November 2016 - Devel - libvirt List Archives

[libvirt] [PATCH] Fix scheduler support check
by Martin Kletzander 18 Nov '16

18 Nov '16

Commit 94cc577807ba tried fixing build on systems that did not have SCHED_BATCH or SCHED_IDLE defined. But instead of changing it to conditional support, it rather completely disabled the support for setting any scheduler. Since then, such old systems are not supported, but rather than reverting that commit, let's change that to the conditional support. That way any addition to the list of schedulers can follow the same style so that we're consistent in the future. Signed-off-by: Martin Kletzander <mkletzan(a)redhat.com> --- Notes: Of course there are various ways how to address that, I went with case. Also defining undefined SCHED_* to -1 makes some gnulib syntax-check go haywire. src/util/virprocess.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) diff --git a/src/util/virprocess.c b/src/util/virprocess.c index 718c4a2664e1..39d6b30c40f2 100644 --- a/src/util/virprocess.c +++ b/src/util/virprocess.c @@ -1183,7 +1183,7 @@ virProcessExitWithStatus(int status) exit(value); } -#if HAVE_SCHED_SETSCHEDULER && defined(SCHED_BATCH) && defined(SCHED_IDLE) +#if HAVE_SCHED_SETSCHEDULER static int virProcessSchedTranslatePolicy(virProcessSchedPolicy policy) @@ -1196,10 +1196,18 @@ virProcessSchedTranslatePolicy(virProcessSchedPolicy policy) return SCHED_BATCH; case VIR_PROC_POLICY_IDLE: +# ifdef SCHED_IDLE return SCHED_IDLE; +# else + return -1; +# endif case VIR_PROC_POLICY_FIFO: +# ifdef SCHED_FIFO return SCHED_FIFO; +# else + return -1; +# endif case VIR_PROC_POLICY_RR: return SCHED_RR; @@ -1225,6 +1233,13 @@ virProcessSetScheduler(pid_t pid, if (!policy) return 0; + if (pol < 0) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + _("Scheduler '%s' is not supported on this platform"), + virProcessSchedPolicyTypeToString(policy)); + return -1; + } + if (pol == SCHED_FIFO || pol == SCHED_RR) { int min = 0; int max = 0; -- 2.10.2

2 2

[libvirt] [PATCH] Gathering network interface statistics with openvswitch
by Mehdi Abaakouk 18 Nov '16

18 Nov '16

Hi, To follow up, my question about vhostuser/ovs interfaces statistics. This is my porposal to implement this feature. Regards, Mehdi Abaakouk (1): virstats: Gathering net interface stats with ovs src/util/virstats.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 96 insertions(+), 3 deletions(-) -- 2.10.2

2 2

[libvirt] [PATCH RFC 0/7] Run qemu under its own namespace
by Michal Privoznik 17 Nov '16

17 Nov '16

This is just an proof of concept of what has been agreed on here: https://www.redhat.com/archives/libvir-list/2016-November/msg00285.html There is still a lot of to be done: - set up seclabels - implement hot(un-)plug - implement other devices, not just disks I'm sending these in a hope that somebody will at least take a quick look. I'm not looking for a code cleanliness (but if you find some issues feel free to raise them), more than design confirmation. If I'm going in wrong direction I'd rather stop now before digging any deeper. Michal Privoznik (7): virprocess: Introduce virProcessSetupPrivateNS virfile: Introduce virFilePopulateDevices virfile: Introduce virFileSetupDev virfile: Introduce virFileSetupDevPTS virfile: Introduce virFileBindMountDevice qemu: Spawn qemu under mount namespace qemu: Prepare disks when starting a domain src/libvirt_private.syms | 5 + src/lxc/lxc_container.c | 20 +--- src/lxc/lxc_controller.c | 82 +++---------- src/qemu/qemu_domain.c | 306 +++++++++++++++++++++++++++++++++++++++++++++++ src/qemu/qemu_domain.h | 12 ++ src/qemu/qemu_process.c | 13 ++ src/util/virfile.c | 116 ++++++++++++++++++ src/util/virfile.h | 21 ++++ src/util/virprocess.c | 24 ++++ src/util/virprocess.h | 2 + 10 files changed, 517 insertions(+), 84 deletions(-) -- 2.8.4

2 15

[libvirt] [PATCH v1] RFC Generating Qemu parameter -cpu features
by Collin L. Walling 17 Nov '16

17 Nov '16

I'm working on the CPU model checking support in libvirt for s390x and I've found when generating the -cpu +feature,-feature command, Qemu does not like the ± syntax when adding cpu features for s390x machines and expects us to use key=value pairs (e.g. feature=on,feature=off). Since x86 Qemu code supports both the ± and key=value methods of enabling CPU features, could we switch to just using key=value syntax in libvirt or do we need to maintain the ± format as well? I assume we could do something more sophisticated than wrapping the check in an if cpu->arch == VIR_ARCH_S390X? Another option would be to just change Qemu to support both formats, but I suspect we would get push back as key=value seems to be the preferred interface. Thanks for your time. - Collin Collin L. Walling (1): qemu: command: rework cpu feature argument support src/qemu/qemu_command.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) -- 2.7.4

4 9

[libvirt] vhostuser and openvswitch interface statistics
by Mehdi Abaakouk 17 Nov '16

17 Nov '16

Hi, I'm looking for interface statistics for a vhostuser network interface (that uses ovs-dpdk on the other socket side). The "virsh domifstat <domain> <ifname>" obviously don't works because it looks at "/proc/net/dev" for statistics. When we use vhostuser, we don't create network interface on the host but a socket. Many tools currently rely on the domainInterfaceStats() API to retrieve this kind of information. So I would like to known if their is a plan to support ovs as network interface statistics source (with something like "ovs-vsctl get Interface <ifname> statistics") ? I have started to write a PoC code that fallback to OVS in virNetInterfaceStats() when we don't found the interface in /proc/net/dev. Does this kind of contribution is welcome/match your plan ? Regards, -- Mehdi Abaakouk mail: sileht(a)sileht.net irc: sileht

2 1

[libvirt] [PATCH for-2.9] qmp: Report QOM type name on query-cpu-definitions
by Eduardo Habkost 17 Nov '16

17 Nov '16

The new typename attribute on query-cpu-definitions will be used to help management software use device-list-properties to check which properties can be set using -cpu or -global for the CPU model. Signed-off-by: Eduardo Habkost <ehabkost(a)redhat.com> --- qapi-schema.json | 4 +++- target-arm/helper.c | 1 + target-i386/cpu.c | 1 + target-ppc/translate_init.c | 1 + target-s390x/cpu_models.c | 1 + 5 files changed, 7 insertions(+), 1 deletion(-) diff --git a/qapi-schema.json b/qapi-schema.json index b0b4bf6..9a3bdd4 100644 --- a/qapi-schema.json +++ b/qapi-schema.json @@ -3216,6 +3216,8 @@ # @unavailable-features: #optional List of properties that prevent # the CPU model from running in the current # host. (since 2.8) +# @typename: Type name that can be used as argument to @device-list-properties, +# to introspect properties configurable using -cpu or -global. # # @unavailable-features is a list of QOM property names that # represent CPU model attributes that prevent the CPU from running. @@ -3237,7 +3239,7 @@ ## { 'struct': 'CpuDefinitionInfo', 'data': { 'name': 'str', '*migration-safe': 'bool', 'static': 'bool', - '*unavailable-features': [ 'str' ] } } + '*unavailable-features': [ 'str' ], 'typename': 'str' } } ## # @query-cpu-definitions: diff --git a/target-arm/helper.c b/target-arm/helper.c index b5b65ca..3fc01b5 100644 --- a/target-arm/helper.c +++ b/target-arm/helper.c @@ -5207,6 +5207,7 @@ static void arm_cpu_add_definition(gpointer data, gpointer user_data) info = g_malloc0(sizeof(*info)); info->name = g_strndup(typename, strlen(typename) - strlen("-" TYPE_ARM_CPU)); + info->q_typename = g_strdup(typename); entry = g_malloc0(sizeof(*entry)); entry->value = info; diff --git a/target-i386/cpu.c b/target-i386/cpu.c index 6eec5dc..725f6cb 100644 --- a/target-i386/cpu.c +++ b/target-i386/cpu.c @@ -2239,6 +2239,7 @@ static void x86_cpu_definition_entry(gpointer data, gpointer user_data) info->name = x86_cpu_class_get_model_name(cc); x86_cpu_class_check_missing_features(cc, &info->unavailable_features); info->has_unavailable_features = true; + info->q_typename = g_strdup(object_class_get_name(oc)); entry = g_malloc0(sizeof(*entry)); entry->value = info; diff --git a/target-ppc/translate_init.c b/target-ppc/translate_init.c index 208fa1e..42b9274 100644 --- a/target-ppc/translate_init.c +++ b/target-ppc/translate_init.c @@ -10305,6 +10305,7 @@ CpuDefinitionInfoList *arch_query_cpu_definitions(Error **errp) info = g_malloc0(sizeof(*info)); info->name = g_strdup(alias->alias); + info->q_typename = g_strdup(object_class_get_name(oc)); entry = g_malloc0(sizeof(*entry)); entry->value = info; diff --git a/target-s390x/cpu_models.c b/target-s390x/cpu_models.c index c1e729d..5b66d33 100644 --- a/target-s390x/cpu_models.c +++ b/target-s390x/cpu_models.c @@ -290,6 +290,7 @@ static void create_cpu_model_list(ObjectClass *klass, void *opaque) info->has_migration_safe = true; info->migration_safe = scc->is_migration_safe; info->q_static = scc->is_static; + info->q_typename = g_strdup(object_class_get_name(klass)); entry = g_malloc0(sizeof(*entry)); -- 2.7.4

3 4

[libvirt] fix for recent gnutls behavior change
by Christian Ehrhardt 17 Nov '16

17 Nov '16

TL;DR: please think if more needs to be changed to follow new gnutls behavior Hi, I found that recently our builds broke with these errors: TEST: virnettlssessiontest [...] 10) TLS Session servercertreq.filename + clientcertreq.filename ... FAILED 11) TLS Session servercertreq.filename + clientcertreq.filename ... FAILED 12) TLS Session servercertreq.filename + clientcertreq.filename ... OK 13) TLS Session servercertreq.filename + clientcertreq.filename ... FAILED 14) TLS Session servercertreq.filename + clientcertreq.filename ... FAILED 15) TLS Session servercertlevel3areq.filename + clientcertlevel2breq.filename ... OK FAIL virnettlssessiontest (exit status: 1) I was able to track that down to a change in gnutls which can be read at https://gitlab.com/gnutls/gnutls/issues/111 which changed the behavior of dname en- and decoding to follow RFC4514. More details on the debugging - if anyone want to reproduce - can be found on https://bugs.launchpad.net/ubuntu/+source/libvirt/+bug/1641615 But the primary purpose of the cover letter is a call to everybody to think if that change could imply the need for more changes in libvirt than just to make the tests work again. Christian Ehrhardt (1): tests: adapt to gnutls change in dname en-/decoding tests/virnettlssessiontest.c | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+) -- 2.7.4

2 3

[libvirt] [PATCH v3 00/18] Use mode PCIe less legacy PCI
by Laine Stump 16 Nov '16

16 Nov '16

(this time with auto-root-port-add goodness!) Last month I posted a short patch series that attempted to auto-assign addresses on PCI Express controllers to devices that were PCI Express devices. https://www.redhat.com/archives/libvir-list/2016-August/msg00732.html In particular, it would assign PCI virtio-1.0 devices, e1000e network devices, and nec-usb-xhci devices to hotpluggable PCIe express ports *if any were available*. However, a basic PCIe machine doesn't have any hotpluggable PCIe ports, and the patchset I sent had no provision for automatically adding any. This new patchset takes care of that limitation by automatically adding pci-root-ports as they are needed, and also automatically adds a dmi-to-pci-bridge device (to create a legacy PCI bus hierarchy) when needed so that we can do away with the code that *always* adds one (and there is a patch that *does* do away with that code :-). Once all of that was done, it turned out that virt-manager could create an *almost* legacy-PCI-free Q35 domain config - the only legacy PCI device was the sound device. Since the ich9 sound device is integrated into the Intel ich9 chip (which is part of real Q35 hardware), as a curiousity I made an RFC patch that attempts to place any ich9 audio device at 00:1B.0, which is the address where real Q35 hardware puts this device. With that patch in place, all you have to do to get a legacy-free Q35 config out of virt-manager is switch the sound device model from ich6 to ich9. (I don't expect that patch will be pushed, but it's nice to see this result). Although Andrea had ACKed most of the patches in the last patchset, I hadn't wanted to push them without accompanying patches to auto-add the pcie-root-ports (since doing so would render virt-manager + new libvirt unusable for Q35 domains). Since then I've decided on a cleaner manner for setting device connectFlags, so all but Patch 1 of the last set was discarded and re-written from scratch. Although there are still a couple more things I'd like to do, these patches can be pushed without a serious regression in functionality for Q35 domains: 1) currently I put each new pcie-root-port on its own slot, rather than putting 8 of them on the different functions of a single slot. This means you can only get ~30 devices before you have to start manually adding pcie controllers. That will be remedied soon. 2) I don't have a patch yet to read a PCI device's capabilities to see whether or not it is an Express device and change the connectFlags accordingly. In the meantime if you want to plug in a vfio assigned device, you'll either need to manually address it to a PCIe port, or you'll need to manually add a pci-bridge device to the config. 3) I don't do anything to assure there are any unused pcie-root-ports available for hotplug. We need to fix that, but I'm not sure how many to "reserve". Suggestions I've heard have been "1", "2", "4", and "just add pcie-root-ports 8 at a time and you'll have 'something between 1 and 8' available". Laine Stump (18): conf: restrict what type of buses will accept a pci-bridge qemu: replace a lot of "def->controllers[i]" with equivalent "cont" qemu: new functions qemuDomainMachineHasPCI[e]Root() qemu: new functions qemuDomainDeviceConnectFlags*() conf: new function virDomainPCIAddressReserveNextAddr() qemu: use virDomainPCIAddressReserveNextAddr in qemuDomainAssignDevicePCISlots conf: make virDomainPCIAddressGetNextSlot() a local static function qemu: replace calls to virDomainPCIAddressReserveNext*() with static function qemu: set/use info->pciConnectFlags during qemuDomainAssignDevicePCISlots qemu: set/use proper pciConnectFlags during hotplug qemu: assign virtio devices to PCIe slot when appropriate qemu: assign e1000e network devices to PCIe slots when appropriate qemu: assign nec-xhci (USB3) controller to a PCIe address when appropriate qemu: only force an available legacy-PCI slot on domains with pci-root qemu: auto-add pcie-root-port/dmi-to-pci-bridge controllers as needed qemu: don't force-add a dmi-to-pci-bridge just on principle qemu: add a USB3 controller to Q35 domains by default [RFC] qemu: try to put ich9 sound device at 00:1B.0 src/conf/device_conf.h | 5 + src/conf/domain_addr.c | 180 ++++- src/conf/domain_addr.h | 15 +- src/libvirt_private.syms | 2 +- src/qemu/qemu_domain.c | 50 +- src/qemu/qemu_domain.h | 2 + src/qemu/qemu_domain_address.c | 802 +++++++++++++++------ src/qemu/qemu_domain_address.h | 4 + src/qemu/qemu_hotplug.c | 25 +- tests/qemuxml2argvdata/qemuxml2argv-autoindex.args | 10 +- tests/qemuxml2argvdata/qemuxml2argv-pcie-root.args | 3 +- .../qemuxml2argv-q35-default-devices-only.args | 22 + .../qemuxml2argv-q35-default-devices-only.xml | 18 + .../qemuxml2argv-q35-pcie-autoadd.args | 56 ++ .../qemuxml2argv-q35-pcie-autoadd.xml | 51 ++ tests/qemuxml2argvdata/qemuxml2argv-q35-pcie.args | 58 ++ tests/qemuxml2argvdata/qemuxml2argv-q35-pcie.xml | 67 ++ .../qemuxml2argv-q35-virt-manager-basic.args | 56 ++ .../qemuxml2argv-q35-virt-manager-basic.xml | 76 ++ .../qemuxml2argv-q35-virtio-pci.args | 58 ++ .../qemuxml2argv-q35-virtio-pci.xml | 1 + tests/qemuxml2argvtest.c | 123 ++++ .../qemuxml2xmlout-autoindex.xml | 10 +- .../qemuxml2xmlout-pcie-root.xml | 4 - .../qemuxml2xmlout-q35-default-devices-only.xml | 40 + .../qemuxml2xmlout-q35-pcie-autoadd.xml | 143 ++++ .../qemuxml2xmloutdata/qemuxml2xmlout-q35-pcie.xml | 152 ++++ .../qemuxml2xmlout-q35-virt-manager-basic.xml | 116 +++ .../qemuxml2xmlout-q35-virtio-pci.xml | 152 ++++ tests/qemuxml2xmltest.c | 108 +++ 30 files changed, 2117 insertions(+), 292 deletions(-) create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-default-devices-only.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-default-devices-only.xml create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pcie-autoadd.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pcie-autoadd.xml create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pcie.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pcie.xml create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-virt-manager-basic.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-virt-manager-basic.xml create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-virtio-pci.args create mode 120000 tests/qemuxml2argvdata/qemuxml2argv-q35-virtio-pci.xml create mode 100644 tests/qemuxml2xmloutdata/qemuxml2xmlout-q35-default-devices-only.xml create mode 100644 tests/qemuxml2xmloutdata/qemuxml2xmlout-q35-pcie-autoadd.xml create mode 100644 tests/qemuxml2xmloutdata/qemuxml2xmlout-q35-pcie.xml create mode 100644 tests/qemuxml2xmloutdata/qemuxml2xmlout-q35-virt-manager-basic.xml create mode 100644 tests/qemuxml2xmloutdata/qemuxml2xmlout-q35-virtio-pci.xml -- 2.7.4

6 67

[libvirt] [PATCH v2] logical: Check for existing vgname before building
by John Ferlan 16 Nov '16

16 Nov '16

https://bugzilla.redhat.com/show_bug.cgi?id=1373711 Add a check for an existing volume group name before trying to build the volume group. Since the process of building a vg involves wiping the first 512 bytes and using pvcreate on each source device path before creating the vg - we could conceivably overwrite something that we shouldn't be. Also, once a pv is part of a vg, the pvcreate would fail unless we chose to overwrite the volume. Signed-off-by: John Ferlan <jferlan(a)redhat.com> --- Difference w/ v2 is that I'm taking a different approach - disallow the second pool-build since the vg already exists. NB: libvirt has no API to extend an existing vg - that's left to an admin anyway. src/storage/storage_backend_logical.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-) diff --git a/src/storage/storage_backend_logical.c b/src/storage/storage_backend_logical.c index ca05fe1..b241495 100644 --- a/src/storage/storage_backend_logical.c +++ b/src/storage/storage_backend_logical.c @@ -682,14 +682,31 @@ virStorageBackendLogicalBuildPool(virConnectPtr conn ATTRIBUTE_UNUSED, virStoragePoolObjPtr pool, unsigned int flags) { - virCommandPtr vgcmd; + virCommandPtr vgcmd = NULL; int fd; char zeros[PV_BLANK_SECTOR_SIZE]; int ret = -1; size_t i; + virStoragePoolSourceList sourceList; virCheckFlags(0, -1); + /* Let's make sure the about to be created vg doesn't already exist */ + memset(&sourceList, 0, sizeof(sourceList)); + sourceList.type = VIR_STORAGE_POOL_LOGICAL; + + if (virStorageBackendLogicalGetPoolSources(&sourceList) < 0) + return -1; + + for (i = 0; i < sourceList.nsources; i++) { + if (STREQ(sourceList.sources[i].name, pool->def->source.name)) { + virReportError(VIR_ERR_STORAGE_POOL_BUILT, + _("volume group with name '%s' already exists"), + pool->def->source.name); + goto cleanup; + } + } + memset(zeros, 0, sizeof(zeros)); vgcmd = virCommandNewArgList(VGCREATE, pool->def->source.name, NULL); @@ -751,6 +768,9 @@ virStorageBackendLogicalBuildPool(virConnectPtr conn ATTRIBUTE_UNUSED, ret = 0; cleanup: + for (i = 0; i < sourceList.nsources; i++) + virStoragePoolSourceClear(&sourceList.sources[i]); + VIR_FREE(sourceList.sources); virCommandFree(vgcmd); return ret; } -- 2.7.4

2 2

[libvirt] [PATCH] logical: Allow [no]overwrite option for poolBuild
by John Ferlan 16 Nov '16

16 Nov '16

https://bugzilla.redhat.com/show_bug.cgi?id=1373711 When using pool-create --build or pool-define and pool-build to create/define and build/initialize the logical pool, the 'pvcreate' command could fail if some previous attempt had been used on the source path previously. So add support for the --override option in order to force usage of the magic override option "-ff" and the option "-y" in order to force creation and answer any questions with yes. NB: Although the reality is part of the process of building the logical pool involves wiping out the first 512 bytes of the disk block to be added (e.g. the partition table) because pvcreate requires that. So, to a degree the device being added is already altered. I suspect the knowlege of whether the disk was in a vg already could be contained outside the range of the first 512 bytes. Signed-off-by: John Ferlan <jferlan(a)redhat.com> --- src/storage/storage_backend_logical.c | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-) diff --git a/src/storage/storage_backend_logical.c b/src/storage/storage_backend_logical.c index ca05fe1..9c76156 100644 --- a/src/storage/storage_backend_logical.c +++ b/src/storage/storage_backend_logical.c @@ -682,13 +682,18 @@ virStorageBackendLogicalBuildPool(virConnectPtr conn ATTRIBUTE_UNUSED, virStoragePoolObjPtr pool, unsigned int flags) { - virCommandPtr vgcmd; + virCommandPtr vgcmd = NULL; int fd; char zeros[PV_BLANK_SECTOR_SIZE]; int ret = -1; size_t i; - virCheckFlags(0, -1); + virCheckFlags(VIR_STORAGE_POOL_BUILD_OVERWRITE | + VIR_STORAGE_POOL_BUILD_NO_OVERWRITE, ret); + + VIR_EXCLUSIVE_FLAGS_GOTO(VIR_STORAGE_POOL_BUILD_OVERWRITE, + VIR_STORAGE_POOL_BUILD_NO_OVERWRITE, + cleanup); memset(zeros, 0, sizeof(zeros)); @@ -731,10 +736,21 @@ virStorageBackendLogicalBuildPool(virConnectPtr conn ATTRIBUTE_UNUSED, /* * Initialize the physical volume because vgcreate is not * clever enough todo this for us :-( + * + * If this API is called twice for the same device, then because + * vgcmd adds the device to the volume group, the pvcreate will + * fail since the pv is already part of the vg. Allow use of the + * override option inorder to overrule! */ - pvcmd = virCommandNewArgList(PVCREATE, - pool->def->source.devices[i].path, - NULL); + if (flags & VIR_STORAGE_POOL_BUILD_OVERWRITE) + pvcmd = virCommandNewArgList(PVCREATE, + "-ff", "-y", + pool->def->source.devices[i].path, + NULL); + else + pvcmd = virCommandNewArgList(PVCREATE, + pool->def->source.devices[i].path, + NULL); if (virCommandRun(pvcmd, NULL) < 0) { virCommandFree(pvcmd); goto cleanup; -- 2.7.4

2 3