Pointed out during review on one or two places, but it actually appears in lot
more places. So let's be consistent.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
After processing the processEvent->data for a qemuProcessEventHandler
callout, it's expected that the called processEvent->eventType helper
will perform the proper free on the data field. In this case it's
a qemuMonitorEventPanicInfoPtr.
We've been building up to this. This adds support for cputune/cachetune
settings for domains in the QEMU driver. The addition into
qemuProcessSetupVcpu() automatically adds support for hotplug. For hot-unplug
we need to remove the allocation only if all the vCPUs were unplugged. But
since the threads are left running, we can't really do much about it now.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1289368
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This wires up the previously added OEM strings XML schema to be able to
generate comamnd line args for QEMU. This requires QEMU >= 2.12 release
containing this patch:
commit 2d6dcbf93fb01b4a7f45a93d276d4d74b16392dd
Author: Daniel P. Berrange <berrange@redhat.com>
Date: Sat Oct 28 21:51:36 2017 +0100
smbios: support setting OEM strings table
Reviewed-by: John Ferlan <jferlan@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
We recently added a generic XHCI USB3 controller to QEMU, and libvirt
supports adding that controller rather than the NEC XHCI USB3
controller, but when auto-adding a USB controller to Q35 domains we
were still adding the vendor-specific NEC controller. This patch
changes to add the generic controller instead, if it's available in
the QEMU binary that will be used.
Signed-off-by: Laine Stump <laine@laine.org>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
Whenever a different kernel is booted, some capabilities related to KVM
(such as CPUID bits) may change. We need to refresh the cache to see the
changes.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
qemuDomainDefValidateVideo() (called from qemuDomainDefValidate()) is
just a loop performing various checks on each video device. Rather
than maintaining this separate function, just fold the validations
into qemuDomainDeviceDefValidateVideo(), which is called once for each
video device.
Commit 10c73bf1 fixed a bug that I had introduced back in commit
70249927 - if a vhost-scsi device had no manually assigned PCI
address, one wouldn't be assigned automatically. There was a slight
problem with the logic of the fix though - in the case of domains with
pcie-root (e.g. those with a q35 machinetype),
qemuDomainDeviceCalculatePCIConnectFlags() will attempt to determine
if the host-side PCI device is Express or legacy by examining sysfs
based on the host-side PCI address stored in
hostdev->source.subsys.u.pci.addr, but that part of the union is only
valid for PCI hostdevs, *not* for SCSI hostdevs. So we end up trying
to read sysfs for some probably-non-existent device, which fails, and
the function virPCIDeviceIsPCIExpress() returns failure (-1).
By coincidence, the return value is being examined as a boolean, and
since -1 is true, we still end up assigning the vhost-scsi device to
an Express slot, but that is just by chance (and could fail in the
case that the gibberish in the "hostside PCI address" was the address
of a real device that happened to be legacy PCI).
Since (according to Paolo Bonzini) vhost-scsi devices appear just like
virtio-scsi devices in the guest, they should follow the same rules as
virtio devices when deciding whether they should be placed in an
Express or a legacy slot. That's accomplished in this patch by
returning early with virtioFlags, rather than erroneously using
hostdev->source.subsys.u.pci.addr. It also adds a test case for PCIe
to assure it doesn't get broken in the future.
https://bugzilla.redhat.com/show_bug.cgi?id=1536461
This reverts commit aeda1b8c56.
Problem is that we need mon->lastError to be set because it's
used all over the place. Also, there's nothing wrong with
reporting error if one occurred. I mean, if there's a thread
executing an API and which currently is talking on monitor it
definitely wants the error reported.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
When migrating a shutoff domain (i.e., offline migration), we have no
statistics to report and thus jobInfo will be NULL in
qemuMigrationFinish.
Broken by me in v3.10.0-183-ge8784e7868.
https://bugzilla.redhat.com/show_bug.cgi?id=1536351
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
We read from QEMU until seeing a \r\n pair to indicate a completed reply
or event. To avoid memory denial-of-service though, we must have a size
limit on amount of data we buffer. 10 MB is large enough that it ought
to cope with normal QEMU replies, and small enough that we're not
consuming unreasonable mem.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Commit 7a931a4204 refactored the code and probably forgot to add
this line.
Signed-off-by: Marc Hartmayer <mhartmay@linux.vnet.ibm.com>
Reviewed-by: Boris Fiuczynski <fiuczy@linux.vnet.ibm.com>
Add a check if it's a iSCSI hostdev and if it's not then don't use the
union member 'iscsi'. The segmentation fault occured when accessing
secinfo->type, but this can vary from case to case.
Signed-off-by: Marc Hartmayer <mhartmay@linux.vnet.ibm.com>
Reviewed-by: Bjoern Walk <bwalk@linux.vnet.ibm.com>
Reviewed-by: Boris Fiuczynski <fiuczy@linux.vnet.ibm.com>
Libvirt 3.7.0 and earlier libvirt reported a migration job as completed
immediately after QEMU finished sending migration data at which point
migration was not really complete yet. Commit v3.7.0-29-g3f2d6d829e
fixed this, but caused a regression in reporting statistics for
completed jobs which started reporting the job as still running. This
happened because the completed job statistics including the job status
are copied from the running job before we finally mark it as completed.
Let's make sure QEMU_DOMAIN_JOB_STATUS_COMPLETED is always set in the
completed job info even when the job has not finished yet.
https://bugzilla.redhat.com/show_bug.cgi?id=1523036
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
When reconnecting to a running domain with host-model CPU started by old
libvirt which did not store the actual CPU in the status XML, we need to
ignore the fallback attribute to make sure we can translate the detected
host CPU model to a model which is supported by the running QEMU.
https://bugzilla.redhat.com/show_bug.cgi?id=1532980
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
virSecurityManagerDomainSetPathLabel is used to make a path known
to the security modules, but today is used interchangably for
- paths to files/dirs to be accessed directly
- paths to a dir, but the access will actually be to files therein
Depending on the security module it is important to know which of
these types it will be.
The argument allowSubtree augments the call to the implementations of
DomainSetPathLabel that can - per security module - decide if extra
actions shall be taken.
For now dac/selinux handle this as before, but apparmor will make
use of it to add a wildcard to the path that was passed.
Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1527740
Users might use a block device as UEFI VAR store. Or even have
OVMF stored there. Therefore, when starting a domain and separate
mount namespace is used, we have to create all the /dev entries
that are configured for the domain.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
Commit id '162efa1a' added support hotplug a redirdev, but
did not add the hot unplug. This patch will add that support
to allow usage of the detach-device --live on the device.
Reviewed-by: John Ferlan <jferlan@redhat.com>
Signed-off-by: Chen Hanxiao <chenhanxiao@gmail.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1528502
So imagine you have /dev/blah symlink which points to /dev/sda.
You attach /dev/blah as disk to your domain. Libvirt correctly
creates the /dev/blah -> /dev/sda symlink in the qemu namespace.
However, then you detach the disk, change the symlink so that it
points to /dev/sdb and tries to attach the disk again. This time,
however, the attach fails (well, qemu attaches wrong disk)
because the code assumes that symlinks don't change. Well they
do.
This is inspired by test fix written by Eduardo Habkost.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
When the -machine pseries,max-cpu-compat=X is supported use
machine parameter instead of -cpu host,compat=X parameter as
that is deprecated now with qemu >= v2.10.
Fixes https://bugzilla.redhat.com/show_bug.cgi?id=1519146
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Since we have user aliases it may happen that users want to
change it using 'update-device'. Instead of ignoring it silently,
error out loudly. Note that we don't limit the check just for
"ua-" prefixes because users might try to change libvirt
generated aliases too.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
The qemuMonitorJSONMakeCommand can properly handle a NULL string
by using the "S:" parameter instead of "s:", so let's use that
of having in if/else condition that only adds the "s:".
A microcode update can cause the CPUID bits to change; an example
from the past was the update that disabled TSX on several Haswell
and Broadwell machines.
Therefore, place microcode version in the virQEMUCaps struct and
XML, and rebuild the cache if the versions do not match.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
When qemuDomainFindOrCreateSCSIDiskController adds a controller,
let's use the same model as a currently found controller under the
assumption that the reason to add the controller in hotplug is
because virDomainHostdevAssignAddress determined that there were
too many devices on the existing controller, but only assigned a
new controller index and did not add a new controller and we
desire to use the same controller model as any existing controller
and not take a chance that qemuDomainSetSCSIControllerModel would
use a default that may be incompatible.
All calls to virDomainAuditCgroupPath() were passing 'rc == 0' as
argument, when it was supposed to pass the 'rc' value directly.
As a consequence, the audit events that were supposed to be
logged (actual cgroup changes) were never being logged, and bogus
audit events were logged when using regular files as disk image.
Fix all calls to use the return value of
virCgroup{Allow,Deny}Device*() directly as the 'rc' argument.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1448149
If a domain has no numa nodes, that means we don't put any
memory-backend-file onto the qemu command line. That in turn
means we can't set access='shared'. Therefore, we should produce
an error instead of ignoring the setting silently.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
The PROBE macro used in qemuMonitorIOProcess and the VIR_DEBUG message
in qemuMonitorJSONIOProcess create a lot of logging churn when debug
logging is enabled during monitor communication.
The messages logged from the PROBE macro are rather useless since they
are reporting the partial state of receiving the reply from qemu. The
actual full reply is still logged in qemuMonitorJSONIOProcessLine once
the full message is received.
QEMU 2.7 and newer don't allow guests to start unless the initial
vCPUs count is a multiple of the vCPU hotplug granularity, so
validate it and report an error if needed.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1283700
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
While at the moment we're only performing a single check that is
connected to vCPU hotplugging, we're going to introduce a second
one soon. Move the topology check underneath the capability check
to make that easier; since, after this change, the 'topologycpus'
variable doesn't need to have function scope, we move its
declaration to the inner scope as well.
The comments around the check are modified in order to explain
the different QEMU versions involved.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Similar to qemuDomainAddChardevTLSObjects let's move the chardev
source must be TCP and it has the @haveTLS flag set checks before
trying to delete the TLS objects.
For the Chr device this represents no change; however, for RNG device
this is an additionaly check that was missed in commit id '68808516'.
Before adding the objects, TCP and haveTLS are checked.
Let's make a comment deletion helper similar to the Add helper
that can be called after the ExitMonitor.
The modify qemuDomainRemoveChrDevice and qemuDomainRemoveRNGDevice
to call the helper instead of inlining the copy and pasted code.
ebtables/iptables processing is skipped for any interface connected to
Open vSwitch (they have their own packet filtering), likewise for
midonet (according to
http://blog.midokura.com/2016/04/midonet-rule-chains), but libvirt
would allow adding a <filterref> to interfaces connected in these
ways, so the user might mistakenly believe they were being protected.
This patch checks for a non-NULL <virtualport> element for an
interface (or its network) and logs an error if <virtualport> and
<filterref> are both present. This could cause some previously working
domains to no longer start, but that's really the whole point of this
patch - to warn people that their filterref isn't protecting them as
they might have thought.
I don't bother checking this during post-parse validation, because
such a check would be incomplete - it's possible that a network would
have a <virtualport> that would be applied to an interface, and you
can't know that until the domain is started.
Resolves: https://bugzilla.redhat.com/1502754
When the <bandwidth> of an interface is changed with update-device,
the old settings are cleared with tc, then new settings added with
tc. But if the <bandwidth has been removed, the old settings weren't
being removed, so the bandwidth restrictions would still be active on
the interface although the interface status in libvirt showed that
they had been removed.
This patch fixes it by calling virNetDevBandwidthClear() if the
"modification" to the interface bandwidth was to completely clear
it.
An alternative could have been to modify virNetDevBandwidthSet() to
always clear existing bandwith settings at the beginning of the
function (currently it short circuits in that case, doing nothing),
but that would have led to cases where virNetDevBandwidthClear() was
now being called in cases where it previously wasn't, and while many
of those cases would be NOPs, there could be cases where it would
cause an error. The way this patch works, the ...Clear() function is
only called in cases where the ...Set() function had previously been
called successfully, so the risk of regression is minimized.
Resolves: https://bugzilla.redhat.com/1454709
Also call qemuDomainRemoveInputDevice if we receive the
event after the Detach API ends.
Commit 67486bb failed to include this.
https://bugzilla.redhat.com/show_bug.cgi?id=1524837
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
VM drivers may need to store additional private data to the status XML
so that it can be restored after libvirtd restart. Since not everything
is needed add a callback infrastructure, where VM drivers can add only
stuff they need.
Note that the private data is formatted as a <privateData> sub-element
of the <disk> or <backingStore> <source> sub-element. This is done since
storing it out of band (in the VM private data) would require a complex
matching process to allow to put the data into correct place.
Commit id '70249927b' neglected to cover this case because the test
had taken the "shortcut" to already add the <address>; however, when
the PCI address assignment code was adjusted by commit id '70249927'
the vhost-scsi (VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_SCSI_HOST) wasn't
covered thus returning a 0 for pciFlags. So I altered the tests too
to make sure it doesn't happen again.
Previously the qemuxml2xmloutdata was a softlink to the source
qemuxml2argvdata, so I unlinked and recreated the output file to
force generation of the adddress. Without the test changes, an
address generation returns:
libvirt: Domain Config error : internal error: Cannot automatically
add a new PCI bus for a device with connect flags 00
if an address was supplied in the test, a restart of libvirtd or
edit of a guest would display the following opaque message:
warning : qemuDomainCollectPCIAddress:1237 :
qemuDomainDeviceCalculatePCIConnectFlags() thinks that the device
with PCI address 0000:00:09.0 should not have a PCI address
where the address is related to the guest PCI address provided.
Commit id 'c5c96545' neglected to validate that the srcPriv was
non-NULL before dereferencing. Similar problem to what was fixed
by commit id '8056721c' but missed during multiple rebases and
code reworks.
ncpus would be -1 on error and the cleanup for loop would not be skipped
in this case.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1522706
If domain is active, but the undefine API was called without the
VIR_DOMAIN_UNDEFINE_KEEP_NVRAM flag set, the following incorrect
error message is produced:
error: Requested operation is not valid: cannot delete inactive domain with nvram
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Move the IDE controller check from command line building to
controller def validation. Also explicitly include the avoidance
check for the implicit IDE controller from qemuBuildSkipController.
Cause the IDE case for command line building to generate a
failure if called to add an IDE since that shouldn't happen
if the Validate code did the right thing.
Move the call to qemuDomainCheckCCWS390AddressSupport from
qemuBuildControllerDevStr to qemuDomainDeviceDefValidateController.
This means we will get the qemuCaps from the driver opaque
variable passed to qemuDomainDeviceDefValidate.
When reconnecting to a running domain started by old libvirt, which did
not change host-model into a custom CPU definition, we replace the CPU
definition with a specific CPU model from host capabilities. However,
that CPU model may not be supported by the running qemu process. We need
to translate the CPU model to one of the models which libvirt could have
used when starting the domain.
https://bugzilla.redhat.com/show_bug.cgi?id=1521202
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
virQEMUCapsProbeQMPCPUDefinitions is now a small wrapper which fills in
qemuCaps with CPU models fetched by virQEMUCapsFetchCPUDefinitions.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
