November 2012 - Devel - libvirt List Archives

[libvirt] Remote connection to KVM hypervisor failed using libvirt 0.10.2
by Ido Levy 23 Nov '12

23 Nov '12

Hello, I compiled libvirt 0.10.2 from source providing extra parameters to the configure procedure to support phyp and hyper-v driver support, and python binding. When running virsh from the tools directory (libvirt-0.10.2/tools) and trying to connect to remote KVM hypervisor using connect qemu+ssh://user@hostname/system I got the following error error: Failed to connect to the hypervisor error: End of file while reading data: : Input/output error The same scenario working without any issue using virsh tool that is shipped with RHEL distribution. Any advise how can I figure out how to resolve this. Thanks, Ido

3 2

[libvirt] QMP Capability Detection Issues with GIT version of QEMU
by Viktor Mihajlovski 23 Nov '12

23 Nov '12

I ran into trouble today trying to use the GIT level of QEMU. In a nutshell: the capability detection with QMP is failing and the fallback using -help isn't working with the GIT level probably due to help text reformatting. The failure reason is that QEMU cannot bind to the QMP monitor socket in the /var/lib/libvirt/qemu directory. That's because the child process is stripped of all capabilities and this directory is chown'ed to qemu:qemu by the QEMU driver. Note that this is failing with the release QEMU as well, with the difference that the fallback is working there. I am willing to provide a patch, however I'd like to get feedback on the approach to use: 1. Add back Linux capabilities CAP_DAC_OVERRIDE, CAP_DAC_READ_SEARCH to allow QEMU to bind to the monitor socket. This seems to be hacky/dodging the problem. 2. Use a separate directory for the QMP probing instance of QEMU. 3. Run the QMP QEMU under the configured qemu user. This would be my favorite. 4. Other ideas? -- Mit freundlichen Grüßen/Kind Regards Viktor Mihajlovski IBM Deutschland Research & Development GmbH Vorsitzender des Aufsichtsrats: Martin Jetter Geschäftsführung: Dirk Wittkopp Sitz der Gesellschaft: Böblingen Registergericht: Amtsgericht Stuttgart, HRB 243294

2 3

[libvirt] [PATCH] Correct include-password option for domdisplay
by Martin Kletzander 23 Nov '12

23 Nov '12

The 'virsh domdisplay' command is able to display the password configured for spice, but it was missing for vnc type graphics. This is just a simple patch for that to work properly. --- tools/virsh-domain.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/tools/virsh-domain.c b/tools/virsh-domain.c index cc47383..18aa869 100644 --- a/tools/virsh-domain.c +++ b/tools/virsh-domain.c @@ -7073,6 +7073,23 @@ cmdDomDisplay(vshControl *ctl, const vshCmd *cmd) if (STREQ(scheme[iter], "vnc")) { /* VNC protocol handlers take their port number as 'port' - 5900 */ port -= 5900; + + if (vshCommandOptBool(cmd, "include-password")) { + /* Create our XPATH lookup for the SPICE password */ + virAsprintf(&xpath, + "string(/domain/devices/graphics" + "[@type='%s']/@passwd)", + scheme[iter]); + if (!xpath) { + virReportOOMError(); + goto cleanup; + } + + /* Attempt to get the SPICE password */ + passwd = virXPathString(xpath, ctxt); + VIR_FREE(xpath); + } + } else if (STREQ(scheme[iter], "spice")) { /* Create our XPATH lookup for the SPICE TLS Port */ virAsprintf(&xpath, "string(/domain/devices/graphics[@type='%s']" -- 1.8.0

5 10

[libvirt] FYI: Upgrade from pc-0.12 / pc-014 vs. savevm section "kvmclock"
by Philipp Hahn 23 Nov '12

23 Nov '12

Hello, I'm using libvirt to manage my KVM instances. I created a VM with qemu-kvm-0.12, later upgraded to qemu-kvm-0.14 and took a snapshot using libvirt. As the original VM was created with qemu-kvm-0.12, libvirt stored "pc-0.12" with its XML data. Now I upgraded to qemu-kvm-1.1.2, where reverting to that snapshot fails with a the message > Unknown savevm section or instance 'kvmclock' 0 > load of migration failed I tracked that down to hw/pc_piix.c:590 where > static QEMUMachine pc_machine_v0_12 = { > .name = "pc-0.12", > .desc = "Standard PC", > .init = pc_init_pci_no_kvmclock, is defined. If I change .init to pc_init_pci, I'm able to load the old snapshot. I think this is because kvm-0.14 always created the kvmclock device, while that was only later changed (0ec329da) to be created "on demand" for pc-0.14s onward. So the snapshot is no longer a pure pc-0.12, but some pc-0.12+something, which qemu-kvm-1.1.2 refuses to load. Now that I know what the problem is I just want to inform others, who might experience the same problem. My workaround is a patched qemu-kvm, where I changed that .init mentioned above, since an additional enabled kvmclock does not seem to do any harm. If someone has a better fix, I'm open to suggestions. Sincerely Philipp -- Philipp Hahn Open Source Software Engineer hahn(a)univention.de Univention GmbH be open. fon: +49 421 22 232- 0 Mary-Somerville-Str.1 D-28359 Bremen fax: +49 421 22 232-99 http://www.univention.de/

1 0

[libvirt] [PATCH] node_device: Fix the improper logic when enumerating SRIOV VF
by Osier Yang 23 Nov '12

23 Nov '12

pciGetVirtualFunctions returns 0 even if there is no "virtfn" entry under the device sysfs path. And pciGetVirtualFunctions returns -1 when it tries to get the PCI config space of the VF, however, with keeping the the VFs already detected. That's why udevProcessPCI and gather_pci_cap use logic like: if (!pciGetVirtualFunctions(syspath, &data->pci_dev.virtual_functions, &data->pci_dev.num_virtual_functions) || data->pci_dev.num_virtual_functions > 0) data->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_VIRTUAL_FUNCTION; to tag the PCI device with "virtual_function" cap. This results in a VF will aslo get "virtual_function" cap. This patch fixes it by: * Ignoring the VF which has failure of getting PCI config space of the device (given that the succesfully detected VFs are kept , it makes sense to not give up on the failure of one VF too) with a warning, so pciGetVirtualFunctions will always return 0 except out of memory. * Free the allocated *virtual_functions when out of memory And thus the logic can be changed to: /* Out of memory */ if (pciGetVirtualFunctions(syspath, &data->pci_dev.virtual_functions, &data->pci_dev.num_virtual_functions) < 0) goto out; else if (!pciGetVirtualFunctions(syspath, &data->pci_dev.virtual_functions, &data->pci_dev.num_virtual_functions) && (data->pci_dev.num_virtual_functions > 0)) data->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_VIRTUAL_FUNCTION; This also fixes a memory leak in the old codes (it realloc the *virtual_functions first, and doesn't shrink it when it fails on getting the PCI config space). --- src/node_device/node_device_hal.c | 13 +++++++++-- src/node_device/node_device_udev.c | 11 +++++++-- src/util/pci.c | 37 +++++++++++++++++++++++------------ 3 files changed, 42 insertions(+), 19 deletions(-) diff --git a/src/node_device/node_device_hal.c b/src/node_device/node_device_hal.c index ff73db0..69e6ebf 100644 --- a/src/node_device/node_device_hal.c +++ b/src/node_device/node_device_hal.c @@ -151,10 +151,17 @@ static int gather_pci_cap(LibHalContext *ctx, const char *udi, if (!pciGetPhysicalFunction(sysfs_path, &d->pci_dev.physical_function)) d->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_PHYSICAL_FUNCTION; - if (!pciGetVirtualFunctions(sysfs_path, &d->pci_dev.virtual_functions, - &d->pci_dev.num_virtual_functions) || - d->pci_dev.num_virtual_functions > 0) + if (!pciGetVirtualFunctions(sysfs_path, + &d->pci_dev.virtual_functions, + &d->pci_dev.num_virtual_functions) < 0) { + VIR_FREE(sysfs_path); + return -1; + } else if (!pciGetVirtualFunctions(sysfs_path, + &d->pci_dev.virtual_functions, + &d->pci_dev.num_virtual_functions) && + (d->pci_dev.num_virtual_functions > 0)) d->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_VIRTUAL_FUNCTION; + } VIR_FREE(sysfs_path); } diff --git a/src/node_device/node_device_udev.c b/src/node_device/node_device_udev.c index acd78f2..72b8850 100644 --- a/src/node_device/node_device_udev.c +++ b/src/node_device/node_device_udev.c @@ -484,9 +484,14 @@ static int udevProcessPCI(struct udev_device *device, if (!pciGetPhysicalFunction(syspath, &data->pci_dev.physical_function)) data->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_PHYSICAL_FUNCTION; - if (!pciGetVirtualFunctions(syspath, &data->pci_dev.virtual_functions, - &data->pci_dev.num_virtual_functions) || - data->pci_dev.num_virtual_functions > 0) + /* Out of memory */ + if (pciGetVirtualFunctions(syspath, &data->pci_dev.virtual_functions, + &data->pci_dev.num_virtual_functions) < 0) + goto out; + else if (!pciGetVirtualFunctions(syspath, + &data->pci_dev.virtual_functions, + &data->pci_dev.num_virtual_functions) && + (data->pci_dev.num_virtual_functions > 0)) data->pci_dev.flags |= VIR_NODE_DEV_CAP_FLAG_PCI_VIRTUAL_FUNCTION; ret = 0; diff --git a/src/util/pci.c b/src/util/pci.c index d1ad121..3b4d96a 100644 --- a/src/util/pci.c +++ b/src/util/pci.c @@ -1930,6 +1930,7 @@ pciGetVirtualFunctions(const char *sysfs_path, unsigned int *num_virtual_functions) { int ret = -1; + int i; DIR *dir = NULL; struct dirent *entry = NULL; char *device_link = NULL; @@ -1952,6 +1953,7 @@ pciGetVirtualFunctions(const char *sysfs_path, *num_virtual_functions = 0; while ((entry = readdir(dir))) { if (STRPREFIX(entry->d_name, "virtfn")) { + struct pci_config_address *config_addr = NULL; if (virBuildPath(&device_link, sysfs_path, entry->d_name) == -1) { virReportOOMError(); @@ -1960,24 +1962,23 @@ pciGetVirtualFunctions(const char *sysfs_path, VIR_DEBUG("Number of virtual functions: %d", *num_virtual_functions); - if (VIR_REALLOC_N(*virtual_functions, - (*num_virtual_functions) + 1) != 0) { - virReportOOMError(); - VIR_FREE(device_link); - goto out; - } if (pciGetPciConfigAddressFromSysfsDeviceLink(device_link, - &((*virtual_functions)[*num_virtual_functions])) != + &config_addr) != SRIOV_FOUND) { - /* We should not get back SRIOV_NOT_FOUND in this - * case, so if we do, it's an error. */ - virReportError(VIR_ERR_INTERNAL_ERROR, - _("Failed to get SR IOV function from device " - "link '%s'"), device_link); + VIR_WARN("Failed to get SRIOV function from device " + "link '%s'", device_link); VIR_FREE(device_link); - goto out; + continue; } else { + if (VIR_ALLOC_N(*virtual_functions, + *num_virtual_functions + 1) < 0) { + virReportOOMError(); + VIR_FREE(config_addr); + goto out; + } + + (*virtual_functions)[*num_virtual_functions] = config_addr; (*num_virtual_functions)++; } VIR_FREE(device_link); @@ -1985,8 +1986,18 @@ pciGetVirtualFunctions(const char *sysfs_path, } ret = 0; + goto cleanup; out: + if (*virtual_functions) { + for (i = 0; i < *num_virtual_functions; i++) + VIR_FREE((*virtual_functions)[i]); + VIR_FREE(*virtual_functions); + } + +cleanup: + if (device_link) + VIR_FREE(device_link); if (dir) closedir(dir); -- 1.7.7.6

1 0

[libvirt] [PATCH 0/3] IPv6 enhancements plus dnsmasq conf-file
by Gene Czarcinski 23 Nov '12

23 Nov '12

The three patches are submitted with "one" because there is a dependency between them. That is, patch 2 (DHCPv6) assumes that patch 1 (guest-to-guest IPV6) has been applied and patch 3 (dnsmasq conf-file) assumes that patch 2 has been applied. While other files are also changed, most of the changes occur in src/network/bridge_driver.c and result in the conflicts. Yes, patch 1 was previously submitted as a stand-alone patch but it became obvious that it was not really standalone. This submittal is a "re-do" of my previous submittal which involved similar changes. Specifically, the order of the patches have been changed so that the conf-file changes are after the DHCPv6 changes. In addition, some code has been cleaned up. For example, there is no need to modify the radvd configuration since both DHCPv6 and dnsmasq's handling of RA are both keyed to dnsmasq version 2.64 or later. Gene Czarcinski (3): v1: allow guest to guest IPv6 without gateway definition v7.1 add support for DHCPv6 v7.2: put dnsmasq parameters into conf-file docs/formatnetwork.html.in | 126 +++++- src/conf/network_conf.c | 100 +++-- src/conf/network_conf.h | 1 + src/network/bridge_driver.c | 461 +++++++++++++++------ src/network/bridge_driver.h | 7 +- src/util/dnsmasq.c | 9 +- tests/networkxml2argvdata/dhcp6-network.argv | 17 + tests/networkxml2argvdata/dhcp6-network.xml | 16 + tests/networkxml2argvdata/isolated-network.argv | 25 +- tests/networkxml2argvdata/nat-network-dhcp6.argv | 20 + tests/networkxml2argvdata/nat-network-dhcp6.xml | 26 ++ .../networkxml2argvdata/nat-network-dns-hosts.argv | 15 +- .../nat-network-dns-srv-record-minimal.argv | 37 +- .../nat-network-dns-srv-record.argv | 37 +- .../nat-network-dns-txt-record.argv | 31 +- tests/networkxml2argvdata/nat-network.argv | 29 +- tests/networkxml2argvdata/netboot-network.argv | 29 +- .../networkxml2argvdata/netboot-proxy-network.argv | 26 +- .../routed-network-dhcphost.argv | 15 + .../routed-network-dhcphost.xml | 19 + tests/networkxml2argvdata/routed-network.argv | 13 +- tests/networkxml2argvtest.c | 46 +- 22 files changed, 790 insertions(+), 315 deletions(-) create mode 100644 tests/networkxml2argvdata/dhcp6-network.argv create mode 100644 tests/networkxml2argvdata/dhcp6-network.xml create mode 100644 tests/networkxml2argvdata/nat-network-dhcp6.argv create mode 100644 tests/networkxml2argvdata/nat-network-dhcp6.xml create mode 100644 tests/networkxml2argvdata/routed-network-dhcphost.argv create mode 100644 tests/networkxml2argvdata/routed-network-dhcphost.xml -- 1.7.11.7

1 5

[libvirt] [PATCHv2] qemu: Stop recursive detection of image chains when an image is missing
by Peter Krempa 23 Nov '12

23 Nov '12

Commit e0c469e58b93f852a72265919703cb6abd3779f8 that fixes the detection of image chain wasn't complete. Iteration through the backing image chain has to stop at the last existing image if some of the images are missing otherwise the backing chain that is cached contains entries with paths being set to NULL resulting to: error: Unable to allow access for disk path (null): Bad address Fortunately stat() is kind enough not to crash when it's presented with a NULL argument. At least on Linux. --- Left meta->backingStoreRaw in place to mark broken chains, added a comment and a warning. --- src/util/storage_file.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/src/util/storage_file.c b/src/util/storage_file.c index 2249212..4417404 100644 --- a/src/util/storage_file.c +++ b/src/util/storage_file.c @@ -728,8 +728,13 @@ virStorageFileGetMetadataFromBuf(int format, meta->backingStore = absolutePathFromBaseFile(path, backing); if (meta->backingStore == NULL) { /* the backing file is (currently) unavailable, treat this - * file as standalone */ + * file as standalone: + * backingStoreRaw is kept to mark broken image chains */ + meta->backingStoreIsFile = false; backingFormat = VIR_STORAGE_FILE_NONE; + VIR_WARN("Backing file '%s' of image '%s' is missing.", + meta->backingStoreRaw, path); + } } VIR_FREE(backing); -- 1.8.0

2 2

[libvirt] [PATCH] build: trivial fix error: implicit declaration of function 'malloc'
by Natanael Copa 23 Nov '12

23 Nov '12

Fixes this error when building with -Werror on Alpine Linux: util/processinfo.c: In function 'virProcessInfoSetAffinity': util/processinfo.c:52:5: error: implicit declaration of function 'malloc' [-Werror=implicit-function-declaration] Signed-off-by: Natanael Copa <ncopa(a)alpinelinux.org> --- src/util/processinfo.c | 1 + 1 file changed, 1 insertion(+) diff --git a/src/util/processinfo.c b/src/util/processinfo.c index c3648d4..b8c60eb 100644 --- a/src/util/processinfo.c +++ b/src/util/processinfo.c @@ -21,6 +21,7 @@ #include <config.h> +#include <stdlib.h> #include <sched.h> #include "processinfo.h" -- 1.8.0

3 5

[libvirt] RFC: Enable unprivileged SG_IO
by Osier Yang 22 Nov '12

22 Nov '12

Hi, http://lwn.net/Articles/524720/ introduces new sysfs knob (unpriv_sgio) for SCSI device to allow the unprivileged SG_IO. I don't have solid thought on how the libvirt interface should be yet. It shouldn't be a XML entry of disk device, as the device can be shared by multiple guests, and the configuration should be kept same for all of them, having a XML entry for it will make things a mess. What Paolo suggested is to add an entry in qemu.conf, just like "cgroup_device_acl": sgio_device_acl = [ "/dev/sda" ] When libvirtd starting, set the sysfs knob "unpriv_sgio" of the devices listed to 1, and 0 when libvirtd exists. I don't quite agree with this approach, as entries in qemu.conf generally should be configuration for the whole qemu driver, however, the SG_IO setting is at the device layer, or not higher than guest layer. What I'm thinking about is to have a public API to tune the knob independantly with domain/driver, that means it's up to management apps to manage the knob's value, setting it to 1 before domain(s) starting, and 0 when no domain is using it. Any thoughts? Regards, Osier

3 4

[libvirt] [PATCH v2] conf: Report sensible error for invalid disk name
by Martin Kletzander 22 Nov '12

22 Nov '12

The error "... but the cause is unknown" appeared for XMLs similar to this: <disk type='file' device='cdrom'> <driver name='qemu' type='raw'/> <source file='/dev/zero'/> <target dev='sr0'/> </disk> Notice unsupported disk type (for the driver), but also no address specified. The first part is not a problem and we should not abort immediately because of that, but the combination with the address unknown was causing an unspecified error. While fixing this, I added an error to one place where this return value was not managed properly. --- v2: - Error moved from virDiskNameToIndex(a)util/util.c to virDomainDiskDefAssignAddress(a)conf/domain_conf.c - One more error added into qemuParseCommandLine(a)qemu/qemu_command.c src/conf/domain_conf.c | 6 +++++- src/qemu/qemu_command.c | 6 +++++- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c index ed21f0f..3a1be02 100644 --- a/src/conf/domain_conf.c +++ b/src/conf/domain_conf.c @@ -3063,8 +3063,12 @@ int virDomainDiskDefAssignAddress(virCapsPtr caps, virDomainDiskDefPtr def) { int idx = virDiskNameToIndex(def->dst); - if (idx < 0) + if (idx < 0) { + virReportError(VIR_ERR_XML_ERROR, + _("Unknown disk name '%s' and no address specified"), + def->dst); return -1; + } switch (def->bus) { case VIR_DOMAIN_DISK_BUS_SCSI: diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c index 22bb209..097de9b 100644 --- a/src/qemu/qemu_command.c +++ b/src/qemu/qemu_command.c @@ -8342,8 +8342,12 @@ virDomainDefPtr qemuParseCommandLine(virCapsPtr caps, !disk->dst) goto no_memory; - if (virDomainDiskDefAssignAddress(caps, disk) < 0) + if (virDomainDiskDefAssignAddress(caps, disk) < 0) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("Cannot assign address for device name '%s'"), + disk->dst); goto error; + } if (VIR_REALLOC_N(def->disks, def->ndisks+1) < 0) goto no_memory; -- 1.8.0

2 2