Now that we have a capability for each domain net backend we can
start validating user's selection against QEMU capabilities.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Since -netdev user can be disabled during QEMU compilation, we
can't blindly expect it to just be there. We need a capability
that tracks its presence.
For qemu-4.2.0 we are not able to detect the capability so do the
next best thing - assume the capability is there. This is
consistent with our current behaviour where we blindly assume the
capability, anyway.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
When enabling switchover-ack on qemu from libvirt, the .party value
was set to both source and target; however, qemuMigrationParamsCheck()
only takes that into account to validate that the remote side of the
migration supports the flag if it is marked optional or auto/always on.
In the case of switchover-ack, when enabled on only the dst and not
the src, the migration will fail if the src qemu does not support
switchover-ack, as the dst qemu will issue a switchover-ack msg:
qemu/migration/savevm.c ->
loadvm_process_command ->
migrate_send_rp_switchover_ack(mis) ->
migrate_send_rp_message(mis, MIG_RP_MSG_SWITCHOVER_ACK, 0, NULL)
Since the src qemu doesn't understand messages with header_type ==
MIG_RP_MSG_SWITCHOVER_ACK, qemu will kill the migration with error:
qemu-kvm: RP: Received invalid message 0x0007 length 0x0000
qemu-kvm: Unable to write to socket: Bad file descriptor
Looking at the original commit [1] for optional migration capabilities,
it seems that the spirit of optional handling was to enhance a given
existing capability where possible. Given that switchover-ack
exclusively depends on return-path, adding it as optional to that cap
feels right.
[1] 61e34b08568 ("qemu: Add support for optional migration capabilities")
Fixes: 1cc7737f69e ("qemu: add support for qemu switchover-ack")
Signed-off-by: Jon Kohler <jon@nutanix.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Avihai Horon <avihaih@nvidia.com>
Cc: Jiri Denemark <jdenemar@redhat.com>
Cc: Markus Armbruster <armbru@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: YangHang Liu <yanghliu@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Now that the logic for detecting supported launchSecurity types
has been moved to domain capabilities generation, we can just use
it when validating launchSecurity type. Just like we do for
device models and so on.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
The inspiration for these rules comes from
qemuValidateDomainDef().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
While it's very unlikely to have QEMU that supports SEV-SNP but
doesn't support plain SEV, for completeness sake we ought to
query SEV capabilities if QEMU supports either. And similarly to
QEMU_CAPS_SEV_GUEST we need to clear the capability if talking to
QEMU proves SEV is not really supported.
This in turn removes the 'sev-snp-guest' capability from one of
our test cases as Peter's machine he uses to refresh capabilities
is not SEV capable. But that's okay. It's consistent with
'sev-guest' capability.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
An iSCSI device with zero hosts will result in a segmentation fault. This patch
adds a check for the number of hosts, which must be one in the case of iSCSI.
Minimal reproducing XML:
<domain type='qemu'>
<name>MyGuest</name>
<uuid>4dea22b3-1d52-d8f3-2516-782e98ab3fa0</uuid>
<os>
<type arch='x86_64'>hvm</type>
</os>
<memory>4096</memory>
<devices>
<disk type='network'>
<source name='dummy' protocol='iscsi'/>
<target dev='vda'/>
</disk>
</devices>
</domain>
Signed-off-by: Rayhan Faizel <rayhan.faizel@gmail.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Add plumbing for QEMU's switchover-ack migration capability, which
helps lower the downtime during VFIO migrations. This capability is
enabled by default as long as both the source and destination support
it.
Note: switchover-ack depends on the return path capability, so this may
not be used when VIR_MIGRATE_TUNNELLED flag is set.
Extensive details about the qemu switchover-ack implementation are
available in the qemu series v6 cover letter [1] where the highlight is
the extreme reduction in guest visible downtime. In addition to the
original test results below, I saw a roughly ~20% reduction in downtime
for VFIO VGPU devices at minimum.
=== Test results ===
The below table shows the downtime of two identical migrations. In the
first migration swithcover ack is disabled and in the second it is
enabled. The migrated VM is assigned with a mlx5 VFIO device which has
300MB of device data to be migrated.
+----------------------+-----------------------+----------+
| Switchover ack | VFIO device data size | Downtime |
+----------------------+-----------------------+----------+
| Disabled | 300MB | 1900ms |
| Enabled | 300MB | 420ms |
+----------------------+-----------------------+----------+
Switchover ack gives a roughly 4.5 times improvement in downtime.
The 1480ms difference is time that is used for resource allocation for
the VFIO device in the destination. Without switchover ack, this time is
spent when the source VM is stopped and thus the downtime is much
higher. With switchover ack, the time is spent when the source VM is
still running.
[1] https://patchwork.kernel.org/project/qemu-devel/cover/20230621111201.29729-1-avihaih@nvidia.com/
Signed-off-by: Jon Kohler <jon@nutanix.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Avihai Horon <avihaih@nvidia.com>
Cc: Markus Armbruster <armbru@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: YangHang Liu <yanghliu@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Commit 7c8e606b64c73ca56d7134cb16d01257f39c53ef attempted to fix
the specification of the ramfb property for vfio-pci devices, but it
failed when ramfb is explicitly set to 'off'. This is because only the
'vfio-pci-nohotplug' device supports the 'ramfb' property. Since we use
the base 'vfio-pci' device unless ramfb is enabled, attempting to set
the 'ramfb' parameter to 'off' this will result in an error like the
following:
error: internal error: QEMU unexpectedly closed the monitor
(vm='rhel'): 2024-06-06T04:43:22.896795Z qemu-kvm: -device
{"driver":"vfio-pci","host":"0000:b1:00.4","id":"hostdev0","display":"on
","ramfb":false,"bus":"pci.7","addr":"0x0"}: Property 'vfio-pci.ramfb'
not found.
This also more closely matches what is done for mdev devices.
Resolves: https://issues.redhat.com/browse/RHEL-28808
Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The attribute 'discard_no_unref' of <disk/> is not allowed to be
changed while the virtual machine is running.
Resolves: https://issues.redhat.com/browse/RHEL-37542
Signed-off-by: Adam Julis <ajulis@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The firmware descriptors have 'amd-sev-snp` feature which
describes whether firmware is suitable for SEV-SNP guests.
Provide necessary implementation to detect the feature and pick
the right firmware if guest is SEV-SNP enabled.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Pretty straightforward as qemu has 'sev-snp-guest' object which
attributes maps pretty much 1:1 to our XML model. Except for
@vcek where QEMU has 'vcek-disabled`, an inverted boolean, while
we model it as virTristateBool. But that's easy to map too.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
SEV-SNP is an enhancement of SEV/SEV-ES and thus it shares some
fields with it. Nevertheless, on XML level, it's yet another type
of <launchSecurity/>.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
This capability tracks sev-snp-guest object availability.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In QEMU commit v9.0.0-1155-g59d3740cb4 the return type of
'query-sev' monitor command changed to accommodate SEV-SNP. Even
though we currently support launching plain SNP guests, this will
soon change.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In a few instances there is a plain if() check for
_virDomainSecDef::sectype. While this works perfectly for now,
soon there'll be another type and we can utilize compiler to
identify all the places that need adaptation. Switch those if()
statements to switch().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
The sectype member of _virDomainSecDef struct is already declared
as of virDomainLaunchSecurity type. There's no need to typecast
it to the very same type when passing it to switch().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Some parts of SEV are to be shared with SEV SNP. In order to
reuse XML parsing / formatting code cleanly, let's move those
common bits into a new struct (virDomainSEVCommonDef) and adjust
rest of the code.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
While working on qemuMonitorJSONGetSEVMeasurement() and
qemuMonitorJSONGetSEVInfo() I've noticed that if these functions
fail, they do so without appropriate error set. Fill in error
reporting.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
When a VM terminates itself while it's being migrated in running state
libvirt would report wrong error:
error: cannot get locked memory limit of process 2502057: No such file or directory
rather than the proper error:
error: operation failed: domain is not running
Remember the error on error paths in qemuMigrationSrcConfirmPhase and
qemuMigrationSrcPerformPhase.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
'qemuProcessStop()' clears the 'current' job data. While the code under
the 'error' label in 'qemuMigrationSrcRun()' does check that the VM is
active before accessing the job, it also invokes multiple helper
functions to clean up the migration including
'qemuMigrationSrcNBDCopyCancel()' which calls 'qemuDomainObjWait()'
invalidating the result of the liveness check as it unlocks the VM.
Duplicate the liveness check and explain why. The rest of the code e.g.
accessing the monitor is safe as 'qemuDomainEnterMonitorAsync()'
performs a liveness check. The cleanup path just ignores the return
values of those functions.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The function is a pointless wrapper on top of
qemuMigrationDstWaitForCompletion.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Similarly to the one change in commit 4d1a1fdffda19a62d62fa2457d162362
we should be checking that the VM is not being yet destroyed if we've
invoked qemuDomainObjWait().
Use the new helper qemuDomainObjIsActive().
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The helper checks whether VM is active including the internal qemu
state. This helper will become useful in situations when an async job
is in use as VIR_JOB_DESTROY can run along async jobs thus both checks
are necessary.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Prevent the possibility that a VM could be considered as alive while
inside qemuProcessStop.
A recently fixed bug which unlocked the domain object while inside
qemuProcessStop showed that there's possibility to confuse the state of
the VM to be considered active while 'qemuProcessStop' is processing
shutdown of the VM. Ensure that this doesn't happen by clearing the
'beingDestroyed' flag only after the VM id is cleared.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
There are few function calls done while cleaning up a stopped VM which
do require the old VM id, to e.g. clean up paths containing the 'short'
domain name in the path.
Anything else, which doesn't strictly require it can be moved after
clearing the 'id' in order to decrease likelyhood of potential bugs.
This patch moves all the code which does not require the 'id' (except
for the log entry and closing the monitor socket) after the statement
clearing the id and adds a comment explaining that anything in the
section must not unlock the VM object.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
'qemuDomainObjStopWorker()' which is meant to dispose of the event loop
thread for the monitor unlocks the VM object while disposing the thread
to prevent possible deadlocks with events waiting on the monitor thread.
Unfortunately 'qemuDomainObjStopWorker()' is called *before* the VM is
marked as inactive by clearing 'vm->def->id', but at the same time it's
no longer marked as 'beingDestroyed' when we're inside
'qemuProcessStop()'.
If 'vm' would be kept locked this wouldn't be a problem. Same way it's
not a problem for anything that uses non-ASYNC VM jobs, or when the
monitor is accessed in an async job, as the 'destroy' job interlocks
with those.
It is a problem for code inside an async job which uses
'qemuDomainObjWait()' though. The API contract of qemuDomainObjWait()
ensures the caller that the VM on successful return from it, but in this
specific reason it's not the case, as both 'beingDestroyed' is already
false, and 'vm->def->id' is not yet cleared.
To fix the issue move the 'qemuDomainObjStopWorker()' call *after*
clearing 'vm->def->id' and also add a note stating what the function is
doing.
Fixes: 860a999802d3c82538373bb3f314f92a2e258754
Closes: https://gitlab.com/libvirt/libvirt/-/issues/640
Reported-by: luzhipeng <luzhipeng@cestc.cn>
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Document why this function exists and meaning of return values.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Clear the 'disk' member of 'blockjob' as we're freeing the disk object
at this point. While this should not normally happen it was observed
when other bug allowed the VM to be cleared while other threads didn't
yet finish.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Similarly to other blockjob handlers, if there's no disk associated with
the blockjob the handler needs to behave correctly. This is needed as
the disk might have been de-associated on unplug or other operations.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Sometimes in release hook it is useful to know if the VM shutdown was graceful
or not. This is especially useful to do cleanup based on the VM shutdown failure
reason in release hook. This patch proposes to use the last argument 'extra'
to pass VM shutoff reason in the call to release hook.
Making this change for Qemu and LXC.
Signed-off-by: Swapnil Ingle <swapnil.ingle@nutanix.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Resolves: https://issues.redhat.com/browse/RHEL-23833
Signed-off-by: Adam Julis <ajulis@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Although virDomainDeviceDefValidate() is called as a part of
parsing device XML routine, it validates only that single device.
The virDomainDefValidate() function performs a more comprehensive
check. It should detect errors resulting from dependencies
between devices, or a device and some other part of XML config.
Therefore, a call to virDomainDefValidate() is added at the end
of qemuDomainAttachDeviceConfig().
Signed-off-by: Adam Julis <ajulis@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
In one of my previous commits I've made us substract isolcpus
from all online CPUs when setting affinity on QEMU threads. See
commit below for more info on that. Nevertheless, this is
something that surely deserves an entry in log. I've chosen INFO
priority for now. We can promote that to a regular WARN if users
complain.
Fixes: da95bcb6b2d9b04958e0f2603202801dd29debb8
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
We currently hardcode the systemd sysusersdir, but it is desirable to be
able to choose a different location in some cases. For example, Fedora
flatpak builds change the RPM %_sysusersdir macro, but we can't currently
honour that.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Reported-by: Yaakov Selkowitz <yselkowi@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The support will be dropped soon by qemu, and libvirt is not rejecting
such configurations. Add validation of this explicitly requested config.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Everywhere we use TPM 2.0 as our default, the chances of TPM
1.2 being supported by the guest OS are very slim. Just reject
such configurations outright.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
TPM 1.2 is a pretty bad default these days, especially for
architectures which were introduced when TPM 2.0 already existed.
We're already carving out exceptions for several scenarios, but
that's basically backwards: at this point, using TPM 1.2 is the
exception.
Restructure the code so that it reflects reality and we don't
have to remember to update it every time a new architecture is
introduced.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
While __attribute((sentinel)) (exposed by glib under
G_GNUC_NULL_TERMINATED macro) is a gcc extension, it's supported
by clang too. It's already being used throughout our code but
some functions that take variadic arguments and expect NULL at
the end were lacking such annotation. Fill them in.
After this, there are still some functions left untouched because
they expect a different sentinel than NULL. Unfortunately, glib
does not provide macro for different sentinels. We may come up
with our own, but let's save that for future work.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When hot-plugging a FS device with un-assigned address with a bootindex
the recently-added validation check would fail as validation on hotplug
is done prior to address assignment.
To fix this problem we can simply relax the check to also pass on _NONE
addresses. Unsupported configurations will still be caught as previous
commit re-checks the definition after address assignment prior to
hotplug.
Resolves: https://issues.redhat.com/browse/RHEL-39271
Fixes: 4690058b6d3dab672bd18ff69c83392245253024
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Some of the checks make sense only after the address is allocated and
thus we need to re-do the validation after the address is assigned.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
In one of my recent commits, I've introduced
virDomainInterfaceClearQoS() which is a helper that either calls
virNetDevBandwidthClear() ('tc' implementation) or
virNetDevOpenvswitchInterfaceClearQos() (for ovs ifaces). But I
made a micro optimization which leads to a bug: the function
checks whether passed iface has any QoS set and returns early if
it has none. In majority of cases this is right thing to do, but
when removing QoS on virDomainUpdateDeviceFlags() this is
problematic. The new definition (passed as argument to
virDomainInterfaceClearQoS()) contains no QoS (because user
requested its removal) and thus instead of removing the old QoS
setting nothing is done.
Fortunately, the fix is simple - pass olddev which contains the
old QoS setting.
Fixes: 812a146dfe784315edece43d09f8d9e432f8230e
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
The code cleaning up virStorageSource doesn't free data allocated by
virStorageSourceInit() so we need to call virStorageSourceDeinit()
explicitly.
Fixes: 8e664737813378d2a1bdeacc2ca8e942327e2cab
Resolves: https://issues.redhat.com/browse/RHEL-33044
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
The virStateDriver struct has .stateInitialize callback which is
declared to return virDrvStateInitResult enum. But some drivers
return a plain int in their implementation which is UB.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
There was no test for this and we mistakenly used 'B' rather than 'T'
when constructing the json value for this parameter. Thus, a value of
'off' was VIR_TRISTATE_SWITCH_OFF=2, which was translated to a boolean
value of 'true'.
Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Unlike other input types, evdev is not a true device since it's backed by
'-object'. We must use object-add/object-del monitor commands instead of
device-add/device-del in this particular case.
This patch adds support for handling live attachment and
detachment of evdev type devices.
Resolves: https://gitlab.com/libvirt/libvirt/-/issues/529
Signed-off-by: Rayhan Faizel <rayhan.faizel@gmail.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
