January 2014 - Devel - libvirt List Archives

[libvirt] [PATCH v2 0/4] Fix deadlock in nwfilter code
by Daniel P. Berrange 29 Jan '14

29 Jan '14

Since we introduced fine grained locking into the QEMU driver so that VM start can run in parallel, we appear to have caused a race with the nwfilter code. In particular since we no longer hold the global QEMU driver lock for the duration of VM startup we have a lock ordering flaw. This results in deadlock when nwfilter operations happen in parallel with VM startup. This also affects the LXC driver. This patch series attempts to address the problem https://bugzilla.redhat.com/show_bug.cgi?id=929412 the removal of the windows thread impl isn't strictly required, I just didn't want to waste time creating a read/write lock impl for Windows threads. See also this callgraph http://berrange.fedorapeople.org/nwfilter.ps In v2: - Re-ordered patches and squashed two together - Add missing locking in state reload function - Don't remove locking from virNWFilterInstantiateFilterLate Daniel P. Berrange (4): Add a read/write lock implementation Remove windows thread implementation in favour of pthreads Add a mutex to serialize updates to firewall Push nwfilter update locking up to top level configure.ac | 15 +- src/Makefile.am | 4 - src/conf/nwfilter_conf.c | 23 +- src/conf/nwfilter_conf.h | 3 +- src/libvirt_private.syms | 8 +- src/lxc/lxc_driver.c | 6 + src/nwfilter/nwfilter_driver.c | 15 +- src/nwfilter/nwfilter_gentech_driver.c | 23 +- src/nwfilter/nwfilter_gentech_driver.h | 2 +- src/qemu/qemu_driver.c | 6 + src/uml/uml_driver.c | 4 + src/util/virthread.c | 291 +++++++++++++++++++++++- src/util/virthread.h | 51 ++++- src/util/virthreadpthread.c | 278 ----------------------- src/util/virthreadpthread.h | 49 ---- src/util/virthreadwin32.c | 396 --------------------------------- src/util/virthreadwin32.h | 53 ----- 17 files changed, 404 insertions(+), 823 deletions(-) delete mode 100644 src/util/virthreadpthread.c delete mode 100644 src/util/virthreadpthread.h delete mode 100644 src/util/virthreadwin32.c delete mode 100644 src/util/virthreadwin32.h -- 1.8.4.2

3 21

[libvirt] [RFC PATCHv2 0/4] Adding 'config' driver
by Adam Walters 29 Jan '14

29 Jan '14

This patchset adds a driver named 'config' which provides access to configuration data, such as secret and storage definitions, without requiring a connection to a hypervisor. This complements my previous patchset, which resolves the race condition on libvirtd startup. The rationale behind this idea is that there exist circumstances under which a driver may require access to things such as secrets during a time at which there is no active connection to a hypervisor. Currently, that data can not be accessed. An example of this in libvirt today is that the storage driver requires a connection in order to access secret data to auto-start an RBD storage pool that uses CephX authentication. My previous patchset breaks the ability to auto-start that type of storage pool, and this patchset fixes it again. This driver is technically what one may call a hypervisor driver, but it does not implement any domain operations. It simply exists to handle requests by drivers for access to information that would otherwise by unattainable. The URI provided by this driver is 'config:///' and has been tested working on four different machines of mine, running three different Linux distributions (Archlinux, Gentoo, and CentOS). Being a very simple driver, I would expect it to work pretty much anywhere. I welcome comments and suggestions related to this driver. At the very least, this patchset combined with my previous patchset resolves the current race condition present on startup of libvirtd without any loss of functionality. Adam Walters (4): config: Adding config driver configure: Implement config driver libvirtd: Reorder load of secrets driver storage: Change hardcoded QEMU connection to use config driver configure.ac | 11 ++ daemon/libvirtd.c | 21 ++-- include/libvirt/virterror.h | 2 + po/POTFILES.in | 1 + src/Makefile.am | 25 +++++ src/config/config_driver.c | 238 +++++++++++++++++++++++++++++++++++++++++++ src/config/config_driver.h | 44 ++++++++ src/storage/storage_driver.c | 13 +-- src/util/virerror.c | 2 + 9 files changed, 345 insertions(+), 12 deletions(-) create mode 100644 src/config/config_driver.c create mode 100644 src/config/config_driver.h -- 1.8.5.2

2 7

[libvirt] [PATCHv2 0/9] Snapshots on gluster
by Peter Krempa 29 Jan '14

29 Jan '14

Peter Krempa (9): storage: Add new argument for createVol backend API storage: gluster: Introduce dummy functions for creating a volume storage: Add internal API to create temporary storage pools and vols snapshot: Add support for specifying snapshot disk backing type snapshot: Test snapshot disk type specification storage: Implement ephemeral storage APIs in the generic storage driver storage: gluster: Support conversion of gluster volumes to temp volumes qemu: snapshot: Switch snapshot file deletion to the new storage API qemu: snapshot: Add support for external active snapshots on gluster docs/formatsnapshot.html.in | 15 + docs/hvsupport.pl | 3 + docs/schemas/domainsnapshot.rng | 76 +++- src/Makefile.am | 3 +- src/check-aclrules.pl | 3 + src/check-drivername.pl | 1 + src/conf/snapshot_conf.c | 42 ++- src/conf/snapshot_conf.h | 15 +- src/driver.h | 13 + src/libvirt_internal.c | 70 ++++ src/libvirt_internal.h | 22 ++ src/libvirt_private.syms | 3 + src/qemu/qemu_command.c | 2 +- src/qemu/qemu_command.h | 9 + src/qemu/qemu_conf.c | 3 - src/qemu/qemu_driver.c | 227 ++++++++++-- src/storage/storage_backend.h | 3 +- src/storage/storage_backend_disk.c | 3 +- src/storage/storage_backend_fs.c | 30 +- src/storage/storage_backend_gluster.c | 119 +++++- src/storage/storage_backend_logical.c | 6 +- src/storage/storage_backend_rbd.c | 3 +- src/storage/storage_backend_sheepdog.c | 3 +- src/storage/storage_driver.c | 405 ++++++++++++++++++++- tests/domainsnapshotxml2xmlin/disk_snapshot.xml | 18 + .../disk_driver_name_null.xml | 2 +- tests/domainsnapshotxml2xmlout/disk_snapshot.xml | 22 +- .../disk_snapshot_redefine.xml | 6 +- 28 files changed, 1007 insertions(+), 120 deletions(-) create mode 100644 src/libvirt_internal.c -- 1.8.5.2

2 22

[libvirt] [libvirt-glib 0/3] Fixes to glib/libvirt mainloop integration
by Christophe Fergeau 29 Jan '14

29 Jan '14

While reviewing danpb's win32 patch for the mainloop integration code, I found 2 more issues in the existing code (one leak, one bug). These patches fix this (first patch is just some cleanup). Christophe

2 6

[libvirt] [PATCH] Honor blacklist for modprobe command
by John Ferlan 29 Jan '14

29 Jan '14

https://bugzilla.redhat.com/show_bug.cgi?id=1045124 When loading modules, libvirt does not honor the modprobe blacklist. By adding a "-b" to the modprobe command libvirt will fail to load a module if it's on the blacklist Check if the failure to load a driver was due to it being on the blacklist using the output of a "modprobe -c" searching for "blacklist <driver_Name>" where driver_Name is possibly a modified string of the input driver changing all '-' into '_' since that's what modprobe does. Signed-off-by: John Ferlan <jferlan(a)redhat.com> --- src/util/virpci.c | 48 ++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 42 insertions(+), 6 deletions(-) diff --git a/src/util/virpci.c b/src/util/virpci.c index e2d222e..18b85f2 100644 --- a/src/util/virpci.c +++ b/src/util/virpci.c @@ -979,6 +979,10 @@ virPCIProbeStubDriver(const char *driver) { char *drvpath = NULL; bool probed = false; + size_t i; + char *drvblklst = NULL; + char *outbuf = NULL; + virCommandPtr cmd = NULL; recheck: if (virPCIDriverDir(&drvpath, driver) == 0 && virFileExists(drvpath)) { @@ -990,18 +994,49 @@ recheck: VIR_FREE(drvpath); if (!probed) { - const char *const probecmd[] = { MODPROBE, driver, NULL }; + cmd = virCommandNewArgList(MODPROBE, "-b", driver, NULL); probed = true; - if (virRun(probecmd, NULL) < 0) { + if (virCommandRun(cmd, NULL) < 0) { char ebuf[1024]; VIR_WARN("failed to load driver %s: %s", driver, virStrerror(errno, ebuf, sizeof(ebuf))); - return -1; + goto cleanup; } + virCommandFree(cmd); + cmd = NULL; goto recheck; } + /* All error path code - purpose is to determine whether the failure + * occurs because device is on blacklist in order to add an error + * message to help detect why load failed + */ + if (virAsprintfQuiet(&drvblklst, "blacklist %s", driver) < 0) + goto cleanup; + + /* Although driver may have dash, 'modprobe -c' adjusts to use underscore */ + for (i = 0; i < drvblklst[i]; i++) + if (drvblklst[i] == '-') + drvblklst[i] = '_'; + + cmd = virCommandNewArgList(MODPROBE, "-c", NULL); + virCommandSetOutputBuffer(cmd, &outbuf); + if (virCommandRun(cmd, NULL) < 0) + goto cleanup; + + /* Find driver on blacklist? */ + if (strstr(outbuf, drvblklst)) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("Failed to load PCI stub module %s: " + "administratively prohibited"), + driver); + } + +cleanup: + virCommandFree(cmd); + VIR_FREE(drvblklst); + VIR_FREE(outbuf); return -1; } @@ -1313,9 +1348,10 @@ virPCIDeviceDetach(virPCIDevicePtr dev, virPCIDeviceList *inactiveDevs) { if (virPCIProbeStubDriver(dev->stubDriver) < 0) { - virReportError(VIR_ERR_INTERNAL_ERROR, - _("Failed to load PCI stub module %s"), - dev->stubDriver); + if (virGetLastError() == NULL) + virReportError(VIR_ERR_INTERNAL_ERROR, + _("Failed to load PCI stub module %s"), + dev->stubDriver); return -1; } -- 1.8.4.2

2 3

[libvirt] [PATCHv2 0/6] Add BlkIO and CPU/mem stat API implementations for lxc
by Thorsten Behrens 29 Jan '14

29 Jan '14

This patch set adds block io, memory and domain cpu statistics API slot implementations to the LXC driver, in order to get linux container monitoring and accounting a bit closer to qemu standards. The last patch is a tad quirky (happy to hear suggestions on alternative ways), in that it widens the permissible value set at the .domainBlockStats slot: for lxc guests, it is relatively likely to have zero disk devices, since host filesystems can be used via passthrough bind mounts. Therefore, passing the zero-length string as device path, is interpreted as 'return summary stats for the entire domains's block io'. Thorsten Behrens (6): Add util virCgroupGetBlkioIo*Serviced methods. Implement domainMemoryStats API slot for LXC driver. Make qemuGetDomainTotalCPUStats a virCgroup function. Implement domainGetCPUStats for lxc driver. Implemet lxcDomainBlockStats for lxc driver Widening API change - accept empty path for virDomainBlockStats src/libvirt.c | 8 +- src/libvirt_private.syms | 3 + src/lxc/lxc_driver.c | 239 +++++++++++++++++++++++++++++++++++ src/qemu/qemu_driver.c | 54 +------- src/util/vircgroup.c | 291 +++++++++++++++++++++++++++++++++++++++++++ src/util/vircgroup.h | 17 +++ tools/virsh-domain-monitor.c | 11 +- tools/virsh.pod | 5 +- 8 files changed, 569 insertions(+), 59 deletions(-) -- 1.8.4

2 16

[libvirt] [PATCH v3 0/2] Enhance virnetdevbandwidthtest
by Michal Privoznik 29 Jan '14

29 Jan '14

yet another version - this time with virBuffer approach. Michal Privoznik (2): virCommand: Introduce virCommandSetDryRun virnetdevbandwidthtest: Introduce testVirNetDevBandwidthSet src/libvirt_private.syms | 1 + src/util/vircommand.c | 57 +++++++++++++++++++++++++++++-- src/util/vircommand.h | 2 ++ tests/virnetdevbandwidthtest.c | 76 ++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 134 insertions(+), 2 deletions(-) -- 1.8.5.2

2 4

Re: [libvirt] [Qemu-devel] [PATCH v4 4/5] qmp: full introspection support for QMP
by Amos Kong 29 Jan '14

29 Jan '14

CC Libvirt-list Original discussion: http://marc.info/?l=qemu-devel&m=139048842504757&w=2 [Qemu-devel] [PATCH v4 0/5] QMP full introspection On Fri, Jan 24, 2014 at 06:48:31PM +0800, Fam Zheng wrote: > On Thu, 01/23 22:46, Amos Kong wrote: > > This patch introduces a new monitor command to query QMP schema > > information, the return data is a range of schema structs, which > > contains the useful metadata to help management to check supported > > features, QMP commands detail, etc. > > > > We use qapi-introspect.py to parse all json definition in > > qapi-schema.json, and generate a range of dictionaries with metadata. > > The query command will visit the dictionaries and fill the data > > to allocated struct tree. Then QMP infrastructure will convert > > the tree to json string and return to QMP client. > > > > TODO: > > Wenchao Xia is working to convert QMP events to qapi-schema.json, > > then event can also be queried by this interface. > > > > I will introduce another command 'query-qga-schema' to query QGA > > schema information, it's easy to add this support based on this > > patch. > > > > Signed-off-by: Amos Kong <akong(a)redhat.com> > > --- > > qapi-schema.json | 11 +++ > > qmp-commands.hx | 42 +++++++++++ > > qmp.c | 215 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 268 insertions(+) > > > > diff --git a/qapi-schema.json b/qapi-schema.json > > index c63f0ca..6033383 100644 > > --- a/qapi-schema.json > > +++ b/qapi-schema.json > > @@ -4411,3 +4411,14 @@ > > 'reference-type': 'String', > > 'type': 'DataObjectType', > > 'unionobj': 'DataObjectUnion' } } > > + > > +## > > +# @query-qmp-schema > > +# > > +# Query QMP schema information > > +# > > +# @returns: list of @DataObject > > +# > > +# Since: 1.8 > > +## > > +{ 'command': 'query-qmp-schema', 'returns': ['DataObject'] } > > diff --git a/qmp-commands.hx b/qmp-commands.hx > > index 02cc815..b83762d 100644 > > --- a/qmp-commands.hx > > +++ b/qmp-commands.hx > > @@ -3291,6 +3291,48 @@ Example: > > } > > > > EQMP > > + { > > + .name = "query-qmp-schema", > > + .args_type = "", > > + .mhandler.cmd_new = qmp_marshal_input_query_qmp_schema, > > + }, > > + > > + > > +SQMP > > +query-qmp-schema > > +---------------- > > + > > +query qmp schema information > > + > > +Return a json-object with the following information: > > + > > +- "name": qmp schema name (json-string) > > +- "type": qmp schema type, it can be 'comand', 'type', 'enum', 'union' > > +- "returns": return data of qmp command (json-object, optional) > > + > > +Example: > > + > > +-> { "execute": "query-qmp-schema" } > > +-> { "return": [ > > + { > > + "name": "query-name", > > + "type": "command", > > + "returns": { > > + "name": "NameInfo", > > + "type": "type", > > + "data": [ > > + { > > + "name": "name", > > + "optional": true, > > + "recursive": false, > > + "type": "str" > > + } > > + ] > > + } > > + } > > + } > > + > > +EQMP > > > > { > > .name = "blockdev-add", > > diff --git a/qmp.c b/qmp.c > > index 0f46171..a64ae6d 100644 > > --- a/qmp.c > > +++ b/qmp.c > > @@ -27,6 +27,8 @@ > > #include "qapi/qmp/qobject.h" > > #include "qapi/qmp-input-visitor.h" > > #include "hw/boards.h" > > +#include "qapi/qmp/qjson.h" > > +#include "qapi-introspect.h" > > > > NameInfo *qmp_query_name(Error **errp) > > { > > @@ -488,6 +490,219 @@ CpuDefinitionInfoList *qmp_query_cpu_definitions(Error **errp) > > return arch_query_cpu_definitions(errp); > > } > > > > +static strList *qobject_to_strlist(QObject *data) > > +{ > > + strList *list = NULL; > > + strList **plist = &list; > > + QList *qlist; > > + const QListEntry *lent; > > + > > + qlist = qobject_to_qlist(data); > > + for (lent = qlist_first(qlist); lent; lent = qlist_next(lent)) { > > + strList *entry = g_malloc0(sizeof(strList)); > > + entry->value = g_strdup(qobject_get_str(lent->value)); > > + *plist = entry; > > + plist = &entry->next; > > + } > > + > > + return list; > > +} > > + > > +static DataObject *qobject_to_dataobj(QObject *data); > > + > > +static DataObjectMember *qobject_to_dataobjmem(QObject *data) > > +{ > > + > > + DataObjectMember *member = g_malloc0(sizeof(DataObjectMember)); > > + > > + member->type = g_malloc0(sizeof(DataObjectMemberType)); > > + if (data->type->code == QTYPE_QDICT) { > > + member->type->kind = DATA_OBJECT_MEMBER_TYPE_KIND_EXTEND; > > + member->type->extend = qobject_to_dataobj(data); > > + } else { > > + member->type->kind = DATA_OBJECT_MEMBER_TYPE_KIND_REFERENCE; > > + member->type->reference = g_strdup(qobject_get_str(data)); > > + } > > + > > + return member; > > +} > > + > > +static DataObjectMemberList *qobject_to_dict_memlist(QObject *data) > > +{ > > + DataObjectMemberList *list = NULL; > > + DataObjectMemberList **plist = &list; > > + QDict *qdict = qobject_to_qdict(data); > > + const QDictEntry *dent; > > + > > + for (dent = qdict_first(qdict); dent; dent = qdict_next(qdict, dent)) { > > + DataObjectMemberList *entry = g_malloc0(sizeof(DataObjectMemberList)); > > + entry->value = qobject_to_dataobjmem(dent->value); > > + > > + entry->value->has_optional = true; > > + entry->value->has_name = true; > > + if (dent->key[0] == '*') { > > + entry->value->optional = true; > > + entry->value->name = g_strdup(dent->key + 1); > > + } else { > > + entry->value->name = g_strdup(dent->key); > > + } > > + *plist = entry; > > + plist = &entry->next; > > + } > > + > > + return list; > > +} > > + > > +static DataObjectMemberList *qobject_to_list_memlist(QObject *data) > > +{ > > + const QListEntry *lent; > > + DataObjectMemberList *list = NULL; > > + DataObjectMemberList **plist = &list; > > + QList *qlist = qobject_to_qlist(data); > > + > > + for (lent = qlist_first(qlist); lent; lent = qlist_next(lent)) { > > + DataObjectMemberList *entry = g_malloc0(sizeof(DataObjectMemberList)); > > + entry->value = qobject_to_dataobjmem(lent->value); > > + entry->value->has_optional = true; > > + entry->value->has_name = true; > > + *plist = entry; > > + plist = &entry->next; > > + } > > + > > + return list; > > +} > > + > > +static DataObjectMemberList *qobject_to_memlist(QObject *data) > > This whole converting is cumbersome. You already did all the traversing through > the type jungle in python when generating this, it's not necessary to do the > similar thing again here. We can parse raw schemas and generate json string table, we can't directly return the string / qobject to monitor, C code has to convert the json to qobject, we have to revisit the qobject and convert them to DataObject/DataObjectMember/DataObject... structs. > Alternatively, I think we have a good reason to extend QMP framework as > necessary here, as we are doing "QMP introspection", which is a part of the > framework: > > * Define final output into qmp_schema_table[], no need to box it like: > > "{'_obj_member': 'False', '_obj_type': 'enum', '_obj_name': > 'ErrorClass', '_obj_data': {'data': ... > > just put it content of "qmp-introspection.output.txt" as a long string in > the header, > like you would generate in qobject_to_memlist: > > const char *qmp_schema_table = > "{ 'name': 'ErrorClass', 'type': 'enumeration', 'data': [...]}," > "{ 'name': '...', ...}," The keys are used for metadata might be 'recursive', 'optional', etc. It might exists problem in namespace, let's use '_obj_' or '_' prefix for the metadata keys. I used a nested dictionary to describe a DataObject, because we can store the metadata and definition to different level, it's helpful in parse the output by Libvirt. example: "{ 'type': 'NameInfo', 'data': {'*name': 'str', '*job': 'str'} }" It's good to store _type, _name, data to same level, but the metadata of items of data's value dictionary can't be appended to same level. "{ '_name': 'NameInfo', '_type': 'type', 'data': { 'name': 'str', '_name_optional': 'True', 'job': 'str', '_job_optional': 'True' } }" A better solution, but I don't know if it will cause trouble for Libvirt to parse the output. "{'_type': 'type', '_name': 'NameInfo', 'data': { 'job': {'_value': 'str', '_recursive': 'True'}, 'name': {'_value': 'str', '_recursive': 'True'} }, '_recursive': 'False' }" When we describe a DataObject (dict/list/str, one schema, extened schema, schema member, etc), so I we generally use a nested dictionary to describe a DataObject, it will split the metadata with original data two different dictionary level, it's convenient for parse of Libvirt. Here I just use a dict member as an example, actually it's more complex to parse all kinds of data objects. > ... > ; > > * Add a new type of qmp command, that returns a QString as a json literal. > query-qmp-schema is defined as this type. (This wouldn't be much code, but > may be abused in the future, I'm afraid. However we can review, limit its > use to introspection only) > > * And return qmp_schema_table from query-qmp-shema, which will be copied to > the wire. > > Feel free to disagree, it's not a perfect solution. But I really think we need > to avoid duplicating "enum", "base", "type", "union", "discriminator", ... In the past, we didn't consider to extend and add metadata by Python, so Libvirt wasn't happy to just get a raw schema(not extended, no metadata). But now, we already successfully to transfer this work to Python, and we can adjust to make management to be happy for the metadata and format. The problem is that Libvirt should parse the output twice, the json items are also json object string. Eric, Is it acceptabled? Example: * as suggested by Fam to put the metadta with schema data in same dict level * process some special cases by nested dictionary (eg: '*tls': 'bool' ==> 'tls': {'_value': 'bool', '_optional': 'True'} ) * return a strList to Libvirt, the content of string is json object, that contains the metadata. { "return": [ "{'_type': 'enum', '_name': 'ErrorClass', 'data': ['GenericError', 'CommandNotFound', 'DeviceEncrypted', 'DeviceNotActive', 'DeviceNotFound', 'KVMMissingCap'], '_recursive': 'False'}", "{'_type': 'command', '_name': 'add_client', 'data': {'tls': {'_value': 'bool', '_optional': 'True'}, 'skipauth': {'_value': 'bool', '_optional': 'True'}, 'protocol': 'str', 'fdname': 'str'}, '_recursive': 'False'}", "{'_type': 'type', '_name': 'NameInfo', 'data': {'job': {'_value': 'str', '_optional': 'True'}, 'name': {'_value': 'str', '_optional': 'True'}}, '_recursive': 'False'}", "{'returns': {'_type': 'type', '_recursive': 'False', 'data': {'job': {'_value': 'str', '_optional': 'True'}, 'name': {'_value': 'str', '_optional': 'True'}}, '_name': 'NameInfo'}, '_name': 'query-name', '_type': 'command', '_recursive': 'False'}", "......", "......", "......" ] } > Fam -- Amos.

4 9

[libvirt] libvirt 1.2.1 vs. Xen 4.1.4
by Guido Günther 28 Jan '14

28 Jan '14

Hi, a user of Debian's libvirt backport to wheezy using Xen reported libvirt 1.2.1 failing for him like: info : libvirt version: 1.2.1 error : virIdentityGetSystem:173 : Unable to lookup SELinux process context: Invalid argument error : xenHypervisorDoV2Sys:1011 : Unable to issue hypervisor ioctl 3166208: Permission denied error : xenHypervisorDoV2Sys:1011 : Unable to issue hypervisor ioctl 3166208: Permission denied error : xenHypervisorDoV2Sys:1011 : Unable to issue hypervisor ioctl 3166208: Permission denied error : xenHypervisorDoV2Sys:1011 : Unable to issue hypervisor ioctl 3166208: Permission denied error : xenHypervisorDoV2Sys:1011 : Unable to issue hypervisor ioctl 3166208: Permission denied error : xend_req:302 : internal error: failed to parse Xend response return code error : xend_req:302 : internal error: failed to parse Xend response return code error : virNetSocketReadWire:1454 : End of file while reading data: Input/output error It seems there weren't any significant changes in this code area recently so I wonder what might be triggering this? Since I currently don't have a xen test setup it'd be happy about any pointer. We don't have libxl enabled. https://bugzilla.redhat.com/show_bug.cgi?id=1055165 has a similar error message. Cheers, -- Guido

3 3

[libvirt] [PATCH RFC 1/2] support compression when --memory-only option is specified
by Qiao Nuohan 28 Jan '14

28 Jan '14

--memory-only option is introduced without compression supported. Therefore, this is a freature regression of virsh dump. This patch is used to add "--compress" and "[--compression-format] <string>" to the command "virsh dump --memory-only" and send dump-guest-memory command to qemu with dump format specified to one of elf, kdump-zlib, kdump-lzo and kdump-snappy. Signed-off-by: Qiao Nuohan <qiaonuohan(a)cn.fujitsu.com> --- include/libvirt/libvirt.h.in | 18 +++++++++++++----- src/libvirt.c | 15 +++++++++++++++ src/qemu/qemu_driver.c | 21 +++++++++++++++++---- src/qemu/qemu_monitor.c | 6 +++--- src/qemu/qemu_monitor.h | 3 ++- src/qemu/qemu_monitor_json.c | 4 +++- src/qemu/qemu_monitor_json.h | 3 ++- tests/qemumonitorjsontest.c | 2 +- tools/virsh-domain.c | 29 +++++++++++++++++++++++++++++ 9 files changed, 85 insertions(+), 16 deletions(-) diff --git a/include/libvirt/libvirt.h.in b/include/libvirt/libvirt.h.in index a448411..cb8989e 100644 --- a/include/libvirt/libvirt.h.in +++ b/include/libvirt/libvirt.h.in @@ -1165,11 +1165,19 @@ typedef virDomainMemoryStatStruct *virDomainMemoryStatPtr; /* Domain core dump flags. */ typedef enum { - VIR_DUMP_CRASH = (1 << 0), /* crash after dump */ - VIR_DUMP_LIVE = (1 << 1), /* live dump */ - VIR_DUMP_BYPASS_CACHE = (1 << 2), /* avoid file system cache pollution */ - VIR_DUMP_RESET = (1 << 3), /* reset domain after dump finishes */ - VIR_DUMP_MEMORY_ONLY = (1 << 4), /* use dump-guest-memory */ + VIR_DUMP_CRASH = (1 << 0), /* crash after dump */ + VIR_DUMP_LIVE = (1 << 1), /* live dump */ + VIR_DUMP_BYPASS_CACHE = (1 << 2), /* avoid file system cache pollution */ + VIR_DUMP_RESET = (1 << 3), /* reset domain after dump finishes */ + VIR_DUMP_MEMORY_ONLY = (1 << 4), /* use dump-guest-memory */ + VIR_DUMP_COMPRESS = (1 << 5), /* dump guest memory in + kdump-compressed format */ + VIR_DUMP_COMPRESS_ZLIB = (1 << 6), /* kdump-compressed format, with + zlib-compressed */ + VIR_DUMP_COMPRESS_LZO = (1 << 7), /* kdump-compressed format, with + lzo-compressed */ + VIR_DUMP_COMPRESS_SNAPPY = (1 << 8), /* kdump-compressed format, with + snappy-compressed */ } virDomainCoreDumpFlags; /* Domain migration flags. */ diff --git a/src/libvirt.c b/src/libvirt.c index 666526e..318296e 100644 --- a/src/libvirt.c +++ b/src/libvirt.c @@ -2969,6 +2969,21 @@ virDomainCoreDump(virDomainPtr domain, const char *to, unsigned int flags) goto error; } + if ((flags & VIR_DUMP_COMPRESS) && !(flags & VIR_DUMP_MEMORY_ONLY)) { + virReportInvalidArg(flags, "%s", + _("compress flag cannot be set without memory-only " + "flag")); + goto error; + } + + if ((flags & (VIR_DUMP_COMPRESS_ZLIB | VIR_DUMP_COMPRESS_LZO | + VIR_DUMP_COMPRESS_SNAPPY)) && !(flags & VIR_DUMP_COMPRESS)) { + virReportInvalidArg(flags, "%s", + _("compression-format cannot be set without " + "compress flag")); + goto error; + } + if (conn->driver->domainCoreDump) { int ret; char *absolute_to; diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index df4f5b5..c4e1aef 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -3380,7 +3380,8 @@ cleanup: } static int qemuDumpToFd(virQEMUDriverPtr driver, virDomainObjPtr vm, - int fd, enum qemuDomainAsyncJob asyncJob) + int fd, enum qemuDomainAsyncJob asyncJob, + const char* dump_format) { qemuDomainObjPrivatePtr priv = vm->privateData; int ret = -1; @@ -3400,7 +3401,7 @@ static int qemuDumpToFd(virQEMUDriverPtr driver, virDomainObjPtr vm, if (qemuDomainObjEnterMonitorAsync(driver, vm, asyncJob) < 0) return -1; - ret = qemuMonitorDumpToFd(priv->mon, fd); + ret = qemuMonitorDumpToFd(priv->mon, fd, dump_format); qemuDomainObjExitMonitor(driver, vm); return ret; @@ -3418,6 +3419,7 @@ doCoreDump(virQEMUDriverPtr driver, virFileWrapperFdPtr wrapperFd = NULL; int directFlag = 0; unsigned int flags = VIR_FILE_WRAPPER_NON_BLOCKING; + const char *dump_format; /* Create an empty file with appropriate ownership. */ if (dump_flags & VIR_DUMP_BYPASS_CACHE) { @@ -3441,7 +3443,16 @@ doCoreDump(virQEMUDriverPtr driver, goto cleanup; if (dump_flags & VIR_DUMP_MEMORY_ONLY) { - ret = qemuDumpToFd(driver, vm, fd, QEMU_ASYNC_JOB_DUMP); + if (dump_flags & VIR_DUMP_COMPRESS_ZLIB) + dump_format = "kdump-zlib"; + else if (dump_flags & VIR_DUMP_COMPRESS_LZO) + dump_format = "kdump-lzo"; + else if (dump_flags & VIR_DUMP_COMPRESS_SNAPPY) + dump_format = "kdump-snappy"; + else + dump_format = "elf"; + ret = qemuDumpToFd(driver, vm, fd, QEMU_ASYNC_JOB_DUMP, + dump_format); } else { ret = qemuMigrationToFile(driver, vm, fd, 0, path, qemuCompressProgramName(compress), false, @@ -3517,7 +3528,9 @@ static int qemuDomainCoreDump(virDomainPtr dom, virCheckFlags(VIR_DUMP_LIVE | VIR_DUMP_CRASH | VIR_DUMP_BYPASS_CACHE | VIR_DUMP_RESET | - VIR_DUMP_MEMORY_ONLY, -1); + VIR_DUMP_MEMORY_ONLY | VIR_DUMP_COMPRESS | + VIR_DUMP_COMPRESS_ZLIB | VIR_DUMP_COMPRESS_LZO | + VIR_DUMP_COMPRESS_SNAPPY, -1); if (!(vm = qemuDomObjFromDomain(dom))) return -1; diff --git a/src/qemu/qemu_monitor.c b/src/qemu/qemu_monitor.c index a968901..c2c4a1e 100644 --- a/src/qemu/qemu_monitor.c +++ b/src/qemu/qemu_monitor.c @@ -2338,10 +2338,10 @@ int qemuMonitorMigrateCancel(qemuMonitorPtr mon) } int -qemuMonitorDumpToFd(qemuMonitorPtr mon, int fd) +qemuMonitorDumpToFd(qemuMonitorPtr mon, int fd, const char *dump_format) { int ret; - VIR_DEBUG("mon=%p fd=%d", mon, fd); + VIR_DEBUG("mon=%p fd=%d dump_format=%s", mon, fd, dump_format); if (!mon) { virReportError(VIR_ERR_INVALID_ARG, "%s", @@ -2361,7 +2361,7 @@ qemuMonitorDumpToFd(qemuMonitorPtr mon, int fd) if (qemuMonitorSendFileHandle(mon, "dump", fd) < 0) return -1; - ret = qemuMonitorJSONDump(mon, "fd:dump"); + ret = qemuMonitorJSONDump(mon, "fd:dump", dump_format); if (ret < 0) { if (qemuMonitorCloseFileHandle(mon, "dump") < 0) diff --git a/src/qemu/qemu_monitor.h b/src/qemu/qemu_monitor.h index eabf000..f2e5763 100644 --- a/src/qemu/qemu_monitor.h +++ b/src/qemu/qemu_monitor.h @@ -495,7 +495,8 @@ int qemuMonitorMigrateToUnix(qemuMonitorPtr mon, int qemuMonitorMigrateCancel(qemuMonitorPtr mon); int qemuMonitorDumpToFd(qemuMonitorPtr mon, - int fd); + int fd, + const char *dump_format); int qemuMonitorGraphicsRelocate(qemuMonitorPtr mon, int type, diff --git a/src/qemu/qemu_monitor_json.c b/src/qemu/qemu_monitor_json.c index ec3b958..a16ef44 100644 --- a/src/qemu/qemu_monitor_json.c +++ b/src/qemu/qemu_monitor_json.c @@ -2636,7 +2636,8 @@ int qemuMonitorJSONMigrateCancel(qemuMonitorPtr mon) int qemuMonitorJSONDump(qemuMonitorPtr mon, - const char *protocol) + const char *protocol, + const char *dump_format) { int ret; virJSONValuePtr cmd = NULL; @@ -2645,6 +2646,7 @@ qemuMonitorJSONDump(qemuMonitorPtr mon, cmd = qemuMonitorJSONMakeCommand("dump-guest-memory", "b:paging", false, "s:protocol", protocol, + "s:format", dump_format, NULL); if (!cmd) return -1; diff --git a/src/qemu/qemu_monitor_json.h b/src/qemu/qemu_monitor_json.h index a93c51e..7691356 100644 --- a/src/qemu/qemu_monitor_json.h +++ b/src/qemu/qemu_monitor_json.h @@ -148,7 +148,8 @@ int qemuMonitorJSONGetSpiceMigrationStatus(qemuMonitorPtr mon, int qemuMonitorJSONMigrateCancel(qemuMonitorPtr mon); int qemuMonitorJSONDump(qemuMonitorPtr mon, - const char *protocol); + const char *protocol, + const char *dump_format); int qemuMonitorJSONGraphicsRelocate(qemuMonitorPtr mon, int type, diff --git a/tests/qemumonitorjsontest.c b/tests/qemumonitorjsontest.c index 2152e4a..256655a 100644 --- a/tests/qemumonitorjsontest.c +++ b/tests/qemumonitorjsontest.c @@ -1154,7 +1154,7 @@ GEN_TEST_FUNC(qemuMonitorJSONSetMigrationDowntime, 1) GEN_TEST_FUNC(qemuMonitorJSONMigrate, QEMU_MONITOR_MIGRATE_BACKGROUND | QEMU_MONITOR_MIGRATE_NON_SHARED_DISK | QEMU_MONITOR_MIGRATE_NON_SHARED_INC, "tcp:localhost:12345") -GEN_TEST_FUNC(qemuMonitorJSONDump, "dummy_protocol") +GEN_TEST_FUNC(qemuMonitorJSONDump, "dummy_protocol", "dummy_dump_format") GEN_TEST_FUNC(qemuMonitorJSONGraphicsRelocate, VIR_DOMAIN_GRAPHICS_TYPE_SPICE, "localhost", 12345, 12346, NULL) GEN_TEST_FUNC(qemuMonitorJSONAddNetdev, "some_dummy_netdevstr") diff --git a/tools/virsh-domain.c b/tools/virsh-domain.c index b4ca4c3..652c7ec 100644 --- a/tools/virsh-domain.c +++ b/tools/virsh-domain.c @@ -4519,6 +4519,14 @@ static const vshCmdOptDef opts_dump[] = { .type = VSH_OT_BOOL, .help = N_("dump domain's memory only") }, + {.name = "compress", + .type = VSH_OT_BOOL, + .help = N_("dump domain's memory in kdump-compressed format") + }, + {.name = "compression-format", + .type = VSH_OT_DATA, + .help = N_("specify the compression format of kdump-compressed format") + }, {.name = NULL} }; @@ -4533,6 +4541,7 @@ doDump(void *opaque) sigset_t sigmask, oldsigmask; const char *name = NULL; const char *to = NULL; + const char *compression_format = NULL; unsigned int flags = 0; sigemptyset(&sigmask); @@ -4556,6 +4565,26 @@ doDump(void *opaque) flags |= VIR_DUMP_RESET; if (vshCommandOptBool(cmd, "memory-only")) flags |= VIR_DUMP_MEMORY_ONLY; + if (vshCommandOptBool(cmd, "compress")) + flags |= VIR_DUMP_COMPRESS; + + if (vshCommandOptString(cmd, "compression-format", &compression_format)) { + if (STREQ(compression_format, "zlib")) + flags |= VIR_DUMP_COMPRESS_ZLIB; + else if (STREQ(compression_format, "lzo")) + flags |= VIR_DUMP_COMPRESS_LZO; + else if (STREQ(compression_format, "snappy")) + flags |= VIR_DUMP_COMPRESS_SNAPPY; + else { + vshError(ctl, _("compression format %s is not supported, " + "expecting 'zlib', 'lzo' or 'snappy'."), + compression_format); + goto out; + } + } else { + if (flags & VIR_DUMP_COMPRESS) + flags |= VIR_DUMP_COMPRESS_ZLIB; + } if (virDomainCoreDump(dom, to, flags) < 0) { vshError(ctl, _("Failed to core dump domain %s to %s"), name, to); -- 1.8.3.1

1 0