- Devel - libvirt List Archives

[libvirt] [PATCH 1/3] LXC: remove unnecessary check on root filesystem
by Gao feng 21 May '13

21 May '13

After commit c131525bec5af248e3843224bc5ce8d6435760f0 "Auto-add a root <filesystem> element to LXC containers on startup" for libvirt lxc, root must be existent. Signed-off-by: Gao feng <gaofeng(a)cn.fujitsu.com> --- src/lxc/lxc_container.c | 18 +++++++----------- src/lxc/lxc_controller.c | 11 ----------- 2 files changed, 7 insertions(+), 22 deletions(-) diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c index a1b6aff..b4be4cd 100644 --- a/src/lxc/lxc_container.c +++ b/src/lxc/lxc_container.c @@ -1940,17 +1940,13 @@ static int lxcContainerChild(void *data) root = virDomainGetRootFilesystem(vmDef); if (argv->nttyPaths) { - if (root) { - const char *tty = argv->ttyPaths[0]; - if (STRPREFIX(tty, "/dev/pts/")) - tty += strlen("/dev/pts/"); - if (virAsprintf(&ttyPath, "%s/%s.devpts/%s", - LXC_STATE_DIR, vmDef->name, tty) < 0) { - virReportOOMError(); - goto cleanup; - } - } else if (VIR_STRDUP(ttyPath, argv->ttyPaths[0]) < 0) { - goto cleanup; + const char *tty = argv->ttyPaths[0]; + if (STRPREFIX(tty, "/dev/pts/")) + tty += strlen("/dev/pts/"); + if (virAsprintf(&ttyPath, "%s/%s.devpts/%s", + LXC_STATE_DIR, vmDef->name, tty) < 0) { + virReportOOMError(); + goto cleanup; } } else if (VIR_STRDUP(ttyPath, "/dev/null") < 0) { goto cleanup; diff --git a/src/lxc/lxc_controller.c b/src/lxc/lxc_controller.c index 730236e..2f01958 100644 --- a/src/lxc/lxc_controller.c +++ b/src/lxc/lxc_controller.c @@ -1280,22 +1280,11 @@ cleanup: static int virLXCControllerSetupDevPTS(virLXCControllerPtr ctrl) { - virDomainFSDefPtr root = virDomainGetRootFilesystem(ctrl->def); char *mount_options = NULL; char *opts = NULL; char *devpts = NULL; int ret = -1; - if (!root) { - if (ctrl->nconsoles != 1) { - virReportError(VIR_ERR_CONFIG_UNSUPPORTED, - _("Expected exactly one console, but got %zu"), - ctrl->nconsoles); - return -1; - } - return 0; - } - VIR_DEBUG("Setting up private /dev/pts"); /* -- 1.8.1.4

3 6

[libvirt] [PATCH] qemuDomainChangeEjectableMedia: Unlock domain while waiting for event
by Michal Privoznik 21 May '13

21 May '13

In 84c59ffa I've tried to fix changing ejectable media process. The process should go like this: 1) we need to call 'eject' on the monitor 2) we should wait for 'DEVICE_TRAY_MOVED' event 3) now we can issue 'change' command However, while waiting in step 2) the domain monitor was locked. So even if qemu reported the desired event, the proper callback was not called immediately. The monitor handling code needs to lock the monitor in order to read the event. So that's the first lock we must not hold while waiting. The second one is the domain lock. When monitor handling code reads an event, the appropriate callback is called then. The first thing that each callback does is locking the corresponding domain as a domain or its device is about to change state. So we need to unlock both monitor and VM lock. Well, holding any lock while sleep()-ing is not the best thing to do anyway. --- src/qemu/qemu_hotplug.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) diff --git a/src/qemu/qemu_hotplug.c b/src/qemu/qemu_hotplug.c index 77d9f4f..88c3a6c 100644 --- a/src/qemu/qemu_hotplug.c +++ b/src/qemu/qemu_hotplug.c @@ -94,15 +94,20 @@ int qemuDomainChangeEjectableMedia(virQEMUDriverPtr driver, qemuDomainObjEnterMonitor(driver, vm); ret = qemuMonitorEjectMedia(priv->mon, driveAlias, force); + qemuDomainObjExitMonitor(driver, vm); + virObjectRef(vm); /* we don't want to report errors from media tray_open polling */ while (retries--) { if (origdisk->tray_status == VIR_DOMAIN_DISK_TRAY_OPEN) break; + virObjectUnlock(vm); VIR_DEBUG("Waiting 500ms for tray to open. Retries left %d", retries); usleep(500 * 1000); /* sleep 500ms */ + virObjectLock(vm); } + virObjectUnref(vm); if (disk->src) { /* deliberately don't depend on 'ret' as 'eject' may have failed for the @@ -115,7 +120,7 @@ int qemuDomainChangeEjectableMedia(virQEMUDriverPtr driver, if (retries <= 0) { virReportError(VIR_ERR_OPERATION_FAILED, "%s", _("Unable to eject media before changing it")); - goto exit_monitor; + goto audit; } if (disk->type != VIR_DOMAIN_DISK_TYPE_DIR) { @@ -124,13 +129,13 @@ int qemuDomainChangeEjectableMedia(virQEMUDriverPtr driver, else if (origdisk->format > 0) format = virStorageFileFormatTypeToString(origdisk->format); } + qemuDomainObjEnterMonitor(driver, vm); ret = qemuMonitorChangeMedia(priv->mon, driveAlias, disk->src, format); + qemuDomainObjExitMonitor(driver, vm); } -exit_monitor: - qemuDomainObjExitMonitor(driver, vm); - +audit: virDomainAuditDisk(vm, origdisk->src, disk->src, "update", ret >= 0); if (ret < 0) -- 1.8.2.1

2 1

[libvirt] ANNOUNCE: libvirt 1.0.5.1 maintenance release
by Cole Robinson 21 May '13

21 May '13

libvirt 1.0.5.1 maintenance release is now available. This is libvirt 1.0.5 with additional bugfixes that have accumulated upstream since the initial release. This release can be downloaded at: http://libvirt.org/sources/stable_updates/libvirt-1.0.5.1.tar.gz Changes in this version: * tests: use portable shell code * qemu: Fix cgroup handling when setting VCPU BW * daemon: fix leak after listing all volumes * Fix iohelper usage with streams opened for read * util: fix virFileOpenAs return value and resulting error logs * iscsi: don't leak portal string when starting a pool * don't mention disk controllers in generic controller errors * conf: don't crash on a tpm device with no backends * tests: files named '.*-invalid.xml' should fail validation * qemu: allocate network connections sooner during domain startup * Make detect_scsi_host_caps a function on all architectures * Fixup rpcgen code on kFreeBSD too * Fix release of resources with lockd plugin * build: avoid non-portable cast of pthread_t * Fix potential use of undefined variable in remote dispatch code * build: fix mingw build of virprocess.c * Fix F_DUPFD_CLOEXEC operation args * spec: proper soft static allocation of qemu uid * build: clean up stray files found by 'make distcheck' * build: always include libvirt_lxc.syms in tarball * qemu: fix stupid typos in VFIO cgroup setup/teardown * build: always include sanitytest in tarball * virInitctlRequest: unbreak make syntax check * virInitctlRequest: unbreak make syntax check * network: fix network driver startup for qemu:///session For info about past maintenance releases, see: http://wiki.libvirt.org/page/Maintenance_Releases Thanks, Cole

1 0

[libvirt] ANNOUNCE: libvirt 0.10.2.5 maintenance release
by Cole Robinson 21 May '13

21 May '13

libvirt 0.10.2.5 maintenance release is now available. This is libvirt 0.10.2 with additional bugfixes that have accumulated upstream since the initial release. This release can be downloaded at: http://libvirt.org/sources/stable_updates/libvirt-0.10.2.5.tar.gz Changes in this version: * Fix TLS tests with gnutls 3 * daemon: fix leak after listing all volumes * spec: proper soft static allocation of qemu uid * spec: Fix minor changelog issues * spec: Avoid using makeinstall relic * audit: properly encode device path in cgroup audit * storage: Fix lvcreate parameter for backingStore. For info about past maintenance releases, see: http://wiki.libvirt.org/page/Maintenance_Releases Thanks, Cole

1 0

[libvirt] [PATCHv3 00/11] Multiple TX queue support
by Michal Privoznik 21 May '13

21 May '13

Fixed version with Laine's comments worked in [1]. Patches 2, 3 and 5 has been ACKed already. BTW, patch 4 is really a different to patch 3. The only thing they share is a commit message body. Kernel and subsequently QEMU learned multiple transmit queues a while ago. The feature has a nice advantage, it allows a single guest to transmit multiple flows of network data using multiple CPUs simultaneously which increase traffic bandwidth. A lot. The documentation how to use this is available at [2] or [3]. 1: https://www.redhat.com/archives/libvir-list/2013-May/msg00943.html 2: https://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/tree/Docume… 3: http://git.qemu.org/?p=qemu.git;a=blob;f=qemu-options.hx;hb=HEAD#l1363 Michal Privoznik (11): Introduce /domain/devices/interface/driver/@queues attribute qemu: Move interface cmd line construction into a separate function qemu: Make qemuMonitorAddNetdev to pass multiple FDs qemu: Make qemuMonitorAddHostNetwork to pass multiple FDs qemu: Adapt command line generation to multiqueue net util: Learn virNetDevTapCreate multiqueue network qemu: Allow multiple vhost-net openings qemu: Rework qemuNetworkIfaceConnect qemu: Adapt qemuDomainAttachNetDevice to multiqueue net qemu: Adapt qemuBuildInterfaceCommandLine to to multiqueue net qemu: Enable multiqueue network docs/formatdomain.html.in | 12 +- docs/schemas/domaincommon.rng | 5 + src/conf/domain_conf.c | 15 + src/conf/domain_conf.h | 1 + src/network/bridge_driver.c | 2 +- src/qemu/qemu_command.c | 447 +++++++++++++-------- src/qemu/qemu_command.h | 13 +- src/qemu/qemu_hotplug.c | 103 +++-- src/qemu/qemu_monitor.c | 78 ++-- src/qemu/qemu_monitor.h | 8 +- src/uml/uml_conf.c | 5 +- src/util/virnetdevtap.c | 113 +++--- src/util/virnetdevtap.h | 2 + .../qemuxml2argvdata/qemuxml2argv-vhost_queues.xml | 51 +++ tests/qemuxml2xmltest.c | 1 + 15 files changed, 568 insertions(+), 288 deletions(-) create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-vhost_queues.xml -- 1.8.2.1

2 25

[libvirt] [PATCH v2] openvzDomainSetNetwork: use virCommand
by Michal Privoznik 21 May '13

21 May '13

Currently, the openvzDomainSetNetwork function constructs an array of strings representing a command line for VZCTL binary. This is a overkill since our virCommand APIs can cover all the functionality. Moreover, the function is not following our structure where return value is set to -1 initially, and after all operations succeeded then it is set to zero. --- src/openvz/openvz_driver.c | 80 +++++++++++----------------------------------- 1 file changed, 19 insertions(+), 61 deletions(-) diff --git a/src/openvz/openvz_driver.c b/src/openvz/openvz_driver.c index 129e328..d46baab 100644 --- a/src/openvz/openvz_driver.c +++ b/src/openvz/openvz_driver.c @@ -88,15 +88,6 @@ static void openvzDriverUnlock(struct openvz_driver *driver) struct openvz_driver ovz_driver; -static void cmdExecFree(const char *cmdExec[]) -{ - int i=-1; - while (cmdExec[++i]) { - VIR_FREE(cmdExec[i]); - } -} - - static int openvzDomainDefPostParse(virDomainDefPtr def, virCapsPtr caps ATTRIBUTE_UNUSED, @@ -827,22 +818,13 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, virDomainNetDefPtr net, virBufferPtr configBuf) { - int rc = 0, narg; - const char *prog[OPENVZ_MAX_ARG]; + int rc = -1; char macaddr[VIR_MAC_STRING_BUFLEN]; virMacAddr host_mac; char host_macaddr[VIR_MAC_STRING_BUFLEN]; struct openvz_driver *driver = conn->privateData; char *opt = NULL; - -#define ADD_ARG_LIT(thisarg) \ - do { \ - if (narg >= OPENVZ_MAX_ARG) \ - goto no_memory; \ - if ((prog[narg++] = strdup(thisarg)) == NULL) \ - goto no_memory; \ - } while (0) - + virCommandPtr cmd = NULL; if (net == NULL) return 0; @@ -852,18 +834,11 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, return -1; } - for (narg = 0; narg < OPENVZ_MAX_ARG; narg++) - prog[narg] = NULL; + if (net->type != VIR_DOMAIN_NET_TYPE_BRIDGE && + net->type != VIR_DOMAIN_NET_TYPE_ETHERNET) + return 0; - narg = 0; - - if (net->type == VIR_DOMAIN_NET_TYPE_BRIDGE || - net->type == VIR_DOMAIN_NET_TYPE_ETHERNET) { - ADD_ARG_LIT(VZCTL); - ADD_ARG_LIT("--quiet"); - ADD_ARG_LIT("set"); - ADD_ARG_LIT(vpsid); - } + cmd = virCommandNewArgList(VZCTL, "--quiet", "set", vpsid, NULL); virMacAddrFormat(&net->mac, macaddr); virDomainNetGenerateMAC(driver->xmlopt, &host_mac); @@ -875,9 +850,6 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, virBuffer buf = VIR_BUFFER_INITIALIZER; int veid = openvzGetVEID(vpsid); - /* --netif_add ifname[,mac,host_ifname,host_mac] */ - ADD_ARG_LIT("--netif_add") ; - /* if user doesn't specify guest interface name, * then we need to generate it */ if (net->data.ethernet.dev == NULL) { @@ -885,8 +857,7 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, if (net->data.ethernet.dev == NULL) { virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("Could not generate eth name for container")); - rc = -1; - goto exit; + goto cleanup; } } @@ -897,8 +868,7 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, if (net->ifname == NULL) { virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("Could not generate veth name")); - rc = -1; - goto exit; + goto cleaup; } } @@ -919,40 +889,28 @@ openvzDomainSetNetwork(virConnectPtr conn, const char *vpsid, } } - if (!(opt = virBufferContentAndReset(&buf))) - goto no_memory; + if (!(opt = virBufferContentAndReset(&buf))) { + virReportOOMError(); + goto cleanup; + } - ADD_ARG_LIT(opt) ; + /* --netif_add ifname[,mac,host_ifname,host_mac] */ + virCommandAddArgList(cmd, "--netif_add", opt, NULL); VIR_FREE(opt); } else if (net->type == VIR_DOMAIN_NET_TYPE_ETHERNET && net->data.ethernet.ipaddr != NULL) { /* --ipadd ip */ - ADD_ARG_LIT("--ipadd") ; - ADD_ARG_LIT(net->data.ethernet.ipaddr) ; + virCommandAddArgList(cmd, "--ipadd", net->data.ethernet.ipaddr, NULL); } /* TODO: processing NAT and physical device */ - if (prog[0] != NULL) { - ADD_ARG_LIT("--save"); - if (virRun(prog, NULL) < 0) { - rc = -1; - goto exit; - } - } + virCommandAddArg(cmd, "--save"); + ret = virCommandRun(cmd, NULL); - exit: - cmdExecFree(prog); + cleanup: + virCommandFree(cmd); return rc; - - no_memory: - VIR_FREE(opt); - virReportError(VIR_ERR_INTERNAL_ERROR, - _("Could not put argument to %s"), VZCTL); - cmdExecFree(prog); - return -1; - -#undef ADD_ARG_LIT } -- 1.8.2.1

3 4

[libvirt] [PATCH 0/2] qemu: Rework CD/DVD changing
by Michal Privoznik 21 May '13

21 May '13

The first patch is bare bug fix (without any bug reported though). The second is again a bug fix, but not so easy to spot. Basically, when we change a media, we should issue 'eject' first, then wait until the tray gets open, after which we can finally issue 'change' command. Even for bare 'eject' we ought to check status, shouldn't we? Michal Privoznik (2): qemu_monitor: Fix tray-open attribute in query-block qemu_hotplug: Rework media changing process src/qemu/qemu_hotplug.c | 51 +++++++++++++++++++++++++++++++++++++++++--- src/qemu/qemu_monitor_json.c | 2 +- src/qemu/qemu_monitor_text.c | 6 +++--- 3 files changed, 52 insertions(+), 7 deletions(-) -- 1.8.0.2

6 22

[libvirt] two nasty races
by Eric Blake 21 May '13

21 May '13

I found root cause on two nasty races today, but ran out of time to write the patches until next week. 1. We have a TOCTTOU race when moving processes between cgroups. virCgroupMoveTask can fail if listing the threads in the parent group picks up multiple threads, then one of those threads exits before we then write to the emulator subgroup to relocate those threads; the chain reaction from this failure is that the domain fails to start. qemu spawns short-lived threads all the time (possibly due to glibc using threads under the hood to implement aio as qemu first probes a disk image); with a 3-second-delay hook script, I was able to hit this race about 5% of the time when repeatedly starting multiple domains in parallel. The fix should be ignoring any ESRCH failures when writing during virCgroupAddTAskStrController, but I plan to test that before posting my patch. Present at least since 0.10.2 (Fedora 18); probably earlier (but I didn't check how far back). 2. We have a (rare) deadlock due to the use of a non-async-signal-safe getpwuid_r in between fork and exec. I was extremely surprised when I hit it in real life while trying to debug the first race; but sure enough, the backtrace confirms that glibc grabs a mutex inside getpwuid_r, and if some other thread in the parent process is interrupted in the middle of getpwuid_r while we are forking a child in our thread, then our child is hung; and since the parent depends on a handshake with the child, the parent's worker thread was also hung. I think the fix to this is to call getpwuid_r prior to fork, and change virSetUIDGID to take the list of supplemental groups as a parameter rather than trying to learn it on the fly. Again, something I plan to test once I have the patch written. Present at least since commit f42cf7cb (Dec 2010) with the introduction of virSetUIDGID, and probably further back. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org

2 4

[libvirt] [PATCH] esx: Implement esxDomainSetMemoryFlags
by Matthias Bolte 21 May '13

21 May '13

--- src/esx/esx_driver.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/src/esx/esx_driver.c b/src/esx/esx_driver.c index b55ef45..3962bf1 100644 --- a/src/esx/esx_driver.c +++ b/src/esx/esx_driver.c @@ -2187,6 +2187,27 @@ esxDomainSetMemory(virDomainPtr domain, unsigned long memory) +static int +esxDomainSetMemoryFlags(virDomainPtr domain, unsigned long memory, + unsigned int flags) +{ + if (flags != VIR_DOMAIN_MEM_CURRENT && + flags != VIR_DOMAIN_MEM_LIVE && + flags != (VIR_DOMAIN_MEM_CURRENT | VIR_DOMAIN_MEM_MAXIMUM) && + flags != (VIR_DOMAIN_MEM_LIVE | VIR_DOMAIN_MEM_MAXIMUM)) { + virReportError(VIR_ERR_INVALID_ARG, _("unsupported flags: (0x%x)"), flags); + return -1; + } + + if (flags & VIR_DOMAIN_MEM_MAXIMUM) { + return esxDomainSetMaxMemory(domain, memory); + } else { + return esxDomainSetMemory(domain, memory); + } +} + + + /* * libvirt exposed virtual CPU usage in absolute time, ESX doesn't provide this * information in this format. It exposes it in 20 seconds slots, but it's hard @@ -5286,6 +5307,7 @@ static virDriver esxDriver = { .domainGetMaxMemory = esxDomainGetMaxMemory, /* 0.7.0 */ .domainSetMaxMemory = esxDomainSetMaxMemory, /* 0.7.0 */ .domainSetMemory = esxDomainSetMemory, /* 0.7.0 */ + .domainSetMemoryFlags = esxDomainSetMemoryFlags, /* 1.0.6 */ .domainSetMemoryParameters = esxDomainSetMemoryParameters, /* 0.8.6 */ .domainGetMemoryParameters = esxDomainGetMemoryParameters, /* 0.8.6 */ .domainGetInfo = esxDomainGetInfo, /* 0.7.0 */ -- 1.7.9.5

2 1

[libvirt] [PATCH] esx: Accept VIR_DOMAIN_AFFECT_CURRENT in addition to VIR_DOMAIN_AFFECT_LIVE
by Matthias Bolte 21 May '13

21 May '13

Basically autotranslate VIR_DOMAIN_AFFECT_CURRENT to VIR_DOMAIN_AFFECT_LIVE. Modification impact is handled lax anyway. There is no distinction between a live and a stored config in the VMware world. A strict modification impact handling would require to check the virtual machine state each time and enforce VIR_DOMAIN_AFFECT_LIVE-only for running and VIR_DOMAIN_AFFECT_CONFIG-only for inactive domains. But that would be useless overhead. --- src/esx/esx_driver.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/src/esx/esx_driver.c b/src/esx/esx_driver.c index f53b9c4..85640ce 100644 --- a/src/esx/esx_driver.c +++ b/src/esx/esx_driver.c @@ -2531,7 +2531,7 @@ esxDomainSetVcpusFlags(virDomainPtr domain, unsigned int nvcpus, esxVI_TaskInfoState taskInfoState; char *taskInfoErrorMessage = NULL; - if (flags != VIR_DOMAIN_AFFECT_LIVE) { + if (flags != VIR_DOMAIN_AFFECT_CURRENT && flags != VIR_DOMAIN_AFFECT_LIVE) { virReportError(VIR_ERR_INVALID_ARG, _("unsupported flags: (0x%x)"), flags); return -1; } @@ -2615,7 +2615,8 @@ esxDomainGetVcpusFlags(virDomainPtr domain, unsigned int flags) esxVI_ObjectContent *hostSystem = NULL; esxVI_DynamicProperty *dynamicProperty = NULL; - if (flags != (VIR_DOMAIN_AFFECT_LIVE | VIR_DOMAIN_VCPU_MAXIMUM)) { + if (flags != (VIR_DOMAIN_AFFECT_CURRENT | VIR_DOMAIN_VCPU_MAXIMUM) && + flags != (VIR_DOMAIN_AFFECT_LIVE | VIR_DOMAIN_VCPU_MAXIMUM)) { virReportError(VIR_ERR_INVALID_ARG, _("unsupported flags: (0x%x)"), flags); return -1; } @@ -5222,7 +5223,7 @@ esxDomainGetMetadata(virDomainPtr domain, int type, esxVI_String *propertyNameList = NULL; esxVI_ObjectContent *virtualMachine = NULL; - if (flags != VIR_DOMAIN_AFFECT_LIVE) { + if (flags != VIR_DOMAIN_AFFECT_CURRENT && flags != VIR_DOMAIN_AFFECT_LIVE) { virReportError(VIR_ERR_INVALID_ARG, _("unsupported flags: (0x%x)"), flags); return NULL; } -- 1.7.9.5

2 1