When doing a peer-to-peer migration it may happen that the
connection to the destination disappears. If that happens,
there's no point in trying to unregister the close callback
because the connection is closed already. It results only in
polluting logs with this message:
error : virNetSocketReadWire:1814 : End of file while reading data: : Input/output error
and the reason for that is unregistering a connection callback
results in RPC (among other things).
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1918211
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
qemuMigrationSrcRunPrepareBlockDirtyBitmaps receives the flags parameter
from qemuMigrationSrcRun, where flags are based on the main API enum
values. Similar to commit f58349c9c6, use the main API enum instead of
internal driver enum when checking flags in
qemuMigrationSrcRunPrepareBlockDirtyBitmaps.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Signaling the condition before vm->def->id is reset to -1 is dangerous:
in case a waiting thread wakes up, it does not see anything interesting
(the domain is still marked as running) and just enters virDomainObjWait
where it waits forever because the condition will never be signalled
again.
Originally it was impossible to get into such situation because the vm
object was locked all the time between signaling the condition and
resetting vm->def->id, but after commit 860a999802 released in 6.8.0,
qemuDomainObjStopWorker called in qemuProcessStop between
virDomainObjBroadcast and setting vm->def->id to -1 unlocks the vm
object giving other threads a chance to wake up and possibly hang.
In real world, this can be easily reproduced by killing, destroying, or
just shutting down (from the guest OS) a domain while it is being
migrated somewhere else. The migration job would never finish.
So let's make sure we delay signaling the domain condition to the point
when a woken up thread can detect the domain is not active anymore.
https://bugzilla.redhat.com/show_bug.cgi?id=1949869
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
This simplyfies the code a bit and removes one "goto", one "VIR_FREE",
and one "VIR_INSERT_ELEMENT_COPY".
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
If the queried QMP command doesn't exist qemuMonitorGetTPMModels returns
0 but sets the string list to NULL which isn't accepted by
g_strv_contains.
Fixes: a5bc5f0ecf8
Reported-by: Olaf Hering <olaf@aepfle.de>
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
If the attempt to attach a device failed, we erased the
unattached device from the namespace. This resulted in erasing an
already attached device in case of a duplicate. We need to check
for existing file in the namespace in order to determine erasing
it in case of a failure.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1780508
Signed-off-by: Kristina Hanicova <khanicov@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
`virHashNew` cannot return NULL, the check is not needed.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The 'storageMigration' flag is supposed to be true if storage migration
is requested, which is based on VIR_MIGRATE_NON_SHARED_DISK or
VIR_MIGRATE_NON_SHARED_INC flags. The assignment to the variable used
QEMU_MONITOR_MIGRATE_NON_SHARED_INC (0x04) instead of
VIR_MIGRATE_NON_SHARED_INC (0x80), caused libvirtd to skip the actual
copy of data.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1978526
Fixes: da69f4b2084bff140238e450e264d6036ebef898
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Remember whether the user passed an explicit index when registering the
event so that we can avoid the top level event when it isn't needed.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
When users register the threshold event for the top level image with an
explicit index (e.g. vda[3]) they are clearly expecting the index in the
event.
This flag will help avoiding emission of the second event without the
index when the client clearly requested one with the index.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
When qos is set or delete, we have to check if the port is an ovs managed
port. If true, call the virNetDevOpenvswitchInterfaceSetQos function when qos
is set, and call the virNetDevOpenvswitchInterfaceClearQos function when
the interface is to be destroyed.
Signed-off-by: Jinsheng Zhang <zhangjl02@inspur.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Return 0 directly if the port is ovs managed. When the ovs port is set
noqueue, qos config on this port will not work.
Signed-off-by: Jinsheng Zhang <zhangjl02@inspur.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
When seeing a guest with a sound device, and no audio backend, we
automatically add an audio backend XML element based on the historical
QEMU driver behaviour. Unfortunately when we live migrate back to an
old libvirt, it may not understand the audio driver type we configured.
We thus need to strip the default audio backend when migrating.
Fixes https://gitlab.com/libvirt/libvirt/-/issues/179
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
If guest is configured to use memfd then the function that build
memory-backend-* part of command line will put
memory-backend-memfd, always. Even for NVDIMMs. This is not
correct, because NVDIMMs need a backing path (usually to a real
host NVDIMM device). Therefore, regardless of memfd being
requested, we have to stick with memory-backend-file.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
We already reject TPM 1.2 in a number of scenarios; let's add
ARM virt guests to the list.
https://bugzilla.redhat.com/show_bug.cgi?id=1970310
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Tested-by: Liu Yiding <liuyd.fnst@fujitsu.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The TPM 2.0 specification predates ARM virtualization, and so
implementing TPM 1.2 support on ARM was not considered a useful
endeavor.
This is technically a breaking change, but TPM support on ARM was
only introduced fairly recently (libvirt 7.1.0) and the previous
default resulted in non working TPM devices; anyone who has a
working configuration is not going to be affected.
https://bugzilla.redhat.com/show_bug.cgi?id=1970310
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Tested-by: Liu Yiding <liuyd.fnst@fujitsu.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Originally qemuDomainAttachNetDevice() would wait until the cleanup at
the very end of the function to add newly hotplugged interfaces to the
domain's nets list. commit 7b8bec4560 modified it to add the new
interface to the nets list earlier (but not all the way at the
beginning of the function either, because there are some operations
(PCI address assignment in particular) that need the new device to not
yet be visible in the domaindef).
But hostdev interfaces short-circuit past most of the body of
qemuDomainAttachNetDevice() (since none of it applies to hostdev
interfaces). In the past that was okay, but since the line that adds
the new interface to the domaindef's nets list is in that "most of the
body", after that commit hotplugged hostdev interfaces are no longer
being properly added to the domaindef nets list, so they don't show up
in the status XML or the virsh domiflist output.
It really *is* important to add interfaces to the nets list earlier,
so we can't revert commit 7b8bec4560, and we also can't move the
insert to common code *earlier* in the function, so instead this patch
duplicates the VIR_APPEND_ELEMENT_COPY() just before the code path for
hostdev interfaces jumps to cleanup.
Resolves: https://bugzilla.redhat.com/1972468
Fixes: 7b8bec45601b6570f6a7413e94d291986d2663f1
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
When writing the memory snapshot into an existing file don't remove it
if the snapshot fails later.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
'file' is too generic to know what's going on. Rename it to
'memorysnapshotfile'.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Use the snapshot disk type from the definition now that we validate that
it matches.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The code executed later when creating a snapshot makes all decisions
based on the configured type rather than the actual type of the existing
file, while the check whether the file exists is based solely on the
on-disk type.
Since a block device is allowed to exist even when not reusing existing
files in contrast to regular files this creates a potential for a block
device to squeak past the check but then be influenced by other code
executed later. Specifically this is a problem when creating a snapshot
with the following XML:
<domainsnapshot>
<disks>
<disk name='vdb' type='file'>
<source file='/dev/sdb'/>
</disk>
</disks>
</domainsnapshot>
If the snapshot creation fails, '/dev/sdb' will be removed because it's
considered to be a regular file by the cleanup code.
Add a check that will force that the configured type matches the on-disk
state.
Additional supporting reason is that qemu stopped to accept block
devices with the 'file' backend, thus the above configuration will not
work any more. This allows us to fail sooner.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1972145
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In case when the snapshot target is of VIR_STORAGE_TYPE_BLOCK type and
doesn't exist libvirt won't be able to create it. Reject such a config
sooner.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Separate the 'else if' branches into nested conditions so that it's more
obvious when we'll be adding additional checks later.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
virQEMUCapsFillDomainDeviceGraphicsCaps fills data needed both for
validation of the graphics type and also for correct display in the
(dom)capablities XML.
Signal the support for egl-headless only when qemu has the capability.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
egl-headless graphics can be compiled out in qemu so we need to be able
to know whether the given qemu version support it.
Base the capability on the presence of the 'egl-headless' member in
'query-display-options' or imply it if 'query-display-options' is not
supported as we implied it before for all versions.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
The DAC security driver has an option to register a callback that
is called instead of chown(). So far QEMU is the only user of
this feature and it's used to set labels on non-local disks (like
gluster), where exists notion of owners but regular chown() can't
be used.
However, this callback (if set) is called always, even for local
disks. And thus the QEMU's implementation duplicated parts of the
DAC driver to deal with chown().
If the DAC driver would call the callback only for non-local
disks then the QEMU's callback can be shorter.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The NVRAM label is set in qemuSecuritySetAllLabel(). There's no
need to set its label upfront. In fact, setting it twice creates
an imbalance because it's unset only once which mangles seclabel
remembering. However, plain removal of the
qemuSecurityDomainSetPathLabel() undoes the fix for the original
bug (when dynamic ownership is off then the NVRAM is not created
with cfg->user and cfg->group but as root:root). Therefore, we
have to switch to virFileOpenAs() and pass cfg->user and
cfg->group and VIR_FILE_OPEN_FORCE_OWNER flag. There's no need to
pass VIR_FILE_OPEN_FORCE_MODE because the file will be created
with the proper mode.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1969347
Fixes: bcdaa91a27b5b2d103535270a6a287efe6cd8bfb
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
