V2 of the libxl soft reset patch, which was pushed as commit da4b0fd9,
dropped the hunk that disposed of the libxl_domain_config object. Add
the missing hunk to properly dispose the object.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
The pvops Linux kernel implements machine_ops.crash_shutdown as
static void xen_hvm_crash_shutdown(struct pt_regs *regs)
{
native_machine_crash_shutdown(regs);
xen_reboot(SHUTDOWN_soft_reset);
}
but currently the libxl driver does not handle the soft reset
shutdown event. As a result, the guest domain never proceeds
past xen_reboot(), making it impossible for HVM domains to save
a crash dump using kexec.
This patch adds support for handling the soft reset event by
calling libxl_domain_soft_reset() and re-enabling domain death
events, which is similar to the xl tool handling of soft reset
shutdown event.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
There are too many goto labels in libxlDomainShutdownThread. Convert the
'destroy' and 'restart' labels to helper functions, leaving only the
commonly used pattern of 'endjob' and 'cleanup' labels.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
In libxlDomainShutdownThread, virObjectEventStateQueue is needlessly
called in the destroy and restart labels. The cleanup label aready
queues whatever event was created based on libxl_shutdown_reason.
There is no need to handle destroy and restart differently.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
When libxlDomainMigrationDstPrepare adds the @args to an
virNetSocketAddIOCallback using libxlMigrateDstReceive as
the target of the virNetSocketIOFunc @func with the knowledge
that the libxlMigrateDstReceive will virObjectUnref @args
at the end thus not needing to Unref during normal processing
for libxlDomainMigrationDstPrepare.
However, Coverity believes there's an issue with this. The
problem is there can be @nsocks virNetSocketAddIOCallback's
added, but only one virObjectUnref. That means the first
one done will Unref and the subsequent callers may not get
the @args (or @opaque) as they expected. If there's only
one socket returned from virNetSocketNewListenTCP, then sure
that works. However, if it returned more than one there's
going to be a problem.
To resolve this, since we start with 1 reference from the
virObjectNew for @args, we will add 1 reference for each
time @args is used for virNetSocketAddIOCallback. Then
since libxlDomainMigrationDstPrepare would be done with
@args, move it's virObjectUnref from the error: label to
the done: label (since error: falls through). That way
once the last IOCallback is done, then @args will be freed.
Signed-off-by: John Ferlan <jferlan@redhat.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
If available, use b_info->nested_hvm instead of
b_info->u.hvm.nested_hvm. This will make nested HVM config available
also for PVH domains.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Otherwise starting PVH guest will result in "arch_setup_bootlate:
mapping shared_info failed (pfn=..., rc=-1, errno: 12): Internal error".
After this change the behavior is the same as in `xl`.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
All of the ones being removed are pulled in by internal.h. The only
exception is sanlock which expects the application to include <stdint.h>
before sanlock's headers, because sanlock prototypes use fixed width
int, but they don't include stdint.h themselves, so we have to leave
that one in place.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Acked-by: Michal Privoznik <mprivozn@redhat.com>
Currently the libxl driver claims support for Xen >= 4.4, but
Xen 4.4 and 4.5 are no longer supported upstream. Let's increase
the minimum supported Xen version to 4.6 and change the defined
LIBXL_API_VERSION to 0x040500, which is the API version defined
when Xen 4.6 was released.
Since Xen 4.6 contains a pkgconfig file, drop the now unused code
that falls back to using LIBVIRT_CHECK_LIB in the absence of
pkgconfig file. In addition, bumping the LIBXL_API_VERSION
required adjusting the calls to libxl_set_vcpuaffinity to account
for the extra parameter in the 0x040500 version of the API.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
It is possible the incoming VM is not fully started when the finish
phase of migration is executed. In libxlDomainMigrationDstFinish,
wait for the thread receiving the VM to complete before executing
finish phase tasks.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
The libxlDomainMigrationDst* functions are a bit flawed in their
handling of modify jobs. A job begins when the destination host
begins receiving the incoming VM and ends after the VM is started.
The finish phase contains another BeginJob/EndJob sequence.
This patch changes the logic to begin a job for the incoming VM
in the prepare phase and end the job in the finish phase.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
The libxlDomainMigrationSrc* functions are a bit flawed in their
handling of modify jobs. A job begins at the start of the begin
phase but ends before the phase completes. No job is running for
the remaining phases of migration on the source host.
Change the logic to keep the job running after a successful begin
phase, and end the job in the confirm phase. The job must also end
in the perform phase in the case of error since confirm phase would
not be executed.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
libxlDoMigrateSrcP2P() performs all phases of the migration
protocol for peer-to-peer migration. Unfortunately the logic
was a bit flawed since it is possible to skip the confirm
phase after a successfull begin and prepare phase. Fix the
logic to always call the confirm phase after a successful begin
and perform. Skip the confirm phase if begin or perform fail.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
If for any reason the restore of a VM fails on the destination host
in a migration operation, the VM is removed (if not persistent) from
the virDomainObjList, meaning it is no longer available for additional
cleanup or processing in the finish phase. Defer removing the VM from
the virDomainObjList until the finish phase, which already contains
logic to remove the VM.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
ACKed-by: Michal Privoznik <mprivozn@redhat.com>
When libvirtd is started, initialize domain objects state with its real
state, not only RUNNING/SHUTOFF.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Currently, the functions return a pointer to the
destination buffer on success or NULL on failure.
Not only does this kind of error handling look quite
alien in the context of libvirt, where most functions
return zero on success and a negative int on failure,
but it's also somewhat pointless because unless there's
been a failure the returned pointer will be the same
one passed in by the user, thus offering no additional
value.
Change the functions so that they return an int
instead.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Since 2.10 QEMU supports a new display type egl-headless which uses the
drm nodes for OpenGL rendering copying back the rendered bits back to
QEMU into a dma-buf which can be accessed by standard "display" apps
like VNC or SPICE. Although this display type can be used on its own,
for any practical use case it makes sense to pair it with either VNC or
SPICE display. The clear benefit of this display is that VNC gains
OpenGL support, which it natively doesn't have, and SPICE gains remote
OpenGL support (native OpenGL support only works locally through a UNIX
socket, i.e. listen type=socket/none).
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Erik Skultety <eskultet@redhat.com>
And replace all calls with virObjectEventStateQueue such that:
libxlDomainEventQueue(driver, event);
becomes:
virObjectEventStateQueue(driver->domainEventState, event);
And remove NULL checking from all callers.
Signed-off-by: Anya Harter <aharter@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Per-domain log files were introduced in commit a30b08b717. The FILE
objects associated with these log files are stored in a hash table
using domid as a key. When a domain is shutdown, destroyed, or
otherwise powered-off, the FILE object is removed from the hash table,
where the free function will close the FILE.
Unfortunately the call to remove the FILE from the hash table occurs
after setting domid=-1 in the libxlDomainCleanup() function. The
object is never removed from the hash table, the free function is
never called, and the underlying fd is leaked. Fix by removing the
FILE object from the hash table before setting domid=-1.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When computing a baseline CPU for a specific hypervisor we have to make
sure to include only CPU features supported by the hypervisor. Otherwise
the computed CPU could not be used for starting a new domain.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Collin Walling <walling@linux.ibm.com>
This is required for virCPUBaseline to accept a list of guest CPU
definitions since they do not have arch set.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Collin Walling <walling@linux.ibm.com>
libxlDoDomainSave() is used in both the save and managedsave code
paths but was unconditionally setting hasManagedSave to true on
success. As a result, undefine would fail after a non-managed
save/restore operation. E.g.
virsh define; virsh start
virsh save; virsh restore
virsh shutdown
virsh undefine
error: Refusing to undefine while domain managed save image exists
Modify libxlDoDomainSave() to take an additional parameter to
specify managed vs non-managed save, and change callers to use it.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
When adding a new object to the domain object list, there should
have been 2 virObjectRef calls made one for each list into which
the object was placed to match the 2 virObjectUnref calls that
would occur during Remove as part of virHashRemoveEntry when
virObjectFreeHashData is called when the element is removed from
the hash table as set up in virDomainObjListNew.
Some drivers (libxl, lxc, qemu, and vz) handled this inconsistency
by calling virObjectRef upon successful return from virDomainObjListAdd
in order to use virDomainObjEndAPI when done with the returned @vm.
While others (bhyve, openvz, test, and vmware) handled this via only
calling virObjectUnlock upon successful return from virDomainObjListAdd.
This patch will "unify" the approach to use virDomainObjEndAPI
for any @vm successfully returned from virDomainObjListAdd.
Because list removal is so tightly coupled with list addition,
this patch fixes the list removal algorithm to return the object
as entered - "locked and reffed". This way, the callers can then
decide how to uniformly handle add/remove success and failure.
This removes the onus on the caller to "specially handle" the
@vm during removal processing.
The Add/Remove logic allows for some logic simplification such
as in libxl where we can Remove the @vm directly rather than
needing to set a @remove_dom boolean and removing after the
libxlDomainObjEndJob completes as the @vm is locked/reffed.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Since the @dconn reference via args->conn will be used via a thread
or callback, let's make sure memory associated with it isn't free'd
unexpectedly before we use it. The Unref will be done when the object
is Dispose'd.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Reviewed-by: Marc Hartmayer <mhartmay@linux.vnet.ibm.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
When adding the @vm to the @args for usage during a thread or
callback, let's add the reference to it at the time of adding to
ensure nothing else deletes it. The corresponding Unref is then
added to the Dispose function.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Reviewed-by: Marc Hartmayer <mhartmay@linux.vnet.ibm.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Rework the code such that virDomainObjListFindByID will always
return a locked/ref counted object so that the callers can
always do the same cleanup logic to call virDomainObjEndAPI.
Makes accessing the objects much more consistent.
NB:
There were 2 callers (lxcDomainLookupByID and qemuDomainLookupByID)
that were already using the ByID name, but not virDomainObjEndAPI -
these were changed as well in this update/patch.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Now that every caller is using virDomainObjListFindByUUIDRef,
let's just remove it and keep the name as virDomainObjListFindByUUID.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
So far we are repeating the following lines over and over:
if (!(virSomeObjectClass = virClassNew(virClassForObject(),
"virSomeObject",
sizeof(virSomeObject),
virSomeObjectDispose)))
return -1;
While this works, it is impossible to do some checking. Firstly,
the class name (the 2nd argument) doesn't match the name in the
code in all cases (the 3rd argument). Secondly, the current style
is needlessly verbose. This commit turns example into following:
if (!(VIR_CLASS_NEW(virSomeObject,
virClassForObject)))
return -1;
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Convert CPU features policy into libxl cpuid policy settings. Use new
("libxl") syntax, which allow to enable/disable specific bits, using
host CPU as a base. For this reason, only "host-passthrough" mode is
accepted.
Libxl do not have distinction between "force" and "required" policy
(there is only "force") and also between "forbid" and "disable" (there
is only "disable"). So, merge them appropriately. If anything, "require"
and "forbid" should be enforced outside of specific driver.
Nested HVM (vmx and svm features) is handled separately, so exclude it
from translation.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Introduce global libxl option for enabling nested HVM feature, similar
to kvm module parameter. This will prevent enabling experimental feature
by mere presence of <cpu mode='host-passthrough'> element in domain
config, unless explicitly enabled. <cpu mode='host-passthrough'> element
may be used to configure other features, like NUMA, or CPUID.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
When support for mode=custom will be added in the future, semantics of
current config will change. Reduce the surprise by emitting a warning.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Preparation for global nestedhvm configuration - libxlMakeDomBuildInfo
needs access to libxlDriverConfig.
No functional change.
Adjusting tests require slightly more mockup functions, because of
libxlDriverConfigNew() call.
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
libxlDriverConfigNew() use libxlDriverConfigDispose() for cleanup in
case of errors. Do not call libxlLoggerFree() on not allocated logger
(NULL).
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Avoid the need for the drivers to explicitly check for a NULL path by
making sure it is at least the empty string.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Ensuring that we don't call the virDrvConnectOpen method with a NULL URI
means that the drivers can drop various checks for NULL URIs. These were
not needed anymore since the probe functionality was split
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Declare what URI schemes a driver supports in its virConnectDriver
struct. This allows us to skip trying to open the driver entirely
if the URI scheme doesn't match.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Add a localOnly flag to the virConnectDriver struct which allows a
driver to indicate whether it is local-only, or permits remote
connections. Stateful drivers running inside libvirtd are generally
local only. This allows us to remote the check for uri->server != NULL
from most drivers.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Currently the virDrvConnectOpen method is supposed to handle both
opening an explicit URI and auto-probing a driver if no URI is
given. Introduce a dedicated virDrvConnectURIProbe method to enable the
probing functionality to be split from the driver opening functionality.
It is still possible for NULL to be passed to the virDrvConnectOpen
method after this change, because the remote driver needs special
handling to enable probing of the URI against a remote libvirtd daemon.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Historically we have used a bare xen:/// URI for connecting to the
legacy Xen driver. The new libxl Xen driver follows the new practice
of allowing '/system' as a path, as well as bare '/' for compat with
the old Xen driver.
This documents xen:///system as the preferred format for Xen, leaving
xen:/// as an undocumented feature just for back-compat.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
All Xen PV and HVM with PV driver support a memory balloon device,
which cannot be disabled through the toolstack. Model the device
in the libxl driver, similar to the recently removed xend-based
driver.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Long ago in commit dfa1e1dd53 the scheduler weight was accidentally
hardcoded to 1000. Weight is a setting with no unit since it is
relative to the weight of other domains. If no weight is specified,
libxl defaults to 256.
Instead of hardcoding the weight to 1000, honor any <shares> specified
in <cputune>. libvirt's notion of shares is synonomous to libxl's
scheduler weight setting. If shares is unspecified, defer default
weight setting to libxl.
Removing the hardcoded weight required some test fixup. While at it,
add an explicit test for <shares> conversion to scheduler weight.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
Inspired by commit ffb7954f to improve readability of the libxl
migration APIs.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In libxlDomainMigrationPrepare it is possible to dereference a NULL
libxlDomainObjPrivatePtr in early error paths. Check for a valid
'priv' before using it.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
Similar to other uses of virDomainObjListAdd, on success add a ref to the
virDomainObj so that virDomainObjEndAPI can be called as usual.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
If starting the domain fails in libxlDomainCreateXML, we mistakenly
jumped to cleanup without calling libxlDomainObjEndJob. Remove the
jump to 'cleanup'.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: John Ferlan <jferlan@redhat.com>
