April 2009 - Devel - libvirt List Archives

[libvirt] live-migration is not working with libvirt-0.6.2 on RHEL5.3
by Gerrit Slomma 21 Apr '09

21 Apr '09

So i still do not get live migration up and running with libvirt-0.6.2 on RHEL5.3 My setup is as following: Host#1 rr016: RHEL5.3 2.6.18-128.el5.x86_64; 4 GB RAM; C2D T8300 Host#2 rr017: RHEL5.3 2.6.18-128.el5.x86_64; 3 GB RAM; C2D E6550 on both hosts: kvm-85 compiled rpms from sourceforge.org (current release) qemu-0.10.0 (from kvm-85) qemu-img 0.9.1-11 from EPEL libvirt-0.6.2 compiled rpms from libvirt.org (current release) + applied RHEL5-build-patch Network: bridged network-if eth0 on bridge sw0 via /etc/sysconfig/network-scripts/ifcfg-eth0 and /etc/sysconfig/network-scripts/ifcfg-sw0, nothing in /etc/libvirt/qemu/networks besides the (untouched) default.xml. iptables: disabled to avoid side-effects Images: exported from a third host via iscsi and accessible from both involved hosts if i bring up a VM on host#2 that is not defined on host#1 this VM comes up fine and i could ssh into it. [root@rr017 ~]# virsh dominfo rr019v3 Id: 5 Name: rr019v3 UUID: ff1ba599-5801-fca9-ab33-8962c2dfa46c OS Typ: hvm Status: laufend CPU(s): 2 CPU-Zeit: 26,9s Max Speicher: 1572864 kB Verwendeter Speicher: 1572864 kB Automatischer Start: deaktiviert if i try to migrate it without URI specified the migration fails with an "unknown failure" whereas the failure is obvious to me: wrong port [root@rr017 ~]# virsh migrate rr019v3 qemu+tcp://192.168.1.119:16509/system Please enter your authentication name:root Please enter your password: Fehler: Unknown failure [root@rr017 ~]# tail -f /var/log/libvirt/qemu/rr019v3.log info balloon Using KVM without synchronous MMU, ballooning disabled stop migrate "tcp:rr016:49152" migration failed cont [root@rr016 ~]# tail -f /var/log/libvirt/qemu/rr019v3.log LC_ALL=C PATH=/sbin:/usr/sbin:/bin:/usr/bin HOME=/root USER=root LOGNAME=root /usr/bin/kvm -S -M pc -m 1536 -smp 2 -name rr019v3 -uuid ff1ba599-5801-fca9-ab33-8962c2dfa46c -monitor pty -pidfile /var/run/libvirt/qemu//rr019v3.pid -boot c -drive file=/dev/disk/by-path/ip-192.168.1.1:3260-iscsi-rr010:01-lun-6,if=ide,index=0 -net nic,macaddr=00:16:3e:69:94:05,vlan=0 -net tap,fd=17,script=,vlan=0,ifname=vnet0 -serial none -parallel none -usb -vnc 127.0.0.1:0 -k de -incoming tcp:0.0.0.0:49152 char device redirected to /dev/pts/3 info cpus * CPU #0: pc=0x00000000000ffff0 thread_id=8550 CPU #1: pc=0x00000000000ffff0 thread_id=8551 balloon 1536 trying --live as a parameter the same applies: [root@rr017 ~]# virsh migrate --live rr019v3 qemu+tcp://192.168.1.119:16509/system Please enter your authentication name:root Please enter your password: Fehler: Unknown failure [root@rr017 ~]# tail -f /var/log/libvirt/qemu/rr019v3.log info balloon Using KVM without synchronous MMU, ballooning disabled migrate "tcp:rr016:49153" migration failed [root@rr016 ~]# tail -f /var/log/libvirt/qemu/rr019v3.log C_ALL=C PATH=/sbin:/usr/sbin:/bin:/usr/bin HOME=/root USER=root LOGNAME=root /usr/bin/kvm -S -M pc -m 1536 -smp 2 -name rr019v3 -uuid ff1ba599-5801-fca9-ab33-8962c2dfa46c -monitor pty -pidfile /var/run/libvirt/qemu//rr019v3.pid -boot c -drive file=/dev/disk/by-path/ip-192.168.1.1:3260-iscsi-rr010:01-lun-6,if=ide,index=0 -net nic,macaddr=00:16:3e:69:94:05,vlan=0 -net tap,fd=17,script=,vlan=0,ifname=vnet0 -serial none -parallel none -usb -vnc 127.0.0.1:0 -k de -incoming tcp:0.0.0.0:49153 char device redirected to /dev/pts/3 info cpus * CPU #0: pc=0x00000000000ffff0 thread_id=8641 CPU #1: pc=0x00000000000ffff0 thread_id=8642 balloon 1536 Specifying the URI migration works, and the VM appears on the target-host [root@rr017 ~]# virsh migrate rr019v3 qemu+tcp://192.168.1.119:16509/system tcp:192.168.1.119:16509 Please enter your authentication name:root Please enter your password: [root@rr016 ~]# virsh list Id Name Status ---------------------------------- 3 rr019v3 laufend But the status is laufend - that means "running" in english - whereas i have to suspend and resume the VM to get it up, then it is running on the target-host. Backmigration is not possible and hangs endless, the libvirtd must be restarted on both hosts to get them back up. Specifying --live as a parameter for the migration does have the same effect on the VM as omitting the parameter. When doing the same on the commandline everything works fine, i even start the VM via ssh from the source-host on the target-host and migrate over the net with minimal downtime. Why doesn't this work? Has anybody got a working setup with RHEL5 up and running?

1 0

[libvirt] PATCH: Remove PID file before starting QEMU, and after shutdown
by Daniel P. Berrange 21 Apr '09

21 Apr '09

We spawn QEMU with a -pidfile option to write its PID out to a file. This file is never removed though, so next time the guest starts there is small race condition where we might be unlucky enough to read the old PID out of the file, instead of the new QEMU PID. This patch adds a call to remove the pidfile from disk when a guest is shut down. As an extra preventative measure, we also delete it before starting a guest Daniel diff -r ba0066a358b6 src/qemu_driver.c --- a/src/qemu_driver.c Mon Apr 20 12:10:15 2009 +0100 +++ b/src/qemu_driver.c Mon Apr 20 13:52:47 2009 +0100 @@ -1389,6 +1389,14 @@ static int qemudStartVMDaemon(virConnect if (qemuPrepareHostDevices(conn, vm->def) < 0) goto cleanup; + if ((ret = virFileDeletePid(driver->stateDir, vm->def->name)) != 0) { + virReportSystemError(conn, ret, + _("Cannot remove stale PID file for %s"), + vm->def->name); + goto cleanup; + } + + vm->def->id = driver->nextvmid++; if (qemudBuildCommandLine(conn, driver, vm->def, qemuCmdFlags, &argv, &progenv, @@ -1512,6 +1520,8 @@ cleanup: static void qemudShutdownVMDaemon(virConnectPtr conn ATTRIBUTE_UNUSED, struct qemud_driver *driver, virDomainObjPtr vm) { + int ret; + if (!virDomainIsActive(vm)) return; @@ -1556,6 +1566,12 @@ static void qemudShutdownVMDaemon(virCon VIR_WARN(_("Failed to remove domain status for %s"), vm->def->name); } + if ((ret = virFileDeletePid(driver->stateDir, vm->def->name)) != 0) { + char ebuf[1024]; + VIR_WARN(_("Failed to remove PID file for %s: %s"), + vm->def->name, virStrerror(errno, ebuf, sizeof ebuf)); + } + vm->pid = -1; vm->def->id = -1; vm->state = VIR_DOMAIN_SHUTOFF; -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

2 2

[libvirt] PATCH: Make VirtualBox domain IDs start from 1 instead of 0
by Daniel P. Berrange 21 Apr '09

21 Apr '09

The VirtualBox driver currently returns domain IDs starting from 0. The domain ID 0 is reserved for a special scenario. It refers to a guest which is also the host OS. ie its the guest in which libvirt is running. This guest will only exist in virtualization technology where there is a separate hypervisor / host OS, eg Xen or MicroSoft Hyper-V. Thus, it should be avoided for drivers like QEMU or VirtualBox. This patch simply rebases the virtualbox domain IDs to start from 1. Daniel Index: src/vbox/vbox_tmpl.c =================================================================== RCS file: /data/cvs/libvirt/src/vbox/vbox_tmpl.c,v retrieving revision 1.1 diff -u -p -r1.1 vbox_tmpl.c --- src/vbox/vbox_tmpl.c 17 Apr 2009 16:09:07 -0000 1.1 +++ src/vbox/vbox_tmpl.c 21 Apr 2009 10:11:48 -0000 @@ -434,7 +434,7 @@ static int vboxListDomains(virConnectPtr if ((state == MachineState_Running) || (state == MachineState_Paused) ) { ret++; - ids[j++] = i; + ids[j++] = i + 1; } } } @@ -535,6 +535,15 @@ static virDomainPtr vboxDomainLookupByID PRUint32 state; int i; + /* Internal vbox IDs start from 0, the public libvirt ID + * starts from 1, so refuse id==0, and adjust the rest*/ + if (id == 0) { + vboxError(conn, VIR_ERR_NO_DOMAIN, + _("no domain with matching id %d"), id); + return NULL; + } + id = id - 1; + if(data->vboxObj) { rc = data->vboxObj->vtbl->GetMachines(data->vboxObj, &machineCnt, &machines); if (NS_FAILED(rc)) { @@ -568,7 +577,7 @@ static virDomainPtr vboxDomainLookupByID dom = virGetDomain(conn, machineName, iidl); if (dom) - dom->id = id; + dom->id = id + 1; /* Cleanup all the XPCOM allocated stuff here */ g_pVBoxFuncs->pfnComUnallocMem(iid); @@ -645,7 +654,7 @@ static virDomainPtr vboxDomainLookupByUU if (dom) if ((state == MachineState_Running) || (state == MachineState_Paused) ) - dom->id = i; + dom->id = i + 1; } if (iid) { @@ -725,7 +734,7 @@ static virDomainPtr vboxDomainLookupByNa if (dom) if ((state == MachineState_Running) || (state == MachineState_Paused) ) - dom->id = i; + dom->id = i + 1; } if (machineName) { @@ -2311,7 +2320,7 @@ static int vboxDomainCreate(virDomainPtr ret = -1; } else { /* all ok set the domid */ - dom->id = i; + dom->id = i + 1; ret = 0; } } -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

3 3

[libvirt] Supporting dynamic bridge names
by Soren Hansen 21 Apr '09

21 Apr '09

I'm fairly sure we've discussed this before, but I couldn't find it in my archives.. A rather long time ago (0.4.0 timeframe, I think) we switched the default network in Ubuntu to use a bridge whose name was defined as "virbr%d". This was done to be able to actually supply a default, enabled network without the risk of interfering with an existing bridge called "virbr0" (ignoring the possible implications of unconditionally using 192.168.122.0/24 (about which, I might add, I've never had /any/ complaints)). Now, since 0.5.0 (or thereabouts, I think), libvirt doesn't support this, which a) makes it rather difficult to achieve our goal of not clashing with existing bridges named "virbr0", and b) causes some amount of grief for updates. To overcome this, I've changed virNetworkAllocateBridge like so: --- libvirt-0.6.1.orig/src/network_conf.c 2009-03-03 09:23:22.000000000 +0100 +++ libvirt-0.6.1/src/network_conf.c 2009-04-16 20:36:43.660996644 +0200 @@ -875,16 +875,20 @@ } char *virNetworkAllocateBridge(virConnectPtr conn, - const virNetworkObjListPtr nets) + const virNetworkObjListPtr nets, + const char *template) { int id = 0; char *newname; + if (!template) + template = "virbr%d"; + do { char try[50]; - snprintf(try, sizeof(try), "virbr%d", id); + snprintf(try, sizeof(try), template, id); if (!virNetworkBridgeInUse(nets, try, NULL)) { if (!(newname = strdup(try))) { --- libvirt-0.6.1.orig/src/network_conf.h 2009-03-03 09:23:22.000000000 +0100 +++ libvirt-0.6.1/src/network_conf.h 2009-04-16 15:36:46.975452201 +0200 @@ -174,7 +174,8 @@ const char *skipname); char *virNetworkAllocateBridge(virConnectPtr conn, - const virNetworkObjListPtr nets); + const virNetworkObjListPtr nets, + const char *template); int virNetworkSetBridgeName(virConnectPtr conn, const virNetworkObjListPtr nets, And virNetworkSetBridgeName like so: --- libvirt-0.6.1.orig/src/network_conf.c 2009-03-03 09:23:22.000000000 +0100 +++ libvirt-0.6.1/src/network_conf.c 2009-04-16 20:36:43.660996644 +0200 @@ -909,7 +913,7 @@ int ret = -1; - if (def->bridge) { + if (def->bridge && !strstr(def->bridge, "%d")) { if (virNetworkBridgeInUse(nets, def->bridge, def->name)) { networkReportError(conn, NULL, NULL, VIR_ERR_INTERNAL_ERROR, _("bridge name '%s' already in use."), @@ -918,7 +922,7 @@ } } else { /* Allocate a bridge name */ - if (!(def->bridge = virNetworkAllocateBridge(conn, nets))) + if (!(def->bridge = virNetworkAllocateBridge(conn, nets, def->bridge))) goto error; } And finally virNetworkLoadConfig like so: --- libvirt-0.6.1.orig/src/network_conf.c 2009-03-03 09:23:22.000000000 +0100 +++ libvirt-0.6.1/src/network_conf.c 2009-04-16 20:36:43.660996644 +0200 @@ -747,7 +747,7 @@ /* Generate a bridge if none is found, but don't check for collisions * if a bridge is hardcoded, so the network is at least defined */ - if (!def->bridge && !(def->bridge = virNetworkAllocateBridge(conn, nets))) + if (!(def->bridge = virNetworkAllocateBridge(conn, nets, def->bridge))) goto error; if (!(net = virNetworkAssignDef(conn, nets, def))) This is far from perfect, though. a) As used to be the case for the VNC port, "virsh net-dumpxml" will give you /current/ bridge name, so e.g. "virsh net-edit" will cause you to hardcode the bridge name if you're not careful to replace the current name with one that says '%d' somewhere. This can be fixed by e.g. either adding a "template" or "current" attribute on the bridge element in the network XML (depending on whether you want the "name" attribute to be set to the current value or the template value). b) Even if you /do/ remember to put '%d' in place of the assigned number during editing, the bridge number will be increased, because virNetworkBridgeInUse thinks the bridge name is in use. This should be easy to fix, though. I'm perfectly willing to work on this, but I'd like to get you guys' gut feeling on this first. -- Soren Hansen | Lead Virtualisation Engineer | Ubuntu Server Team Canonical Ltd. | http://www.ubuntu.com/

3 5

[libvirt] Dummy tag in xml desc?
by Jean-Michel Guillaume 21 Apr '09

21 Apr '09

Hello Everybody! I'd like to find a way to pass information between a machine & a virtual machine managed by libvirt (like we can with VirtualBox or VMware thanks to guest tools for instance ). I thought that I would have been able to write a comment section or something like that within the xml description of the right domain, but every new comments or xml tags are erased once I save them... Maybe somebody's got an idea to help me :D... JMi

2 1

[Libvirt] Python version
by Christian Weyermann 21 Apr '09

21 Apr '09

Hello everybody, we are doing a project for university where we want to use libvirt. We decided to use the Python bindings. The question is, which python versions are supported by libvirt 0.6.2? We are especially interested in Python 3.0. Thanks and best Regards, Chris

2 1

[libvirt] [RFC] Storage volume clone API
by Cole Robinson 21 Apr '09

21 Apr '09

Hi all, Attached is a stab at an API for cloning a storage volume. This patch implements the command for FS volumes, along with a virsh command 'vol-clone'. This builds on the previously posted 'drop pool lock during volume creation' work. The new API call looks like: /** * virStorageVolClone: * @vol: pointer to storage vol to clone * @xmldesc: description of volume to create * @flags: flags for creation (unused, pass 0) * * Clone a storage volume within the parent pool. * Information for the new volume (name, perms) * are passed via a typical volume XML description * Not all pools support cloning of volumes * * return the storage volume, or NULL on error */ virStorageVolPtr virStorageVolClone(virStorageVolPtr vol, const char *xmldesc, unsigned int flags) The user passes information about the new volume via <volume> XML. Currently we only use the passed name and permissions values, but backingStore info could also be relevant, and 'format' if we allow cloning between image formats. Current limitations: - Parsing the 'clone' xml will complain if the 'capacity' element hasn't been specified, even though it is meaningless for the new volume. Not too sure what the right thing to do here is. - A clone of a raw sparse file will turn out fully allocated. Could probably borrow code from 'cp' to try and get this right. - Doesn't handle backing stores. I think we would need to call out to 'qemu-img convert' to have any chance of getting this right. And if we do that, we could also allow cloning from one image format to another. A lot of the patch is just plumbing, which I will break out into separate patches on the next post. For now, most of the interesting bits are at near the bottom. Feedback appreciated. Thanks, Cole

3 4

[libvirt] Question about the specification of "virsh vcpupin" and "virsh vcpuinfo" in libvirt-0.6.1.
by Takahashi Tomohiro 21 Apr '09

21 Apr '09

Hi, I have questions. I would like to know the specification of "virsh vcpupin" and "virsh vcpuinfo" in libvirt-0.6.1. "virsh vcpupin" and "virsh vcpuinfo" can not be executed to inactive domains. I think it is the specification. Is it correct ? I would like to know whether the libvirt-community have a support plan for pinning inactive domains or not. Because "xm vcpu-pin" and "vcpu-list" can be excuted to inactive domains in Xen3.3. Thanks, Tomohiro Takahashi

3 4

[libvirt] Dummy tag in xml desc?
by Jean-Michel Guillaume 21 Apr '09

21 Apr '09

Hello Everybody! I'd like to find a way to pass information between a machine & a virtual machine managed by libvirt (like we can with VirtualBox or VMware thanks to guest tools for instance ). I thought that I would have been able to write a comment section or something like that within the xml description of the right domain, but every new comments or xml tags are erased once I save them... Maybe somebody's got an idea to help me :D... JMi -- Guillaume Jean-Michel E-Mail : jean-michel.guillaume(a)activeeon.com Phone : + 33 (0)4 97 15 53 84 ACTIVEEON SAS SAS au capital de 75.000 € - RCS de Grasse n° 500 807 284 http://www.activeeon.com 2004, route des Lucioles – B.P.93

1 0

[libvirt] [PATCH 2/4] VirtualBox support
by Pritesh Kothari 21 Apr '09

21 Apr '09

Hi All, I have attached a patch which when applied on the HEAD as of today would allow virtualbox support in libvirt. It takes cares of all the stuff mentioned on the list earlier. Still if I have missed anything, please do tell me. The patches are organized as below: Patch 0/4: contains sample xml file Patch 1/4: contains diff of files already in libvirt. Patch 2/4: contains new files needed for VirtualBox support. Patch 3/4: contains support for host only and internal network. Patch 4/4: contains support for rdp in libvirt as mentioned by danpb (also had sent it separately earlier today) Regards, Pritesh

3 8