
+ QEMU >= 2.12.0
+ </li>
+ </ul>
+ <p>
+ To confirm that the virtualization stack supports SEV, run the following:
+ </p>
+ <pre>
+# virsh domcapabilities
+<domainCapabilities>
+...
+ <features>
+ ...
+ <sev supported='yes'>
+ <cbitpos>47</cbitpos>
+ <reducedPhysBits>1</reducedPhysBits>
+ </sev>
+ ...
+ </features>
+</domainCapabilities></pre>
+ <p>
+ Note that if libvirt was already installed and libvirtd running before enabling
SEV in the kernel followed by the host reboot you need to force libvirtd
+ to re-probe both the host and QEMU capabilities. First stop libvirtd:
+ </p>
+ <pre>
+# systemctl stop libvirtd.service
+ </pre>
+ <p>
+ Now you need to clean the capabilities cache:
+ </p>
+ <pre>
+# rm -f /var/cache/libvirt/qemu/capabilities/*
+ </pre>
+ <p>
+ If you now restart libvirtd, it will re-probe the capabilities and if
+ you now run:
+ </p>
+ <pre>
+# virsh domcapabilities
+ </pre>
+ <p>
+ SEV should be listed as supported. If you still see:
+ </p>
+ <pre>
+<sev supported='no'/>
+ </pre>
+ <p>
+ it means one of two things:
+ <ol>
+ <li>
+ libvirt does support SEV, but either QEMU or the host does not
+ </li>
+ <li>
+ you have libvirt <=5.1.0 which suffered from getting a
+ <code>'Permission denied'</code> on
<code>/dev/sev</code> because
+ of the default permissions on the character device which prevented
+ QEMU from opening it during capabilities probing - you can either
+ manually tweak the permissions so that QEMU has access to it or
+ preferably install libvirt 5.1.0 or higher
+ </li>
+ </ol>
+ </p>
+ <h2>
+ <a id="Configuration">VM Configuration</a>
+ <a class="headerlink" href="#Configuration"
title="Permalink to this headline">¶</a>
+ </h2>
+ <p>
+ SEV is enabled in the XML by specifying the
+ <a
href="https://libvirt.org/formatdomain.html#launchSecurity">...
</a> element. However, specifying <code>launchSecurity</code> isn't
+ enough to boot an SEV VM. Further configuration requirements are discussed
+ below.
+ </p>
+
+ <h3>Machine type</h3>
+ <p>
+ Even though both Q35 and legacy PC machine types (for PC see also
+ "virtio") can be used with SEV, usage of the legacy PC machine type
is
+ strongly discouraged, since depending on how your OVMF package was
+ built (e.g. including features like SecureBoot or SMM) Q35 may even be
+ required.
+ </p>
+
+ <h5>Q35</h5>
+<pre>
+...
+<os>
+ <type arch='x86_64'
machine='pc-q35-3.0'>hvm</type>
+ ...
+</os>
+...</pre>
+
+ <h5>i440fx (discouraged)</h5>
+ <pre>
+...
+<os>
+ <type arch='x86_64'
machine='pc-i440fx-3.0'>hvm</type>
+ ...
+</os>
+...
+ </pre>
+
+ <h3>Boot loader</h3>
+ <p>
+ SEV is only going to work with OVMF (UEFI), so you'll need to point libvirt
to
+ the correct OVMF binary.
+ </p>
+ <pre>
+...
+<os>
+ <type arch='x86_64'
machine='pc-q35-3.0'>hvm</type>
+ <loader readonly='yes'
type='pflash'>/usr/share/edk2/ovmf/OVMF_CODE.fd</loader>
+</os>
+...</pre>
+
+ <h3>Memory</h3>
+ <p>
+ Internally, SEV expects that the encrypted memory pages won't be swapped
out or move
+ around so the VM memory needs to be pinned in physical RAM which will be
+ handled by QEMU. Apart from that, certain memory regions allocated by QEMU
+ itself (UEFI pflash, device ROMs, video RAM, etc.) have to be encrypted as
+ well. This causes a conflict in how libvirt tries to protect the host.
+ By default, libvirt enforces a memory hard limit on each VM's cgroup in
order
+ to protect the host from malicious QEMU to allocate and lock all the available
+ memory. This limit corresponds to the total memory allocation for the VM given
+ by <code><currentMemory></code> element. However,
trying to account for the additional
+ memory regions QEMU allocates when calculating the limit in an automated
manner
+ is non-deterministic. One way to resolve this is to set the hard limit
manually.
+
+ <p>
+ Note: Figuring out the right number so that your guest boots and isn't
killed is
+ challenging, but 256MiB extra memory over the total guest RAM should suffice
for
+ most workloads and may serve as a good starting point.
+
+ For example, a domain with 4GB memory with a 256MiB extra hard limit would
look
+ like this:
+ </p>
+ </p>
+ <pre>
+# virsh edit <domain>
+<domain>
+ ...
+ <currentMemory unit='KiB'>4194304</currentMemory>
+ <memtune>
+ <hard_limit unit='KiB'>4456448</hard_limit>
+ </memtune>
+ ...
+</domain></pre>
+ <p>
+ There's another, preferred method of taking care of the limits by
+ using the<code><memoryBacking></code> element along
with the
+ <code><locked/></code> subelement:
+ </p>
+ <pre>
+<domain>
+ ...
+ <memoryBacking>
+ <locked/>
+ </memoryBacking>
+ ...
+</domain></pre>
+ <p>
+ What that does is that it tells libvirt not to force any hard limit (well,
+ unlimited) upon the VM cgroup. The obvious advantage is that one doesn't
need
+ to determine the hard limit for every single SEV-enabled VM. However, there is
+ a significant security-related drawback to this approach. Since no hard limit
+ is applied, a malicious QEMU could perform a DoS attack by locking all of the
+ host's available memory. The way to avoid this issue and to protect the
host is
+ to enforce a bigger hard limit on the master cgroup containing all of the VMs
+ - on systemd this is <code>machine.slice</code>.
+ </p>
+ <pre>
+# systemctl set-property machine.slice MemoryHigh=<value></pre>
+ <p>
+ To put even stricter measures in place which would involve the OOM killer, use
+ <pre>
+# systemctl set-property machine.slice MemoryMax=<value></pre>
+ instead. Alternatively, you can create a systemd config (don't forget
+ to reload systemd configuration in this case):
+ <pre>
+# cat << EOF >
/etc/systemd/system.control/machine.slice.d/90-MemoryMax.conf
+MemoryMax=<value>
+EOF</pre>
+ The trade-off to keep in mind with the second approach is that the VMs
+ can still perform DoS on each other.
+ </p>
+ <h3>Virtio</h3>
+ <p>
+ In order to make virtio devices work, we need to enable emulated IOMMU
+ on the devices so that virtual DMA can work.
+ </p>
+ <pre>
+# virsh edit <domain>
+<domain>
+ ...
+ <controller type='virtio-serial' index='0'>
+ <driver iommu='on'/>
+ </controller>
+ <controller type='scsi' index='0'
model='virtio-scsi'>
+ <driver iommu='on'/>
+ </controller>
+ ...
+ <memballoon model='virtio'>
+ <driver iommu='on'/>
+ </memballoon>
+ <rng model='virtio'>
+ <backend model='random'>/dev/urandom</backend>
+ <driver iommu='on'/>
+ </rng>
+ ...
+<domain></pre>
+ <p>
+ If you for some reason want to use the legacy PC machine type, further changes
+ to the virtio
+ configuration is required, because SEV will not work with Virtio <1.0. In
+ libvirt, this is handled by using the virtio-non-transitional device model
+ (libvirt >= 5.2.0 required).
+
+ <p>
+ Note: some devices like video devices don't
+ support non-transitional model, which means that virtio GPU cannot be used.
+ </p>
+ </p>
+ <pre>
+<domain>
+ ...
+ <devices>
+ ...
+ <memballoon model='virtio-non-transitional'>
+ <driver iommu='on'/>
+ </memballoon>
+ </devices>
+ ...
+</domain></pre>
+
+
+ <h2>
+ <a id="Limitations">Limitations</a>
+ <a class="headerlink" href="#Limitations"
title="Permalink to this headline">¶</a>
+ </h2>
+ <p>
+ Currently, the boot disk cannot be of type virtio-blk, instead, virtio-scsi
+ needs to be used if virtio is desired. This limitation is expected to be lifted
+ with future releases of kernel (the kernel used at the time of writing the
+ article is 5.0.14).
+ If you still cannot start an SEV VM, it could be because of wrong SELinux label
on the <code>/dev/sev</code> device with selinux-policy <3.14.2.40
which prevents QEMU from touching the device. This can be resolved by upgrading the
package, tuning the selinux policy rules manually to allow svirt_t to access the device
(see <code>audit2allow</code> on how to do that) or putting SELinux into
permissive mode (discouraged).
+ </p>
+ <h2>
+ <a id="Examples">Full domain XML examples</a>
+ <a class="headerlink" href="#Examples" title="Permalink
to this headline">¶</a>
+ </h2>
+ <h5>Q35 machine</h5>
+ <pre>
+<domain type='kvm'>
+ <name>sev-dummy</name>
+ <memory unit='KiB'>4194304</memory>
+ <currentMemory unit='KiB'>4194304</currentMemory>
+ <memoryBacking>
+ <locked/>
+ </memoryBacking>
+ <vcpu placement='static'>4</vcpu>
+ <os>
+ <type arch='x86_64'
machine='pc-q35-3.0'>hvm</type>
+ <loader readonly='yes'
type='pflash'>/usr/share/edk2/ovmf/OVMF_CODE.fd</loader>
+
<nvram>/var/lib/libvirt/qemu/nvram/sev-dummy_VARS.fd</nvram>
+ </os>
+ <features>
+ <acpi/>
+ <apic/>
+ <vmport state='off'/>
+ </features>
+ <cpu mode='host-model' check='partial'>
+ <model fallback='allow'/>
+ </cpu>
+ <clock offset='utc'>
+ <timer name='rtc' tickpolicy='catchup'/>
+ <timer name='pit' tickpolicy='delay'/>
+ <timer name='hpet' present='no'/>
+ </clock>
+ <on_poweroff>destroy</on_poweroff>
+ <on_reboot>restart</on_reboot>
+ <on_crash>destroy</on_crash>
+ <pm>
+ <suspend-to-mem enabled='no'/>
+ <suspend-to-disk enabled='no'/>
+ </pm>
+ <devices>
+ <emulator>/usr/bin/qemu-kvm</emulator>
+ <disk type='file' device='disk'>
+ <driver name='qemu' type='qcow2'/>
+ <source file='/var/lib/libvirt/images/sev-dummy.qcow2'/>
+ <target dev='sda' bus='scsi'/>
+ <boot order='1'/>
+ </disk>
+ <controller type='virtio-serial' index='0'>
+ <driver iommu='on'/>
+ </controller>
+ <controller type='scsi' index='0'
model='virtio-scsi'>
+ <driver iommu='on'/>
+ </controller>
+ <interface type='network'>
+ <mac address='52:54:00:cc:56:90'/>
+ <source network='default'/>
+ <model type='virtio'/>
+ <driver iommu='on'/>
+ </interface>
+ <graphics type='spice' autoport='yes'>
+ <listen type='address'/>
+ <gl enable='no'/>
+ </graphics>
+ <video>
+ <model type='qxl'/>
+ </video>
+ <memballoon model='virtio'>
+ <driver iommu='on'/>
+ </memballoon>
+ <rng model='virtio'>
+ <driver iommu='on'/>
+ </rng>
+ </devices>
+ <launchSecurity type='sev'>
+ <cbitpos>47</cbitpos>
+ <reducedPhysBits>1</reducedPhysBits>
+ <policy>0x0003</policy>
+ </launchSecurity>
+</domain></pre>
+
+ <h5>PC-i440fx machine:</h5>
+ <pre>
+<domain type='kvm'>
+ <name>sev-dummy-legacy</name>
+ <memory unit='KiB'>4194304</memory>
+ <currentMemory unit='KiB'>4194304</currentMemory>
+ <memtune>
+ <hard_limit unit='KiB'>5242880</hard_limit>
+ </memtune>
+ <vcpu placement='static'>4</vcpu>
+ <os>
+ <type arch='x86_64'
machine='pc-i440fx-3.0'>hvm</type>
+ <loader readonly='yes'
type='pflash'>/usr/share/edk2/ovmf/OVMF_CODE.fd</loader>
+
<nvram>/var/lib/libvirt/qemu/nvram/sev-dummy_VARS.fd</nvram>
+ <boot dev='hd'/>
+ </os>
+ <features>
+ <acpi/>
+ <apic/>
+ <vmport state='off'/>
+ </features>
+ <cpu mode='host-model' check='partial'>
+ <model fallback='allow'/>
+ </cpu>
+ <clock offset='utc'>
+ <timer name='rtc' tickpolicy='catchup'/>
+ <timer name='pit' tickpolicy='delay'/>
+ <timer name='hpet' present='no'/>
+ </clock>
+ <on_poweroff>destroy</on_poweroff>
+ <on_reboot>restart</on_reboot>
+ <on_crash>destroy</on_crash>
+ <pm>
+ <suspend-to-mem enabled='no'/>
+ <suspend-to-disk enabled='no'/>
+ </pm>
+ <devices>
+ <emulator>/usr/bin/qemu-kvm</emulator>
+ <disk type='file' device='disk'>
+ <driver name='qemu' type='qcow2'/>
+ <source
file='/var/lib/libvirt/images/sev-dummy-seabios.qcow2'/>
+ <target dev='sda' bus='sata'/>
+ </disk>
+ <interface type='network'>
+ <mac address='52:54:00:d8:96:c8'/>
+ <source network='default'/>
+ <model type='virtio-non-transitional'/>
+ </interface>
+ <serial type='pty'>
+ <target type='isa-serial' port='0'>
+ <model name='isa-serial'/>
+ </target>
+ </serial>
+ <console type='pty'>
+ <target type='serial' port='0'/>
+ </console>
+ <input type='tablet' bus='usb'>
+ <address type='usb' bus='0' port='1'/>
+ </input>
+ <input type='mouse' bus='ps2'/>
+ <input type='keyboard' bus='ps2'/>
+ <graphics type='spice' autoport='yes'>
+ <listen type='address'/>
+ <gl enable='no'/>
+ </graphics>
+ <video>
+ <model type='qxl' ram='65536' vram='65536'
vgamem='16384' heads='1' primary='yes'/>
+ </video>
+ <memballoon model='virtio-non-transitional'>
+ <driver iommu='on'/>
+ </memballoon>
+ <rng model='virtio-non-transitional'>
+ <driver iommu='on'/>
+ </rng>
+ </devices>
+ <launchSecurity type='sev'>
+ <cbitpos>47</cbitpos>
+ <reducedPhysBits>1</reducedPhysBits>
+ <policy>0x0003</policy>
+ </launchSecurity>
+</domain></pre>
+ </div>
+</body>
+</html>
--
2.21.0