Commit Graph

15540 Commits

Author SHA1 Message Date
Cédric Bosdonnat
a41680f8c5 LXC from native: migrate fstab and lxc.mount.entry
Tmpfs relative size and default 50% size values aren't supported as
we have no idea of the available memory at the conversion time.
2014-02-12 17:52:46 +00:00
Cédric Bosdonnat
197b13e5d9 LXC from native: import rootfs
LXC rootfs can be either a directory or a block device or an image
file. The first two types have been implemented, but the image file is
still to be done since LXC auto-guesses the file format at mount time
and the LXC driver doesn't support the 'auto' format.
2014-02-12 17:52:46 +00:00
Cédric Bosdonnat
7195c807b2 LXC driver: started implementing connectDomainXMLFromNative
This function aims at converting LXC configuration into a libvirt
domain XML description to help users migrate from LXC to libvirt.

Here is an example of how the lxc configuration works:
virsh -c lxc:/// domxml-from-native lxc-tools /var/lib/lxc/migrate_test/config

It is possible that some parts couldn't be properly mapped into a
domain XML fragment, so users should carefully review the result
before creating the domain.

fstab files in lxc.mount lines will need to be merged into the
configuration file as lxc.mount.entry.

As we can't know the amount of memory of the host, we have to set a
default value for max_balloon that users will probably want to adjust.
2014-02-12 17:52:46 +00:00
Cédric Bosdonnat
3daa14834a Improve virConf parse to handle LXC config format
virConf now honours a VIR_CONF_FLAG_LXC_FORMAT flag to handle LXC
configuration files. The differences are that property names can
contain '.' character and values are all strings without any bounding
quotes.

Provide a new virConfWalk function calling a handler on all non-comment
values. This function will be used by the LXC conversion code to loop
over LXC configuration lines.
2014-02-12 17:52:46 +00:00
Eric Blake
6831c1d327 event: pass reason for PM events
Commit 57ddcc23 (v0.9.11) introduced the pmwakeup event, with
an optional 'reason' field reserved for possible future expansion.
But it failed to wire the field through RPC, so even if we do
add a reason in the future, we will be unable to get it back
to the user.

Worse, commit 7ba5defb (v1.0.0) repeated the same mistake with
the pmsuspend_disk event.

As long as we are adding new RPC calls, we might as well fix
the events to actually match the signature so that we don't have
to add yet another RPC in the future if we do decide to start
using the reason field.

* src/remote/remote_protocol.x
(remote_domain_event_callback_pmwakeup_msg)
(remote_domain_event_callback_pmsuspend_msg)
(remote_domain_event_callback_pmsuspend_disk_msg): Add reason
field.
* daemon/remote.c (remoteRelayDomainEventPMWakeup)
(remoteRelayDomainEventPMSuspend)
(remoteRelayDomainEventPMSuspendDisk): Pass reason to client.
* src/conf/domain_event.h (virDomainEventPMWakeupNewFromDom)
(virDomainEventPMSuspendNewFromDom)
(virDomainEventPMSuspendDiskNewFromDom): Require additional
parameter.
* src/conf/domain_event.c (virDomainEventPMClass): New class.
(virDomainEventPMDispose): New function.
(virDomainEventPMWakeupNew*, virDomainEventPMSuspendNew*)
(virDomainEventPMSuspendDiskNew*)
(virDomainEventDispatchDefaultFunc): Use new class.
* src/remote/remote_driver.c (remoteDomainBuildEvent*PM*): Pass
reason through.
* src/remote_protocol-structs: Regenerate.

Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-12 10:48:16 -07:00
Eric Blake
158795d20e event: convert remaining domain events to new style
Following the patterns established by lifecycle events, this
creates all the new RPC calls needed to pass callback IDs
for every domain event, and changes the limits in client and
server codes to use modern style when possible.

I've tested all combinations: both 'old client and new server'
and 'new client and old server' continue to work with the old
RPCs, and 'new client and new server' benefit from server-side
filtering with the new RPCs.

* src/remote/remote_protocol.x (REMOTE_PROC_DOMAIN_EVENT_*): Add
REMOTE_PROC_DOMAIN_EVENT_CALLBACK_* counterparts.
* daemon/remote.c (remoteRelayDomainEvent*): Send callbackID via
newer RPC when used with new-style registration.
(remoteDispatchConnectDomainEventCallbackRegisterAny): Extend to
cover all domain events.
* src/remote/remote_driver.c (remoteDomainBuildEvent*): Add new
Callback and Helper functions.
(remoteEvents): Match order of RPC numbers, register new handlers.
(remoteConnectDomainEventRegisterAny)
(remoteConnectDomainEventDeregisterAny): Extend to cover all
domain events.
* src/remote_protocol-structs: Regenerate.

Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-12 10:48:16 -07:00
Eric Blake
355ea62650 event: client RPC protocol tweaks for domain lifecycle events
The counterpart to the server RPC additions; here, a single
function can serve both old and new calls, while incoming
events must be serviced by two different functions.  Again,
some wise choices in our XDR made it easier to share code
managing similar events.

While this only supports lifecycle events, it covers the
harder part of how Register and RegisterAny interact; the
remaining 15 events will be a mechanical change in a later
patch.  For Register, we now have a callbackID locally for
more efficient cleanup if the RPC fails; we also prefer to
use the newer RPC where we know it is supported (the older
RPC must be used if we don't know if RegisterAny is
supported).

* src/remote/remote_driver.c (remoteEvents): Register new RPC
event handler.
(remoteDomainBuildEventLifecycle): Move guts...
(remoteDomainBuildEventLifecycleHelper): ...here.
(remoteDomainBuildEventCallbackLifecycle): New function.
(remoteConnectDomainEventRegister)
(remoteConnectDomainEventDeregister)
(remoteConnectDomainEventRegisterAny)
(remoteConnectDomainEventDeregisterAny): Use new RPC when supported.
2014-02-12 10:48:16 -07:00
Eric Blake
caaf6ba1b6 event: prepare client to track domain callbackID
We want to convert over to server-side events, even for older
APIs.  To do that, the client side of the remote driver wants
to distinguish between legacy virConnectDomainEventRegister and
normal virConnectDomainEventRegisterAny, while knowing the
client callbackID and the server's serverID for both types of
registration.  The client also needs to probe whether the
server supports server-side filtering.  However, for ease of
review, we don't actually use the new RPCs until a later patch.

* src/conf/object_event_private.h (virObjectEventStateCallbackID):
Add parameter.
* src/conf/object_event.c (virObjectEventCallbackListAddID)
(virObjectEventStateRegisterID): Separate legacy from callbackID.
(virObjectEventStateCallbackID): Pass through parameter.
(virObjectEventCallbackLookup): Let legacy and global domain
lifecycle events share a common remoteID.
* src/conf/network_event.c (virNetworkEventStateRegisterID):
Update caller.
* src/conf/domain_event.c (virDomainEventStateRegister)
(virDomainEventStateRegisterID, virDomainEventStateDeregister):
Likewise.
(virDomainEventStateRegisterClient)
(virDomainEventStateCallbackID): Implement new functions.
* src/conf/domain_event.h (virDomainEventStateRegisterClient)
(virDomainEventStateCallbackID): New prototypes.
* src/remote/remote_driver.c (private_data): Add field.
(doRemoteOpen): Probe server feature.
(remoteConnectDomainEventRegister)
(remoteConnectDomainEventRegisterAny): Use new function.

Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-12 10:48:15 -07:00
Eric Blake
0372295770 event: server RPC protocol tweaks for domain lifecycle events
This patch adds some new RPC call numbers, but for ease of review,
they sit idle until a later patch adds the client counterpart to
drive the new RPCs.  Also for ease of review, I limited this patch
to just the lifecycle event; although converting the remaining
15 domain events will be quite mechanical.  On the server side,
we have to have a function per RPC call, largely with duplicated
bodies (the key difference being that we store in our callback
opaque pointer whether events should be fired with old or new
style); meanwhile, a single function can drive multiple RPC
messages.  With a strategic choice of XDR struct layout, we can
make the event generation code for both styles fairly compact.

I debated about adding a tri-state witness variable per
connection (values 'unknown', 'legacy', 'modern').  It would start
as 'unknown', move to 'legacy' if any RPC call is made to a legacy
event call, and move to 'modern' if the feature probe is made;
then the event code could issue an error if the witness state is
incorrect (a legacy RPC call while in 'modern', a modern RPC call
while in 'unknown' or 'legacy', and a feature probe while in
'legacy' or 'modern').  But while it might prevent odd behavior
caused by protocol fuzzing, I don't see that it would prevent
any security holes, so I considered it bloat.

Note that sticking @acl markers on the new RPCs generates unused
functions in access/viraccessapicheck.c, because there is no new
API call that needs to use the new checks; however, having a
consistent .x file is worth the dead code.

* src/libvirt_internal.h (VIR_DRV_FEATURE_REMOTE_EVENT_CALLBACK):
New feature.
* src/remote/remote_protocol.x
(REMOTE_PROC_CONNECT_DOMAIN_EVENT_CALLBACK_REGISTER_ANY)
(REMOTE_PROC_CONNECT_DOMAIN_EVENT_CALLBACK_DEREGISTER_ANY)
(REMOTE_PROC_DOMAIN_EVENT_CALLBACK_LIFECYCLE): New RPCs.
* daemon/remote.c (daemonClientCallback): Add field.
(remoteDispatchConnectDomainEventCallbackRegisterAny)
(remoteDispatchConnectDomainEventCallbackDeregisterAny): New
functions.
(remoteDispatchConnectDomainEventRegisterAny)
(remoteDispatchConnectDomainEventDeregisterAny): Mark legacy use.
(remoteRelayDomainEventLifecycle): Change message based on legacy
or new use.
(remoteDispatchConnectSupportsFeature): Advertise new feature.
* src/remote_protocol-structs: Regenerate.

Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-12 10:48:15 -07:00
Eric Blake
047fd2e83e event: dynamically manage server-side RPC domain events
This patch continues the earlier conversion made for network
events, with a goal of introducing server-side event filtering
in a later patch.  Actual behavior is unchanged without
further RPC changes.

* daemon/libvirtd.h (daemonClientPrivate): Alter the tracking of
domain events.
* daemon/remote.c (remoteClientInitHook, remoteClientFreeFunc)
(remoteRelayDomainEvent*)
(remoteDispatchConnectDomainEventRegister)
(remoteDispatchConnectDomainEventRegisterAny): Track domain
callbacks dynamically.
2014-02-12 10:48:15 -07:00
Michael Chapman
74cf8202d2 storage: handle NULL return from virGetStorageVol
virGetStorageVol can return NULL on out-of-memory. If it does, cleanly
abort the volume clone operation.

Signed-off-by: Michael Chapman <mike@very.puzzling.org>
2014-02-12 15:18:43 +01:00
Ján Tomko
7236a473f0 Revert "storage: disk: Separate creating of the volume from building"
This reverts commit 67ccf91bf2.
We only generate the volume key after we've built it, but the storage
driver expects it to be filled after createVol finishes.
Squash the volume building back with creating to fulfill this
expectation.
2014-02-12 14:54:05 +01:00
Ján Tomko
42bbde2d06 Revert "storage: lvm: Separate creating of the volume from building"
This reverts commit af1fb38f55.
With it, creating new logical volumes fails:
https://www.redhat.com/archives/libvir-list/2014-February/msg00658.html

In the storage driver, we expect CreateVol to fill out the volume key,
but the LVM backend fills the key with the uuid reported by lvs after the
logical volume is created.
2014-02-12 14:51:05 +01:00
Ján Tomko
5b0dc11db8 Fix leaks in vircapstest
Coverity complains about cell_cpus being leaked on error
and valgrind shows 'caps' is leaked on success.

Introduced in eb64e87.
2014-02-12 14:41:47 +01:00
Cédric Bosdonnat
d385239260 Fixed build with clang.
Two unused global variables, and DBUS_TYPE_INVALID used as a const
char*.

Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-12 06:36:17 -07:00
Cédric Bosdonnat
4f20084b12 AppArmor: Fix the place where the template should be installed
The security driver expects /etc/apparmor.d/libvirt/TEMPLATE but we
installed it to /etc/apparmor.d/libvirtd/TEMPLATE. Move the template to
the expected place since that code was here long before.
2014-02-12 06:34:32 -07:00
Oleg Strikov
69fba97f63 qemu: Implement a stub cpuArchDriver.baseline() handler for aarch64
Openstack Nova calls virConnectBaselineCPU() during initialization
of the instance to get a full list of CPU features.
This patch adds a stub to aarch64-specific code to handle
this request (no actual work is done). That's enough to have
this stub with limited functionality because qemu/kvm backend
supports only 'host-passthrough' cpu mode on aarch64.

Signed-off-by: Oleg Strikov <oleg.strikov@canonical.com>
2014-02-11 17:34:55 -05:00
Jim Fehlig
2fbfedeb0d libxl: fix libxlDoDomainSave documentation
Update the function's comment, which was missed when removing use of
the driver lock everywhere.
2014-02-11 11:03:53 -07:00
Jim Fehlig
3d8a3d6e5b libxl: register for domain events immediately after creation
A small fix for the possiblitiy of jumping to an error path before
registering for domain events, preventing receiving important ones
like shutdown and death.
2014-02-11 11:03:53 -07:00
Jim Fehlig
e20bf46741 libxl: rename libxlCreateDomEvents to libxlDomEventsRegister
libxlDomEventsRegister better reflects its purpose: register for
domain events from libxl.
2014-02-11 11:03:53 -07:00
Ján Tomko
47fa97a799 Rename 'index' in virCapabilitiesGetCpusForNode
This shadows the index function on some systems (RHEL-6.4, FreeBSD 9):
../../src/conf/capabilities.c: In function 'virCapabilitiesGetCpusForNode':
../../src/conf/capabilities.c:1005: warning: declaration of'index'
      shadows a global declaration [-Wshadow]
/usr/include/strings.h:57: warning: shadowed declaration is here [-Wshadow]
2014-02-11 16:35:33 +01:00
Pradipta Kr. Banerjee
eb64e8752b vircapstest: Introduce virCapabilitiesGetCpusForNodemask test
This test creates a Fake NUMA topology with non-sequential cell ids
to check if libvirt properly handles the same

Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Signed-off-by: Pradipta Kr. Banerjee <bpradip@in.ibm.com>
2014-02-11 14:44:20 +00:00
Pradipta Kr. Banerjee
cd921cf077 Handle non-sequential NUMA node numbers
On some platforms like IBM PowerNV the NUMA node numbers can be
non-sequential. For eg. numactl --hardware o/p from such a machine looks
as given below

node distances:
   node   0   1  16  17
     0:  10  40  40  40
     1:  40  10  40  40
    16:  40  40  10  40
    17:  40  40  40  10

The NUMA nodes are 0,1,16,17

Libvirt uses sequential index as NUMA node numbers and this can
result in crash or incorrect results.

Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Signed-off-by: Pradipta Kr. Banerjee <bpradip@in.ibm.com>
2014-02-11 14:44:20 +00:00
Peter Krempa
037ffda3c7 storage: gluster: Set volume metadata in a separate function
Extract the metadata setting code into a separate function for future
use.
2014-02-11 13:46:32 +01:00
Martin Kletzander
d27e6bc40f qemu: introduce spiceport chardev backend
Add a new backend for any character device.  This backend uses channel
in spice connection.  This channel is similar to spicevmc, but
all-purpose in contrast to spicevmc.

Apart from spicevmc, spiceport-backed chardev will not be formatted
into the command-line if there is no spice to use (with test for that
as well).  For this I moved the def->graphics counting to the start
of the function so its results can be used in rest of the code even in
the future.

Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
2014-02-11 13:43:55 +01:00
Martin Kletzander
296a4791eb qemu: remove pointless condition
This patch is here just to ease the code review and make related
changes look more sensible.  Apart from removing the condition this is
merely a whitespace (indentation) change.

Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
2014-02-11 13:43:55 +01:00
Martin Kletzander
a53e504052 qemu: rework '-serial none'
Limiting ourselves to qemu without QEMU_CAPS_DEVICE capability, we
used '-serial none' only if there was no serial device defined in the
domain XML.  This means that if we want to have a possibility of the
device being defined in XML, but not used in the command-line
(e.g. when it's pointless), we'll fail to attach '-serial none' to the
command-line (when skipping the device's command-line building and the
device being the only one).

Since there is no such device, this patch doesn't actually do
anything, but enables easier future additions in this manner.

Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
2014-02-11 13:43:55 +01:00
Martin Kletzander
5b189541ac conf: introduce spiceport chardev backend
Add a new character device backend called 'spiceport' that uses
spice's channel for communications and apart from spicevmc can be used
as a backend for any character device from libvirt's point of view.

Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
2014-02-11 13:43:55 +01:00
Wido den Hollander
0227889ab0 rbd: Use rbd_create3 to create RBD format 2 images by default
This new RBD format supports snapshotting and cloning. By having
libvirt create images in format 2 end-users of the created images
can benefit from the new RBD format.

Older versions of libvirt can work with this new RBD format as long
as librbd supports format 2. RBD format is supported by librbd since
version 0.56 (Ceph Bobtail).

Signed-off-by: Wido den Hollander <wido@widodh.nl>
2014-02-11 12:10:22 +00:00
Joel SIMOES
9741006333 Libvirt lose sheepdogs volumes on pool refresh or restart.
When restarting sheepdog pool, all volumes are missing.
This patch add automatically all volume from the added pool.

Adding last Daniel P. Berrange's syntaxes correction.
Adding vol on separeted function 'inspired' from parallels_storage :
parallelsAddDiskVolume
2014-02-11 11:32:04 +00:00
Laine Stump
0144d72963 build: correctly check for SOICGIFVLAN GET_VLAN_VID_CMD command
In order to make a client-only build successful on RHEL4 (yes, you
read that correctly!), commit 3ed2e54 modified src/util/virnetdev.c so
that the functional version of virNetDevGetVLanID() was only compiled
if GET_VLAN_VID_CMD was defined. However, it is *never* defined, but
is only an enum value, so the proper version was no longer compiled
even on platforms that support it. This resulted in the vlan tag not
being properly set for guest traffic on VEPA mode guest macvtap
interfaces that were bound to a vlan interface (that's the only place
that libvirt currently uses virNetDevGetVLanID)

Since there is no way to compile conditionally based on the presence
of an enum value, this patch modifies configure.ac to check for said
enum value with AC_CHECK_DECLS(), which #defines
HAVE_DECL_GET_VLAN_VID_CMD to 1 if it's successful compiling a test
program that uses GET_VLAN_VID_CMD (and still #defines it, but to 0,
if it's not successful).  We can then make the compilation of
virNetDevGetVLanID() conditional on the value of
HAVE_DECL_GET_VLAN_VID_CMD.
2014-02-11 01:43:38 +02:00
Chen Hanxiao
b95fde2cf7 docs: grammar fixes in formatdomain
Some grammar fixes.

s/namespace,set/namespace, set
s/container being allowed/container are allowed
s/the <code>uid/The <code>uid

Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
2014-02-10 14:14:43 -07:00
Yuri Myasoedov
cc25e45158 maint: fix line numbers in check-aclrules reports
Reset line numbering on each input file in check-aclrules.pl. Otherwise
it reports wrong line numbers in its error messages.

Signed-off-by: Yuri Myasoedov <ymyasoedov@yandex.ru>
Signed-off-by: Roman Bogorodskiy <bogorodskiy@gmail.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
2014-02-10 14:07:22 -07:00
Michal Privoznik
28900766d5 virNetworkLoadState: Disallow mangled 'floor' element
In the network status XML we may have the <floor/> element with the
'sum' attribute. The attribute represents sum of all 'floor'-s of
computed over each interface connected to the network (this is needed to
guarantee certain bandwidth for certain domain). The sum is therefore a
number. However, if the number was mangled (e.g. by an user's
interference to network status file), we've just ignored it without
refusing to parse such file. This was all due to 'goto error' missing.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
2014-02-10 19:26:16 +01:00
Peter Krempa
9bf629ab60 qemu: Use correct permissions when determining the image chain
The code took into account only the global permissions. The domains now
support per-vm DAC labels and per-image DAC labels. Use the most
specific label available.
2014-02-10 15:49:59 +01:00
Michal Privoznik
e209c07760 networkStartNetwork: Be more verbose
The lack of debug printings might be frustrating in the future.
Moreover, this function doesn't follow the usual pattern we have in the
rest of the code:

  int ret = -1;
  /* do some work */
  ret = 0;
cleanup:
  /* some cleanup work */
  return ret;

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
2014-02-10 11:47:24 +01:00
Peter Krempa
600bca592b qemu: hyperv: Add support for timer enlightenments
Add a new <timer> for the HyperV reference time counter enlightenment
and the iTSC reference page for Windows guests.

This feature provides a paravirtual approach to track timer events for
the guest (similar to kvmclock) with the option to use real hardware
clock on systems with a iTSC with compensation across various hosts.
2014-02-10 11:30:10 +01:00
Peter Krempa
8ffaa42d7b conf: Enforce supported options for certain timers
According to the documentation various timer options are only supported
by certain timer types. Add a post parse check to verify that the user
didn't specify invalid options.

Also fix the qemu command line parsing function to set correct default
values for the kvmclock timer so that it passes the new check.
2014-02-10 11:17:32 +01:00
Peter Krempa
bbd392ff86 schema: Fix guest timer specification schema according to the docs
According to the documentation describing various tunables for domain
timers not all the fields are supported by all the driver types. Express
these in the RNG:

- rtc, platform: Only these support the "track" attribute.
- tsc: only one to support "frequency" and "mode" attributes
- hpet, pit: tickpolicy/catchup attribute/element
- kvmclock: no extra attributes are supported

Additionally the attributes of the <catchup> element for
tickpolicy='catchup' are optional according to the parsing code. Express
this in the XML and fix a spurious space added while formatting the
<catchup> element and add tests for it.
2014-02-10 11:09:14 +01:00
Felix Geyer
f88a3d9b0c apparmor: Improve profiles
Tested on Debian unstable.
The profile updates are partly taken from the Ubuntu trusty libvirt package.

Signed-off-by: Guido Günther <agx@sigxcpu.org>
2014-02-07 19:16:51 +01:00
John Ferlan
b60644f38f virpci: Resolve coverity issues
Coverity complains about "USE_AFTER_FREE" due to how virPCIDeviceSetStubDriver
"could" return either -1, 0, or 1 from the VIR_STRDUP() and then possibly makes
a call to virPCIDeviceDetach().

The only way this could happen is if NULL were passed as the "driver" name
and virStrdup() returned 0.  Since the calling functions check < 0 on the
initial function call, the 0 possibility causes Coverity to complain.

To fix this - enforce that the second parameter is not NULL using
ATTRIBUTE_NONNULL(2) for the function prototype, then in virPCIDeviceDetach
add an sa_assert(dev->stubDriver). This will result in Coverity not complaining
any more.
2014-02-07 10:58:24 -05:00
Christophe Fergeau
f336b1cccb Add glusterfs to VIR_CONNECT_LIST_STORAGE_POOLS_FILTERS_POOL_TYPE
If it's not present in this list, we won't be able to get only
glusterfs pools when using virConnectListAllStoragePools.
2014-02-07 10:26:46 +01:00
Christophe Fergeau
1b9c673936 build-sys: Removed unused variable from configure.ac
LIBGLUSTER_LIBS is emptied before gluster is enabled/disabled, but nothing
else sets/uses this variable, so it can be removed.
2014-02-07 10:26:45 +01:00
Martin Kletzander
440a1aa508 qemu: keep pre-migration domain state after failed migration
Couple of codepaths shared the same code which can be moved out to a
function and on one of such places, qemuMigrationConfirmPhase(), the
domain was resumed even if it wasn't running before the migration
started.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1057407

Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
2014-02-07 10:07:38 +01:00
Jim Fehlig
630b645695 libxl: remove unneeded locking of driver when restoring
libxlDomainRestoreFlags acquires the driver lock while reading the
domain config from the save file and adding it to
libxlDriverPrivatePtr->domains.  But virDomainObjList provides
self-locking APIs, so remove the needless driver locking.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
2014-02-06 10:39:32 -07:00
Jim Fehlig
778067e195 libxl: improve subprocess handling
If available, let libxl handle reaping any children it creates by
specifying libxl_sigchld_owner_libxl_always_selective_reap.  This
feature was added to improve subprocess handling in libxl when used
in an application that does not install a SIGCHLD handler like
libvirt

http://lists.xen.org/archives/html/xen-devel/2014-01/msg01555.html

Prior to this patch, it is possible to hit asserts in libxl when
reaping subprocesses, particularly during simultaneous operations
on multiple domains.  With this patch, and the corresponding changes
to libxl, I no longer see the asserts.  Note that the libxl changes
will be included in Xen 4.4.0.  Previous Xen versions will be
susceptible to hitting the asserts even with this patch applied to
the libvirt libxl driver.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
2014-02-06 10:20:31 -07:00
Jim Fehlig
03b3f8940a libxl: handle domain shutdown events in a thread
Handling the domain shutdown event within the event handler seems
a bit unfair to libxl's event machinery.  Domain "shutdown" could
take considerable time.  E.g. if the shutdown reason is reboot,
the domain must be reaped and then started again.

Spawn a shutdown handler thread to do this work, allowing libxl's
event machinery to go about its business.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
2014-02-06 10:17:58 -07:00
Jim Fehlig
eaa8d9b2c7 libxl: remove list of timer registrations from libxlDomainObjPrivate
Due to some misunderstanding of requirements libxl places on timer
handling, I introduced the half-brained idea of maintaining a list
of timeouts that the driver could force to expire before freeing a
libxlDomainObjPrivate (and hence libxl_ctx).  But testing all
the latest versions of Xen supported by the libxl driver (4.2.3,
4.3.1, 4.4.0 RC3), I see that libxl will handle this just fine and
there is no need to force expiration behind libxl's back.  Indeed it
may be harmful to do so.

This patch removes the timer list, allowing libxl to handle cleanup
of its timer registrations.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
2014-02-06 10:08:11 -07:00
Jim Fehlig
cda52dbfe5 libxl: fix leaking libxlDomainObjPrivate
When libxl registers an FD with the libxl driver, the refcnt of the
associated libxlDomainObjPrivate object is incremented. The refcnt
is decremented when libxl deregisters the FD.  But some FDs are only
deregistered when their libxl ctx is freed, which unfortunately is
done in the libxlDomainObjPrivate dispose function.  With references
held by the FDs, libxlDomainObjPrivate is never disposed.

I added the ref/unref in FD registration/deregistration when adding
the same in timer registration/deregistration.  For timers, this
is a simple approach to ensuring the libxlDomainObjPrivate is not
disposed prior to their expirtation, which libxl guarantees will
occur.  It is not needed for FDs, and only causes
libxlDomainObjPrivate to leak.

This patch removes the reference on libxlDomainObjPrivate for FD
registrations, but retains them for timer registrations.  Tested on
the latest releases of Xen supported by the libxl driver:  4.2.3,
4.3.1, and 4.4.0 RC3.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
2014-02-06 10:06:26 -07:00
Matthieu Coudron
0778fc1ab9 qemu_driver: Introduce <filesystem/> support in device attach/detach
This commit allows to attach/detach a <filesystem> device in qemu. For
this purpose I'm introducing two new functions: virDomainFSInsert() and
virDomainFSRemove() and adding necessary code in the qemu driver.  It
compares filesystems based on their "destination" folder. So if two
filesystems share the same destination, they are considered equal and
the qemu driver would reject the insertion.

Signed-off-by: Matthieu Coudron <mattator@gmail.com>
2014-02-06 17:20:03 +01:00