Call qemuExtVhostUserGPUPrepareDomain() to fill the domain with the
location of the vhost-user binary to start.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Check qemu capability, and accept 3d acceleration. 3d acceleration
support is checked when looking for a suitable vhost-user helper.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
The fact that qemu is capable -netdev socket is not enough to
start a migratable domain. It also needs dbus-vmstate capability.
Since there are already some qemu releases which have
net-socket-dgram capability and don't have dbus-vmstate we need
to check for dbus-vmstate.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
If managed='no', then the tap device must already exist, and setting
of MAC address and online status (IFF_UP) is skipped.
NB: we still set IFF_VNET_HDR and IFF_MULTI_QUEUE as appropriate,
because those bits must be properly set in the TUNSETIFF we use to set
the tap device name of the handle we've opened - if IFF_VNET_HDR has
not been set and we set it the request will be honored even when
running libvirtd unprivileged; if IFF_MULTI_QUEUE is requested to be
different than how it was created, that will result in an error from
the kernel. This means that you don't need to pay attention to
IFF_VNET_HDR when creating the tap devices, but you *do* need to set
IFF_MULTI_QUEUE if you're going to use multiple queues for your tap
device.
NB2: /dev/vhost-net normally has permissions 600, so it can't be
opened by an unprivileged process. This would normally cause a warning
message when using a virtio net device from an unprivileged
libvirtd. I've found that setting the permissions for /dev/vhost-net
permits unprivileged libvirtd to use vhost-net for virtio devices, but
have no idea what sort of security implications that has. I haven't
changed libvrit's code to avoid *attempting* to open /dev/vhost-net -
if you are concerned about the security of opening up permissions of
/dev/vhost-net (probably a good idea at least until we ask someone who
knows about the code) then add <driver name='qemu'/> to the interface
definition and you'll avoid the warning message.
Note that virNetDevTapCreate() is the correct function to call in the
case of an existing device, because the same ioctl() that creates a
new tap device will also open an existing tap device.
Resolves: https://bugzilla.redhat.com/1723367 (partially)
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In some places where virDomainObjListForEach() is called the
passed callback calls virDomainObjListRemoveLocked(). Well, this
is unsafe, because the former only grabs a read lock but the
latter modifies the list.
I've identified the following unsafe calls:
- qemuProcessReconnectAll()
- libxlReconnectDomains()
The rest seem to be safe.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When the network interface is of "user" type, and QEMU has the "-net
socket,fd=" datagram support, call qemuInterfacePrepareSlirp() to
probe and associate a slirp-helper with the interface.
The usage of automated slirp-helper can be prevented with
disableSlirp (in particular when resuming a
VM that didn't start with slirp-helper before).
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
If a slirp-helper is associated with a network interface,
prepare/start/stop the process via qemu-extdevice.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
pid filenames (from swtpm and other helpers from this series) are
based on VM shortname, which is derived from VM id. If the id is reset
to early, the state filenames will not be found.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Once QEMU is started, the qemuDomainLogContext is owned by it, and can
no longer be used from libvirt. Instead, use
qemuDomainLogAppendMessage() which will redirect the log.
This is not strictly necessary for swtpm, but the following patches
are going to reuse qemuExtDeviceLogCommand().
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
With blockdev we must issue the block_set_io_throttle QMP command to
setup disk throttling as we currently can't do it with the 'throttle'
layer.
Unfortunately there's nothing we can do if it fails.
https://bugzilla.redhat.com/show_bug.cgi?id=1733163
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
QEMU-4.1 supports 'Direct Mode' for Hyper-V synthetic timers
(hv-stimer-direct CPU flag): Windows guests can request that timer
expiration notifications are delivered as normal interrupts (and not
VMBus messages). This is used by Hyper-V on KVM.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Since qemuDomainDefPostParse callback requires qemuCaps, we need to make
sure it gets the capabilities stored in the domain's private data if the
domain is running. Passing NULL may cause QEMU capabilities probing to
be triggered in case QEMU binary changed in the meantime. When this
happens while a running domain object is locked, QMP event delivered to
the domain before QEMU capabilities probing finishes will deadlock the
event loop.
This patch fixes all paths leading to virDomainDefPostParse.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Since qemuDomainDefPostParse callback requires qemuCaps, we need to make
sure it gets the capabilities stored in the domain's private data if the
domain is running. Passing NULL may cause QEMU capabilities probing to
be triggered in case QEMU binary changed in the meantime. When this
happens while a running domain object is locked, QMP event delivered to
the domain before QEMU capabilities probing finishes will deadlock the
event loop.
Several general functions from domain_conf.c were lazily passing NULL as
the parseOpaque pointer instead of letting their callers pass the right
data. This patch fixes all paths leading to virDomainDefCopy to do the
right thing.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Since qemuDomainDefPostParse callback requires qemuCaps, we need to make
sure it gets the capabilities stored in the domain's private data if the
domain is running. Passing NULL may cause QEMU capabilities probing to
be triggered in case QEMU binary changed in the meantime. When this
happens while a running domain object is locked, QMP event delivered to
the domain before QEMU capabilities probing finishes will deadlock the
event loop.
This patch fixes all paths leading to qemuDomainDefFormatBufInternal.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
'default monitor of an allocation' is defined as the resctrl
monitor group that created along with an resctrl allocation,
which is created by resctrl file system. If the monitor group
specified in domain configuration file is happened to be a
default monitor group of an allocation, then it is not necessary
to create monitor group since it is already created. But if
an monitor group is not an allocation default group, you
should create the group under folder
'/sys/fs/resctrl/mon_groups' and fill the vcpu PIDs to 'tasks'
file.
Signed-off-by: Wang Huaqiang <huaqiang.wang@intel.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
hv-spinlocks is not a CPUID feature and should not be checked as such.
While starting a domain with hv-spinlocks enabled, we would report a
warning about unsupported hyperv spinlocks feature even though it was
set properly.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Originally the names of the hyperv CPU features were only used
internally for looking up their CPUID bits. So we used "__kvm_hv_"
prefix for them to make sure the names do not collide with normal CPU
features stored in our CPU map.
But with QEMU 4.1 we check which features were enabled or disabled by a
freshly started QEMU process using their names rather than their CPUID
bits (mostly because of MSR features). Thus we need to change our made
up internal names into the actual names used by QEMU. Most of the names
are only used with QEMU 4.1 and newer and the reset was introduced with
QEMU recently enough to already support spelling with "-". Thus we don't
need to define them as "hv_*" with a translation to "hv-*" for new QEMU.
Without this patch libvirt would mistakenly report all hyperv features
as unavailable and refuse to start any domain using them with QEMU 4.1.
Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In case of an incoming migration we do not need to run swtpm_setup
with all the parameters but only want to get the benefit of it
creating a TPM state file for us that we can then label with an
SELinux label. The actual state will be overwritten by the in-
coming state. So we have to pass an indicator for incomingMigration
all the way to the command line parameter generation for swtpm_setup.
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
If we are using -blockdev, then node names are always available
(because we set them). But when not using it, we have to scrape node
names from QMP, and want to do so as infrequently as possible. We
were scraping node names after reconnecting a new libvirtd to an
existing guest (see qemuProcessReconnect), and after any block job
that may have changed the set of node names we care about (legacy
block jobs), but forgot to scrape the names when first starting a
guest. Do so now in order to allow the checkpoint code to always have
access to a node name without having to repeat a node name scrape
itself.
Future patches may need to clean up qemuDomainSetBlockThreshold (if
node names are always available, then it doesn't need to repeat a
scrape) and/or hotplug and media changes (if the addition of new nodes
can result in a null node name, then scraping at that point in time
would be appropriate). But for now, this patch addresses only the
most common instance of a missing node name.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
The block job event handler qemuProcessHandleBlockJob looks at the block
job data to see whether the job requires synchronous handling. Since the
block job event may arrive before we continue the job handling (if the
job has no data to copy) we could hit the state when the job is still
set as QEMU_BLOCKJOB_STATE_NEW (as we move it to the
QEMU_BLOCKJOB_STATE_RUNNING state only after returning from monitor).
If the event handler uses qemuBlockJobStartupFinalize it would
unregister and free the job. Thankfully this is not a big problem for
legacy blockjobs as we don't need much data for them but since we'd
re-instantiate the job data structure we'd report wrong job type for
active commit as qemu reports it as a regular commit job.
Fix it by not using qemuBlockJobStartupFinalize function in
qemuProcessHandleBlockJob as it is not starting the job anyways.
https://bugzilla.redhat.com/show_bug.cgi?id=1721375
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The PR manager is a property of the format layer in qemu so we need to
be able to track it also in the chains of orphaned block jobs.
Add a helper for qemu to look also into the blockjob state.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Refresh the state of the jobs and process any events that might have
happened while libvirt was not running.
The job state processing requires some care to figure out if a job
needs to be bumped.
For any invalid job try doing our best to cancel it.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Add support for handling the event either synchronously or
asynchronously using the event thread.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
With blockdev we'll need to use the JOB_STATUS_CHANGE so gate the old
events by the blockdev capability.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Now that block job data is stored in the status XML portion we need to
make sure that everything which changes the state also saves the status
XML. The job registering function is used while parsing the status XML
so in that case we need to skip the XML saving.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Add the job structure to the table when instantiating a new job and
remove it when it terminates/fails.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Pass an xmlopt argument through all the needed network conf
functions, like is done for domain XML handling. No functional
change for now
Reviewed-by: Laine Stump <laine@laine.org>
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Filter out the given capabilities and set domain taint if we've done so.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
For testing purposes it's sometimes desired to be able to control the
presence of capabilities of qemu. This adds the possibility to do this
via the qemu namespace.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When connecting to virtlogd fails e.g. due to wrong libvirtd selinux
process label we'd report an utterly useless error message:
$ virsh start upstream
error: Failed to start domain upstream
error: Cannot recv data: Connection reset by peer
Use virLastErrorPrefixMessage in the correct place to give a better
sense of what's going on:
$ virsh start upstream
error: Failed to start domain upstream
error: can't connect to virtlogd: Cannot recv data: Connection reset by peer
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
Without "unavailable-features" CPU property we cannot properly detect
whether a specific MSR feature we asked for (either explicitly or
implicitly via a CPU model) was disabled by QEMU for some reason.
Because this could break migration, snapshots, and save/restore
operaions, it's better to just forbid any use of MSR features with QEMU
which lacks "unavailable-features" CPU property.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Always assume JSON monitor was requested, since all the callers
pass true anyway.
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Acked-by: Peter Krempa <pkrempa@redhat.com>
Now that we no longer support the HMP monitor, remove some dead code.
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Acked-by: Peter Krempa <pkrempa@redhat.com>
Now that the virDomainQemuAttach API returns an error, we can remove the
unused qemuProcessAttach function as well, deleting the only user
that possibly could have requested to open a non-JSON monitor.
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Acked-by: Peter Krempa <pkrempa@redhat.com>
If spawning qemu fails then we report an error and proceed to
writing status XML onto the disk. This is unnecessary as we are
sure that the domain is not running.
At the same time, if virPidFileReadPath() fails it returns
-errno. Use it in the error message. It may explain what went
wrong.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When updating guest CPU definition according to the vCPU actually
created by QEMU, we want to use the generic qemuMonitorGetGuestCPU to
get both CPUID and MSR features.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
It was never implemented or used for anything else anyway. Mainly
because it uses CPUID features bits. The function is renamed as
qemuMonitorGetGuestCPUx86 to make this explicit.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Properly filter features which should not be passed to QEMU because they
were never supported by QEMU or they did nothing and QEMU dropped them.
Currently they are just silently ignored by the command line generator.
Let's make this process more visible and clean by dropping the features
from the domain's active definition in qemuProcessUpdateGuestCPU.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
If libvirt receives DISCONNECTED event and prDaemonRunning is set
to false, and qemuDomainRemoveDiskDevice() is performing in the
meantime, then qemuDomainRemoveDiskDevice() will fail to remove
pr-helper object because prDaemonRunning is false. But removing
that check from qemuHotplugRemoveManagedPR() is not enough,
because after removing the object through monitor the
qemuProcessKillManagedPRDaemon() is called which contains the
same check. Thus the pr-helper process might be left behind.
Signed-off-by: Jie Wang <wangjie88@huawei.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The hash table returned by qemuMonitorGetAllBlockJobInfo is organized by
the frontend name (which skipps the 'drive-' prefix). While our code
properly matches the jobs to the disk, qemu needs the full job name
including the 'drive-' prefix to be able to identify jobs.
Fix this by adding an argument to qemuMonitorGetAllBlockJobInfo which
does not modify the job name before filling the hash.
This fixes a regression where users would not be able to cancel/pivot
block jobs after restarting libvirtd while a blockjob is running.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Commit 2f2254c7f4 attempted to fix a memory leak by ensuring
cpumapToSet is always a freshly allocated bitmap, but regrettably
introduced a NULL pointer access while doing so, because it called
virBitmapCopy() without allocating the destination bitmap first.
Solve the issue by using virBitmapNewCopy() instead.
Reported-by: John Ferlan <jferlan@redhat.com>
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
We're using VIR_AUTOPTR() for everything now, plus the
cleanup section was not doing anything useful anyway.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In two out of three scenarios we are cleaning up properly after
ourselves, but commit 5f2212c062 has changed the remaining one
in a way that caused it to start leaking cpumapToSet.
Refactor the logic so that cpumapToSet is always a freshly
allocated bitmap that gets cleaned up automatically thanks to
VIR_AUTOPTR(); this also allows us to remove the hostcpumap
variable.
Reported-by: John Ferlan <jferlan@redhat.com>
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Ever since the feature was introduced with commit 0f8e7ae33a,
it has contained a logic error in that it attempted to use a NUMA
node map where a CPU map was expected.
Because of that, guests using <numatune> might fail to start:
# virsh start guest
error: Failed to start domain guest
error: cannot set CPU affinity on process 40055: Invalid argument
This was particularly easy to trigger on POWER 8 machines, where
secondary threads always show up as offline in the host: having
<numatune>
<memory mode='strict' placement='static' nodeset='1'/>
</numatune>
in the guest configuration, for example, would result in libvirt
trying to set the process affinity so that it would prefer
running on CPU 1, but since that's a secondary thread and thus
shows up as offline, the operation would fail, and so would
starting the guest.
Use the newly introduced virNumaNodesetToCPUset() to convert the
NUMA node map to a CPU map, which in the example above would be
48,56,64,72,80,88 - a valid input for virProcessSetAffinity().
https://bugzilla.redhat.com/show_bug.cgi?id=1703661
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When migrating a domain with invtsc CPU feature enabled, the TSC
frequency of the destination host must match the frequency used when the
domain was started on the source host or the destination host has to
support TSC scaling.
If the frequencies do not match and the destination host does not
support TSC scaling, QEMU will fail to set the right TSC frequency when
starting vCPUs on the destination and thus migration will fail. However,
this is quite late since both host might have spent significant time
transferring memory and perhaps even storage data.
By adding the check to libvirt we can let migration fail before any data
starts to be sent over. If for some reason libvirt is unable to detect
the host's TSC frequency or scaling support, we'll just let QEMU try and
the migration will either succeed or fail later.
Luckily, we mandate TSC frequency to be explicitly set in the domain XML
to even allow migration of domains with invtsc. We can just check
whether the requested frequency is compatible with the current host
before starting QEMU.
https://bugzilla.redhat.com/show_bug.cgi?id=1641702
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
If the scheduler is set before vCPU0 cannot be moved into its cpu,cpuacct
cgroup. While it is not yet known whether this is a bug or not, it makes sense
for us to do that later as otherwise the scheduler would be inherited by vCPU
and I/O Threads even when they do not have any such setting specified.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
It's funny how this went unnoticed for such a long time. Long
story short, if a domain is configured with
VIR_DOMAIN_NUMATUNE_MEM_STRICT libvirt doesn't really honour
that. This is because of 7e72ac7878 after which libvirt allowed
qemu to allocate memory just anywhere and only after that it used
some magic involving cpuset.memory_migrate and cpuset.mems to
move the memory to desired NUMA nodes. This was done in order to
work around some KVM bug where KVM would fail if there wasn't a
DMA zone available on the NUMA node. Well, while the work around
might stopped libvirt tickling the KVM bug it also caused a bug
on libvirt side: if there is not enough memory on configured NUMA
node(s) then any attempt to start a domain must fail. Because of
the way we play with guest memory domains can start just happily.
The solution is to move the child we've just forked into emulator
cgroup, set up cpuset.mems and exec() qemu only after that.
This basically reverts 7e72ac7878 which was a workaround
for kernel bug. This bug was apparently fixed because I've tested
this successfully with recent kernel.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
