This patch adds a new domain lookup helper qemuDomObjFromDomainDriver
that lookups the domain and leaves the driver locked. The driver is
returned as the second argument of that function. If the lookup fails
the driver is unlocked to help avoid cleanup codepaths.
This patch also improves docs for the helpers.
When a qemu domain is backed by huge pages, apparmor needs to grant the domain
rw access to files under the hugetlbfs mount point. Add a hook, called in
qemu_process.c, which ends up adding the read-write access through
virt-aa-helper. Qemu will be creating a randomly named file under the
mountpoint and unlinking it as soon as it has mmap()d it, therefore we
cannot predict the full pathname, but for the same reason it is generally
safe to provide access to $path/**.
Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
This patch exports qemuMigrationIsAllowed and adds a new parameter to it
to denote if it's a remote migration or a local migration. Local
migrations are used in snapshots and saving of the machine state and
have fewer restrictions. This patch also adjusts callers of the function
and tweaks some error messages to be more universal.
Currently there is no way to detect it via QMP and requesting "-sandbox
off" works correctly even if it was compiled out, so this will work
unless someone both requests the sandbox in qemu.conf and builds QEMU
without the support for it.
These classes can borrow unused bandwidth. Basically,
only egress qdsics can have classes, therefore we can
do this kind of traffic shaping only on host's outgoing,
that is domain's incoming traffic.
When the disk snapshot part of an external system checkpoint fails the
memory image is retained. This patch adds code to remove the image in
such case.
In case the snapshot code isn't able to restart CPUs after an external
checkpoint we would leak a copy of the domains XML definition. This
patch fixes the cleanup path.
False positive, but it breaks the build with gcc-4.6.3.
qemu/qemu_migration.c:2931:37: error: 'offline' may be used
uninitialized in this function [-Werror=uninitialized]
qemu/qemu_migration.c:2887:10: note: 'offline' was declared here
When restarting CPUs after an external snapshot, the restarting function
was called without the appropriate async job type. This caused that a
new sync job wasn't created and allowed races in the monitor.
Offline migration transfers inactive definition of a domain (which may
or may not be active). After successful completion, the domain remains
in its current state on source host and is defined but inactive on
destination host. It's a bit more clever than virDomainGetXMLDesc() on
source host followed by virDomainDefineXML() on destination host, as
offline migration will run pre-migration hook to update the domain XML
on destination host. Currently, copying non-shared storage is not
supported during offline migration.
Offline migration can be requested with a new migration flag called
VIR_MIGRATE_OFFLINE (which has to be combined with
VIR_MIGRATE_PERSIST_DEST flag).
This fixes a problem that showed up during testing of:
https://bugzilla.redhat.com/show_bug.cgi?id=881480
Due to a logic error in the function that gets the name of the bridge
an interface connects to, any time a bridge was specified directly
(type='bridge') rather than indirectly (type='network'), An error
would be logged (although the operation would then complete
successfully):
Network type 6 is not supported
The final virReportError() in the function
qemuDomainNetGetBridgeName() was apparently avoided in the past with a
"goto cleanup" at the end of each case, but the case of bridge somehow
no longer has that final goto cleanup.
The proper solution is anyway to not rely on goto's, but put the error
log inside an else {} clause, so that it's executed only if the type
is neither bridge nor network (in reality, this function should only
ever be called for those two types, that's why this is an internal
error).
While making this change, the error message was also tuned to be more
correct (since it's not really the type of the network, but the type
of the interface, and it *is* otherwise supported, it's just that the
interface type in question doesn't *have* a bridge device associated
with it, or at least we don't know how to get it).
If a network interface model is not specified, libvirt will run
into an unchecked NULL pointer coredump. On the other hand if
the empty model is ignored, a PCI bus address would be generated,
which is not supported by S390.
Since the only valid network type model for S390 is virtio,
we use this as the default value, which is the same for QEMU.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Things are supposed to look like:
<machine canonical='pc-0.12'>pc</machine>
But are currently swapped. This can cause many VMs to revert to having
machine type='pc' which will affect save/restore across qemu upgrades.
QEMU supports setting vendor and product strings for disk since
1.2.0 (only scsi-disk, scsi-hd, scsi-cd support it), this patch
exposes it with new XML elements <vendor> and <product> of disk
device.
Unmanaged PCI devices were only leaked if pciDeviceListAdd failed but
managed devices were always leaked. And leaking PCI device is likely to
leave PCI config file descriptor open. This patch fixes
qemuReattachPciDevice to either free the PCI device or add it to the
inactivePciHostdevs list.
An attempt to attach device that is already attached to a domain results
in the following error:
virsh # attach-device rhel6 pci2 --persistent
error: Failed to attach device from pci2
error: invalid argument: device is already in the domain configuration
The "invalid argument" error code looks wrong, we usually use "operation
invalid" when the action cannot be done in current state.
Only one error in qemu_monitor was already using the relatively
new OPERATION_UNSUPPORTED error, even though it is a better fit
for all of the messages related to options that are unsupported
due to the version of qemu in use rather than due to a user's
XML or .conf file choice. Suggested by Osier Yang.
* src/qemu/qemu_monitor.c (qemuMonitorSendFileHandle)
(qemuMonitorAddHostNetwork, qemuMonitorRemoveHostNetwork)
(qemuMonitorAttachDrive, qemuMonitorDiskSnapshot)
(qemuMonitorDriveMirror, qemuMonitorTransaction)
(qemuMonitorBlockCommit, qemuMonitorDrivePivot)
(qemuMonitorBlockJob, qemuMonitorSystemWakeup)
(qemuMonitorGetVersion, qemuMonitorGetMachines)
(qemuMonitorGetCPUDefinitions, qemuMonitorGetCommands)
(qemuMonitorGetEvents, qemuMonitorGetKVMState)
(qemuMonitorGetObjectTypes, qemuMonitorGetObjectProps)
(qemuMonitorGetTargetArch): Use better error category.
Without this patch, attempts to create a disk snapshot when qemu
is too old results in a cryptic message:
virsh # snapshot-create 23 --disk-only
error: operation failed: Failed to take snapshot: unknown command: 'snapshot_blkdev'
Now it reports:
virsh # snapshot-create 23 --disk-only
error: unsupported configuration: live disk snapshot not supported with this QEMU binary
All versions of qemu that support live disk snapshot also support
QMP (basically upstream qemu 1.1 and later, and backports to RHEL 6.2).
* src/qemu/qemu_capabilities.h (QEMU_CAPS_DISK_SNAPSHOT): New
capability.
* src/qemu/qemu_capabilities.c (qemuCaps): Track it.
(qemuCapsProbeQMPCommands): Set it.
* src/qemu/qemu_driver.c (qemuDomainSnapshotCreateDiskActive): Use
it.
* src/qemu/qemu_monitor.c (qemuMonitorDiskSnapshot): Simplify.
* src/qemu/qemu_monitor_json.c (qemuMonitorJSONDiskSnapshot):
Likewise.
* src/qemu/qemu_monitor_text.h (qemuMonitorTextDiskSnapshot):
Delete.
* src/qemu/qemu_monitor_text.c (qemuMonitorTextDiskSnapshot):
Likewise.
Currently to deal with auto-shutdown libvirtd must periodically
poll all stateful drivers. Thus sucks because it requires
acquiring both the driver lock and locks on every single virtual
machine. Instead pass in a "inhibit" callback to virStateInitialize
which drivers can invoke whenever they want to inhibit shutdown
due to existance of active VMs.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Since we can't (currently) rely on the ability to provide blanket
support for all possible network changes by calling the toplevel
netdev hostside disconnect/connect functions (due to qemu only
supporting a lockstep between initialization of host side and guest
side of devices), in order to support live change of an interface's
nwfilter we need to make a special purpose function to only call the
nwfilter teardown and setup functions if the filter for an interface
(or its parameters) changes. The pattern is nearly identical to that
used to change the bridge that an interface is connected to.
This patch was inspired by a request from Guido Winkelmann
<guido@sagersystems.de>, who tested an earlier version.
The fact that only the guest agent, or ACPI flag can be used
when requesting reboot/shutdown is merely a limitation of the
QEMU driver impl at this time. Thus it should not be in
libvirt.c code
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The default machine type must be stored in the first element of
the caps->machineTypes array. This was done for help output
parsing but not for QMP probing.
Added a helper function qemuSetDefaultMachine to apply the same
fix up for both probing methods.
Further, it was necessary to set caps->nmachineTypes after QMP
probing.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
https://bugzilla.redhat.com/show_bug.cgi?id=872292
Libvirt should not attempt to call a QMP command that has not been
documented in qemu.git - if future qemu introduces a command by the
same name but with subtly different semantics, then libvirt will be
broken when trying to use that command.
We also had some code that could never be reached - some of our
commands have an alternate for new vs. old qemu HMP commands; but
if we are new enough to support QMP, we only need a fallback to
the new HMP counterpart, and don't need to try for a QMP counterpart
for the old HMP version.
See also this attempt to convert the three snapshot commands to QMP:
https://lists.gnu.org/archive/html/qemu-devel/2012-07/msg01597.html
although it looks like that will still not happen before qemu 1.3.
That thread eventually decided that qemu would use the name
'save-vm' rather than 'savevm', which mitigates the fact that
libvirt's attempt to use a QMP 'savevm' would be broken, but we
might not be as lucky on the other commands.
* src/qemu/qemu_monitor_json.c (qemuMonitorJSONSetCPU)
(qemuMonitorJSONAddDrive, qemuMonitorJSONDriveDel)
(qemuMonitorJSONCreateSnapshot, qemuMonitorJSONLoadSnapshot)
(qemuMonitorJSONDeleteSnapshot): Use only HMP fallback for now.
(qemuMonitorJSONAddHostNetwork, qemuMonitorJSONRemoveHostNetwork)
(qemuMonitorJSONAttachDrive, qemuMonitorJSONGetGuestDriveAddress):
Delete; QMP implies QEMU_CAPS_DEVICE, which prefers AddNetdev,
RemoveNetdev, and AddDrive anyways (qemu_hotplug.c has all callers).
* src/qemu/qemu_monitor.c (qemuMonitorAddHostNetwork)
(qemuMonitorRemoveHostNetwork, qemuMonitorAttachDrive): Reflect
deleted commands.
* src/qemu/qemu_monitor_json.h (qemuMonitorJSONAddHostNetwork)
(qemuMonitorJSONRemoveHostNetwork, qemuMonitorJSONAttachDrive):
Likewise.
https://bugzilla.redhat.com/show_bug.cgi?id=876828
Commit 38c4a9cc introduced a regression in hot unplugging of disks
from qemu, where cgroup device ACLs were no longer being revoked
(thankfully not a security hole: cgroup ACLs only prevent open()
of the disk; so reverting the ACL prevents future abuse but doesn't
stop abuse from an fd that was already opened before the ACL change).
Commit 1b2ebf95 overlooked that there were two spots affected.
* src/qemu/qemu_hotplug.c (qemuDomainDetachDiskDevice):
Transfer backing chain before deletion.
* src/qemu/qemu_driver.c (qemuDomainDetachDeviceDiskLive): Fix
spacing (partly to ensure a different-looking patch).
This patch adds two labels and gets rid of a ton of duplicated code.
This patch also fixes some error message and switches most of them to
proper error reporting functions.
This patch adds macros to help retrieve configuration values from qemu
driver's configuration. Some configuration options are grouped
together in the process.
The virStateInitialize method and several cgroups methods were
using an 'int privileged' parameter or similar for dual-state
values. These are better represented with the bool type.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
As of 1a50ba2cb0 we fail to connect to the
monitor instead of getting an exit status != 0 from qemu itself. This
breaks capabilities probing for the non QMP case.
Replace the following names
* struct qemu_snap_remove with virQEMUSnapRemovePtr
* struct qemu_snap_reparent with virQEMUSnapReparentPtr
* struct qemu_save_header with virQEMUSaveHeaderPtr
* enum qemu_save_formats with virQEMUSaveFormat
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Remove the obsolete 'qemud' naming prefix and underscore
based type name. Introduce virQEMUDriverPtr as the replacement,
in common with LXC driver naming style
Throughout the code, we've always used VIR_DOMAIN_SHUTDOWN* flags
even for virDomainReboot() API and its implementation. Fortunately,
the appropriate macros has the same value. But if we want to keep
things consistent, we should be using the correct macros. This
patch doesn't break anything, luckily.
Error messages produced while dispatching guest agent commands didn't
have an apparent reference to the fact that they are dealing with guest
agent commands. This patch fixes up some of the messages to contain that
reference.
using qemu guest agent. As said in previous patch,
@mountPoint must be NULL and @flags zero because
qemu guest agent doesn't support these arguments
yet. If qemu learns them, we can start supporting
them as well.
With QMP capability probing, the version was not set.
virsh version returns:
...
Cannot extract running QEMU hypervisor version
This is fixed by computing caps->version from QMP major,
minor, micro values.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
QMP Capability probing will fail if QEMU cannot bind to the
QMP monitor socket in the qemu_driver->libDir directory.
That's because the child process is stripped of all
capabilities and this directory is chown'ed to the configured
QEMU user/group (normally qemu:qemu) by the QEMU driver.
To prevent this from happening, the driver startup will now pass
the QEMU uid and gid down to the capability probing code.
All capability probing invocations of QEMU will be run with
the configured QEMU uid instead of libvirtd's.
Furter, the pid file handling is moved to libvirt, as QEMU
cannot write to the qemu_driver->runDir (root:root). This also
means that the libvirt daemonizing must be used.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
If qemuMonitorOpenUnix is called without a related pid, i.e. for
QMP probing, a connect failure can happen as the result of a race.
Without a pid there is no retry and thus we give up too early.
This changes the code to retry if no pid is supplied.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Change some legacy function names to use 'qemu' as their
prefix instead of 'qemud' which was a hang over from when
the QEMU driver ran inside a separate daemon
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=876828
Commit 38c4a9cc introduced a regression in hot unplugging of disks
from qemu, where cgroup device ACLs were no longer being revoked
(thankfully not a security hole: cgroup ACLs only prevent open()
of the disk; so reverting the ACL prevents future abuse but doesn't
stop abuse from an fd that was already opened before the ACL change).
The actual regression is due to a latent bug. The hot unplug code
was computing the set of files needing cgroup ACL revocation based
on the XML passed in by the user, rather than based on the domain's
details on which disk was being deleted. As long as the revoke
path was always recomputing the backing chain, this didn't really
matter; but now that we want to compute the chain exactly once and
remember that computation, we need to hang on to the backing chain
until after the revoke has happened.
* src/qemu/qemu_hotplug.c (qemuDomainDetachPciDiskDevice):
Transfer backing chain before deletion.
