On Thu, Feb 05, 2015 at 02:25:41PM -0700, Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

Otherwise domainCreateWithFlags via remote returns -1 as domid.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_driver.c | 3 +++ 1 file changed, 3 insertions(+)

diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c index c95b387..ce3a99b 100644 --- a/src/libxl/libxl_driver.c +++ b/src/libxl/libxl_driver.c @@ -2375,6 +2375,9 @@ libxlDomainCreateWithFlags(virDomainPtr dom, }

ret = libxlDomainStart(driver, vm, (flags & VIR_DOMAIN_START_PAUSED) != 0, -1); + if (ret < 0) + goto cleanup; + dom->id = vm->def->id;

It looks at though this is the only callsite of libxlDomainStart that needs this change. Have you noticed this with other APIs that call libxlDomainStart?

Not directly. But there is a problem if domain gets restarted by some other client - the first one will still use old domid. IIUC for now the only workaround is to listen to lifecycle events for every domain. -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?

Marek Marczykowski-Górecki wrote:

At least Xen supports backend drivers in another domain (aka "driver domain"). This patch introduces XML config option for such setting as 'domain' attribute of 'source' element. Verification its content is left for the driver.

In the future same option will be needed for USB devices (hostdev objects), but for now libxl doesn't have support for PVUSB.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - describe in docs/formatdomain.html.in - enforce empty domain tag (only 'name' attribute allowed)

Changes in v3: - change <domain name='xx'/> element to domain='' attribute of source element - this is more logical place - grammar of docs

docs/formatdomain.html.in | 29 +++++++++++++++++++++++++++-- docs/schemas/domaincommon.rng | 20 ++++++++++++++++++++ src/conf/domain_conf.c | 33 +++++++++++++++++++++++++++++---- src/conf/domain_conf.h | 1 + src/util/virstoragefile.c | 4 +++- src/util/virstoragefile.h | 1 + 6 files changed, 81 insertions(+), 7 deletions(-)

diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in index c5ad6f4..38c42d5 100644 --- a/docs/formatdomain.html.in +++ b/docs/formatdomain.html.in @@ -1907,13 +1907,17 @@ <span class="since">since 0.0.3</span></dt> <dd> The <code>file</code> attribute specifies the fully-qualified - path to the file holding the disk. + path to the file holding the disk. The optional + <code>domain</code> attribute allows specifying a backend domain

Interesting approach. It almost feels like this is a different type of disk, e.g. <disk type='brokered'> :). There's a fair bit of experience on the list wrt extending the XML. Let's see if other have suggestions. Regards, Jim

+ (aka driver domain) where the file is stored. </dd> <dt><code>type='block'</code> <span class="since">since 0.0.3</span></dt> <dd> The <code>dev</code> attribute specifies the path to the - host device to serve as the disk. + host device to serve as the disk. The optional + <code>domain</code> attribute allows specifying a backend domain + (aka driver domain) for the device. </dd> <dt><code>type='dir'</code> <span class="since">since 0.7.5</span></dt> @@ -4336,6 +4340,27 @@ qemu-kvm -net nic,model=? /dev/null element is unspecified is to have the link state <code>up</code>. <span class="since">Since 0.9.5</span> </p> + <h5><a name="elementDomain">Setting up a network backend in a driver domain</a></h5> +<pre> + ... + <devices> + ... + <interface type='bridge'> + <source bridge='br0' <b>domain='netvm'</b>/> + </interface> + ... + </devices> + ...</pre> + + <p> + The optional <code>domain</code> attribute allows specifying a backend + domain (aka driver domain) for the device. Use the <code>name</code> attribute + to specify its name. You can use it to create a direct network link between + domains (so data will not go through host system). Use with type 'ethernet' + to create plain network link, or with 'bridge' to connect to some bridge + inside the driver domain. + <span class="since">Since 1.0.7 (Xen only)</span> + </p>

<h5><a name="ipconfig">IP configuration</a></h5> <pre> diff --git a/docs/schemas/domaincommon.rng b/docs/schemas/domaincommon.rng index d467dce..a4321f1 100644 --- a/docs/schemas/domaincommon.rng +++ b/docs/schemas/domaincommon.rng @@ -1340,6 +1340,11 @@ <attribute name="file"> <ref name="absFilePath"/> </attribute> + <optional> + <attribute name="domain"> + <ref name="domainName"/> + </attribute> + </optional> </optional> <optional> <ref name="storageStartupPolicy"/> @@ -1365,6 +1370,11 @@ </attribute> </optional> <optional> + <attribute name="domain"> + <ref name="domainName"/> + </attribute> + </optional> + <optional> <ref name="storageStartupPolicy"/> </optional> <optional> @@ -2040,6 +2050,11 @@ <attribute name="bridge"> <ref name="deviceName"/> </attribute> + <optional> + <attribute name="domain"> + <ref name="domainName"/> + </attribute> + </optional> <empty/> </element> </optional> @@ -2059,6 +2074,11 @@ <attribute name="dev"> <ref name="deviceName"/> </attribute> + <optional> + <attribute name="domain"> + <ref name="domainName"/> + </attribute> + </optional> <empty/> </element> </optional> diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c index 4251b13..17b699a 100644 --- a/src/conf/domain_conf.c +++ b/src/conf/domain_conf.c @@ -1442,6 +1442,7 @@ void virDomainNetDefFree(virDomainNetDefPtr def) VIR_FREE(def->backend.vhost); VIR_FREE(def->virtPortProfile); VIR_FREE(def->script); + VIR_FREE(def->domain_name); VIR_FREE(def->ifname); VIR_FREE(def->ifname_guest); VIR_FREE(def->ifname_guest_actual); @@ -5520,9 +5521,11 @@ virDomainDiskSourceParse(xmlNodePtr node, switch ((virStorageType)src->type) { case VIR_STORAGE_TYPE_FILE: src->path = virXMLPropString(node, "file"); + src->domain_name = virXMLPropString(node, "domain"); break; case VIR_STORAGE_TYPE_BLOCK: src->path = virXMLPropString(node, "dev"); + src->domain_name = virXMLPropString(node, "domain"); break; case VIR_STORAGE_TYPE_DIR: src->path = virXMLPropString(node, "dir"); @@ -7369,6 +7372,7 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, char *vhostuser_path = NULL; char *vhostuser_type = NULL; char *trustGuestRxFilters = NULL; + char *domain_name = NULL; virNWFilterHashTablePtr filterparams = NULL; virDomainActualNetDefPtr actual = NULL; xmlNodePtr oldnode = ctxt->node; @@ -7423,12 +7427,14 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, def->type == VIR_DOMAIN_NET_TYPE_BRIDGE && xmlStrEqual(cur->name, BAD_CAST "source")) { bridge = virXMLPropString(cur, "bridge"); + domain_name = virXMLPropString(cur, "domain"); } else if (!dev && (def->type == VIR_DOMAIN_NET_TYPE_ETHERNET || def->type == VIR_DOMAIN_NET_TYPE_DIRECT) && xmlStrEqual(cur->name, BAD_CAST "source")) { dev = virXMLPropString(cur, "dev"); mode = virXMLPropString(cur, "mode"); + domain_name = virXMLPropString(cur, "domain"); } else if (!vhostuser_path && !vhostuser_mode && !vhostuser_type && def->type == VIR_DOMAIN_NET_TYPE_VHOSTUSER && xmlStrEqual(cur->name, BAD_CAST "source")) { @@ -7802,6 +7808,10 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, def->script = script; script = NULL; } + if (domain_name != NULL) { + def->domain_name = domain_name; + domain_name = NULL; + } if (ifname != NULL) { def->ifname = ifname; ifname = NULL; @@ -8061,6 +8071,7 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, VIR_FREE(addrtype); VIR_FREE(trustGuestRxFilters); VIR_FREE(ips); + VIR_FREE(domain_name); virNWFilterHashTableFree(filterparams);

return def; @@ -16544,6 +16555,7 @@ virDomainDiskSourceFormatInternal(virBufferPtr buf, virBufferAddLit(buf, "<source"); virBufferEscapeString(buf, " file='%s'", src->path); virBufferEscapeString(buf, " startupPolicy='%s'", startupPolicy); + virBufferEscapeString(buf, " domain='%s'", src->domain_name);

virDomainDiskSourceDefFormatSeclabel(buf, src->nseclabels, src->seclabels, flags, @@ -16554,6 +16566,7 @@ virDomainDiskSourceFormatInternal(virBufferPtr buf, virBufferAddLit(buf, "<source"); virBufferEscapeString(buf, " dev='%s'", src->path); virBufferEscapeString(buf, " startupPolicy='%s'", startupPolicy); + virBufferEscapeString(buf, " domain='%s'", src->domain_name);

virDomainDiskSourceDefFormatSeclabel(buf, src->nseclabels, src->seclabels, flags, @@ -17713,8 +17726,14 @@ virDomainNetDefFormat(virBufferPtr buf, break;

case VIR_DOMAIN_NET_TYPE_ETHERNET: - virBufferEscapeString(buf, "<source dev='%s'/>\n", - def->data.ethernet.dev); + if (def->data.ethernet.dev || def->domain_name) { + virBufferAddLit(buf, "<source"); + virBufferEscapeString(buf, " dev='%s'", + def->data.ethernet.dev); + virBufferEscapeString(buf, " domain='%s'", + def->domain_name); + virBufferAddLit(buf, "/>\n"); + } break;

case VIR_DOMAIN_NET_TYPE_VHOSTUSER: @@ -17730,8 +17749,14 @@ virDomainNetDefFormat(virBufferPtr buf, break;

case VIR_DOMAIN_NET_TYPE_BRIDGE: - virBufferEscapeString(buf, "<source bridge='%s'/>\n", - def->data.bridge.brname); + if (def->data.bridge.brname || def->domain_name) { + virBufferAddLit(buf, "<source"); + virBufferEscapeString(buf, " bridge='%s'", + def->data.bridge.brname); + virBufferEscapeString(buf, " domain='%s'", + def->domain_name); + virBufferAddLit(buf, "/>\n"); + } break;

case VIR_DOMAIN_NET_TYPE_SERVER: diff --git a/src/conf/domain_conf.h b/src/conf/domain_conf.h index 93f2314..28c6920 100644 --- a/src/conf/domain_conf.h +++ b/src/conf/domain_conf.h @@ -995,6 +995,7 @@ struct _virDomainNetDef { unsigned long sndbuf; } tune; char *script; + char *domain_name; /* backend domain name */ char *ifname; char *ifname_guest; char *ifname_guest_actual; diff --git a/src/util/virstoragefile.c b/src/util/virstoragefile.c index 8d3d1f5..c2703b7 100644 --- a/src/util/virstoragefile.c +++ b/src/util/virstoragefile.c @@ -1835,7 +1835,8 @@ virStorageSourceCopy(const virStorageSource *src, VIR_STRDUP(ret->backingStoreRaw, src->backingStoreRaw) < 0 || VIR_STRDUP(ret->snapshot, src->snapshot) < 0 || VIR_STRDUP(ret->configFile, src->configFile) < 0 || - VIR_STRDUP(ret->compat, src->compat) < 0) + VIR_STRDUP(ret->compat, src->compat) < 0 || + VIR_STRDUP(ret->domain_name, src->domain_name)) goto error;

if (src->nhosts) { @@ -2023,6 +2024,7 @@ virStorageSourceClear(virStorageSourcePtr def)

VIR_FREE(def->path); VIR_FREE(def->volume); + VIR_FREE(def->domain_name); virStorageSourcePoolDefFree(def->srcpool); VIR_FREE(def->driverName); virBitmapFree(def->features); diff --git a/src/util/virstoragefile.h b/src/util/virstoragefile.h index b4c3808..a3c7e1e 100644 --- a/src/util/virstoragefile.h +++ b/src/util/virstoragefile.h @@ -241,6 +241,7 @@ struct _virStorageSource { char *snapshot; /* for storage systems supporting internal snapshots */ char *configFile; /* some storage systems use config file as part of the source definition */ + char *domain_name; /* name of domain holding backing storage file/dev */ size_t nhosts; virStorageNetHostDefPtr hosts; virStorageSourcePoolDefPtr srcpool;

From: Marek Marczykowski <marmarek@invisiblethingslab.com> This implement handling of <backenddomain name=''/> parameter introduced in previous patch. Works on Xen >= 4.3, because only there libxl supports setting backend domain by name, not XID. Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - rebase on 1.0.6+ - fix indentation - make libxl_name_to_domid switch more defensive Changes in v3: - rebase on 1.2.12+ - leave name->XID resolution to libxl itself, which greatly simplify the code (but requires Xen at least 4.3) Changes in v4: - change back to def->domain_name in disk handling code src/libxl/libxl_conf.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 0555b91..5cf3d8f 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -971,6 +971,18 @@ libxlMakeDisk(virDomainDiskDefPtr l_disk, libxl_device_disk *x_disk) return -1; } + if (l_disk->domain_name) { +#ifdef LIBXL_HAVE_DEVICE_BACKEND_DOMNAME + if (VIR_STRDUP(x_disk->backend_domname, l_disk->src->domain_name) < 0) + return -1; +#else + virReportError(VIR_ERR_XML_DETAIL, "%s", + _("this version of libxenlight does not " + "support backend domain name")); + return -1; +#endif + } + return 0; } @@ -1098,6 +1110,18 @@ libxlMakeNic(virDomainDefPtr def, return -1; } + if (l_nic->domain_name) { +#ifdef LIBXL_HAVE_DEVICE_BACKEND_DOMNAME + if (VIR_STRDUP(x_nic->backend_domname, l_nic->domain_name) < 0) + return -1; +#else + virReportError(VIR_ERR_XML_DETAIL, "%s", + _("this version of libxenlight does not " + "support backend domain name")); + return -1; +#endif + } + return 0; } -- 1.8.3.1

Marek Marczykowski-Górecki wrote:

From: Marek Marczykowski <marmarek@invisiblethingslab.com>

At least Xen supports backend drivers in another domain (aka "driver domain"). This patch introduces XML config option for such setting as 'domain' attribute of 'source' element. Verification its content is left for the driver.

In the future same option will be needed for USB devices (hostdev objects), but for now libxl doesn't have support for PVUSB.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - describe in docs/formatdomain.html.in - enforce empty domain tag (only 'name' attribute allowed)

Changes in v3: - change <domain name='xx'/> element to domain='' attribute of source element - this is more logical place - grammar of docs

Changes in v4: - revert back to separate element, named <backenddomain name='xx'/>

docs/formatdomain.html.in | 29 +++++++++++++++++++++++++++++ docs/schemas/domaincommon.rng | 16 ++++++++++++++++ src/conf/domain_conf.c | 21 +++++++++++++++++++++ src/conf/domain_conf.h | 2 ++ 4 files changed, 68 insertions(+)

diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in index c5ad6f4..16d8b5c 100644 --- a/docs/formatdomain.html.in +++ b/docs/formatdomain.html.in @@ -2357,6 +2357,13 @@ </li> </ul> </dd> + <dt><code>backenddomain</code></dt> + <dd>The optional <code>backenddomain</code> element allows specifying a backend + domain (aka driver domain) for the device. If the real device/file resides + in some other domain on the same host, use the <code>name</code> + attribute to specify its name.

I shortened the long lines and changed the text a bit.

+ <span class="since">Since 1.2.13 (Xen only)</span> + </dd> <dt><code>boot</code></dt> <dd>Specifies that the disk is bootable. The <code>order</code> attribute determines the order in which devices will be tried during @@ -4235,6 +4242,28 @@ qemu-kvm -net nic,model=? /dev/null network device. <span class="since">Since 0.9.10 (QEMU and KVM only)</span>. </p> + <h5><a name="elementDomain">Setting up a network backend in a driver domain</a></h5> +<pre> + ... + <devices> + ... + <interface type='bridge'> + <source bridge='br0'/> + <b><backenddomain name='netvm'/></b> + </interface> + ... + </devices> + ...</pre> + + <p> + The optional <code>backenddomain</code> element allows specifying a backend + domain (aka driver domain) for the device. Use the <code>name</code> attribute + to specify its name. You can use it to create a direct network link between + domains (so data will not go through host system). Use with type 'ethernet' + to create plain network link, or with 'bridge' to connect to some bridge + inside the driver domain.

Same here.

+ <span class="since">Since 1.2.13 (Xen only)</span> + </p>

<h5><a name="elementQoS">Quality of service</a></h5>

diff --git a/docs/schemas/domaincommon.rng b/docs/schemas/domaincommon.rng index d467dce..6721431 100644 --- a/docs/schemas/domaincommon.rng +++ b/docs/schemas/domaincommon.rng @@ -1127,6 +1127,14 @@ </optional> <ref name="target"/> <optional> + <element name="backenddomain"> + <attribute name="name"> + <ref name="domainName"/> + </attribute> + <empty/> + </element> + </optional> + <optional>

I moved this below "devcieBoot", since it seemed to fit there better :-).

<ref name="deviceBoot"/> </optional> <optional> @@ -2328,6 +2336,14 @@ </element> </optional> <optional> + <element name="backenddomain"> + <attribute name="name"> + <ref name="domainName"/> + </attribute> + <empty/> + </element> + </optional> + <optional> <element name="model"> <attribute name="type"> <data type="string"> diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c index 4251b13..657b71f 100644 --- a/src/conf/domain_conf.c +++ b/src/conf/domain_conf.c @@ -1260,6 +1260,7 @@ virDomainDiskDefFree(virDomainDiskDefPtr def) VIR_FREE(def->wwn); VIR_FREE(def->vendor); VIR_FREE(def->product); + VIR_FREE(def->domain_name); virDomainDeviceInfoClear(&def->info);

VIR_FREE(def); @@ -1442,6 +1443,7 @@ void virDomainNetDefFree(virDomainNetDefPtr def) VIR_FREE(def->backend.vhost); VIR_FREE(def->virtPortProfile); VIR_FREE(def->script); + VIR_FREE(def->domain_name); VIR_FREE(def->ifname); VIR_FREE(def->ifname_guest); VIR_FREE(def->ifname_guest_actual); @@ -5725,6 +5727,7 @@ virDomainDiskDefParseXML(virDomainXMLOptionPtr xmlopt, char *discard = NULL; char *mirrorFormat = NULL; char *mirrorType = NULL; + char *domain_name = NULL; int expected_secret_usage = -1; int auth_secret_usage = -1; int ret = 0; @@ -5790,6 +5793,9 @@ virDomainDiskDefParseXML(virDomainXMLOptionPtr xmlopt, if (target && STRPREFIX(target, "ioemu:")) memmove(target, target+6, strlen(target)-5); + } else if (!domain_name && + xmlStrEqual(cur->name, BAD_CAST "backenddomain")) { + domain_name = virXMLPropString(cur, "name"); } else if (xmlStrEqual(cur->name, BAD_CAST "geometry")) { if (virXPathUInt("string(./geometry/@cyls)", ctxt, &def->geometry.cylinders) < 0) { @@ -6514,6 +6520,8 @@ virDomainDiskDefParseXML(virDomainXMLOptionPtr xmlopt, driverName = NULL; def->src->encryption = encryption; encryption = NULL; + def->domain_name = domain_name; + domain_name = NULL; def->serial = serial; serial = NULL; def->wwn = wwn; @@ -6576,6 +6584,7 @@ virDomainDiskDefParseXML(virDomainXMLOptionPtr xmlopt, VIR_FREE(product); VIR_FREE(mirrorType); VIR_FREE(mirrorFormat); + VIR_FREE(domain_name);

ctxt->node = save_ctxt; return def; @@ -7365,6 +7374,7 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, char *mode = NULL; char *linkstate = NULL; char *addrtype = NULL; + char *domain_name = NULL; char *vhostuser_mode = NULL; char *vhostuser_path = NULL; char *vhostuser_type = NULL; @@ -7503,6 +7513,9 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, } else if (!script && xmlStrEqual(cur->name, BAD_CAST "script")) { script = virXMLPropString(cur, "path"); + } else if (!domain_name && + xmlStrEqual(cur->name, BAD_CAST "backenddomain")) { + domain_name = virXMLPropString(cur, "name"); } else if (xmlStrEqual(cur->name, BAD_CAST "model")) { model = virXMLPropString(cur, "type"); } else if (xmlStrEqual(cur->name, BAD_CAST "driver")) { @@ -7802,6 +7815,10 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, def->script = script; script = NULL; } + if (domain_name != NULL) { + def->domain_name = domain_name; + domain_name = NULL; + } if (ifname != NULL) { def->ifname = ifname; ifname = NULL; @@ -8059,6 +8076,7 @@ virDomainNetDefParseXML(virDomainXMLOptionPtr xmlopt, VIR_FREE(mode); VIR_FREE(linkstate); VIR_FREE(addrtype); + VIR_FREE(domain_name); VIR_FREE(trustGuestRxFilters); VIR_FREE(ips); virNWFilterHashTableFree(filterparams); @@ -16820,6 +16838,8 @@ virDomainDiskDefFormat(virBufferPtr buf, def->src->backingStoreRaw, 1) < 0) return -1;

+ virBufferEscapeString(buf, "<backenddomain name='%s'/>\n", def->domain_name); + virDomainDiskGeometryDefFormat(buf, def); virDomainDiskBlockIoDefFormat(buf, def);

@@ -17786,6 +17806,7 @@ virDomainNetDefFormat(virBufferPtr buf,

virBufferEscapeString(buf, "<script path='%s'/>\n", def->script); + virBufferEscapeString(buf, "<backenddomain name='%s'/>\n", def->domain_name); if (def->ifname && !((flags & VIR_DOMAIN_DEF_FORMAT_INACTIVE) && (STRPREFIX(def->ifname, VIR_NET_GENERATED_PREFIX)))) { diff --git a/src/conf/domain_conf.h b/src/conf/domain_conf.h index 93f2314..b2ef8db 100644 --- a/src/conf/domain_conf.h +++ b/src/conf/domain_conf.h @@ -700,6 +700,7 @@ struct _virDomainDiskDef { int sgio; /* enum virDomainDeviceSGIO */ int discard; /* enum virDomainDiskDiscard */ unsigned int iothread; /* unused = 0, > 0 specific thread # */ + char *domain_name; /* backend domain name */ };

@@ -995,6 +996,7 @@ struct _virDomainNetDef { unsigned long sndbuf; } tune; char *script; + char *domain_name; /* backend domain name */ char *ifname; char *ifname_guest; char *ifname_guest_actual;

ACK. I'll squash in the below diff before pushing. Regards, Jim diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in index eb6a412..fb0a0d1 100644 --- a/docs/formatdomain.html.in +++ b/docs/formatdomain.html.in @@ -2373,11 +2373,10 @@ </ul> </dd> <dt><code>backenddomain</code></dt> - <dd>The optional <code>backenddomain</code> element allows specifying a backend - domain (aka driver domain) for the device. If the real device/file resides - in some other domain on the same host, use the <code>name</code> - attribute to specify its name. - <span class="since">Since 1.2.13 (Xen only)</span> + <dd>The optional <code>backenddomain</code> element allows specifying a + backend domain (aka driver domain) hosting the disk. Use the + <code>name</code> attribute to specify the backend domain name. + <span class="since">Since 1.2.13 (Xen only)</span> </dd> <dt><code>boot</code></dt> <dd>Specifies that the disk is bootable. The <code>order</code> @@ -4277,12 +4276,13 @@ qemu-kvm -net nic,model=? /dev/null ...</pre> <p> - The optional <code>backenddomain</code> element allows specifying a backend - domain (aka driver domain) for the device. Use the <code>name</code> attribute - to specify its name. You can use it to create a direct network link between - domains (so data will not go through host system). Use with type 'ethernet' - to create plain network link, or with 'bridge' to connect to some bridge - inside the driver domain. + The optional <code>backenddomain</code> element allows specifying a + backend domain (aka driver domain) for the interface. Use the + <code>name</code> attribute to specify the backend domain name. You + can use it to create a direct network link between domains (so data + will not go through host system). Use with type 'ethernet' to create + plain network link, or with type 'bridge' to connect to a bridge inside + the backend domain. <span class="since">Since 1.2.13 (Xen only)</span> </p> diff --git a/docs/schemas/domaincommon.rng b/docs/schemas/domaincommon.rng index 5476f5d..f41ca43 100644 --- a/docs/schemas/domaincommon.rng +++ b/docs/schemas/domaincommon.rng @@ -1171,6 +1171,9 @@ </optional> <ref name="target"/> <optional> + <ref name="deviceBoot"/> + </optional> + <optional> <element name="backenddomain"> <attribute name="name"> <ref name="domainName"/> @@ -1179,9 +1182,6 @@ </element> </optional> <optional> - <ref name="deviceBoot"/> - </optional> - <optional> <element name="readonly"> <empty/> </element>

On Thu, Feb 05, 2015 at 02:53:44PM -0700, Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

Vfb entries in domain config are used only by PV drivers. Qemu parameters are build based on b_info struct. So fill it with the same data as vfb entries (actually the first one). This will additionally allow graphic-less domain, when no <graphics/> entries are present in domain XML (previously VNC was always enabled).

This should already be handled in libxlMakeVfbList().

Indeed, I've blindly rebased that patch from older version, but...

Is there something missing in that function?

... yes, if there is no graphics defined, the driver will set b_info->u.hvm.nographic to 1. Which isn't enough to disable graphic, because actual condition in libxl is: if (info->nographic && (!info->sdl && !info->vnc)) { flexarray_append(dm_args, "-nographic"); } I'll think about some better solution (perhaps only part about initially setting vnc and sdl to false?).

Regards, Jim

...

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_conf.c | 100 ++++++++++++++++++++++++++++++++++++------------- 1 file changed, 74 insertions(+), 26 deletions(-)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 1811a83..c9f8ad5 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -618,12 +618,53 @@ libxlMakeChrdevStr(virDomainChrDefPtr def, char **buf) _("unsupported chardev '%s'"), type); return -1; } + return 0; +} + +static int +libxlMakeVNCInfo(virPortAllocatorPtr graphicsports, + virDomainGraphicsDefPtr l_vfb, + libxl_vnc_info *x_vnc) +{ + unsigned short port; + const char *listenAddr; + + libxl_defbool_set(&x_vnc->enable, 1); + /* driver handles selection of free port */ + libxl_defbool_set(&x_vnc->findunused, 0); + if (l_vfb->data.vnc.autoport) { + + if (virPortAllocatorAcquire(graphicsports, &port) < 0) + return -1; + l_vfb->data.vnc.port = port; + } + x_vnc->display = l_vfb->data.vnc.port - LIBXL_VNC_PORT_MIN;

+ listenAddr = virDomainGraphicsListenGetAddress(l_vfb, 0); + if (listenAddr) { + /* libxl_device_vfb_init() does VIR_STRDUP("127.0.0.1") */ + VIR_FREE(x_vnc->listen); + if (VIR_STRDUP(x_vnc->listen, listenAddr) < 0) + return -1; + } + return 0; +} + +static int +libxlMakeSDLInfo(virDomainGraphicsDefPtr l_vfb, + libxl_sdl_info *x_sdl) +{ + libxl_defbool_set(&x_sdl->enable, 1); + if (VIR_STRDUP(x_sdl->display, l_vfb->data.sdl.display) < 0) + return -1; + if (VIR_STRDUP(x_sdl->xauthority, l_vfb->data.sdl.xauth) < 0) + return -1; return 0; }

static int libxlMakeDomBuildInfo(virDomainDefPtr def, + virPortAllocatorPtr graphicsports, libxl_ctx *ctx, libxl_domain_config *d_config) { @@ -745,6 +786,35 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, return -1; }

+ /* Disable VNC and SDL until explicitly enabled */ + libxl_defbool_set(&b_info->u.hvm.vnc.enable, 0); + libxl_defbool_set(&b_info->u.hvm.sdl.enable, 0); + + for (i = 0; i < def->ngraphics; i++) { + switch (def->graphics[i]->type) { + case VIR_DOMAIN_GRAPHICS_TYPE_VNC: + if (libxl_defbool_val(b_info->u.hvm.vnc.enable)) + continue; + if (libxlMakeVNCInfo(graphicsports, + def->graphics[i], + &b_info->u.hvm.vnc) < 0) + return -1; + if (def->graphics[i]->data.vnc.keymap && + VIR_STRDUP(b_info->u.hvm.keymap, + def->graphics[i]->data.vnc.keymap) < 0) { + virReportOOMError(); + return -1; + } + break; + case VIR_DOMAIN_GRAPHICS_TYPE_SDL: + if (libxl_defbool_val(b_info->u.hvm.sdl.enable)) + continue; + if (libxlMakeSDLInfo(def->graphics[i], &b_info->u.hvm.sdl) < 0) + return -1; + break; + } + } + /* * The following comment and calculation were taken directly from * libxenlight's internal function libxl_get_required_shadow_memory(): @@ -1171,38 +1241,16 @@ libxlMakeVfb(virPortAllocatorPtr graphicsports, virDomainGraphicsDefPtr l_vfb, libxl_device_vfb *x_vfb) { - unsigned short port; - const char *listenAddr; - libxl_device_vfb_init(x_vfb);

switch (l_vfb->type) { case VIR_DOMAIN_GRAPHICS_TYPE_SDL: - libxl_defbool_set(&x_vfb->sdl.enable, 1); - if (VIR_STRDUP(x_vfb->sdl.display, l_vfb->data.sdl.display) < 0) - return -1; - if (VIR_STRDUP(x_vfb->sdl.xauthority, l_vfb->data.sdl.xauth) < 0) + if (libxlMakeSDLInfo(l_vfb, &x_vfb->sdl) < 0) return -1; break; case VIR_DOMAIN_GRAPHICS_TYPE_VNC: - libxl_defbool_set(&x_vfb->vnc.enable, 1); - /* driver handles selection of free port */ - libxl_defbool_set(&x_vfb->vnc.findunused, 0); - if (l_vfb->data.vnc.autoport) { - - if (virPortAllocatorAcquire(graphicsports, &port) < 0) - return -1; - l_vfb->data.vnc.port = port; - } - x_vfb->vnc.display = l_vfb->data.vnc.port - LIBXL_VNC_PORT_MIN; - - listenAddr = virDomainGraphicsListenGetAddress(l_vfb, 0); - if (listenAddr) { - /* libxl_device_vfb_init() does VIR_STRDUP("127.0.0.1") */ - VIR_FREE(x_vfb->vnc.listen); - if (VIR_STRDUP(x_vfb->vnc.listen, listenAddr) < 0) - return -1; - } + if (libxlMakeVNCInfo(graphicsports, l_vfb, &x_vfb->vnc) < 0) + return -1; if (VIR_STRDUP(x_vfb->keymap, l_vfb->data.vnc.keymap) < 0) return -1; break; @@ -1611,7 +1659,7 @@ libxlBuildDomainConfig(virPortAllocatorPtr graphicsports, if (libxlMakeDomCreateInfo(ctx, def, &d_config->c_info) < 0) return -1;

- if (libxlMakeDomBuildInfo(def, ctx, d_config) < 0) + if (libxlMakeDomBuildInfo(def, graphicsports, ctx, d_config) < 0) return -1;

if (libxlMakeDiskList(def, d_config) < 0)

-- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?

Marek Marczykowski-Górecki wrote:

On Sat, Feb 07, 2015 at 12:22:51PM +0100, Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 05, 2015 at 02:53:44PM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

Vfb entries in domain config are used only by PV drivers. Qemu parameters are build based on b_info struct. So fill it with the same data as vfb entries (actually the first one). This will additionally allow graphic-less domain, when no <graphics/> entries are present in domain XML (previously VNC was always enabled).

This should already be handled in libxlMakeVfbList().

Indeed, I've blindly rebased that patch from older version, but...

...

Is there something missing in that function?

... yes, if there is no graphics defined, the driver will set b_info->u.hvm.nographic to 1. Which isn't enough to disable graphic, because actual condition in libxl is: if (info->nographic && (!info->sdl && !info->vnc)) { flexarray_append(dm_args, "-nographic"); }

I'll think about some better solution (perhaps only part about initially setting vnc and sdl to false?).

The problem is when domain XML contains no <video/> nor <graphics/> elements - libxl defaults then to setting up VNC (on 127.0.0.1). The solution here is to disable VNC/SDL and enable only when domain config says so. This still make VNC set when only <video/> is present (without <graphics type='vnc'/>), but at least there is a way make domain really headless.

Below is patch which implement that approach. Note that it makes a big change in behaviour - it can make user's configuration broken (if he/she relied on this bug).

The current behavior is a regression wrt the legacy Xen driver. In that driver, the absence of <video> and <graphics> results in invoking qemu with -nographic. Since we aim for compatibility with the old Xen driver, I'd argue you are fixing the bug :-).

Disclaimer: I've tested it only with qemu in stubdomain.

I've tested it with a regular HVM domain.

-----8<----- From 37126eea6927f38fcb7f7f1190143fd7de54e9e6 Mon Sep 17 00:00:00 2001 From: Marek Marczykowski <marmarek@invisiblethingslab.com> Date: Sun, 7 Apr 2013 19:48:03 +0200 Subject: [PATCH] libxl: disable VNC and SDL until explicitly enabled Cc: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>

Otherwise when VM have no graphics defined, it is still enabled (with some libxl defaults).

The commit message should indicate this fixes a regression wrt the legacy Xen driver.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_conf.c | 4 ++++ 1 file changed, 4 insertions(+)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 1811a83..7ee4e54 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -745,6 +745,10 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, return -1; }

+ /* Disable VNC and SDL until explicitly enabled */ + libxl_defbool_set(&b_info->u.hvm.vnc.enable, 0); + libxl_defbool_set(&b_info->u.hvm.sdl.enable, 0);

ACK. I'll adjust the commit message and push shortly. Regards, Jim

Marek Marczykowski-Górecki wrote:

Do not silently ignore its value. LibXL support only one address, so refuse multiple IPs.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_conf.c | 5 +++++ src/libxl/libxl_domain.c | 11 +++++++++++ 2 files changed, 16 insertions(+)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index c9f8ad5..4730585 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -1133,6 +1133,11 @@ libxlMakeNic(virDomainDefPtr def, case VIR_DOMAIN_NET_TYPE_ETHERNET: if (VIR_STRDUP(x_nic->script, l_nic->script) < 0) return -1; + if (l_nic->nips > 0) { + x_nic->ip = virSocketAddrFormat(&l_nic->ips[0]->address); + if (!x_nic->ip) + return -1; + }

IP address should be supported on VIR_DOMAIN_NET_TYPE_NETWORK interface types too. The domainXML docs even contain an example http://libvirt.org/formatdomain.html#ipconfig Regards, Jim

break; case VIR_DOMAIN_NET_TYPE_NETWORK: { diff --git a/src/libxl/libxl_domain.c b/src/libxl/libxl_domain.c index 856cfb4..afb0b15 100644 --- a/src/libxl/libxl_domain.c +++ b/src/libxl/libxl_domain.c @@ -482,6 +482,17 @@ libxlDomainDeviceDefPostParse(virDomainDeviceDefPtr dev, STRNEQ(def->os.type, "hvm")) dev->data.chr->targetType = VIR_DOMAIN_CHR_CONSOLE_TARGET_TYPE_XEN;

+ if (dev->type == VIR_DOMAIN_DEVICE_NET && + (dev->data.net->type == VIR_DOMAIN_NET_TYPE_BRIDGE || + dev->data.net->type == VIR_DOMAIN_NET_TYPE_ETHERNET)) { + if (dev->data.net->nips > 1) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + _("multiple IP addresses not supported on device type %s"), + virDomainNetTypeToString(dev->data.net->type)); + return -1; + } + } + if (dev->type == VIR_DOMAIN_DEVICE_HOSTDEV || (dev->type == VIR_DOMAIN_DEVICE_NET && dev->data.net->type == VIR_DOMAIN_NET_TYPE_HOSTDEV)) {

Marek Marczykowski-Górecki wrote:

Updated patch below.

-----8<----- From c1bd7134d4d70a06229243c41284eb37daa35f38 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Marek=20Marczykowski-G=C3=B3recki?= <marmarek@invisiblethingslab.com> Date: Thu, 5 Feb 2015 05:59:30 +0100 Subject: [PATCH] libxl: pass ipaddr to libxl toolstack Cc: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>

Do not silently ignore its value. LibXL support only one address, so refuse multiple IPs.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_conf.c | 11 +++++++++++ src/libxl/libxl_domain.c | 12 ++++++++++++ 2 files changed, 23 insertions(+)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index c9f8ad5..ec0f67a 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -1133,6 +1133,11 @@ libxlMakeNic(virDomainDefPtr def, case VIR_DOMAIN_NET_TYPE_ETHERNET: if (VIR_STRDUP(x_nic->script, l_nic->script) < 0) return -1; + if (l_nic->nips > 0) { + x_nic->ip = virSocketAddrFormat(&l_nic->ips[0]->address); + if (!x_nic->ip) + return -1; + } break; case VIR_DOMAIN_NET_TYPE_NETWORK: { @@ -1150,6 +1155,12 @@ libxlMakeNic(virDomainDefPtr def, return -1; }

+ if (l_nic->nips > 0) { + x_nic->ip = virSocketAddrFormat(&l_nic->ips[0]->address); + if (!x_nic->ip) + return -1; + } + if ((brname = virNetworkGetBridgeName(network))) { if (VIR_STRDUP(x_nic->bridge, brname) < 0) fail = true; diff --git a/src/libxl/libxl_domain.c b/src/libxl/libxl_domain.c index 856cfb4..f0eaf6c 100644 --- a/src/libxl/libxl_domain.c +++ b/src/libxl/libxl_domain.c @@ -482,6 +482,18 @@ libxlDomainDeviceDefPostParse(virDomainDeviceDefPtr dev, STRNEQ(def->os.type, "hvm")) dev->data.chr->targetType = VIR_DOMAIN_CHR_CONSOLE_TARGET_TYPE_XEN;

+ if (dev->type == VIR_DOMAIN_DEVICE_NET && + (dev->data.net->type == VIR_DOMAIN_NET_TYPE_BRIDGE || + dev->data.net->type == VIR_DOMAIN_NET_TYPE_ETHERNET || + dev->data.net->type == VIR_DOMAIN_NET_TYPE_NETWORK)) { + if (dev->data.net->nips > 1) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + _("multiple IP addresses not supported on device type %s"), + virDomainNetTypeToString(dev->data.net->type)); + return -1; + } + } + if (dev->type == VIR_DOMAIN_DEVICE_HOSTDEV || (dev->type == VIR_DOMAIN_DEVICE_NET && dev->data.net->type == VIR_DOMAIN_NET_TYPE_HOSTDEV)) {

ACK and pushed. Sorry I've stalled on reviewing your series. My libvirt timeslices have been consumed by fixing deadlocks and libxl assertions https://www.redhat.com/archives/libvir-list/2015-February/msg00024.html As suggested by Ian Jackson, I'm working on a series to remove the per-domain libxl_ctx. Those are remnants from the Xen 4.1 libxl days and are no longer required in newer libxl. I have a fair bit of testing to finish, but the series is looking promising thus far. Regards, Jim

Marek Marczykowski-Górecki wrote:

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - rebase on 1.2.12+ - multiple devices support

src/libxl/libxl_conf.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 48 insertions(+)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index 4730585..b1131ea 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -815,6 +815,54 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, } }

+ if (def->ninputs) { +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + if (VIR_ALLOC_N(b_info->u.hvm.usbdevice_list, def->ninputs+1) < 0) + return -1; +#else + if (def->ninputs > 1) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only one input device")); + return -1; + } +#endif + for (i = 0; i < def->ninputs; i++) { + if (def->inputs[i]->bus != VIR_DOMAIN_INPUT_BUS_USB) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only USB input")); + return -1; + } + }

+ for (i = 0; i < def->ninputs; i++) { + switch (def->inputs[i]->type) { + case VIR_DOMAIN_INPUT_TYPE_MOUSE: +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + if (VIR_STRDUP(b_info->u.hvm.usbdevice_list[i], "mouse") < 0) + return -1; +#else + VIR_FREE(b_info->u.hvm.usbdevice); + if (VIR_STRDUP(b_info->u.hvm.usbdevice, "mouse") < 0) + return -1; +#endif + break; + case VIR_DOMAIN_INPUT_TYPE_TABLET: +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + if (VIR_STRDUP(b_info->u.hvm.usbdevice_list[i], "tablet") < 0) + return -1; +#else + VIR_FREE(b_info->u.hvm.usbdevice); + if (VIR_STRDUP(b_info->u.hvm.usbdevice, "tablet") < 0) + return -1; +#endif

One of the #ifdef could be dropped with something like for (i=0; i < def->ninputs; i++) { char *temp; #ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST temp = b_info->u.hvm.usbdevice_list[i]; #else temp = b_info->u.hvm.usbdevice; #endif ... VIR_STRDUP(temp, "mouse"); ... } Also, could you look into adding support for usbdevice_list to the parsing/formating code in src/xenconfig? Regards, Jim

+ break; + default: + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("Unknown input device type")); + return -1; + } + } + } + /* * The following comment and calculation were taken directly from * libxenlight's internal function libxl_get_required_shadow_memory():

From: Marek Marczykowski <marmarek@invisiblethingslab.com> Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - rebase on 1.2.12+ - multiple devices support Changes in v3: - reduce code duplication src/libxl/libxl_conf.c | 44 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+) diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index b873622..c7108b0 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -749,6 +749,50 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, libxl_defbool_set(&b_info->u.hvm.vnc.enable, 0); libxl_defbool_set(&b_info->u.hvm.sdl.enable, 0); + if (def->ninputs) { + for (i = 0; i < def->ninputs; i++) { + if (def->inputs[i]->bus != VIR_DOMAIN_INPUT_BUS_USB) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only USB input")); + return -1; + } + } +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + if (VIR_ALLOC_N(b_info->u.hvm.usbdevice_list, def->ninputs+1) < 0) + return -1; +#else + if (def->ninputs > 1) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only one input device")); + return -1; + } +#endif + for (i = 0; i < def->ninputs; i++) { + char **usbdevice; +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + usbdevice = &b_info->u.hvm.usbdevice_list[i]; +#else + usbdevice = &b_info->u.hvm.usbdevice; +#endif + switch (def->inputs[i]->type) { + case VIR_DOMAIN_INPUT_TYPE_MOUSE: + VIR_FREE(*usbdevice); + if (VIR_STRDUP(*usbdevice, "mouse") < 0) + return -1; + break; + case VIR_DOMAIN_INPUT_TYPE_TABLET: + VIR_FREE(*usbdevice); + if (VIR_STRDUP(*usbdevice, "tablet") < 0) + return -1; + break; + default: + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("Unknown input device type")); + return -1; + } + } + } + /* * The following comment and calculation were taken directly from * libxenlight's internal function libxl_get_required_shadow_memory(): -- 1.8.3.1

Marek Marczykowski-Górecki wrote:

In Xen>=4.3, libxl supports new syntax for USB devices: usbdevice=[ "DEVICE", "DEVICE", ... ] Add support for that in xenconfig driver. When only one device is defined, keep using old syntax for backward compatibility.

Cool, thanks for doing this! But I have a "small" nit...

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/xenconfig/xen_common.c | 108 +++++++++++++++++++++++++++++++--------------

xm only supported the usbdevice=DEVICE syntax, but the code in xen_common is used to parse/format both xm and xl config. I think parsing/formating of these devices should be moved to xen-xm and xen-xl. Regards, Jim

1 file changed, 76 insertions(+), 32 deletions(-)

diff --git a/src/xenconfig/xen_common.c b/src/xenconfig/xen_common.c index a2a1474..4494f1d 100644 --- a/src/xenconfig/xen_common.c +++ b/src/xenconfig/xen_common.c @@ -964,6 +964,7 @@ static int xenParseEmulatedDevices(virConfPtr conf, virDomainDefPtr def) { const char *str; + virConfValuePtr val;

if (STREQ(def->os.type, "hvm")) { if (xenConfigGetString(conf, "soundhw", &str, NULL) < 0) @@ -973,31 +974,42 @@ xenParseEmulatedDevices(virConfPtr conf, virDomainDefPtr def) xenParseSxprSound(def, str) < 0) return -1;

- if (xenConfigGetString(conf, "usbdevice", &str, NULL) < 0) - return -1;

- if (str && - (STREQ(str, "tablet") || - STREQ(str, "mouse") || - STREQ(str, "keyboard"))) { - virDomainInputDefPtr input; - if (VIR_ALLOC(input) < 0) - return -1; - - input->bus = VIR_DOMAIN_INPUT_BUS_USB; - if (STREQ(str, "mouse")) - input->type = VIR_DOMAIN_INPUT_TYPE_MOUSE; - else if (STREQ(str, "tablet")) - input->type = VIR_DOMAIN_INPUT_TYPE_TABLET; - else if (STREQ(str, "keyboard")) - input->type = VIR_DOMAIN_INPUT_TYPE_KBD; - if (VIR_ALLOC_N(def->inputs, 1) < 0) { - virDomainInputDefFree(input); + val = virConfGetValue(conf, "usbdevice"); + /* usbdevice can be defined as either a single string or a list */ + if (val && val->type == VIR_CONF_LIST) + val = val->list; + /* otherwise val->next is NULL, so can be handled by the same code */ + while (val) { + if (val->type != VIR_CONF_STRING) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("config value %s was malformed"), + "usbdevice"); return -1; + } + str = val->str; + + if (str && + (STREQ(str, "tablet") || + STREQ(str, "mouse") || + STREQ(str, "keyboard"))) { + virDomainInputDefPtr input; + if (VIR_ALLOC(input) < 0) + return -1;

+ input->bus = VIR_DOMAIN_INPUT_BUS_USB; + if (STREQ(str, "mouse")) + input->type = VIR_DOMAIN_INPUT_TYPE_MOUSE; + else if (STREQ(str, "tablet")) + input->type = VIR_DOMAIN_INPUT_TYPE_TABLET; + else if (STREQ(str, "keyboard")) + input->type = VIR_DOMAIN_INPUT_TYPE_KBD; + if (VIR_APPEND_ELEMENT(def->inputs, def->ninputs, input) < 0) { + virDomainInputDefFree(input); + return -1; + } } - def->inputs[0] = input; - def->ninputs = 1; + val = val->next; } }

@@ -1953,36 +1965,68 @@ static int xenFormatInputDevs(virConfPtr conf, virDomainDefPtr def) { size_t i; + const char *devtype; + virConfValuePtr usbdevices = NULL, lastdev;

if (STREQ(def->os.type, "hvm")) { + if (VIR_ALLOC(usbdevices) < 0) + goto error; + + usbdevices->type = VIR_CONF_LIST; + usbdevices->list = NULL; + lastdev = NULL; for (i = 0; i < def->ninputs; i++) { if (def->inputs[i]->bus == VIR_DOMAIN_INPUT_BUS_USB) { if (xenConfigSetInt(conf, "usb", 1) < 0) - return -1; + goto error;

switch (def->inputs[i]->type) { case VIR_DOMAIN_INPUT_TYPE_MOUSE: - if (xenConfigSetString(conf, "usbdevice", "mouse") < 0) - return -1; - + devtype = "mouse"; break; case VIR_DOMAIN_INPUT_TYPE_TABLET: - if (xenConfigSetString(conf, "usbdevice", "tablet") < 0) - return -1; - + devtype = "tablet"; break; case VIR_DOMAIN_INPUT_TYPE_KBD: - if (xenConfigSetString(conf, "usbdevice", "keyboard") < 0) - return -1; - + devtype = "keyboard"; break; + default: + continue; } - break; + + if (lastdev == NULL) { + if (VIR_ALLOC(lastdev) < 0) + goto error; + usbdevices->list = lastdev; + } else { + if (VIR_ALLOC(lastdev->next) < 0) + goto error; + lastdev = lastdev->next; + } + lastdev->type = VIR_CONF_STRING; + if (VIR_STRDUP(lastdev->str, devtype) < 0) + goto error; } } + if (usbdevices->list != NULL) { + if (usbdevices->list->next == NULL) { + /* for compatibility with Xen <= 4.2, use old syntax when + * only one device present */ + if (xenConfigSetString(conf, "usbdevice", usbdevices->list->str) < 0) + goto error; + virConfFreeValue(usbdevices); + } else { + virConfSetValue(conf, "usbdevice", usbdevices); + } + } else { + VIR_FREE(usbdevices); + } }

return 0; + error: + virConfFreeValue(usbdevices); + return -1; }

Marek Marczykowski-Górecki wrote:

From: Marek Marczykowski <marmarek@invisiblethingslab.com>

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- Changes in v2: - rebase on 1.2.12+ - multiple devices support Changes in v3: - reduce code duplication

src/libxl/libxl_conf.c | 44 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)

diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index b873622..c7108b0 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -749,6 +749,50 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, libxl_defbool_set(&b_info->u.hvm.vnc.enable, 0); libxl_defbool_set(&b_info->u.hvm.sdl.enable, 0);

+ if (def->ninputs) { + for (i = 0; i < def->ninputs; i++) { + if (def->inputs[i]->bus != VIR_DOMAIN_INPUT_BUS_USB) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only USB input")); + return -1; + } + } +#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + if (VIR_ALLOC_N(b_info->u.hvm.usbdevice_list, def->ninputs+1) < 0) + return -1; +#else + if (def->ninputs > 1) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("libxenlight supports only one input device")); + return -1; + } +#endif + for (i = 0; i < def->ninputs; i++) { + char **usbdevice;

I'm no C expert, but I think 'char *' is fine, avoiding the subsequent addressof and dereferencing. Regards, Jim

+#ifdef LIBXL_HAVE_BUILDINFO_USBDEVICE_LIST + usbdevice = &b_info->u.hvm.usbdevice_list[i]; +#else + usbdevice = &b_info->u.hvm.usbdevice; +#endif + switch (def->inputs[i]->type) { + case VIR_DOMAIN_INPUT_TYPE_MOUSE: + VIR_FREE(*usbdevice); + if (VIR_STRDUP(*usbdevice, "mouse") < 0) + return -1; + break; + case VIR_DOMAIN_INPUT_TYPE_TABLET: + VIR_FREE(*usbdevice); + if (VIR_STRDUP(*usbdevice, "tablet") < 0) + return -1; + break; + default: + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("Unknown input device type")); + return -1; + } + } + } + /* * The following comment and calculation were taken directly from * libxenlight's internal function libxl_get_required_shadow_memory():

On Thu, Feb 19, 2015 at 11:43:15AM -0700, Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

It will not be possible to detach such device later. Also improve logging in such cases.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_driver.c | 41 +++++++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c index ce3a99b..005cc96 100644 --- a/src/libxl/libxl_driver.c +++ b/src/libxl/libxl_driver.c @@ -2787,6 +2787,7 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, int actualType; libxl_device_nic nic; int ret = -1; + char mac[VIR_MAC_STRING_BUFLEN];

/* preallocate new slot for device */ if (VIR_REALLOC_N(vm->def->nets, vm->def->nnets + 1) < 0) @@ -2801,6 +2802,14 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver,

actualType = virDomainNetGetActualType(net);

+ /* -2 means "multiple matches" so then fail also */

No longer true after commit 2fbae1b2. I think you just want to check if virDomainNetFindIdx() >= 0, meaning the def already contains a net device with the same mac address.

But here the error is when the device *is* found, so the opposite case than already reported as an error by virDomainNetFindIdx. Is it correct (harmless) to call virReportError when no error occurred (function returns >=0)? Maybe libvirt should check for duplicated devices at higher level (-> for all drivers in the same place)?

...

+ if (virDomainNetFindIdx(vm->def, net) != -1) { + virReportError(VIR_ERR_OPERATION_FAILED, + _("device matching mac address %s already exists"), + virMacAddrFormat(&net->mac, mac)); + return -1; + } + if (actualType == VIR_DOMAIN_NET_TYPE_HOSTDEV) { /* This is really a "smart hostdev", so it should be attached * as a hostdev (the hostdev code will reach over into the @@ -2879,6 +2888,7 @@ libxlDomainAttachDeviceConfig(virDomainDefPtr vmdef, virDomainDeviceDefPtr dev) virDomainHostdevDefPtr hostdev; virDomainHostdevDefPtr found; virDomainHostdevSubsysPCIPtr pcisrc; + char mac[VIR_MAC_STRING_BUFLEN];

switch (dev->type) { case VIR_DOMAIN_DEVICE_DISK: @@ -2896,6 +2906,12 @@ libxlDomainAttachDeviceConfig(virDomainDefPtr vmdef, virDomainDeviceDefPtr dev)

case VIR_DOMAIN_DEVICE_NET: net = dev->data.net; + if (virDomainNetFindIdx(vmdef, net) >= 0) { + virReportError(VIR_ERR_INVALID_ARG, + _("network device with mac %s already exists."), + virMacAddrFormat(&net->mac, mac)); + return -1; + }

Ah, like you've done here :-).

...

if (virDomainNetInsert(vmdef, net)) return -1; dev->data.net = NULL; @@ -3060,8 +3076,18 @@ libxlDomainDetachNetDevice(libxlDriverPrivatePtr driver, char mac[VIR_MAC_STRING_BUFLEN]; int ret = -1;

- if ((detachidx = virDomainNetFindIdx(vm->def, net)) < 0) + if ((detachidx = virDomainNetFindIdx(vm->def, net)) < 0) { + if (detachidx == -2) { + virReportError(VIR_ERR_OPERATION_FAILED, + _("multiple devices matching mac address %s found"), + virMacAddrFormat(&net->mac, mac)); + } else { + virReportError(VIR_ERR_OPERATION_FAILED, + _("network device %s not found"), + virMacAddrFormat(&net->mac, mac)); + }

virDomainNetFindIdx() handles the error reporting now.

...

return -1; + }

detach = vm->def->nets[detachidx];

@@ -3136,6 +3162,7 @@ libxlDomainDetachDeviceConfig(virDomainDefPtr vmdef, virDomainDeviceDefPtr dev) virDomainHostdevDefPtr hostdev, det_hostdev; virDomainNetDefPtr net; int idx; + char mac[VIR_MAC_STRING_BUFLEN];

switch (dev->type) { case VIR_DOMAIN_DEVICE_DISK: @@ -3150,8 +3177,18 @@ libxlDomainDetachDeviceConfig(virDomainDefPtr vmdef, virDomainDeviceDefPtr dev)

case VIR_DOMAIN_DEVICE_NET: net = dev->data.net; - if ((idx = virDomainNetFindIdx(vmdef, net)) < 0) + if ((idx = virDomainNetFindIdx(vmdef, net)) < 0) { + if (idx == -2) { + virReportError(VIR_ERR_OPERATION_FAILED, + _("multiple devices matching mac address %s found"), + virMacAddrFormat(&dev->data.net->mac, mac)); + } else { + virReportError(VIR_ERR_OPERATION_FAILED, + _("network device %s not found"), + virMacAddrFormat(&dev->data.net->mac, mac)); + }

Same here.

Regards, Jim

-- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?

On Thu, Feb 19, 2015 at 01:58:02PM -0700, Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 11:43:15AM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

It will not be possible to detach such device later. Also improve logging in such cases.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_driver.c | 41 +++++++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c index ce3a99b..005cc96 100644 --- a/src/libxl/libxl_driver.c +++ b/src/libxl/libxl_driver.c @@ -2787,6 +2787,7 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, int actualType; libxl_device_nic nic; int ret = -1; + char mac[VIR_MAC_STRING_BUFLEN];

/* preallocate new slot for device */ if (VIR_REALLOC_N(vm->def->nets, vm->def->nnets + 1) < 0) @@ -2801,6 +2802,14 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver,

actualType = virDomainNetGetActualType(net);

+ /* -2 means "multiple matches" so then fail also */

No longer true after commit 2fbae1b2. I think you just want to check if virDomainNetFindIdx() >= 0, meaning the def already contains a net device with the same mac address.

But here the error is when the device *is* found, so the opposite case than already reported as an error by virDomainNetFindIdx.

If you find an idx >= 0, then the domain def already contains a net device with the same mac address, right? In that case, you report an error and return failure from libxlDomainAttachNetDevice().

Right, but if I do not find one (idx == -1), I will proceed with (possibly successful) adding the device, while the error was already reported by virDomainNetFindIdx. -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?

Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 01:58:02PM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 11:43:15AM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

It will not be possible to detach such device later. Also improve logging in such cases.

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> --- src/libxl/libxl_driver.c | 41 +++++++++++++++++++++++++++++++++++++++-- 1 file changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c index ce3a99b..005cc96 100644 --- a/src/libxl/libxl_driver.c +++ b/src/libxl/libxl_driver.c @@ -2787,6 +2787,7 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, int actualType; libxl_device_nic nic; int ret = -1; + char mac[VIR_MAC_STRING_BUFLEN];

/* preallocate new slot for device */ if (VIR_REALLOC_N(vm->def->nets, vm->def->nnets + 1) < 0) @@ -2801,6 +2802,14 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver,

actualType = virDomainNetGetActualType(net);

+ /* -2 means "multiple matches" so then fail also */

No longer true after commit 2fbae1b2. I think you just want to check if virDomainNetFindIdx() >= 0, meaning the def already contains a net device with the same mac address.

But here the error is when the device *is* found, so the opposite case than already reported as an error by virDomainNetFindIdx.

If you find an idx >= 0, then the domain def already contains a net device with the same mac address, right? In that case, you report an error and return failure from libxlDomainAttachNetDevice().

Right, but if I do not find one (idx == -1), I will proceed with (possibly successful) adding the device, while the error was already reported by virDomainNetFindIdx.

Ah, right :-/.

Another option: introduce virDomainHasNet() to detect if the domain def already contains the net device.

A better option would be to fix this in libxl, for the benefit of other libxl apps. Regards, Jim

Marek Marczykowski-Górecki wrote:

On Thu, Feb 19, 2015 at 03:10:13PM -0700, Jim Fehlig wrote:

...

Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 01:58:02PM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 11:43:15AM -0700, Jim Fehlig wrote:

> Marek Marczykowski-Górecki wrote: > > > > >> It will not be possible to detach such device later. Also improve >> logging in such cases. >> >> Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> >> --- >> src/libxl/libxl_driver.c | 41 +++++++++++++++++++++++++++++++++++++++-- >> 1 file changed, 39 insertions(+), 2 deletions(-) >> >> diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c >> index ce3a99b..005cc96 100644 >> --- a/src/libxl/libxl_driver.c >> +++ b/src/libxl/libxl_driver.c >> @@ -2787,6 +2787,7 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, >> int actualType; >> libxl_device_nic nic; >> int ret = -1; >> + char mac[VIR_MAC_STRING_BUFLEN]; >> >> /* preallocate new slot for device */ >> if (VIR_REALLOC_N(vm->def->nets, vm->def->nnets + 1) < 0) >> @@ -2801,6 +2802,14 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, >> >> actualType = virDomainNetGetActualType(net); >> >> + /* -2 means "multiple matches" so then fail also */ >> >> >> >> >> > No longer true after commit 2fbae1b2. I think you just want to check if > virDomainNetFindIdx() >= 0, meaning the def already contains a net > device with the same mac address. > > > > But here the error is when the device *is* found, so the opposite case than already reported as an error by virDomainNetFindIdx.

If you find an idx >= 0, then the domain def already contains a net device with the same mac address, right? In that case, you report an error and return failure from libxlDomainAttachNetDevice().

Right, but if I do not find one (idx == -1), I will proceed with (possibly successful) adding the device, while the error was already reported by virDomainNetFindIdx.

Ah, right :-/.

Another option: introduce virDomainHasNet() to detect if the domain def already contains the net device.

A better option would be to fix this in libxl, for the benefit of other libxl apps.

Actually libxl has no problem with duplicated mac addresses, its libvirt that makes problem.

Yeah, it appears duplicate mac addresses are only valid if on different PCI devices. Back to virDomainHasNet? :-) Or checking for the duplicate directly in libxlDomainAttachNetDevice()? Regards, Jim

Marek Marczykowski-Górecki wrote:

On Thu, Feb 19, 2015 at 03:58:30PM -0700, Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 03:10:13PM -0700, Jim Fehlig wrote:

...

Jim Fehlig wrote:

...

Marek Marczykowski-Górecki wrote:

...

On Thu, Feb 19, 2015 at 01:58:02PM -0700, Jim Fehlig wrote:

> Marek Marczykowski-Górecki wrote: > >> On Thu, Feb 19, 2015 at 11:43:15AM -0700, Jim Fehlig wrote: >> >>> Marek Marczykowski-Górecki wrote: >>> >>>> It will not be possible to detach such device later. Also improve >>>> logging in such cases. >>>> >>>> Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> >>>> --- >>>> src/libxl/libxl_driver.c | 41 +++++++++++++++++++++++++++++++++++++++-- >>>> 1 file changed, 39 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c >>>> index ce3a99b..005cc96 100644 >>>> --- a/src/libxl/libxl_driver.c >>>> +++ b/src/libxl/libxl_driver.c >>>> @@ -2787,6 +2787,7 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, >>>> int actualType; >>>> libxl_device_nic nic; >>>> int ret = -1; >>>> + char mac[VIR_MAC_STRING_BUFLEN]; >>>> >>>> /* preallocate new slot for device */ >>>> if (VIR_REALLOC_N(vm->def->nets, vm->def->nnets + 1) < 0) >>>> @@ -2801,6 +2802,14 @@ libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, >>>> >>>> actualType = virDomainNetGetActualType(net); >>>> >>>> + /* -2 means "multiple matches" so then fail also */ >>>> >>>> >>>> >>>> >>>> >>>> >>> No longer true after commit 2fbae1b2. I think you just want to check if >>> virDomainNetFindIdx() >= 0, meaning the def already contains a net >>> device with the same mac address. >>> >>> >>> >>> >>> >> But here the error is when the device *is* found, so the opposite case >> than already reported as an error by virDomainNetFindIdx. >> >> >> >> > If you find an idx >= 0, then the domain def already contains a net > device with the same mac address, right? In that case, you report an > error and return failure from libxlDomainAttachNetDevice(). > > > > Right, but if I do not find one (idx == -1), I will proceed with (possibly successful) adding the device, while the error was already reported by virDomainNetFindIdx.

Ah, right :-/.

Another option: introduce virDomainHasNet() to detect if the domain def already contains the net device.

A better option would be to fix this in libxl, for the benefit of other libxl apps.

Actually libxl has no problem with duplicated mac addresses, its libvirt that makes problem.

Yeah, it appears duplicate mac addresses are only valid if on different PCI devices.

Is that true for libvirt generally (all drivers)?

...

Back to virDomainHasNet? :-) Or checking for the duplicate directly in libxlDomainAttachNetDevice()?

What do you mean by "directly"? This is exactly what my patch did (until virDomainNetFindIdx stopped reporting duplicates).

I mean coding up the search for an existing mac directly in libxlDomainAttachNetDevice(). E.g. static int libxlDomainAttachNetDevice(libxlDriverPrivatePtr driver, libxlDomainObjPrivatePtr priv, virDomainObjPtr vm, virDomainNetDefPtr net) { int actualType; libxl_device_nic nic; int ret = -1; size_t i; virDomainDefPtr def = vm->def; for (i = 0; i < def->nnets; i++) { if (virMacAddrCmp(&def->nets[i]->mac, &net->mac) == 0) error; } ... } Regards, Jim

On Thu, Feb 19, 2015 at 11:59:38AM -0700, Jim Fehlig wrote:

Marek Marczykowski-Górecki wrote:

...

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> ---

There was a discussion about calling virDomainDefCheckABIStability. The use case I have for this is to start multiple domains from the same savefile.

Interesting use case...

If you're curious about details, it is used for DisposableVMs in Qubes OS: http://theinvisiblethings.blogspot.com/2010/06/disposable-vms.html Basically to speedup domain startup (with some preloaded applications etc). Domain state is discarded after its termination (disk image is read-only).

...

This require changing: - UUID - name - IP address

Did you mean mac address? I don't see where the virtDomain*CheckABIStability() functions check for changes to IP addresses.

Of course, I mean MAC address (in addition to IP address).

...

Those are exactly type of changes blocked by virDomainDefCheckABIStability. Perhaps some "permissive" mode should be introduced to the virDomainDefCheckABIStability function?

I'd like to hear what others think about this proposal. Are UUID and name really part of a domain's ABI?

At least by the implementation of virDomainDefCheckABIStability... Comment says this is to not confuse domain during migration.

Regards, Jim

...

src/libxl/libxl_driver.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/src/libxl/libxl_driver.c b/src/libxl/libxl_driver.c index 005cc96..7539d33 100644 --- a/src/libxl/libxl_driver.c +++ b/src/libxl/libxl_driver.c @@ -1456,11 +1456,6 @@ libxlDomainRestoreFlags(virConnectPtr conn, const char *from, #endif

virCheckFlags(VIR_DOMAIN_SAVE_PAUSED, -1); - if (dxml) { - virReportError(VIR_ERR_ARGUMENT_UNSUPPORTED, "%s", - _("xml modification unsupported")); - return -1; - }

fd = libxlDomainSaveImageOpen(driver, cfg, from, &def, &hdr); if (fd < 0) @@ -1469,6 +1464,18 @@ libxlDomainRestoreFlags(virConnectPtr conn, const char *from, if (virDomainRestoreFlagsEnsureACL(conn, def) < 0) goto cleanup_unlock;

+ if (dxml) { + virDomainDefPtr def2 = NULL; + + if (!(def2 = virDomainDefParseString(dxml, cfg->caps, driver->xmlopt, + 1 << VIR_DOMAIN_VIRT_XEN, + VIR_DOMAIN_XML_INACTIVE))) { + goto cleanup; + } + virDomainDefFree(def); + def = def2; + } + if (!(vm = virDomainObjListAdd(driver->domains, def, driver->xmlopt, VIR_DOMAIN_OBJ_LIST_ADD_LIVE |

-- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?

Marek Marczykowski-Górecki wrote:

Xen have feature of having device model in separate domain (called stub domain). Add a 'type' attribute to 'emulator' element to allow selecting such a configuration.

Or maybe 'mode', describing the mode in which the emulator runs: as a process or as a VM.

Emulator path is still used for qemu running in dom0 (if any). Libxl currently do not allow to select stubdomain path.

That seems to overload a single <emulator>. Would it be better to have multiple <emulators>? E.g. <emulator>/usr/lib/xen/bin/qemu-system-i386</emulator> <emulator mode='stubdomain'>/usr/lib/xen/bin/stubdom-dm</emulator>

Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> ---

I think it would be good idea to introduce the same change to capabilities XML. The problem is I can't include domain_conf.h from capabilities.h, so probably that enum declaration needs to be moved to capabilities.h. Is it the right way? Or it should be done somehow different?

docs/formatdomain.html.in | 14 ++++++++++++++ docs/schemas/domaincommon.rng | 23 ++++++++++++++++++++++- src/conf/domain_conf.c | 36 ++++++++++++++++++++++++++++++++++-- src/conf/domain_conf.h | 9 +++++++++ src/libxl/libxl_conf.c | 17 +++++++++++++++++ 5 files changed, 96 insertions(+), 3 deletions(-)

diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in index 38c42d5..4f539e2 100644 --- a/docs/formatdomain.html.in +++ b/docs/formatdomain.html.in @@ -1652,6 +1652,20 @@ The <a href="formatcaps.html">capabilities XML</a> specifies the recommended default emulator to use for each particular domain type / architecture combination. + + <span class="since">Since 1.2.13</span>, the <code>emulator</code> + element may contain <code>type</code> attribute. Possible values are: + <dl> + <dt><code>type='default'</code></dt> + <dd>Equivalent to not setting <code>type</code> attribute at all. + </dd> + + <dt><code>type='stubdom'</code></dt> + <dd>Launch emulator in stub domain (Xen only). The emulator path + still indicate which binary is used in dom0 - there is no control + which binary is used as a stub domain. + </dd> + </dl> </dd> </dl>

diff --git a/docs/schemas/domaincommon.rng b/docs/schemas/domaincommon.rng index a4321f1..2a12073 100644 --- a/docs/schemas/domaincommon.rng +++ b/docs/schemas/domaincommon.rng @@ -2519,7 +2519,28 @@ --> <define name="emulator"> <element name="emulator"> - <ref name="absFilePath"/> + <choice> + <group> + <optional> + <attribute name="type"> + <choice> + <value>qemu</value> + <value>stubdom</value> + </choice> + </attribute> + </optional> + <ref name="absFilePath"/> + </group> + <group> + <attribute name="type"> + <choice> + <value>qemu</value> + <value>stubdom</value> + </choice> + </attribute> + <empty/> + </group> + </choice> </element> </define> <!-- diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c index 17b699a..c268091 100644 --- a/src/conf/domain_conf.c +++ b/src/conf/domain_conf.c @@ -741,6 +741,10 @@ VIR_ENUM_IMPL(virDomainDiskTray, VIR_DOMAIN_DISK_TRAY_LAST, "closed", "open");

+VIR_ENUM_IMPL(virDomainEmulatorType, VIR_DOMAIN_EMULATOR_TYPE_LAST, + "qemu", + "stubdom"); + VIR_ENUM_IMPL(virDomainRNGModel, VIR_DOMAIN_RNG_MODEL_LAST, "virtio"); @@ -13712,6 +13716,14 @@ virDomainDefParseXML(xmlDocPtr xml, }

def->emulator = virXPathString("string(./devices/emulator[1])", ctxt); + if ((tmp = virXPathString("string(./devices/emulator/@type)", ctxt))) { + def->emulator_type = virDomainEmulatorTypeTypeFromString(tmp); + VIR_FREE(tmp); + if (def->emulator_type < 0) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + _("Unknown emulator type '%s'"), tmp); + } + }

/* analysis of the disk devices */ if ((n = virXPathNodeSet("./devices/disk", ctxt, &nodes)) < 0) @@ -15690,6 +15702,14 @@ virDomainDefCheckABIStability(virDomainDefPtr src, goto error; }

+ if (src->emulator_type != dst->emulator_type) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, + _("Target domain emulator type %s does not match source %s"), + virDomainEmulatorTypeTypeToString(dst->emulator_type), + virDomainEmulatorTypeTypeToString(src->emulator_type)); + goto error; + } + if (!virDomainDefFeaturesCheckABIStability(src, dst)) goto error;

@@ -19893,8 +19913,20 @@ virDomainDefFormatInternal(virDomainDefPtr def, virBufferAddLit(buf, "<devices>\n"); virBufferAdjustIndent(buf, 2);

- virBufferEscapeString(buf, "<emulator>%s</emulator>\n", - def->emulator); + if (def->emulator || + def->emulator_type != VIR_DOMAIN_EMULATOR_TYPE_DEFAULT) { + virBufferAddLit(buf, "<emulator"); + if (def->emulator_type != VIR_DOMAIN_EMULATOR_TYPE_DEFAULT) { + virBufferAsprintf(buf, " type='%s'", + virDomainEmulatorTypeTypeToString(def->emulator_type)); + } + if (!def->emulator) { + virBufferAddLit(buf, "/>\n"); + } else { + virBufferEscapeString(buf, ">%s</emulator>\n", + def->emulator); + } + }

for (n = 0; n < def->ndisks; n++) if (virDomainDiskDefFormat(buf, def->disks[n], flags) < 0) diff --git a/src/conf/domain_conf.h b/src/conf/domain_conf.h index 28c6920..38b9037 100644 --- a/src/conf/domain_conf.h +++ b/src/conf/domain_conf.h @@ -1918,6 +1918,13 @@ struct _virBlkioDevice { };

typedef enum { + VIR_DOMAIN_EMULATOR_TYPE_DEFAULT, + VIR_DOMAIN_EMULATOR_TYPE_STUBDOM, + + VIR_DOMAIN_EMULATOR_TYPE_LAST +} virDomainEmulatorType; + +typedef enum { VIR_DOMAIN_RNG_MODEL_VIRTIO,

VIR_DOMAIN_RNG_MODEL_LAST @@ -2083,6 +2090,7 @@ struct _virDomainDef {

virDomainOSDef os; char *emulator; + virDomainEmulatorType emulator_type; /* These three options are of type virTristateSwitch, * except VIR_DOMAIN_FEATURE_CAPABILITIES that is of type * virDomainCapabilitiesPolicy */ @@ -2841,6 +2849,7 @@ VIR_ENUM_DECL(virDomainGraphicsSpiceMouseMode) VIR_ENUM_DECL(virDomainGraphicsVNCSharePolicy) VIR_ENUM_DECL(virDomainHyperv) VIR_ENUM_DECL(virDomainKVM) +VIR_ENUM_DECL(virDomainEmulatorType) VIR_ENUM_DECL(virDomainRNGModel) VIR_ENUM_DECL(virDomainRNGBackend) VIR_ENUM_DECL(virDomainTPMModel) diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c index b1131ea..a84ce09 100644 --- a/src/libxl/libxl_conf.c +++ b/src/libxl/libxl_conf.c @@ -768,6 +768,23 @@ libxlMakeDomBuildInfo(virDomainDefPtr def, b_info->device_model_version = libxlDomainGetEmulatorType(def); }

+ /* In case of stubdom there will be two qemu instances: + * - in stubdom (libxl uses hardcoded path for this one), + * - in dom0 as a backend for stubdom (if needed).

Your comment here provoked my thinking above about supporting multiple <emulator>s. ISTM there should be an <emulator> device for each of these qemu instances.

+ * Emulator path control only the second one. It makes a perfect sense + * to use <emulator type='stubdom'/> (yes, without emulator path).

I'm not so sure. Maybe we should first add support in libxl for specifying a stubdomain emulator path. Regards, Jim

+ */ + if (def->emulator_type == VIR_DOMAIN_EMULATOR_TYPE_STUBDOM) + libxl_defbool_set(&b_info->device_model_stubdomain, 1); + + if (def->os.cmdline && def->os.cmdline[0]) { + b_info->extra_hvm = virStringSplit(def->os.cmdline, " ", 0); + if (b_info->extra_hvm == NULL) { + virReportOOMError(); + return -1; + } + } + if (def->nserials) { if (def->nserials > 1) { virReportError(VIR_ERR_CONFIG_UNSUPPORTED,