- Devel - libvirt List Archives

[libvirt] discrepancies in issuing virsh capabilities with different user-accounts
by Gerrit Slomma 06 May '09

06 May '09

Hello When doing a virsh capabilities with root i get a different output than when issuing the command with my own unprivileged account roadrunner. Furthermore the first issuing of the command as roadrunner throws a error-message. Restarting of the libvirt-daemon gives the correct output. Sample: Last login: Wed May 6 21:00:56 2009 from 192.168.1.120 [root@rr018 ~]# virsh capabilities|grep kvm [root@rr018 ~]# su - roadrunner [roadrunner@rr018 ~]$ virsh capabilities|grep kvm libvir: Remote Fehler : unable to connect to '@/home/roadrunner/.libvirt/libvirt-sock': Verbindungsaufbau abgelehnt Fehler: Verbindung zum Hypervisor scheiterte Failed to bind socket to '@/home/roadrunner/.libvirt/libvirt-sock': Address already in use Failed to bind socket to '@/home/roadrunner/.libvirt/libvirt-sock': Address already in use Failed to bind socket to '@/home/roadrunner/.libvirt/libvirt-sock': Address already in use Failed to bind socket to '@/home/roadrunner/.libvirt/libvirt-sock': Address already in use ^C [roadrunner@rr018 ~]$ virsh capabilities|grep kvm <domain type='kvm'> <emulator>/usr/bin/kvm</emulator> <domain type='kvm'> [roadrunner@rr018 ~]$ exit logout [root@rr018 ~]# virsh capabilities|grep kvm [root@rr018 ~]# /etc/init.d/libvirtd restart libvirtd-Daemon beenden: [ OK ] libvirtd-Daemon starten: [ OK ] [root@rr018 ~]# virsh capabilities|grep kvm <domain type='kvm'> <emulator>/usr/bin/kvm</emulator> <domain type='kvm'> [root@rr018 ~]#

1 1

Re: [libvirt] KVM processes -- should we be able to attach them to the libvirtd process?
by Daniel P. Berrange 06 May '09

06 May '09

On Wed, May 06, 2009 at 08:56:18PM +0200, Gerrit Slomma wrote: > Daniel P. Berrange schrieb: > >On Tue, May 05, 2009 at 11:38:13PM -0500, Matthew Farrellee wrote: > > > >>It doesn't appear to be the case that the libvirtd daemon can trivially > >>restart and continue with no interruptions. Right now it loses track of > >>VMs. > >> > > > >That a is a bug then, if you can reproduce it, please file a BZ ticket > >so we can track it down & fix it. > > > > > >>In a scenario where VMs are not deployed and locked to specific physical > >>nodes, it can be highly valuable to have ways to ensure a VM is no > >>longer running when a layer of its management stops functioning. > >> > > > >IMHO this is a problem to be solved by clustering software. If the > >clustering software detects a failure with the management service, > >then it should power fence the entire node. Relying on management > >service failure to kill the VMs will never be reliable enough. > > > I think he is pointing towards a VM that runs on a host where it isn't > defined at via a corresponding *.xml. > If you restart a libvirt i looses connection to this or these specific > VM(s). That is a bug that needs fixing. Even if there is no persistent config, we should not loose track of the running VM, because we always write out the 'live' XML config to /var/run/libvirt explicitly so that it is available at restart. Daniel -- |: 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 :|

1 0

[libvirt] PATCH: Raise log level for dlopen() problems
by Daniel P. Berrange 06 May '09

06 May '09

Some of the problems when dlopen'ing a module really should be reported to the user/admin more readily. So this raises the logging level for the important failure messages, so they're visible by default Daniel diff -r 15c4668d403b src/driver.c --- a/src/driver.c Wed Apr 29 15:29:17 2009 +0100 +++ b/src/driver.c Thu Apr 30 14:49:27 2009 +0100 @@ -54,13 +54,13 @@ virDriverLoadModule(const char *name) return NULL; if (access(modfile, R_OK) < 0) { - DEBUG("Moodule %s not accessible", modfile); + VIR_WARN("Module %s not accessible", modfile); goto cleanup; } handle = dlopen(modfile, RTLD_NOW | RTLD_LOCAL); if (!handle) { - DEBUG("failed to load module %s %s", modfile, dlerror()); + VIR_ERROR("failed to load module %s %s", modfile, dlerror()); goto cleanup; } @@ -70,12 +70,12 @@ virDriverLoadModule(const char *name) regsym = dlsym(handle, regfunc); if (!regsym) { - DEBUG("Missing module registration symbol %s", regfunc); + VIR_ERROR("Missing module registration symbol %s", regfunc); goto cleanup; } if ((*regsym)() < 0) { - DEBUG("Failed module registration %s", regfunc); + VIR_ERROR("Failed module registration %s", regfunc); goto cleanup; } -- |: 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 1

[libvirt] PATCH: More robust name & UUID uniqueness checking for QEMU
by Daniel P. Berrange 06 May '09

06 May '09

When defining a VM config, we need to apply the following logi - If existing VM has same UUID - If name also matches => allow - Else => raise error - Else - If name matches => raise error - Else => allow When creating a live VM, or restoring a VM we need to apply similar, but slightly different logic - If existing VM has same UUID - If name also matches - If existing VM is running => raise error - Else => allow - Else => raise error - Else - If name matches => raise error - Else => allow This patch applies those checks for the QEMU driver Daniel diff -r 4e6a98395da5 src/qemu_driver.c --- a/src/qemu_driver.c Thu Apr 30 14:50:14 2009 +0100 +++ b/src/qemu_driver.c Thu Apr 30 15:03:03 2009 +0100 @@ -2145,22 +2145,37 @@ static virDomainPtr qemudDomainCreate(vi if (virSecurityDriverVerify(conn, def) < 0) goto cleanup; - vm = virDomainFindByName(&driver->domains, def->name); - if (vm) { - qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, - _("domain '%s' is already defined"), - def->name); - goto cleanup; - } + /* See if a VM with matching UUID already exists */ vm = virDomainFindByUUID(&driver->domains, def->uuid); if (vm) { - char uuidstr[VIR_UUID_STRING_BUFLEN]; - - virUUIDFormat(def->uuid, uuidstr); - qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, - _("domain with uuid '%s' is already defined"), - uuidstr); - goto cleanup; + /* UUID matches, but if names don't match, refuse it */ + if (STRNEQ(vm->def->name, def->name)) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + vm->def->name, uuidstr); + goto cleanup; + } + + /* UUID & name match, but if VM is already active, refuse it */ + if (virDomainIsActive(vm)) { + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain is already active as '%s'"), vm->def->name); + goto cleanup; + } + virDomainObjUnlock(vm); + } else { + /* UUID does not match, but if a name matches, refuse it */ + vm = virDomainFindByName(&driver->domains, def->name); + if (vm) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + def->name, uuidstr); + goto cleanup; + } } if (!(vm = virDomainAssignDef(conn, @@ -2348,6 +2363,11 @@ static int qemudDomainDestroy(virDomainP _("no domain with matching uuid '%s'"), uuidstr); goto cleanup; } + if (!virDomainIsActive(vm)) { + qemudReportError(dom->conn, dom, NULL, VIR_ERR_OPERATION_FAILED, + "%s", _("domain is not running")); + goto cleanup; + } qemudShutdownVMDaemon(dom->conn, driver, vm); event = virDomainEventNewFromObj(vm, @@ -3258,17 +3278,36 @@ static int qemudDomainRestore(virConnect goto cleanup; } - /* Ensure the name and UUID don't already exist in an active VM */ + /* See if a VM with matching UUID already exists */ vm = virDomainFindByUUID(&driver->domains, def->uuid); - if (!vm) - vm = virDomainFindByName(&driver->domains, def->name); if (vm) { + /* UUID matches, but if names don't match, refuse it */ + if (STRNEQ(vm->def->name, def->name)) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + vm->def->name, uuidstr); + goto cleanup; + } + + /* UUID & name match, but if VM is already active, refuse it */ if (virDomainIsActive(vm)) { qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_INVALID, _("domain is already active as '%s'"), vm->def->name); goto cleanup; - } else { - virDomainObjUnlock(vm); + } + virDomainObjUnlock(vm); + } else { + /* UUID does not match, but if a name matches, refuse it */ + vm = virDomainFindByName(&driver->domains, def->name); + if (vm) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + def->name, uuidstr); + goto cleanup; } } @@ -3603,18 +3642,41 @@ static virDomainPtr qemudDomainDefine(vi if (virSecurityDriverVerify(conn, def) < 0) goto cleanup; - vm = virDomainFindByName(&driver->domains, def->name); + /* See if a VM with matching UUID already exists */ + vm = virDomainFindByUUID(&driver->domains, def->uuid); if (vm) { + /* UUID matches, but if names don't match, refuse it */ + if (STRNEQ(vm->def->name, def->name)) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + vm->def->name, uuidstr); + goto cleanup; + } + + /* UUID & name match */ virDomainObjUnlock(vm); newVM = 0; + } else { + /* UUID does not match, but if a name matches, refuse it */ + vm = virDomainFindByName(&driver->domains, def->name); + if (vm) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(vm->def->uuid, uuidstr); + qemudReportError(conn, NULL, NULL, VIR_ERR_OPERATION_FAILED, + _("domain '%s' is already defined with uuid %s"), + def->name, uuidstr); + goto cleanup; + } } if (!(vm = virDomainAssignDef(conn, &driver->domains, def))) { - virDomainDefFree(def); - goto cleanup; - } + goto cleanup; + } + def = NULL; vm->persistent = 1; if (virDomainSaveConfig(conn, @@ -3636,6 +3698,7 @@ static virDomainPtr qemudDomainDefine(vi if (dom) dom->id = vm->def->id; cleanup: + virDomainDefFree(def); if (vm) virDomainObjUnlock(vm); if (event) -- |: 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 1

[libvirt] PATCH: Don't reset / detach host PCI devices in test scripts !
by Daniel P. Berrange 06 May '09

06 May '09

The PCI passthrough patches made it so that qemudBuildCommandLine() would actually try to detach your host devices & reset them. Most definitely not what you want when running this via a test case! This patch moves the host device management out into a separate method, so that qemudBuildCommandLine() doesn't do anything except safely build the command line. Daniel qemu_conf.c | 46 ----------------------------------- qemu_driver.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 76 insertions(+), 46 deletions(-) Index: src/qemu_conf.c =================================================================== RCS file: /data/cvs/libvirt/src/qemu_conf.c,v retrieving revision 1.133 diff -u -p -u -p -r1.133 qemu_conf.c --- src/qemu_conf.c 2 Mar 2009 20:22:35 -0000 1.133 +++ src/qemu_conf.c 2 Mar 2009 20:47:36 -0000 @@ -47,7 +47,6 @@ #include "datatypes.h" #include "xml.h" #include "nodeinfo.h" -#include "pci.h" #define VIR_FROM_THIS VIR_FROM_QEMU @@ -1395,52 +1394,7 @@ int qemudBuildCommandLine(virConnectPtr ADD_ARG_LIT("-pcidevice"); ADD_ARG_LIT(pcidev); VIR_FREE(pcidev); - - if (hostdev->managed) { - pciDevice *dev = pciGetDevice(conn, - hostdev->source.subsys.u.pci.domain, - hostdev->source.subsys.u.pci.bus, - hostdev->source.subsys.u.pci.slot, - hostdev->source.subsys.u.pci.function); - if (!dev) - goto error; - - if (pciDettachDevice(conn, dev) < 0) { - pciFreeDevice(conn, dev); - goto error; - } - - pciFreeDevice(conn, dev); - } /* else { - XXX validate that non-managed device isn't in use, eg - by checking that device is either un-bound, or bound - to pci-stub.ko - } */ } - - } - - /* Now that all the PCI hostdevs have be dettached, we can reset them */ - for (i = 0 ; i < vm->def->nhostdevs ; i++) { - virDomainHostdevDefPtr hostdev = vm->def->hostdevs[i]; - pciDevice *dev; - - if (hostdev->mode != VIR_DOMAIN_HOSTDEV_MODE_SUBSYS || - hostdev->source.subsys.type != VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_PCI) - continue; - - dev = pciGetDevice(conn, - hostdev->source.subsys.u.pci.domain, - hostdev->source.subsys.u.pci.bus, - hostdev->source.subsys.u.pci.slot, - hostdev->source.subsys.u.pci.function); - if (!dev) - goto error; - - if (pciResetDevice(conn, dev) < 0) - goto error; - - pciFreeDevice(conn, dev); } if (migrateFrom) { Index: src/qemu_driver.c =================================================================== RCS file: /data/cvs/libvirt/src/qemu_driver.c,v retrieving revision 1.208 diff -u -p -u -p -r1.208 qemu_driver.c --- src/qemu_driver.c 2 Mar 2009 17:39:43 -0000 1.208 +++ src/qemu_driver.c 2 Mar 2009 20:47:36 -0000 @@ -1133,6 +1133,79 @@ static int qemudNextFreeVNCPort(struct q return -1; } +static int qemuPrepareHostDevices(virConnectPtr conn, + virDomainDefPtr def) { + int i; + + /* We have to use 2 loops here. *All* devices must + * be detached before we reset any of them, because + * in some cases you have to reset the whole PCI bus, + * which impacts all devices on it + */ + + for (i = 0 ; i < def->nhostdevs ; i++) { + virDomainHostdevDefPtr hostdev = def->hostdevs[i]; + + if (hostdev->mode != VIR_DOMAIN_HOSTDEV_MODE_SUBSYS) + continue; + if (hostdev->source.subsys.type != VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_PCI) + continue; + + if (!hostdev->managed) { + pciDevice *dev = pciGetDevice(conn, + hostdev->source.subsys.u.pci.domain, + hostdev->source.subsys.u.pci.bus, + hostdev->source.subsys.u.pci.slot, + hostdev->source.subsys.u.pci.function); + if (!dev) + goto error; + + if (pciDettachDevice(conn, dev) < 0) { + pciFreeDevice(conn, dev); + goto error; + } + + pciFreeDevice(conn, dev); + } /* else { + XXX validate that non-managed device isn't in use, eg + by checking that device is either un-bound, or bound + to pci-stub.ko + } */ + } + + /* Now that all the PCI hostdevs have be dettached, we can safely + * reset them */ + for (i = 0 ; i < def->nhostdevs ; i++) { + virDomainHostdevDefPtr hostdev = def->hostdevs[i]; + pciDevice *dev; + + if (hostdev->mode != VIR_DOMAIN_HOSTDEV_MODE_SUBSYS) + continue; + if (hostdev->source.subsys.type != VIR_DOMAIN_HOSTDEV_SUBSYS_TYPE_PCI) + continue; + + dev = pciGetDevice(conn, + hostdev->source.subsys.u.pci.domain, + hostdev->source.subsys.u.pci.bus, + hostdev->source.subsys.u.pci.slot, + hostdev->source.subsys.u.pci.function); + if (!dev) + goto error; + + if (pciResetDevice(conn, dev) < 0) { + pciFreeDevice(conn, dev); + goto error; + } + + pciFreeDevice(conn, dev); + } + + return 0; + +error: + return -1; +} + static virDomainPtr qemudDomainLookupByName(virConnectPtr conn, const char *name); @@ -1210,6 +1283,9 @@ static int qemudStartVMDaemon(virConnect return -1; } + if (qemuPrepareHostDevices(conn, vm->def) < 0) + return -1; + vm->def->id = driver->nextvmid++; if (qemudBuildCommandLine(conn, driver, vm, qemuCmdFlags, &argv, &progenv, -- |: 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 :|

4 5

[libvirt] [PATCH] Refresh QEMU driver caps in getCapabilities
by Cole Robinson 06 May '09

06 May '09

Hi all, The attached patch fixes QEMU getCapabilities to refresh caps before returning them to the user. Currently the capabilities are only refreshed once (at daemon startup), which means libvirtd needs to be restarted to pick up changes if QEMU or KVM are installed while the daemon is running. See: https://bugzilla.redhat.com/show_bug.cgi?id=460649 There are several things 'wrong' with this change: - We reset/rescan fields that won't change (host arch, mac address prefix). This should be fixed at some point, but isn't a big deal since total performance impact is negligible (see below). - We only refresh the capabilities when the user calls getCapabilities, which means we are still carrying around stale caps prior to that, which is what libvirt validates against. In practice, virt-manager and virt-install both call getCapabilities often so this isn't an issue, though the caps internal API should probably address this at some point. To test the performance impact, I used a simple python script: import libvirt conn = libvirt.open("qemu:///system") for i in range(0, 30): conn.getCapabilities() The time difference was on average .02 seconds slower, which I think is negligible. If at somepoint in the future, capabilities generation becomes smarter (searching PATH emulators, scraping device list output, etc.) it might be worth re-checking the time impact. But for now it doesn't seem to be an issue. Thanks, Cole

4 7

[libvirt] [PATCH 1/1] Changes in version detection and moving a global variable in the global struct
by Pritesh Kothari 06 May '09

06 May '09

Hi All, PATCH 1/1: contains changes in version detection and moving a global variable in the global struct, so that no more global vairables exists. Regards, Pritesh

3 4

[libvirt] [PATCH 1/1] Making path detection more robust in VirtualBox
by Pritesh Kothari 06 May '09

06 May '09

Hi All, PATCH 1/1: contains changes in the glue code for making path detection more robust in VirtualBox Regards, Pritesh

3 2

[libvirt] participate in irc chat
by jens 06 May '09

06 May '09

Hello, on http://libvirt.org/contact.html I read about the #virt irc channel. How can I participate in this channel, it seems to be locked because I have to specify a channel key. Greetings Jens

2 2

[libvirt] libvirt / virt-manager issue with NFS share as storage plus a philosophy question - resend w/subject
by mike.hinz＠yr20.com 06 May '09

06 May '09

I had originally posted a plea for help on the virt-manager related list, but the same issue exists with virsh and this list seems to have higher traffic. I'd been chatting on the irc channel with various people about this, but wanted to put this on the mail list for the record and to also raise a philosophical question. First, we've confirmed that Fedora 11 preview (fully updated) has an issue with properly mounting NFS shares without the noacl option. Fedora 10 works fine with regard to mounting NFS shares. That issue causes problems when attempting to write files to an NFS share and more importantly, for this group, causes problems when attempting to create an NFS based storage pool and then subsequently creating volumes on said NFS share. That bug is 499178. This was all figured out with Openfiler support and with the support of the fine folks working on libvirt and virt-manager. There is a somewhat ugly workaround for this and I can describe it in case anyone is interested. Now comes the philosophy piece: It seems like that the way that libvirt and virt-manager want to handle storage is to be able to fully control the mounting process. That seems to me to be a very nice thing as you have a single way to manage storage along with your VMs. However, at least with respect to NFS, there are a number of parameters that people might need to give to the mount command for things like performance optimization and who knows what else. Note that this is a consideration completely aside from the bug issue above. It seems to me that either you have to just use the user created mount points, either manually or by fstab OR you have to allow all the possible options to be passed if you want to fully control the process. Right now, it's tricky to do, if you want to get paras passed for tuning or whatever and requires some tricky handwork to get it done. One way to accomplish this from the user input perspective would be to add an 'options' input field that could take multiple parameters on the step 2 of 2 page of the 'add storage pool' function. So, is my thinking all screwed up about this or is there merit to this concept? I tend to think, at least right now, that the logical thing is to either have the user create the mount points through the normal mechanisms that have in place roughly forever and then you just ask for what these are with respect to the NFS pools or you have to allow for user to pass the required parameters via your mechanism, although there are certainly many, many options when it comes to NFS. Thanks for all the help to date! I hope the above is helpful! Regards, Mike Hinz President YR20 1718 Fry Road, Suite 440 Houston, TX 77084 mike.hinz(a)yr20.com 832-225-1293 (o) 713-594-3095 (m) 832-550-2657 (f) Regards, Mike Hinz President YR20 1718 Fry Road, Suite 440 Houston, TX 77084 mike.hinz(a)yr20.com 832-225-1293 (o) 713-594-3095 (m) 832-550-2657 (f)

2 1