November 2016 - Devel - libvirt List Archives

[libvirt] [PATCH V2] qemu: Allow use of hot plugged host CPUs if no affinity set
by Viktor Mihajlovski 24 Nov '16

24 Nov '16

If the cpuset cgroup controller is disabled in /etc/libvirt/qemu.conf QEMU virtual machines can in principle use all host CPUs, even if they are hot plugged, if they have no explicit CPU affinity defined. However, there's libvirt code supposed to handle the situation where the libvirt daemon itself is not using all host CPUs. The code in qemuProcessInitCpuAffinity attempts to set an affinity mask including all defined host CPUs. Unfortunately, the resulting affinity mask for the process will not contain the offline CPUs. See also the sched_setaffinity(2) man page. That means that even if the host CPUs come online again, they won't be used by the QEMU process anymore. The same is true for newly hot plugged CPUs. So we are effectively preventing that QEMU uses all processors instead of enabling it to use them. It only makes sense to set the QEMU process affinity if we're able to actually grow the set of usable CPUs, i.e. if the process affinity is a subset of the online host CPUs. There's still the chance that for some reason the deliberately chosen libvirtd affinity matches the online host CPU mask by accident. In this case the behavior remains as it was before (CPUs offline while setting the affinity will not be used if they show up later on). Signed-off-by: Viktor Mihajlovski <mihajlov(a)linux.vnet.ibm.com> Tested-by: Matthew Rosato <mjrosato(a)linux.vnet.ibm.com> --- V2: - Avoid memory leak by freeing cpumap if the affinity has to be set. - Made an attempt to better describe the issue at hand src/qemu/qemu_process.c | 33 +++++++++++++++++++++++---------- 1 file changed, 23 insertions(+), 10 deletions(-) diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c index 4758c49..7e55086 100644 --- a/src/qemu/qemu_process.c +++ b/src/qemu/qemu_process.c @@ -2202,6 +2202,7 @@ qemuProcessInitCpuAffinity(virDomainObjPtr vm) int ret = -1; virBitmapPtr cpumap = NULL; virBitmapPtr cpumapToSet = NULL; + virBitmapPtr hostcpumap = NULL; qemuDomainObjPrivatePtr priv = vm->privateData; if (!vm->pid) { @@ -2223,21 +2224,32 @@ qemuProcessInitCpuAffinity(virDomainObjPtr vm) * the spawned QEMU instance to all pCPUs if no map is given in * its config file */ int hostcpus; - - /* setaffinity fails if you set bits for CPUs which - * aren't present, so we have to limit ourselves */ - if ((hostcpus = virHostCPUGetCount()) < 0) + hostcpumap = virHostCPUGetOnlineBitmap(); + cpumap = virProcessGetAffinity(vm->pid); + + if (hostcpumap && cpumap && virBitmapEqual(hostcpumap, cpumap)) { + /* we're using all available CPUs, no reason to set + * mask. If libvirtd is running without explicit + * affinity, we can use hotplugged CPUs for this VM */ + ret = 0; goto cleanup; + } else { + /* setaffinity fails if you set bits for CPUs which + * aren't present, so we have to limit ourselves */ + if ((hostcpus = virHostCPUGetCount()) < 0) + goto cleanup; - if (hostcpus > QEMUD_CPUMASK_LEN) - hostcpus = QEMUD_CPUMASK_LEN; + if (hostcpus > QEMUD_CPUMASK_LEN) + hostcpus = QEMUD_CPUMASK_LEN; - if (!(cpumap = virBitmapNew(hostcpus))) - goto cleanup; + virBitmapFree(cpumap); + if (!(cpumap = virBitmapNew(hostcpus))) + goto cleanup; - virBitmapSetAll(cpumap); + virBitmapSetAll(cpumap); - cpumapToSet = cpumap; + cpumapToSet = cpumap; + } } } @@ -2248,6 +2260,7 @@ qemuProcessInitCpuAffinity(virDomainObjPtr vm) cleanup: virBitmapFree(cpumap); + virBitmapFree(hostcpumap); return ret; } -- 1.9.1

1 0

[libvirt] [PATCH 0/2] Fix probing for no-overwrite on fs storage
by John Ferlan 24 Nov '16

24 Nov '16

Details in the patches, especially #2. I'm not 100% certain the blkid_probe_filter_superblocks_type is really necessary, but left it there with a fixed list of probe'd types rather than removing and finding some corner issue later. John Ferlan (2): fs: Remove virStoragePoolProbeResult and FILESYSTEM_PROBE_ values fs: Fix probing on no-overwrite of target device src/storage/storage_backend_fs.c | 42 +++++++++++++++++++--------------------- src/storage/storage_backend_fs.h | 5 ----- 2 files changed, 20 insertions(+), 27 deletions(-) -- 2.7.4

2 7

[libvirt] [PATCH v4 0/2] List only online cpus for vcpupin/emulatorpin when vcpu placement static
by Nitesh Konkar 24 Nov '16

24 Nov '16

Currently when the vcpu placement is static and cpuset is not specified, CPU Affinity shows 0.. CPUMAX. This patchset will result in display of only online CPU's under CPU Affinity on linux. Fixes the following Bug: virsh dumpxml Fedora <domain type='kvm' id='4'> <name>Fedora</name> <uuid>aecf3e5e-6f9a-42a3-9d6a-223a75569a66</uuid> <maxMemory slots='32' unit='KiB'>3145728</maxMemory> <memory unit='KiB'>524288</memory> <currentMemory unit='KiB'>524288</currentMemory> <vcpu placement='static' current='8'>160</vcpu> <resource> <partition>/machine</partition> </resource> ..................... ....................... ......................... <memballoon model='virtio'> <alias name='balloon0'/> <address type='pci' domain='0x0000' bus='0x00' slot='0x06' function='0x0'/> </memballoon> </devices> <seclabel type='dynamic' model='dac' relabel='yes'> <label>+0:+0</label> <imagelabel>+0:+0</imagelabel> </seclabel> </domain> lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 8 On-line CPU(s) list: 0-2,4-7 Off-line CPU(s) list: 3 Thread(s) per core: 1 Core(s) per socket: 7 Socket(s): 1 .......... .......... NUMA node0 CPU(s): 0-2,4-7 NUMA node1 CPU(s): cat /sys/devices/system/cpu/online 0-2,4-7 Before Patch virsh vcpupin Fedora VCPU: CPU Affinity ---------------------------------- 0: 0-7 1: 0-7 ... ... 158: 0-7 159: 0-7 virsh emulatorpin Fedora emulator: CPU Affinity ---------------------------------- *: 0-7 After Patch virsh vcpupin Fedora VCPU: CPU Affinity ---------------------------------- 0: 0-2,4-7 1: 0-2,4-7 ... ... 158: 0-2,4-7 159: 0-2,4-7 virsh emulatorpin Fedora emulator: CPU Affinity ---------------------------------- *: 0-2,4-7 Nitesh Konkar (2): conf: List only online cpus for virsh vcpupin conf: List only online cpus for virsh emulatorpin src/conf/domain_conf.c | 6 ++++++ src/qemu/qemu_driver.c | 5 +++++ 2 files changed, 11 insertions(+) -- 2.1.0

2 3

[libvirt] [PATCH] util: fix domain object leaks on closecallbacks
by Wang King 24 Nov '16

24 Nov '16

From: wangjing <king.wang(a)huawei.com> The virCloseCallbacksSet method increase object reference for VM, and decrease object reference in virCloseCallbacksUnset. But VM UUID will be deleted from closecallbacks list in virCloseCallbacksRun when connection disconnected, and then object reference cannot be decreased by virCloseCallbacksUnset in callback functions. Signed-off-by: Wang King <king.wang(a)huawei.com> --- src/util/virclosecallbacks.c | 29 +++++++++++++++++------------ 1 file changed, 17 insertions(+), 12 deletions(-) diff --git a/src/util/virclosecallbacks.c b/src/util/virclosecallbacks.c index 891a92b..26d5075 100644 --- a/src/util/virclosecallbacks.c +++ b/src/util/virclosecallbacks.c @@ -300,7 +300,9 @@ virCloseCallbacksGetForConn(virCloseCallbacksPtr closeCallbacks, data.list = list; data.oom = false; + virObjectLock(closeCallbacks); virHashForEach(closeCallbacks->list, virCloseCallbacksGetOne, &data); + virObjectUnlock(closeCallbacks); if (data.oom) { VIR_FREE(list->entries); @@ -329,22 +331,15 @@ virCloseCallbacksRun(virCloseCallbacksPtr closeCallbacks, * them all from the hash. At that point we can release * the lock and run the callbacks safely. */ - virObjectLock(closeCallbacks); list = virCloseCallbacksGetForConn(closeCallbacks, conn); if (!list) return; for (i = 0; i < list->nentries; i++) { - char uuidstr[VIR_UUID_STRING_BUFLEN]; - virUUIDFormat(list->entries[i].uuid, uuidstr); - virHashRemoveEntry(closeCallbacks->list, uuidstr); - } - virObjectUnlock(closeCallbacks); - - for (i = 0; i < list->nentries; i++) { virDomainObjPtr vm; + virDomainObjPtr dom; - if (!(vm = virDomainObjListFindByUUID(domains, + if (!(vm = virDomainObjListFindByUUIDRef(domains, list->entries[i].uuid))) { char uuidstr[VIR_UUID_STRING_BUFLEN]; virUUIDFormat(list->entries[i].uuid, uuidstr); @@ -352,10 +347,20 @@ virCloseCallbacksRun(virCloseCallbacksPtr closeCallbacks, continue; } - vm = list->entries[i].callback(vm, conn, opaque); - if (vm) - virObjectUnlock(vm); + dom = list->entries[i].callback(vm, conn, opaque); + if (dom) + virObjectUnlock(dom); + virObjectUnref(vm); } + + virObjectLock(closeCallbacks); + for (i = 0; i < list->nentries; i++) { + char uuidstr[VIR_UUID_STRING_BUFLEN]; + virUUIDFormat(list->entries[i].uuid, uuidstr); + virHashRemoveEntry(closeCallbacks->list, uuidstr); + } + virObjectUnlock(closeCallbacks); + VIR_FREE(list->entries); VIR_FREE(list); } -- 2.8.3

1 0

[libvirt] [PATCH v3 0/2] vbox: address thread-safety issues
by Dawid Zamirski 23 Nov '16

23 Nov '16

This patch series solves vbox driver thread-safety issues where libvird process would segfault when there was more than one concurrent vbox connection. The reason for the segfault was because the original implementation called pfnComInitialize/Unintialize on each virConnectOpen/Close calls which were setting pointers to ISession and IVirtualBox instances that were held by the global g_pVBoxGlobalData. This caused new connection to overwrite the pointer created by the prior connection. Changes since v2 to address feedback in [1]: * call uninitialize when vboxSdkInitialize or vboxExtractVersion fails. * make sure to set the global vbox_driver pointer to NULL after virObjectUnref releases last reference, to avoid segfaults when VBOX_RELEASE is called in unintialize call. * tested all possible failure paths to make sure libvirtd does not crash, that is, vboxSdkInitialize has 4 code paths that can possibly return -1 and made sure all of them are safely handled. * fixed "unintilalize" typos in code comments and commit messages. In other words, the change boils down to: if (vboxSdkInitialize() < 0 || vboxExtractVersion() < 0) { gVBoxAPI.UPFN.Uninitialize(vbox_driver); // call uninit on failure /* make sure to clear the pointer when last reference was released */ if (!virObjectUnref(vbox_driver)) vbox_driver = NULL; // make sure this is NULL virMutexUnlock(&vbox_driver_lock); return NULL; } [1] https://www.redhat.com/archives/libvir-list/2016-November/msg01132.html Dawid Zamirski (2): vbox: change how vbox API is initialized. vbox: get rid of g_pVBoxGlobalData src/vbox/vbox_common.c | 565 +++++++++++++++++++++++------------------- src/vbox/vbox_common.h | 2 +- src/vbox/vbox_driver.c | 1 + src/vbox/vbox_network.c | 24 +- src/vbox/vbox_storage.c | 20 +- src/vbox/vbox_tmpl.c | 434 ++++++++++++++++---------------- src/vbox/vbox_uniformed_api.h | 110 ++++---- 7 files changed, 616 insertions(+), 540 deletions(-) -- 2.9.3

2 3

[libvirt] [PATCH 0/3] Couple of small fixes and improvements
by Marc Hartmayer 23 Nov '16

23 Nov '16

Fixed trailing '/' for path, removed an obsolete comment, and removed a useless warning. Bjoern Walk (1): virutil: fix trailing '/' for path prefixes Marc Hartmayer (2): virfile: Only generate a warning if there is something to report qemu: Removed an outdated comment in qemuDomainSaveImageStartVM() src/qemu/qemu_driver.c | 1 - src/util/virfile.c | 2 +- src/util/virutil.c | 4 ++-- 3 files changed, 3 insertions(+), 4 deletions(-) -- 2.5.5

2 4

[libvirt] [PATCH 0/2] Fix typos in util and virsh manpage
by Nitesh Konkar 23 Nov '16

23 Nov '16

2 3

[libvirt] [PATCH v2 0/2] vbox: address thread-safety issues
by Dawid Zamirski 23 Nov '16

23 Nov '16

This patch series solves vbox driver thread-safety issues where libvird process would segfault when there was more than one concurrent vbox connection. The reason for the segfault was because the original implementation called pfnComInitialize/Unintialize on each virConnectOpen/Close calls which were setting pointers to ISession and IVirtualBox instances that were held by the global g_pVBoxGlobalData. This caused new connection to overwrite the pointer created by the prior connection. Changes since v1 [1]: * use just vboxDriver object instead of vboxGlobalData and vboxPrivate structs. The vboxDriver takes advantage of libvirt object reference counting and more closely resembles the pattern of "global" drivers used by other drivers such as vz or xen. This also takes care of the original concern mentioned in v1 where allocation of the struct was not guarded by a mutex. * use pfnClientInialize/Unintialize vbox API for newer vbox versions instead of the old pfnComInitialize/Unintialze which are marked as deprecated by upstream vbox. While this series adresses the issues I've been targetting, i.e. avoid libvird sefault with concurrent vbox connections, the libvirt vbox driver still incorrectly calls vbox's init/unint APIs as those must be called from the same thread according to vbox source code and docs. Furthermore, if one is running vbox 'debug' build, it will refuse to intialize as it's not called from the main thread due to asserts [2]. Those asserts are compiled away on 'release' builds so it appears to work, but may cause some unwanted side effects. Unfortunately, I don't think it's possible to have libvirt driver call those APIs from the same thread as libvirt is a client-server architecture and each libvir API call from the connection starts a new thread, so I guess this is not doeable right now :-/ [1] https://www.redhat.com/archives/libvir-list/2016-September/msg01306.html [2] https://github.com/mdaniel/virtualbox-org-svn-vbox-trunk/blob/master/src/VB… Dawid Zamirski (2): vbox: change how vbox API is initialized. vbox: get rid of g_pVBoxGlobalData src/vbox/vbox_common.c | 561 +++++++++++++++++++++++------------------- src/vbox/vbox_common.h | 2 +- src/vbox/vbox_driver.c | 1 + src/vbox/vbox_network.c | 24 +- src/vbox/vbox_storage.c | 20 +- src/vbox/vbox_tmpl.c | 434 ++++++++++++++++---------------- src/vbox/vbox_uniformed_api.h | 110 ++++----- 7 files changed, 612 insertions(+), 540 deletions(-) -- 2.9.3

3 11

[libvirt] [PATCH 0/6] Add deadline scheduler
by Martin Polednik 23 Nov '16

23 Nov '16

The policy SCHED_DEADLINE is available since kernel 3.14 (and most likely backported to older RT_PREEMPT kernels). It is safer to use than fifo or round robin policies due to only limiting part of cpu time for the RT process, leading to lack of lockups of the host. The series adds new vcpusched/iothreadsched called 'deadline' that activates SCHED_DEADLINE for given process. As the scheduler is linux specific, extra implementation was required - it is not possible to use sched_setscheduler, sched_setattr syscall must be used. Martin Polednik (6): conf: add entries for deadline scheduler util: allow virProcessSetScheduler to set SCHED_DEADLINE virDomainFormatSchedDef: factor out subset code conf: add deadline scheduler schema: add deadline scheduler docs: mention deadline scheduler in vcpusched docs/formatdomain.html.in | 15 ++-- docs/schemas/domaincommon.rng | 16 +++++ src/conf/domain_conf.c | 154 +++++++++++++++++++++++++++++++++++------- src/conf/domain_conf.h | 3 + src/qemu/qemu_process.c | 3 +- src/util/virprocess.c | 82 +++++++++++++++++++++- src/util/virprocess.h | 29 +++++++- 7 files changed, 268 insertions(+), 34 deletions(-) -- 2.8.1

3 22

[libvirt] [PATCH 0/2] qemu: Update CGroups on some devices hotplug
by Michal Privoznik 23 Nov '16

23 Nov '16

We are already doing that for some devices (e.g. disks, hostdevs), but completely forgot about some others (chardevs and RNGs). So far the hotplug worked by pure chance as /dev/random and /dev/urandom are enabled in the device CGroups by default. Michal Privoznik (2): qemu: Update cgroup on RNG hotplug qemu: Update cgroup on chardev hotplug src/qemu/qemu_cgroup.c | 108 +++++++++++++++++++++++++++++++++++++++++------- src/qemu/qemu_cgroup.h | 8 ++++ src/qemu/qemu_hotplug.c | 37 ++++++++++++++--- 3 files changed, 131 insertions(+), 22 deletions(-) -- 2.8.4

2 6