- Devel - libvirt List Archives

[libvirt] [PATCH] Make logical pools independent on target path
by Martin Kletzander 27 Jun '13

27 Jun '13

When using logical pools, we had to trust the target->pth provided. This parameter, however, can be completely ommited and we can get the correct path using 'lvdisplay' command. Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=952973 Signed-off-by: Martin Kletzander <mkletzan(a)redhat.com> --- Notes: I'm not sure we can drop the target.path from the XML (see tests), so another variant is to keep it there the same way it was defined by user/mgmt app. If that's desired, mentioning it with an ack is enough for me to know I should drop the conf/storage_conf.c hunk as well as the hunks patching tests. configure.ac | 4 ++ src/conf/storage_conf.c | 19 +++--- src/storage/storage_backend_logical.c | 79 +++++++++++++++------- tests/storagepoolxml2xmlin/pool-logical-create.xml | 1 - tests/storagepoolxml2xmlin/pool-logical.xml | 1 - .../storagepoolxml2xmlout/pool-logical-create.xml | 1 - tests/storagepoolxml2xmlout/pool-logical.xml | 1 - 7 files changed, 67 insertions(+), 39 deletions(-) diff --git a/configure.ac b/configure.ac index 5d1bc6b..753139d 100644 --- a/configure.ac +++ b/configure.ac @@ -1603,6 +1603,7 @@ if test "$with_storage_lvm" = "yes" || test "$with_storage_lvm" = "check"; then AC_PATH_PROG([PVS], [pvs], [], [$PATH:/sbin:/usr/sbin]) AC_PATH_PROG([VGS], [vgs], [], [$PATH:/sbin:/usr/sbin]) AC_PATH_PROG([LVS], [lvs], [], [$PATH:/sbin:/usr/sbin]) + AC_PATH_PROG([LVDISPLAY], [lvdisplay], [], [$PATH:/sbin:/usr/sbin]) if test "$with_storage_lvm" = "yes" ; then if test -z "$PVCREATE" ; then AC_MSG_ERROR([We need pvcreate for LVM storage driver]) ; fi @@ -1617,6 +1618,7 @@ if test "$with_storage_lvm" = "yes" || test "$with_storage_lvm" = "check"; then if test -z "$PVS" ; then AC_MSG_ERROR([We need pvs for LVM storage driver]) ; fi if test -z "$VGS" ; then AC_MSG_ERROR([We need vgs for LVM storage driver]) ; fi if test -z "$LVS" ; then AC_MSG_ERROR([We need lvs for LVM storage driver]) ; fi + if test -z "$LVDISPLAY" ; then AC_MSG_ERROR([We need lvdisplay for LVM storage driver]) ; fi else if test -z "$PVCREATE" ; then with_storage_lvm=no ; fi if test -z "$VGCREATE" ; then with_storage_lvm=no ; fi @@ -1630,6 +1632,7 @@ if test "$with_storage_lvm" = "yes" || test "$with_storage_lvm" = "check"; then if test -z "$PVS" ; then with_storage_lvm=no ; fi if test -z "$VGS" ; then with_storage_lvm=no ; fi if test -z "$LVS" ; then with_storage_lvm=no ; fi + if test -z "$LVDISPLAY" ; then with_storage_lvm=no ; fi if test "$with_storage_lvm" = "check" ; then with_storage_lvm=yes ; fi fi @@ -1648,6 +1651,7 @@ if test "$with_storage_lvm" = "yes" || test "$with_storage_lvm" = "check"; then AC_DEFINE_UNQUOTED([PVS],["$PVS"],[Location of pvs program]) AC_DEFINE_UNQUOTED([VGS],["$VGS"],[Location of vgs program]) AC_DEFINE_UNQUOTED([LVS],["$LVS"],[Location of lvs program]) + AC_DEFINE_UNQUOTED([LVDISPLAY],["$LVDISPLAY"],[Location of lvdisplay program]) fi fi AM_CONDITIONAL([WITH_STORAGE_LVM], [test "$with_storage_lvm" = "yes"]) diff --git a/src/conf/storage_conf.c b/src/conf/storage_conf.c index cc3d3d9..948e190 100644 --- a/src/conf/storage_conf.c +++ b/src/conf/storage_conf.c @@ -1,7 +1,7 @@ /* * storage_conf.c: config handling for storage driver * - * Copyright (C) 2006-2012 Red Hat, Inc. + * Copyright (C) 2006-2013 Red Hat, Inc. * Copyright (C) 2006-2008 Daniel P. Berrange * * This library is free software; you can redistribute it and/or @@ -947,15 +947,16 @@ virStoragePoolDefParseXML(xmlXPathContextPtr ctxt) /* When we are working with a virtual disk we can skip the target * path and permissions */ if (!(options->flags & VIR_STORAGE_POOL_SOURCE_NETWORK)) { - if (!(target_path = virXPathString("string(./target/path)", ctxt))) { - virReportError(VIR_ERR_XML_ERROR, "%s", - _("missing storage pool target path")); - goto error; + if (ret->type != VIR_STORAGE_POOL_LOGICAL) { + if (!(target_path = virXPathString("string(./target/path)", ctxt))) { + virReportError(VIR_ERR_XML_ERROR, "%s", + _("missing storage pool target path")); + goto error; + } + ret->target.path = virFileSanitizePath(target_path); + if (!ret->target.path) + goto error; } - ret->target.path = virFileSanitizePath(target_path); - if (!ret->target.path) - goto error; - if (virStorageDefParsePerms(ctxt, &ret->target.perms, "./target/permissions", DEFAULT_POOL_PERM_MODE) < 0) diff --git a/src/storage/storage_backend_logical.c b/src/storage/storage_backend_logical.c index 944aa0e..66a4e42 100644 --- a/src/storage/storage_backend_logical.c +++ b/src/storage/storage_backend_logical.c @@ -45,6 +45,37 @@ #define PV_BLANK_SECTOR_SIZE 512 +static char * +virStorageBackendGetVolPath(const char *poolname, const char *volname) +{ + char *start = NULL; + char *lvpath = NULL; + char *output = NULL; + + virCommandPtr cmd = virCommandNewArgList(LVDISPLAY, + "--columns", + "--options", "lv_path", + "--noheadings", + "--unbuffered", + NULL); + + virCommandAddArgFormat(cmd, "%s/%s", poolname, volname); + virCommandSetOutputBuffer(cmd, &output); + + if (virCommandRun(cmd, NULL) < 0 || + !(start = strchr(output, '/'))) { + virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("Cannot get LV path")); + goto cleanup; + } + + ignore_value(VIR_STRDUP(lvpath, start)); + + cleanup: + VIR_FREE(output); + virCommandFree(cmd); + return lvpath; +} + static int virStorageBackendLogicalSetActive(virStoragePoolObjPtr pool, int on) @@ -116,11 +147,8 @@ virStorageBackendLogicalMakeVol(virStoragePoolObjPtr pool, } if (vol->target.path == NULL) { - if (virAsprintf(&vol->target.path, "%s/%s", - pool->def->target.path, vol->name) < 0) { - virReportOOMError(); + if (VIR_STRDUP(vol->target.path, groups[9]) < 0) goto cleanup; - } } /* Skips the backingStore of lv created with "--virtualsize", @@ -130,12 +158,11 @@ virStorageBackendLogicalMakeVol(virStoragePoolObjPtr pool, * (lvs outputs "[$lvname_vorigin] for field "origin" if the * lv is created with "--virtualsize"). */ - if (groups[1] && !STREQ(groups[1], "") && (groups[1][0] != '[')) { - if (virAsprintf(&vol->backingStore.path, "%s/%s", - pool->def->target.path, groups[1]) < 0) { - virReportOOMError(); + if (groups[1] && STRNEQ(groups[1], "") && (groups[1][0] != '[')) { + vol->backingStore.path = virStorageBackendGetVolPath(pool->def->source.name, + groups[1]); + if (!vol->backingStore.path) goto cleanup; - } vol->backingStore.format = VIR_STORAGE_POOL_LOGICAL_LVM2; } @@ -277,21 +304,20 @@ virStorageBackendLogicalFindLVs(virStoragePoolObjPtr pool, virStorageVolDefPtr vol) { /* - * # lvs --separator , --noheadings --units b --unbuffered --nosuffix --options \ - * "lv_name,origin,uuid,devices,seg_size,vg_extent_size,size,lv_attr" VGNAME + * # lvs --separator '#' --noheadings --units b --unbuffered --nosuffix --options "lv_name,origin,uuid,devices,seg_size,vg_extent_size,size,lv_path" VGNAME * - * RootLV,,06UgP5-2rhb-w3Bo-3mdR-WeoL-pytO-SAa2ky,/dev/hda2(0),5234491392,33554432,5234491392,-wi-ao - * SwapLV,,oHviCK-8Ik0-paqS-V20c-nkhY-Bm1e-zgzU0M,/dev/hda2(156),1040187392,33554432,1040187392,-wi-ao - * Test2,,3pg3he-mQsA-5Sui-h0i6-HNmc-Cz7W-QSndcR,/dev/hda2(219),1073741824,33554432,1073741824,owi-a- - * Test3,,UB5hFw-kmlm-LSoX-EI1t-ioVd-h7GL-M0W8Ht,/dev/hda2(251),2181038080,33554432,2181038080,-wi-a- - * Test3,Test2,UB5hFw-kmlm-LSoX-EI1t-ioVd-h7GL-M0W8Ht,/dev/hda2(187),1040187392,33554432,1040187392,swi-a- + * RootLV##06UgP5-2rhb-w3Bo-3mdR-WeoL-pytO-SAa2ky#/dev/hda2(0)#5234491392#33554432#5234491392#/dev/VGNAME/RootLV + * SwapLV##oHviCK-8Ik0-paqS-V20c-nkhY-Bm1e-zgzU0M#/dev/hda2(156)#1040187392#33554432#1040187392#/dev/VGNAME/SwapLV + * Test2##3pg3he-mQsA-5Sui-h0i6-HNmc-Cz7W-QSndcR#/dev/hda2(219)#1073741824#33554432#1073741824#/dev/VGNAME/Test2 + * Test3##UB5hFw-kmlm-LSoX-EI1t-ioVd-h7GL-M0W8Ht#/dev/hda2(251)#2181038080#33554432#2181038080#/dev/VGNAME/Test3 + * Test3#Test2#UB5hFw-kmlm-LSoX-EI1t-ioVd-h7GL-M0W8Ht#/dev/hda2(187)#1040187392#33554432#1040187392#/dev/VGNAME/Test3 * * Pull out name, origin, & uuid, device, device extent start #, * segment size, extent size, size, attrs * * NB can be multiple rows per volume if they have many extents * - * NB lvs from some distros (e.g. SLES10 SP2) outputs trailing "," on each line + * NB lvs from some distros (e.g. SLES10 SP2) outputs trailing separator on each line * * NB Encrypted logical volumes can print ':' in their name, so it is * not a suitable separator (rhbz 470693). @@ -314,7 +340,7 @@ virStorageBackendLogicalFindLVs(virStoragePoolObjPtr pool, "--unbuffered", "--nosuffix", "--options", - "lv_name,origin,uuid,devices,segtype,stripes,seg_size,vg_extent_size,size,lv_attr", + "lv_name,origin,uuid,devices,segtype,stripes,seg_size,vg_extent_size,size,lv_path", pool->def->source.name, NULL); if (virStorageBackendRunProgRegex(pool, @@ -702,6 +728,7 @@ virStorageBackendLogicalCreateVol(virConnectPtr conn, int fd = -1; virCommandPtr cmd = NULL; virErrorPtr err; + bool created = false; if (vol->target.encryption != NULL) { virReportError(VIR_ERR_CONFIG_UNSUPPORTED, @@ -717,13 +744,6 @@ virStorageBackendLogicalCreateVol(virConnectPtr conn, VIR_FREE(vol->target.path); } - if (virAsprintf(&vol->target.path, "%s/%s", - pool->def->target.path, - vol->name) == -1) { - virReportOOMError(); - return -1; - } - cmd = virCommandNewArgList(LVCREATE, "--name", vol->name, NULL); @@ -742,9 +762,15 @@ virStorageBackendLogicalCreateVol(virConnectPtr conn, if (virCommandRun(cmd, NULL) < 0) goto error; + created = true; virCommandFree(cmd); cmd = NULL; + vol->target.path = virStorageBackendGetVolPath(pool->def->source.name, + vol->name); + if (!vol->target.path) + goto error; + if ((fd = virStorageBackendVolOpen(vol->target.path)) < 0) goto error; @@ -784,7 +810,8 @@ virStorageBackendLogicalCreateVol(virConnectPtr conn, error: err = virSaveLastError(); VIR_FORCE_CLOSE(fd); - virStorageBackendLogicalDeleteVol(conn, pool, vol, 0); + if (created) + virStorageBackendLogicalDeleteVol(conn, pool, vol, 0); virCommandFree(cmd); virSetError(err); virFreeError(err); diff --git a/tests/storagepoolxml2xmlin/pool-logical-create.xml b/tests/storagepoolxml2xmlin/pool-logical-create.xml index 4c67089..e1bb4db 100644 --- a/tests/storagepoolxml2xmlin/pool-logical-create.xml +++ b/tests/storagepoolxml2xmlin/pool-logical-create.xml @@ -10,7 +10,6 @@ <device path="/dev/sdb3"/> </source> <target> - <path>/dev/HostVG</path> <permissions> <mode>0700</mode> <owner>0</owner> diff --git a/tests/storagepoolxml2xmlin/pool-logical.xml b/tests/storagepoolxml2xmlin/pool-logical.xml index c4bfa07..307e2ab 100644 --- a/tests/storagepoolxml2xmlin/pool-logical.xml +++ b/tests/storagepoolxml2xmlin/pool-logical.xml @@ -9,7 +9,6 @@ <format type='lvm2'/> </source> <target> - <path>/dev/HostVG</path> <permissions> <mode>0700</mode> <owner>0</owner> diff --git a/tests/storagepoolxml2xmlout/pool-logical-create.xml b/tests/storagepoolxml2xmlout/pool-logical-create.xml index 7413f1c..fbce90e 100644 --- a/tests/storagepoolxml2xmlout/pool-logical-create.xml +++ b/tests/storagepoolxml2xmlout/pool-logical-create.xml @@ -12,7 +12,6 @@ <format type='lvm2'/> </source> <target> - <path>/dev/HostVG</path> <permissions> <mode>0700</mode> <owner>0</owner> diff --git a/tests/storagepoolxml2xmlout/pool-logical.xml b/tests/storagepoolxml2xmlout/pool-logical.xml index 07860ef..7fad0d8 100644 --- a/tests/storagepoolxml2xmlout/pool-logical.xml +++ b/tests/storagepoolxml2xmlout/pool-logical.xml @@ -9,7 +9,6 @@ <format type='lvm2'/> </source> <target> - <path>/dev/HostVG</path> <permissions> <mode>0700</mode> <owner>0</owner> -- 1.8.2.1

2 2

[libvirt] [PATCH] add symbol ")" in virsh nodedev-detach help
by xuzhang 27 Jun '13

27 Jun '13

--- tools/virsh-nodedev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/virsh-nodedev.c b/tools/virsh-nodedev.c index da93b8e..49546a2 100644 --- a/tools/virsh-nodedev.c +++ b/tools/virsh-nodedev.c @@ -605,7 +605,7 @@ static const vshCmdOptDef opts_node_device_detach[] = { }, {.name = "driver", .type = VSH_OT_STRING, - .help = N_("pci device assignment backend driver (e.g. 'vfio' or 'kvm'") + .help = N_("pci device assignment backend driver (e.g. 'vfio' or 'kvm'")) }, {.name = NULL} }; -- 1.7.11.7

1 0

[libvirt] [PATCH 0/2] Fix a pair of network-related crashes
by Ján Tomko 27 Jun '13

27 Jun '13

Ján Tomko (2): netdev: accept NULL in virNetDevSetupControl bridge: don't crash on bandwidth unplug with no bandwidth src/network/bridge_driver.c | 5 +++++ src/util/virnetdev.c | 14 ++++++++------ 2 files changed, 13 insertions(+), 6 deletions(-) -- 1.8.1.5

3 11

[libvirt] [PATCH 0/5]New atomic API to delete snapshot object
by Guannan Ren 27 Jun '13

27 Jun '13

Add a new snapshot API to delete snapshot object atomically int virDomainSnapshotDeleteByName(virDomainPtr domain, const char *name, unsigned int flags); The existing virDomainSnapshotDelete API accepts the snapshot object being deleted as an argument that would be not API atomic. Guannan Ren(5) [PATCH 1/5] snapshot: define new API virDomainSnapshotDeleteByName [PATCH 2/5] snapshot: auto generate RPC calls for remoteDomainSnapshotDeleteByName [PATCH 3/5] qemu: implement SnapshotDeleteByName [PATCH 4/5] python: make auto-generated function name nicer [PATCH 5/5] virsh: use virDomainSnapshotDeleteByName in virsh include/libvirt/libvirt.h.in | 5 +++++ python/generator.py | 3 +++ src/driver.h | 6 ++++++ src/libvirt.c | 73 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ src/libvirt_public.syms | 5 +++++ src/qemu/qemu_driver.c | 102 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++------------------ src/remote/remote_driver.c | 1 + src/remote/remote_protocol.x | 13 +++++++++++- src/remote_protocol-structs | 7 +++++++ tools/virsh-snapshot.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++++---------- 10 files changed, 249 insertions(+), 30 deletions(-)

2 7

[libvirt] virsh can create vHBA, but returen error msg "Node device not found"
by Dennis Chen 27 Jun '13

27 Jun '13

Hi, I am using the libvirt 1.0.6 with the patch applied: http://libvirt.org/git/?p=libvirt.git;a=commitdiff;h=371c155 The problem I encountered is, with virsh comand 'nodedev-create /home/kvm/vHBA.xml', the vHBA can be generated as expected on the host, but error msg will show in the virsh shell: libvirt: Node Device Driver error : Node device not found I try to find the reason with gdb, and found that the 'nodedev-create' command will call virNetClientIOEventLoop() function of the remote driver, in this function, code will wait on: repoll: ret = poll(fds, ARRAY_CARDINALITY(fds), timeout); after about 50s, poll() returns, but the replied msg->header.status=VIR_NET_ERROR. I am not familiar with the rpc call in the remote driver, does anybody here can give some clues? BRs, Dennis

1 0

[libvirt] [PATCH] docs: Document hypervisor drivers that support certain timer models
by Peter Krempa 26 Jun '13

26 Jun '13

Not every timer model is supported with each hypervisor. Explicitly mention the driver supporting each timer model. --- docs/formatdomain.html.in | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in index 77126a5..d77bf3f 100644 --- a/docs/formatdomain.html.in +++ b/docs/formatdomain.html.in @@ -1310,8 +1310,10 @@ <dt><code>name</code></dt> <dd> The <code>name</code> attribute selects which timer is - being modified, and can be one of "platform", "hpet", - "kvmclock", "pit", "rtc", or "tsc". + being modified, and can be one of + "platform" (currently unsupported), + "hpet" (libxl, qemu), "kvmclock" (xen, libxl, qemu), + "pit" (qemu), "rtc" (qemu), or "tsc" (libxl). </dd> <dt><code>track</code></dt> <dd> -- 1.8.2.1

3 3

[libvirt] [PATCHv3] pci: new iommu_group functions
by Laine Stump 26 Jun '13

26 Jun '13

Any device which belongs to an "IOMMU group" (used by vfio) will have links to all devices of its group listed in /sys/bus/pci/$device/iommu_group/devices; /sys/bus/pci/$device/iommu_group is actually a link to /sys/kernel/iommu_groups/$n, where $n is the group number (there will be a corresponding device node at /dev/vfio/$n once the devices are bound to the vfio-pci driver) The following functions are added: virPCIDeviceGetIOMMUGroupList Gets a virPCIDeviceList with one virPCIDeviceList for each device in the same IOMMU group as the provided virPCIDevice (a copy of the original device object is included in the list. virPCIDeviceAddressIOMMUGroupIterate Calls the function @actor once for each device in the group that contains the given virPCIDeviceAddress. virPCIDeviceAddressGetIOMMUGroupAddresses Fills in a virPCIDeviceAddressPtr * with an array of virPCIDeviceAddress, one for each device in the iommu group of the provided virPCIDeviceAddress (including a copy of the original). virPCIDeviceAddressGetIOMMUGroupNum Returns the group number as an int (a valid group number will always be 0 or greater). If there is no iommu_group link in the device's directory (usually indicating that vfio isn't loaded), -2 will be returned. On any real error, -1 will be returned. --- changes from v2: 1) rename functions for consistency 2) set errno=0 prior to each call to readdir, and check for non-zero errno at the end. 3) remove some braces. Everything else in the series of 12 was ACKed, and all but this and the patch that adds the <iommuGroup> element to the nodedev-dumpxml output have been pushed. src/libvirt_private.syms | 4 + src/util/virpci.c | 223 +++++++++++++++++++++++++++++++++++++++++++++++ src/util/virpci.h | 15 +++- 3 files changed, 239 insertions(+), 3 deletions(-) diff --git a/src/libvirt_private.syms b/src/libvirt_private.syms index 795e011..f08ac64 100644 --- a/src/libvirt_private.syms +++ b/src/libvirt_private.syms @@ -1693,13 +1693,17 @@ virObjectUnref; # util/virpci.h +virPCIDeviceAddressGetIOMMUGroupAddresses; +virPCIDeviceAddressGetIOMMUGroupNum; virPCIDeviceAddressGetSysfsFile; +virPCIDeviceAddressIOMMUGroupIterate; virPCIDeviceAddressParse; virPCIDeviceCopy; virPCIDeviceDetach; virPCIDeviceFileIterate; virPCIDeviceFree; virPCIDeviceGetIOMMUGroupDev; +virPCIDeviceGetIOMMUGroupList; virPCIDeviceGetManaged; virPCIDeviceGetName; virPCIDeviceGetRemoveSlot; diff --git a/src/util/virpci.c b/src/util/virpci.c index 32a5493..520dbb8 100644 --- a/src/util/virpci.c +++ b/src/util/virpci.c @@ -1876,6 +1876,229 @@ cleanup: return ret; } + +/* virPCIDeviceAddressIOMMUGroupIterate: + * Call @actor for all devices in the same iommu_group as orig + * (including orig itself) Even if there is no iommu_group for the + * device, call @actor once for orig. + */ +int +virPCIDeviceAddressIOMMUGroupIterate(virPCIDeviceAddressPtr orig, + virPCIDeviceAddressActor actor, + void *opaque) +{ + char *groupPath = NULL; + DIR *groupDir = NULL; + int ret = -1; + struct dirent *ent; + + if (virAsprintf(&groupPath, + PCI_SYSFS "devices/%04x:%02x:%02x.%x/iommu_group/devices", + orig->domain, orig->bus, orig->slot, orig->function) < 0) { + virReportOOMError(); + goto cleanup; + } + + if (!(groupDir = opendir(groupPath))) { + /* just process the original device, nothing more */ + ret = (actor)(orig, opaque); + goto cleanup; + } + + while ((errno = 0, ent = readdir(groupDir)) != NULL) { + virPCIDeviceAddress newDev; + + if (ent->d_name[0] == '.') + continue; + + if (virPCIDeviceAddressParse(ent->d_name, &newDev) < 0) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("Found invalid device link '%s' in '%s'"), + ent->d_name, groupPath); + goto cleanup; + } + + if ((actor)(&newDev, opaque) < 0) + goto cleanup; + } + if (errno != 0) { + virReportSystemError(errno, + _("Failed to read directory entry for %s"), + groupPath); + goto cleanup; + } + + ret = 0; + +cleanup: + VIR_FREE(groupPath); + if (groupDir) + closedir(groupDir); + return ret; +} + + +static int +virPCIDeviceGetIOMMUGroupAddOne(virPCIDeviceAddressPtr newDevAddr, void *opaque) +{ + int ret = -1; + virPCIDeviceListPtr groupList = opaque; + virPCIDevicePtr newDev; + + if (!(newDev = virPCIDeviceNew(newDevAddr->domain, newDevAddr->bus, + newDevAddr->slot, newDevAddr->function))) + goto cleanup; + + if (virPCIDeviceListAdd(groupList, newDev) < 0) + goto cleanup; + + newDev = NULL; /* it's now on the list */ + ret = 0; +cleanup: + virPCIDeviceFree(newDev); + return ret; +} + + +/* + * virPCIDeviceGetIOMMUGroupList - return a virPCIDeviceList containing + * all of the devices in the same iommu_group as @dev. + * + * Return the new list, or NULL on failure + */ +virPCIDeviceListPtr +virPCIDeviceGetIOMMUGroupList(virPCIDevicePtr dev) +{ + virPCIDeviceListPtr groupList = virPCIDeviceListNew(); + virPCIDeviceAddress devAddr = { dev->domain, dev->bus, + dev->slot, dev->function }; + + if (!groupList) + goto error; + + if (virPCIDeviceAddressIOMMUGroupIterate(&devAddr, + virPCIDeviceGetIOMMUGroupAddOne, + groupList) < 0) + goto error; + + return groupList; + +error: + virObjectUnref(groupList); + return NULL; +} + + +typedef struct { + virPCIDeviceAddressPtr **iommuGroupDevices; + size_t *nIommuGroupDevices; +} virPCIDeviceAddressList; +typedef virPCIDeviceAddressList *virPCIDeviceAddressListPtr; + +static int +virPCIGetIOMMUGroupAddressesAddOne(virPCIDeviceAddressPtr newDevAddr, void *opaque) +{ + int ret = -1; + virPCIDeviceAddressListPtr addrList = opaque; + virPCIDeviceAddressPtr copyAddr; + + /* make a copy to insert onto the list */ + if (VIR_ALLOC(copyAddr) < 0) + goto cleanup; + + *copyAddr = *newDevAddr; + + if (VIR_APPEND_ELEMENT(*addrList->iommuGroupDevices, + *addrList->nIommuGroupDevices, copyAddr) < 0) { + virReportOOMError(); + goto cleanup; + } + + ret = 0; +cleanup: + VIR_FREE(copyAddr); + return ret; +} + + +/* + * virPCIGetIOMMUGroupAddresses - return a virPCIDeviceList containing + * all of the devices in the same iommu_group as @dev. + * + * Return the new list, or NULL on failure + */ +int +virPCIDeviceAddressGetIOMMUGroupAddresses(virPCIDeviceAddressPtr devAddr, + virPCIDeviceAddressPtr **iommuGroupDevices, + size_t *nIommuGroupDevices) +{ + int ret = -1; + virPCIDeviceAddressList addrList = { iommuGroupDevices, + nIommuGroupDevices }; + + if (virPCIDeviceAddressIOMMUGroupIterate(devAddr, + virPCIGetIOMMUGroupAddressesAddOne, + &addrList) < 0) + goto cleanup; + + ret = 0; +cleanup: + return ret; +} + + +/* virPCIGetIOMMUGroupNum - return the group number of this PCI + * device's iommu_group, or -2 if there is no iommu_group for the + * device (or -1 if there was any other error) + */ +int +virPCIDeviceAddressGetIOMMUGroupNum(virPCIDeviceAddressPtr addr) +{ + char *devName = NULL; + char *devPath = NULL; + char *groupPath = NULL; + const char *groupNumStr; + unsigned int groupNum; + int ret = -1; + + if (virAsprintf(&devName, "%.4x:%.2x:%.2x.%.1x", addr->domain, + addr->bus, addr->slot, addr->function) < 0) { + virReportOOMError(); + goto cleanup; + } + + if (virPCIFile(&devPath, devName, "iommu_group") < 0) + goto cleanup; + if (virFileIsLink(devPath) != 1) { + ret = -2; + goto cleanup; + } + if (virFileResolveLink(devPath, &groupPath) < 0) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("Unable to resolve device %s iommu_group symlink %s"), + devName, devPath); + goto cleanup; + } + + groupNumStr = last_component(groupPath); + if (virStrToLong_ui(groupNumStr, NULL, 10, &groupNum) < 0) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("device %s iommu_group symlink %s has " + "invalid group number %s"), + devName, groupPath, groupNumStr); + ret = -1; + goto cleanup; + } + + ret = groupNum; +cleanup: + VIR_FREE(devName); + VIR_FREE(devPath); + VIR_FREE(groupPath); + return ret; +} + + /* virPCIDeviceGetIOMMUGroupDev - return the name of the device used * to control this PCI device's group (e.g. "/dev/vfio/15") */ diff --git a/src/util/virpci.h b/src/util/virpci.h index b56b89b..0aa6fee 100644 --- a/src/util/virpci.h +++ b/src/util/virpci.h @@ -113,12 +113,21 @@ int virPCIDeviceListFindIndex(virPCIDeviceListPtr list, */ typedef int (*virPCIDeviceFileActor)(virPCIDevicePtr dev, const char *path, void *opaque); - int virPCIDeviceFileIterate(virPCIDevicePtr dev, virPCIDeviceFileActor actor, void *opaque); -char * -virPCIDeviceGetIOMMUGroupDev(virPCIDevicePtr dev); + +typedef int (*virPCIDeviceAddressActor)(virPCIDeviceAddress *addr, + void *opaque); +int virPCIDeviceAddressIOMMUGroupIterate(virPCIDeviceAddressPtr orig, + virPCIDeviceAddressActor actor, + void *opaque); +virPCIDeviceListPtr virPCIDeviceGetIOMMUGroupList(virPCIDevicePtr dev); +int virPCIDeviceAddressGetIOMMUGroupAddresses(virPCIDeviceAddressPtr devAddr, + virPCIDeviceAddressPtr **iommuGroupDevices, + size_t *nIommuGroupDevices); +int virPCIDeviceAddressGetIOMMUGroupNum(virPCIDeviceAddressPtr dev); +char *virPCIDeviceGetIOMMUGroupDev(virPCIDevicePtr dev); int virPCIDeviceIsAssignable(virPCIDevicePtr dev, int strict_acs_check); -- 1.7.11.7

2 1

[libvirt] RFC: Improving unit test coverage
by Daniel P. Berrange 26 Jun '13

26 Jun '13

Over the many years that libvirt has been in existance, we have developed a huge number of features, and rate of change in the codebase shows no sign of slowing down. Unfortunately the rate of regressions and/or new bugs also shows no sign of decreasing. As libvirt is used ever more widely, poor quality releases are having an increasingly serious negative impact on the impressions of libvirt by downstream consumers. In addition the introduction of the new access control facilities will have the unfortunate side-effect of increasing the severity of many bugs. Historically a read-write connection to libvirtd could be considered to be equivalent to root, so a bug affecting an authenticated user would not be considered a privilege escalation, merely denial-of-service. With the introduction of access control, a read-write connection to libvirtd may have substantially fewer privileges than root. The result is that many (arguably all) crasher bugs in libvirtd which were previously considered mere denial of service issues, and not worthy of security alerts, will now be classed as potential privilege escalation flaws. It is clear that the current rate of defects in libvirt code is no longer tenable / acceptable. There are a number of test suites which are targetting libvirt, but this document will focus on the unit testing framework, rather than the integration test harnesses. The rationale for this is - Unit tests are run by all developers daily via 'make check' - Unit tests gate the release process - Unit tests are run every time any RPM is built - Unit tests do not have an dependancy on external system infrastructure / setup - Unit tests are directly synchronized with the code commits since they're in the same GIT repo Unit testing techniques ======================= One of the challenges of unit testing is how to isolate the test suite from the broader environment of the machine being used for development. For example, the test suite must not rely on information located in the filesystem outside the source treee (eg it must not touch /proc or /sys). In addition it must be careful not to negatively impact anything running on the developers' machine (eg it must not spawn libvirtd in a way which would clash with existing running libvirtd) There is no silver bullet to achieve isolation of the test suite from the host OS environment. Instead a variety of different techniques must be applied depending on the scenario to be addressed. Code structure -------------- First, and foremost, code structure has an enourmous impact on the ability to unit test code. Ensuring there are clear boundaries & interfaces between different areas of cdoe is key to allowing pieces to be unit tested in isolation from each other. For example, testing of the QEMU command line generator, can only be achieved because this code for converting from XML to ARGV has been isolated from the code which actually starts up the QEMU process. While some areas of libvirt code are very well modularized, others are not. When writing new unit tests it is not at all unexpected to have to re-factor existing code to further modularize it. Such refactoring is usually good for overall readability of the code, as well as facilitating testing. Intermediate data forms ----------------------- As hinted at above, a key idea when refactoring code to make it more amenable to testing is to introduce an intermediate data form. For example, when generating the QEMU command line arguments from an XML document, the virCommandPtr object serves as the intermediate data form. Instead of having the API which builds the command line, directly execute the QEMU binary, it spits out a virCommandPtr object as an intermediate data form. The contents of this object can then be validated against pre-defined expected command lines. As a counter-example, the network filter code is very hard to unit test as it is currently designed because it does not have any intermediate data form. The methods which process the network filter configuration build up command lines and then directly execute them as they go. There is no way to extract the command line data, without them being executed. To enable unit testing it is neccessary to refactor the network filter code to introduce a clear intermediate data form which can represent the commands that will be executed. Static analysis --------------- While the elephant in the room is Coverity, there is a large amount of static analysis that can be done directly in the libvirt source tree. Indeed all of the 'syntax-check' rules are considered to be static analysis. While these rules all focus on style issues, there is no reason why this should be the case. What makes static analysis tools like Coverity so hard to write is the problem of understanding the semantics of C code. Fortunately, when doing static analysis of a specific project (like libvirt), there is often no need to solve the general case. If there are certain style patterns used in a project, these can make it possible to write project specific analysis rules with little more than a few regular expressions or other simple scripting statements. For example, the two style checks 'src/check-drivername.pl' 'src/check-driverimpls.pl' enforce a consistent naming convention on each internal driver method. With this naming pattern enforced, it is now possible to write a script which validates that every driver method includes a call to the correct access control hook. This is project-specific static analysis that could not be done with Coverity. For more advanced problems it is possible to make use of a framework such as CIL to perform static analysis. CIL is an OCaml project which can parse most C code providing an in-memory representation of its structure. Then it has a number of modules to perform data flow analysis on the code. Combined with knowledge of project coding patterns this tool allows the creation of scripts to validate mutex lock acquisition ordering rules, or correct error reporting on all error exit paths. Monkey patching / Duck Punching ------------------------------- One of the pain-points with a project such as libvirt is finding a way to test parts of the code which rely on external system state, such as /proc, or /sys filesystem data. It is not always possible to refactor the code to test it in isolation from the system services. In such cases, the technique of monkey patching or duck punching (better known from the world of scripting languages) can in fact be used from C. The idea behind the technique is to replace the APIs / objects that are used by the test suite, with set of "mock" objects or API implementations which simulate their operation, without relying on external state. The trick to applying this technique in the C world with native code, is to rely on the use of the dynamic linkers' library preload facility, via the LD_PRELOAD. Any symbol exported by a library listed in the LD_PRELOAD env variable, will override symbols with the same name in the library that the test suite is actually linked to. So, for example, if the preloaded library exports an 'open' symbol, this will replace the 'open' symbol exported by the standard C library. To use this technique it is neccessary to first understand what interfaces need to be replaced. If attempting to test some code which uses files in some location under /sys/, one might change the implementation of 'open', so that when passed a filename starting with '/sys', it will re-write the path to point to a location under the test suite directory. This technique is used in the vircgrouptest.c / vircgroupmock.c files to allow comprehensive testing of libvirt's cgroup management code, without actually ever touching the cgroup filesystem. It is also used by the securityselinuxtest.c / securityselinuxhelper.c files to stub out the sys calls which write SELinux attributes. Fuzz testing ------------ This refers to the technique of taking an API or interface and giving it some data which has some intentionale inaccuracies or edge cases. For example, a API accepting a 'const char *str' might be given a NULL string, or a zero length string, or a string containing non-printable control characters. An API accepting an 'int' might be tested with various boundary conditions such as -1, 0, -MAX_INT, +MAX_INT. In the case of an XML document, one technique would be to incrementally remove attributes and/or elements to see if the parser will ever crash on a NULL pointer (it should either succeed or return a error, never crash). Or alternatively change the values in various attributes as one would with parameters passed to an API. If an attribute appears to be a number, then replace it with a string. If an attribute appears to be an enum, replace it with some bogus value. Try zero length strings for various attributes, or again try non-printable control characters. In the case of an RPC protocol, the idea would be to make (intelligent) changes to packets in an attempt to trick the server into accepting bad data, or even crashing. It is not practical to test all possible wierd data sets, so the trick with fuzz testing is to try to intelligently tailor the data to the interface being tested. eg if a string parameter represents a file path, there's no point just randomly changing characters in the path, since that'll just generate loads of other valid file paths. Instead it is neccessary to make some more planned changes like removing the leading '/' to produce a relative path, inserting a number of '../../../' dirs to trick the application into writing to the incorrect location, and things of that nature. For any moderately complex parser, intelligently designed fuzz testing is likely to identify a wide range of bugs. Error testing ------------- Validating correct operation is great, but there is plenty of code which lives in error handling code paths, which will never be exercised this way. Since they are seldom run, and by their nature deal with unexpected data or events, the error paths are often a good source of bugs. The fuzz testing technique is one very effective way of exercising error handling paths, but should not be considered the only possible one. The monkey patching techniques can also be applied to the task of error checking. Alternatively the code structure may facilitate the injection of error conditions in a well defined manner. For example, all libvirt memory allocation goes via a wrapper API, instead of to malloc() directly. It is possible to tell the wrapper API to fail specific allocation attempts. By running the test suite repeatedly, failing a different memory allocation each time, it is possible to exhaustively test error handling of the tested code. Areas requiring testing ======================= The best way to identify areas lacking code coverage is to simply pick a random libvirt source code file. The odds are that it has insufficient testing. The more optimistic way is to look at the code coverage reports. Run 'autogen.sh' with the --enable-test-coverage arg. Then build it and run 'make check'. Once the tests have completed, then run 'make cov' to generate an HTML code coverage report. Some known areas needing work are: Cgroups setup for QEMU ---------------------- For a given QEMU guest XML configuration there is a well defined set of cgroups parameters that are expected to be set. The vircgroupmock.c file demonstrates how to monkey patch the cgroups filesystem, allowing unit tests to be written. The vircgroupmock.c helper could be reused to create a test suite for validating the operation of the qemu_cgroups.c APIs. Cgroups setup for LXC ---------------------- The same as the QEMU test, but for the lxc_cgroups.c file. SELinux security driver labelling --------------------------------- For a given XML configuration, there is a well defined set of file labels that the SELinux security driver must apply to the filesystem on VM startup. The securityselinuxhelper.c file demonstrates how to monkey patch the SELinux library file labelling APIs from unit tests. Currently the securityselinuxlabeltest.c test suite does checks for labels set on disks and serial devices. There are many other aspects of domain XML that imply labelling. In particular PCI/USB host devices & kernel/initrd boot images. It is also neccessary to deal with backing files for qcow2 format disks. Audit messages -------------- Upon QEMU or LXC guest startup a number of audit records must be logged detailing various pieces of metadata about the guest. By monkey patching the audit_log_user_message() and audit_open() library APIs with an LD_PRELOAD library, it would be possible to perform unit testsing of the domain_audit.c APIs to ensure that correct audit records are logged for various different guest configurations. Network filter -------------- As described earlier, the network filter code is really badly designed from the point of view of unit testing. It needs to be refactored to have an intermediate data form that represents the list of iptables/ebtables rules that are expected for a given XML configuration. Currently it generates shell scripts containing iptables commands, optionally re-directing via the 'firewall-cmd' program for integration with firewalld. Refactoring to unit test this offers a prime opportunity to remove the use of shell entirely The core observation of the nwfilter code is that it consists of lists of commands to be run to create ebtables/iptables rules. Periodically there are commands which are able to rollback the state changes made by earlier commands when an error occurs. The idea for the intermediate data form would thus be to introduce the concept of a "script" object, as a higher level interface than that offered by virCommandPtr. A script would comprise one or more sets of command lines for making changes. At various points it would be possible to add a checkpoint marker, together with one or more sets of command lines that would be execute when errors occur which require rollback. The network filter code would thus output "script" objects whose command sets could be validated by a unit test suite. These script objects can then be translated into virCommandPtr objects for actual execution by separate code. Or even translated into DBus API calls for direct communication with firewalld, bypassing the inefficient firewall-cmd program. The unit test would consist of XML data files with network filter configuration rules, alongside expected script command lines, similar to how the qemu xml -> argv test works. virsh testing ------------- The virsh command has rarely had any serious direct testing of its own. Its code is mostly exercised indirectly by test suites doing testing of QEMU or other drivers. Libvirt has a hypervisor driver that is explicit targetted to testing tools like virsh - the "test" driver. This is a 100% in-memory driver that stubs out all the libvirt APIs with semi-serious semantics. It is used to do automated testing of large parts of virt-install and virt-manager. It can be used to get near 100% testing coverage of the virsh command from within unit tests. Since it is all in-memory there is no scope for it to be affected the host system state. libvirtd testing ---------------- Testing libvirtd is a tricky proposition, because merely starting the daemon will have an effect on host OS state, particularly if any resources are set to autostart. It can also clash with any libvirtd process the developer is already running. The test driver is able to be run inside libvirtd. If it were possible to disable all the other hypervisor drivers in libvirtd, the test driver could be used to execise the RPC dispatch code in libvirtd in a safe contained manner. This could be particularly good for doing scalability testing of libvirtd when large numbers of objects. eg since the test driver doesn't actually run anything, it is possible to start many 1000's of "test" guests with the only memory overhead being that of the parsed XML configuration. This would allow the test suite to identify & validate limits in the RPC protocol or inefficiencies in libvirtd. lock driver client ------------------ The virtlockd client code recently had a bug where it generated invalid RPC packets in response to certain operations. It is not desirable to have the unit tests actually run the full virtlockd daemon, just to validate the client code. It would, however, be possible to have a unit test which provided a fake virtlockd UNIX domain socket for the client to connect to. It would blindly read any data off the socket & validate it against pre-defined RPC packets. This approach is already used with some success in the QEMU JSON monitor test suite. QEMU disk snapshots ------------------- The QEMU snapshot code is some of the most complex code in the QEMU driver. As such it would be desirable to get test coverage of it. The easiest way todo this would be via the libvirt TCK integration test harness. There is, however, value in refactoring some of the code to allow its testing in a unit test environment. One approach would be for the test suite to directly invoke methods in the qemu driver dispatch table. This would require refactoring the qemuStateInitialize code though, to make it possible to setup the basic QEMU driver global state. More desirable would be to refactor the snapshot code so that it was more detached from the public driver APIs. For example if some of it could be isolated in a qemu_snapshot.c file, this may make it practical to unit test some of it. Techniques from the qemumonitorjsontest.c file could be used to stub out the monitor APIs that the snapshot code requires in order to operate. QEMU driver startup ------------------- An important aspect of the QEMU driver is that it is able to reconnect to existing domains when it starts up. It would be desirable to unit test this functionality to validate the sequence of operations done when reconnecting to a guest. Again this would require some refactoring of the qemuStateInitialize method to allow its execution in a test suite. It would also likely require the techniques for stubing the QEMU monitor APIs. XML parser/formatter reliability -------------------------------- While we have good testing of valid XML documents, there have been a number of cases where invalid XML documents have caused crashes or other bad behaviour. It is not reasonable to add test cases for every possible invalid XML document. Instead we need to adopt a fuzzing approach, where we take valid XML documents and then make incremental changes to them & then ensure that the code does not crash. Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|

2 1

[libvirt] [PATCH v4] qemu: Implement CPUs check against machine type's cpu-max
by Michal Novotny 26 Jun '13

26 Jun '13

Implement check whether (maximum) vCPUs doesn't exceed machine type's cpu-max settings. Differences between v3 and v4 (this one): - Rebased to latest libvirt version - Capability XML output extended by maxCpus field - Extended caps-qemu-kvm.xml test by maxCpus for one of test emulators On older versions of QEMU the check is disabled. Signed-off-by: Michal Novotny <minovotn(a)redhat.com> --- docs/schemas/capability.rng | 5 ++++ src/conf/capabilities.c | 4 +++ src/conf/capabilities.h | 1 + src/qemu/qemu_capabilities.c | 41 +++++++++++++++++++++++++++- src/qemu/qemu_capabilities.h | 3 +- src/qemu/qemu_monitor.h | 1 + src/qemu/qemu_monitor_json.c | 6 ++++ src/qemu/qemu_process.c | 27 ++++++++++++++++++ tests/capabilityschemadata/caps-qemu-kvm.xml | 16 +++++------ 9 files changed, 94 insertions(+), 10 deletions(-) diff --git a/docs/schemas/capability.rng b/docs/schemas/capability.rng index 106ca73..65c7c72 100644 --- a/docs/schemas/capability.rng +++ b/docs/schemas/capability.rng @@ -290,6 +290,11 @@ <text/> </attribute> </optional> + <optional> + <attribute name='maxCpus'> + <ref name='unsignedInt'/> + </attribute> + </optional> <text/> </element> </define> diff --git a/src/conf/capabilities.c b/src/conf/capabilities.c index da92c78..5aeb2ab 100644 --- a/src/conf/capabilities.c +++ b/src/conf/capabilities.c @@ -853,6 +853,8 @@ virCapabilitiesFormatXML(virCapsPtr caps) virBufferAddLit(&xml, " <machine"); if (machine->canonical) virBufferAsprintf(&xml, " canonical='%s'", machine->canonical); + if (machine->maxCpus > 0) + virBufferAsprintf(&xml, " maxCpus='%d'", machine->maxCpus); virBufferAsprintf(&xml, ">%s</machine>\n", machine->name); } @@ -871,6 +873,8 @@ virCapabilitiesFormatXML(virCapsPtr caps) virBufferAddLit(&xml, " <machine"); if (machine->canonical) virBufferAsprintf(&xml, " canonical='%s'", machine->canonical); + if (machine->maxCpus > 0) + virBufferAsprintf(&xml, " maxCpus='%d'", machine->maxCpus); virBufferAsprintf(&xml, ">%s</machine>\n", machine->name); } virBufferAddLit(&xml, " </domain>\n"); diff --git a/src/conf/capabilities.h b/src/conf/capabilities.h index abcf6de..22b6fb6 100644 --- a/src/conf/capabilities.h +++ b/src/conf/capabilities.h @@ -46,6 +46,7 @@ typedef virCapsGuestMachine *virCapsGuestMachinePtr; struct _virCapsGuestMachine { char *name; char *canonical; + int maxCpus; }; typedef struct _virCapsGuestDomainInfo virCapsGuestDomainInfo; diff --git a/src/qemu/qemu_capabilities.c b/src/qemu/qemu_capabilities.c index c4e076a..89f41b8 100644 --- a/src/qemu/qemu_capabilities.c +++ b/src/qemu/qemu_capabilities.c @@ -256,6 +256,7 @@ struct _virQEMUCaps { size_t nmachineTypes; char **machineTypes; char **machineAliases; + int *machineMaxCpus; }; struct _virQEMUCapsCache { @@ -335,6 +336,7 @@ virQEMUCapsSetDefaultMachine(virQEMUCapsPtr qemuCaps, { char *name = qemuCaps->machineTypes[defIdx]; char *alias = qemuCaps->machineAliases[defIdx]; + int cpu_max = qemuCaps->machineMaxCpus[defIdx]; memmove(qemuCaps->machineTypes + 1, qemuCaps->machineTypes, @@ -342,8 +344,12 @@ virQEMUCapsSetDefaultMachine(virQEMUCapsPtr qemuCaps, memmove(qemuCaps->machineAliases + 1, qemuCaps->machineAliases, sizeof(qemuCaps->machineAliases[0]) * defIdx); + memmove(qemuCaps->machineMaxCpus + 1, + qemuCaps->machineMaxCpus, + sizeof(qemuCaps->machineMaxCpus[0]) * defIdx); qemuCaps->machineTypes[0] = name; qemuCaps->machineAliases[0] = alias; + qemuCaps->machineMaxCpus[0] = cpu_max; } /* Format is: @@ -390,7 +396,8 @@ virQEMUCapsParseMachineTypesStr(const char *output, } if (VIR_REALLOC_N(qemuCaps->machineTypes, qemuCaps->nmachineTypes + 1) < 0 || - VIR_REALLOC_N(qemuCaps->machineAliases, qemuCaps->nmachineTypes + 1) < 0) { + VIR_REALLOC_N(qemuCaps->machineAliases, qemuCaps->nmachineTypes + 1) < 0 || + VIR_REALLOC_N(qemuCaps->machineMaxCpus, qemuCaps->nmachineTypes + 1) < 0) { VIR_FREE(name); VIR_FREE(canonical); virReportOOMError(); @@ -404,6 +411,8 @@ virQEMUCapsParseMachineTypesStr(const char *output, qemuCaps->machineTypes[qemuCaps->nmachineTypes-1] = name; qemuCaps->machineAliases[qemuCaps->nmachineTypes-1] = NULL; } + /* Value 0 means "unknown" as it's not exposed by QEMU binary */ + qemuCaps->machineMaxCpus[qemuCaps->nmachineTypes-1] = 0; } while ((p = next)); @@ -1764,11 +1773,14 @@ virQEMUCapsPtr virQEMUCapsNewCopy(virQEMUCapsPtr qemuCaps) goto no_memory; if (VIR_ALLOC_N(ret->machineAliases, qemuCaps->nmachineTypes) < 0) goto no_memory; + if (VIR_ALLOC_N(ret->machineMaxCpus, qemuCaps->nmachineTypes) < 0) + goto no_memory; ret->nmachineTypes = qemuCaps->nmachineTypes; for (i = 0; i < qemuCaps->nmachineTypes; i++) { if (VIR_STRDUP(ret->machineTypes[i], qemuCaps->machineTypes[i]) < 0 || VIR_STRDUP(ret->machineAliases[i], qemuCaps->machineAliases[i]) < 0) goto error; + ret->machineMaxCpus[i] = qemuCaps->machineMaxCpus[i]; } return ret; @@ -1789,9 +1801,11 @@ void virQEMUCapsDispose(void *obj) for (i = 0; i < qemuCaps->nmachineTypes; i++) { VIR_FREE(qemuCaps->machineTypes[i]); VIR_FREE(qemuCaps->machineAliases[i]); + qemuCaps->machineMaxCpus[i] = -1; } VIR_FREE(qemuCaps->machineTypes); VIR_FREE(qemuCaps->machineAliases); + VIR_FREE(qemuCaps->machineMaxCpus); for (i = 0; i < qemuCaps->ncpuDefinitions; i++) { VIR_FREE(qemuCaps->cpuDefinitions[i]); @@ -1932,6 +1946,7 @@ int virQEMUCapsGetMachineTypesCaps(virQEMUCapsPtr qemuCaps, if (VIR_STRDUP(mach->name, qemuCaps->machineTypes[i]) < 0) goto error; } + mach->maxCpus = qemuCaps->machineMaxCpus[i]; (*machines)[i] = mach; } @@ -1966,6 +1981,25 @@ const char *virQEMUCapsGetCanonicalMachine(virQEMUCapsPtr qemuCaps, } +int virQEMUCapsGetMachineMaxCpus(virQEMUCapsPtr qemuCaps, + const char *name) +{ + size_t i; + + if (!name) + return 0; + + for (i = 0; i < qemuCaps->nmachineTypes; i++) { + if (!qemuCaps->machineMaxCpus[i]) + continue; + if (STREQ(qemuCaps->machineTypes[i], name)) + return qemuCaps->machineMaxCpus[i]; + } + + return 0; +} + + static int virQEMUCapsProbeQMPCommands(virQEMUCapsPtr qemuCaps, qemuMonitorPtr mon) @@ -2083,6 +2117,10 @@ virQEMUCapsProbeQMPMachineTypes(virQEMUCapsPtr qemuCaps, virReportOOMError(); goto cleanup; } + if (VIR_ALLOC_N(qemuCaps->machineMaxCpus, nmachines) < 0) { + virReportOOMError(); + goto cleanup; + } for (i = 0; i < nmachines; i++) { if (VIR_STRDUP(qemuCaps->machineAliases[i], machines[i]->alias) < 0 || @@ -2090,6 +2128,7 @@ virQEMUCapsProbeQMPMachineTypes(virQEMUCapsPtr qemuCaps, goto cleanup; if (machines[i]->isDefault) defIdx = i; + qemuCaps->machineMaxCpus[i] = machines[i]->cpu_max; } qemuCaps->nmachineTypes = nmachines; diff --git a/src/qemu/qemu_capabilities.h b/src/qemu/qemu_capabilities.h index 64a4b1d..7088747 100644 --- a/src/qemu/qemu_capabilities.h +++ b/src/qemu/qemu_capabilities.h @@ -234,7 +234,8 @@ size_t virQEMUCapsGetMachineTypes(virQEMUCapsPtr qemuCaps, char ***names); const char *virQEMUCapsGetCanonicalMachine(virQEMUCapsPtr qemuCaps, const char *name); - +int virQEMUCapsGetMachineMaxCpus(virQEMUCapsPtr qemuCaps, + const char *name); int virQEMUCapsGetMachineTypesCaps(virQEMUCapsPtr qemuCaps, size_t *nmachines, virCapsGuestMachinePtr **machines); diff --git a/src/qemu/qemu_monitor.h b/src/qemu/qemu_monitor.h index 3d9afa3..06ae4c5 100644 --- a/src/qemu/qemu_monitor.h +++ b/src/qemu/qemu_monitor.h @@ -654,6 +654,7 @@ struct _qemuMonitorMachineInfo { char *name; bool isDefault; char *alias; + int cpu_max; }; int qemuMonitorGetMachines(qemuMonitorPtr mon, diff --git a/src/qemu/qemu_monitor_json.c b/src/qemu/qemu_monitor_json.c index 88a0dc9..1362acf 100644 --- a/src/qemu/qemu_monitor_json.c +++ b/src/qemu/qemu_monitor_json.c @@ -4042,6 +4042,12 @@ int qemuMonitorJSONGetMachines(qemuMonitorPtr mon, if (VIR_STRDUP(info->alias, tmp) < 0) goto cleanup; } + if (virJSONValueObjectHasKey(child, "cpu-max") && + virJSONValueObjectGetNumberInt(child, "cpu-max", &info->cpu_max) < 0) { + virReportError(VIR_ERR_INTERNAL_ERROR, "%s", + _("query-machines reply has malformed 'cpu-max' data")); + goto cleanup; + } } ret = n; diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c index 5a0f18b..3146ce2 100644 --- a/src/qemu/qemu_process.c +++ b/src/qemu/qemu_process.c @@ -3330,6 +3330,30 @@ error: } +static bool +qemuValidateCpuMax(virDomainDefPtr def, virQEMUCapsPtr qemuCaps) +{ + int cpu_max; + + cpu_max = virQEMUCapsGetMachineMaxCpus(qemuCaps, def->os.machine); + if (!cpu_max) + return true; + + if (def->vcpus > cpu_max) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + "%s", _("CPUs greater than specified machine type limit")); + return false; + } + + if (def->maxvcpus > cpu_max) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + "%s", _("Maximum CPUs greater than specified machine type limit")); + return false; + } + + return true; +} + int qemuProcessStart(virConnectPtr conn, virQEMUDriverPtr driver, virDomainObjPtr vm, @@ -3519,6 +3543,9 @@ int qemuProcessStart(virConnectPtr conn, vm->def->emulator))) goto cleanup; + if (!qemuValidateCpuMax(vm->def, priv->qemuCaps)) + goto cleanup; + if (qemuAssignDeviceAliases(vm->def, priv->qemuCaps) < 0) goto cleanup; diff --git a/tests/capabilityschemadata/caps-qemu-kvm.xml b/tests/capabilityschemadata/caps-qemu-kvm.xml index 36c4b49..1fbc22b 100644 --- a/tests/capabilityschemadata/caps-qemu-kvm.xml +++ b/tests/capabilityschemadata/caps-qemu-kvm.xml @@ -33,18 +33,18 @@ <arch name='i686'> <wordsize>32</wordsize> <emulator>/usr/bin/qemu</emulator> - <machine>pc-0.11</machine> - <machine canonical='pc-0.11'>pc</machine> - <machine>pc-0.10</machine> - <machine>isapc</machine> + <machine maxCpus='255'>pc-0.11</machine> + <machine canonical='pc-0.11' maxCpus='255'>pc</machine> + <machine maxCpus='255'>pc-0.10</machine> + <machine maxCpus='1'>isapc</machine> <domain type='qemu'> </domain> <domain type='kvm'> <emulator>/usr/bin/qemu-kvm</emulator> - <machine>pc-0.11</machine> - <machine canonical='pc-0.11'>pc</machine> - <machine>pc-0.10</machine> - <machine>isapc</machine> + <machine maxCpus='255'>pc-0.11</machine> + <machine canonical='pc-0.11' maxCpus='255'>pc</machine> + <machine maxCpus='255'>pc-0.10</machine> + <machine maxCpus='1'>isapc</machine> </domain> </arch> <features> -- 1.7.11.7

2 5

[libvirt] [PATCH] Plug leak in virCgroupMoveTask
by Ján Tomko 26 Jun '13

26 Jun '13

We only break out of the while loop if *content is an empty string. However the buffer has been allocated to BUFSIZ + 1 (8193 in my case), but it gets overwritten in the next for iteration. Move VIR_FREE right before we overwrite it to avoid the leak. ==5777== 16,386 bytes in 2 blocks are definitely lost in loss record 1,022 of 1,027 ==5777== by 0x5296E28: virReallocN (viralloc.c:184) ==5777== by 0x52B0C66: virFileReadLimFD (virfile.c:1137) ==5777== by 0x52B0E1A: virFileReadAll (virfile.c:1199) ==5777== by 0x529B092: virCgroupGetValueStr (vircgroup.c:534) ==5777== by 0x529AF64: virCgroupMoveTask (vircgroup.c:1079) Introduced by 83e4c77. https://bugzilla.redhat.com/show_bug.cgi?id=978352 --- src/util/vircgroup.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/src/util/vircgroup.c b/src/util/vircgroup.c index 0484c71..5e75355 100644 --- a/src/util/vircgroup.c +++ b/src/util/vircgroup.c @@ -1076,6 +1076,7 @@ int virCgroupMoveTask(virCgroupPtr src_group, virCgroupPtr dest_group) * aware that it needs to move. Therefore, we must iterate * until content is empty. */ while (1) { + VIR_FREE(content); rc = virCgroupGetValueStr(src_group, i, "tasks", &content); if (rc != 0) return rc; @@ -1085,8 +1086,6 @@ int virCgroupMoveTask(virCgroupPtr src_group, virCgroupPtr dest_group) rc = virCgroupAddTaskStrController(dest_group, content, i); if (rc != 0) goto cleanup; - - VIR_FREE(content); } } -- 1.8.1.5

2 2