5:58 p.m.
This set adds the necessary bits to be able to minimally define KVM and
XenFV domains. It allows me to create valid KVM guests.
It sidesteps a couple of the issues we'll need to tackle at some point,
like how to convey the appropriate virtual network for a KVM interface, etc.
Comments welcome.
5:58 p.m.
New subject: [PATCH 1 of 5] [RFC] Update VSSD schemas for FullVirt
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200500454 28800
# Node ID 0772ada04c729ecbf3a3854fa63f748311a082e4
# Parent 05fa2ceba145e9e93066bd66c74299680bd7d21d
[RFC] Update VSSD schemas for FullVirt
This adds a flag to the Xen VSSD to enable full-virt or not, as well as the
required boot device information. In the implementation, a "false" or
missing value for the flag will indicate paravirt-ness.
For KVM, I added the boot device as well, since that's something it will
always need. For both, a missing boot device will imply "hd".
Comments? Additions?
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
diff -r 05fa2ceba145 -r 0772ada04c72 schema/VSSD.mof
--- a/schema/VSSD.mof Tue Jan 15 16:24:39 2008 -0800
+++ b/schema/VSSD.mof Wed Jan 16 08:20:54 2008 -0800
@@ -8,8 +8,17 @@ class Xen_VirtualSystemSettingData : CIM
class Xen_VirtualSystemSettingData : CIM_VirtualSystemSettingData
{
+ [Description ("Flag to determine whether this guest is fully-virtualized")]
+ boolean isFullVirt;
+
+ [Description ("The bootloader and arguments to use when in "
+ "para-virtualized mode")]
string Bootloader;
string BootloaderArgs;
+
+ [Description ("The device to boot from when in fully-virtualized mode."
+ "One of hd,fd,cdrom.")]
+ string BootDevice;
};
@@ -20,4 +29,8 @@ class Xen_VirtualSystemSettingData : CIM
]
class KVM_VirtualSystemSettingData : CIM_VirtualSystemSettingData
{
+
+ [Description ("The device to boot from. One of hd,fd,cdrom.")]
+ string BootDevice;
+
};
5:58 p.m.
New subject: [PATCH 2 of 5] Fix xmlgen to include boot parameter for KVM
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200503258 28800
# Node ID be7f34a7f1a879dc220eaa8f5e2eab6dd3a04420
# Parent 0772ada04c729ecbf3a3854fa63f748311a082e4
Fix xmlgen to include boot parameter for KVM
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
diff -r 0772ada04c72 -r be7f34a7f1a8 libxkutil/xmlgen.c
--- a/libxkutil/xmlgen.c Wed Jan 16 08:20:54 2008 -0800
+++ b/libxkutil/xmlgen.c Wed Jan 16 09:07:38 2008 -0800
@@ -462,21 +462,32 @@ static char *_kvm_os_xml(struct domain *
int ret;
char *xml;
char *type;
+ char *boot;
+ struct kv bootattr = {"dev", NULL};
if (os->type == NULL)
os->type = strdup("hvm");
+ if (os->boot == NULL)
+ os->boot = strdup("hd");
+
type = tagify("type", os->type, NULL, 0);
+ bootattr.val = os->boot;
+ boot = tagify("boot", NULL, &bootattr, 1);
+
ret = asprintf(&xml,
"<os>\n"
" %s\n"
+ " %s\n"
"</os>\n",
- type);
+ type,
+ boot);
if (ret == -1)
xml = NULL;
free(type);
+ free(boot);
return xml;
}
5:58 p.m.
New subject: [PATCH 3 of 5] Add network interface parsing for KVM domains
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200504973 28800
# Node ID c505a88eb5e25f5e3b2418312dc48b6602857946
# Parent be7f34a7f1a879dc220eaa8f5e2eab6dd3a04420
Add network interface parsing for KVM domains
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
diff -r be7f34a7f1a8 -r c505a88eb5e2 libxkutil/device_parsing.c
--- a/libxkutil/device_parsing.c Wed Jan 16 09:07:38 2008 -0800
+++ b/libxkutil/device_parsing.c Wed Jan 16 09:36:13 2008 -0800
@@ -61,7 +61,7 @@ static void cleanup_net_device(struct ne
{
free(dev->type);
free(dev->mac);
- free(dev->bridge);
+ free(dev->source);
}
static void cleanup_emu_device(struct emu_device *dev)
@@ -216,18 +216,22 @@ static int parse_net_device(xmlNode *ino
if (ndev->mac == NULL)
goto err;
} else if (XSTREQ(child->name, "source")) {
- ndev->bridge = get_attr_value(child, "bridge");
- if (ndev->bridge == NULL)
- goto err;
+ ndev->source = get_attr_value(child, "bridge");
+ if (ndev->source != NULL)
+ continue;
+ ndev->source = get_attr_value(child, "network");
+ if (ndev->source != NULL)
+ continue;
+ goto err;
}
}
if (ndev->mac == NULL)
goto err;
- if (ndev->bridge == NULL) {
- ndev->bridge = strdup(DEFAULT_BRIDGE);
- printf("No bridge, taking default of `%s'\n",
ndev->bridge);
+ if ((STREQC(ndev->type, "bridge")) && (ndev->source ==
NULL)) {
+ ndev->source = strdup(DEFAULT_BRIDGE);
+ printf("No bridge, taking default of `%s'\n",
ndev->source);
}
vdev->type = VIRT_DEV_NET;
@@ -517,7 +521,7 @@ struct virt_device *virt_device_dup(stru
if (dev->type == VIRT_DEV_NET) {
DUP_FIELD(dev, _dev, dev.net.mac);
DUP_FIELD(dev, _dev, dev.net.type);
- DUP_FIELD(dev, _dev, dev.net.bridge);
+ DUP_FIELD(dev, _dev, dev.net.source);
} else if (dev->type == VIRT_DEV_DISK) {
DUP_FIELD(dev, _dev, dev.disk.type);
DUP_FIELD(dev, _dev, dev.disk.device);
diff -r be7f34a7f1a8 -r c505a88eb5e2 libxkutil/device_parsing.h
--- a/libxkutil/device_parsing.h Wed Jan 16 09:07:38 2008 -0800
+++ b/libxkutil/device_parsing.h Wed Jan 16 09:36:13 2008 -0800
@@ -41,7 +41,7 @@ struct net_device {
struct net_device {
char *type;
char *mac;
- char *bridge;
+ char *source;
};
struct mem_device {
diff -r be7f34a7f1a8 -r c505a88eb5e2 src/Virt_DevicePool.c
--- a/src/Virt_DevicePool.c Wed Jan 16 09:07:38 2008 -0800
+++ b/src/Virt_DevicePool.c Wed Jan 16 09:36:13 2008 -0800
@@ -295,7 +295,7 @@ static char *netpool_member_of(const CMP
for (i = 0; i < count; i++) {
if (STREQ((devs[i].id), dev)) {
result = _netpool_member_of(conn,
- devs[i].dev.net.bridge);
+ devs[i].dev.net.source);
break;
}
}
7:59 p.m.
New subject: [PATCH 3 of 5] Add network interface parsing for KVM domains
Dan Smith wrote:
static void cleanup_emu_device(struct emu_device *dev)
@@ -216,18 +216,22 @@ static int parse_net_device(xmlNode *ino
if (ndev->mac == NULL)
goto err;
} else if (XSTREQ(child->name, "source")) {
- ndev->bridge = get_attr_value(child, "bridge");
- if (ndev->bridge == NULL)
- goto err;
+ ndev->source = get_attr_value(child, "bridge");
+ if (ndev->source != NULL)
+ continue;
+ ndev->source = get_attr_value(child, "network");
+ if (ndev->source != NULL)
+ continue;
+ goto err;
}
}
if (ndev->mac == NULL)
goto err;
- if (ndev->bridge == NULL) {
- ndev->bridge = strdup(DEFAULT_BRIDGE);
- printf("No bridge, taking default of `%s'\n",
ndev->bridge);
+ if ((STREQC(ndev->type, "bridge")) && (ndev->source ==
NULL)) {
+ ndev->source = strdup(DEFAULT_BRIDGE);
+ printf("No bridge, taking default of `%s'\n",
ndev->source);
}
Not a fault of this patch, but this should probably be changed to a
CU_DEBUG() at some point.
Also, might be a good idea to add a comment or a debug message to
indicate that we're not handling all interface types yet.
--
Kaitlin Rupert
IBM Linux Technology Center
karupert(a)us.ibm.com
8:06 p.m.
New subject: [PATCH 3 of 5] Add network interface parsing for KVM domains
KR> Not a fault of this patch, but this should probably be changed to
KR> a CU_DEBUG() at some point.
Ah, yeah, thanks. It was my fault from before, so I'll fix it here
since I'm touching it.
KR> Also, might be a good idea to add a comment or a debug message to
KR> indicate that we're not handling all interface types yet.
Yeah, I've reworked that bit to catch unsupported configurations.
I'll post a new version in a bit.
Thanks!
--
Dan Smith
IBM Linux Technology Center
Open Hypervisor Team
email: danms(a)us.ibm.com
5:58 p.m.
New subject: [PATCH 4 of 5] Fix KVM network xml to use virtual network type
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200504975 28800
# Node ID 81149ab841550817357e38e7d6c88c3049a0fd32
# Parent c505a88eb5e25f5e3b2418312dc48b6602857946
Fix KVM network xml to use virtual network type
...and default the network name for now.
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
diff -r c505a88eb5e2 -r 81149ab84155 libxkutil/xmlgen.c
--- a/libxkutil/xmlgen.c Wed Jan 16 09:36:13 2008 -0800
+++ b/libxkutil/xmlgen.c Wed Jan 16 09:36:15 2008 -0800
@@ -169,7 +169,7 @@ static bool disk_to_xml(char **xml, stru
return true;
}
-static bool net_to_xml(char **xml, struct virt_device *dev)
+static bool xen_net_to_xml(char **xml, struct virt_device *dev)
{
int ret;
char *_xml;
@@ -193,6 +193,43 @@ static bool net_to_xml(char **xml, struc
free(_xml);
return true;
+}
+
+static bool kvm_net_to_xml(char **xml, struct virt_device *dev)
+{
+ int ret;
+ char *_xml;
+ struct net_device *net = &dev->dev.net;
+
+ if (net->source == NULL)
+ net->source = strdup("default");
+
+ ret = asprintf(&_xml,
+ "<interface type='%s'>\n"
+ " <mac address='%s'/>\n"
+ " <source network='%s'/>\n"
+ "</interface>\n",
+ net->type,
+ net->mac,
+ net->source);
+ if (ret == -1)
+ return false;
+ else
+ astrcat(xml, _xml);
+
+ free(_xml);
+
+ return true;
+}
+
+static bool net_to_xml(char **xml, struct virt_device *dev)
+{
+ if (STREQ(dev->dev.net.type, "network"))
+ return kvm_net_to_xml(xml, dev);
+ else if (STREQ(dev->dev.net.type, "bridge"))
+ return xen_net_to_xml(xml, dev);
+ else
+ return false;
}
static bool vcpu_to_xml(char **xml, struct virt_device *dev)
9:37 p.m.
New subject: [PATCH 4 of 5] Fix KVM network xml to use virtual network type
Dan Smith wrote:
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200504975 28800
# Node ID 81149ab841550817357e38e7d6c88c3049a0fd32
# Parent c505a88eb5e25f5e3b2418312dc48b6602857946
Fix KVM network xml to use virtual network type
...and default the network name for now.
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
+
+static bool net_to_xml(char **xml, struct virt_device *dev)
+{
+ if (STREQ(dev->dev.net.type, "network"))
+ return kvm_net_to_xml(xml, dev);
+ else if (STREQ(dev->dev.net.type, "bridge"))
+ return xen_net_to_xml(xml, dev);
+ else
+ return false;
}
static bool vcpu_to_xml(char **xml, struct virt_device *dev)
Is making the Xen/KVM distinction on network type safe? (That's not a
rhetorical way of saying I don't think it is, this is just something I
have basically no info on.)
--
-Jay
9:51 p.m.
New subject: [PATCH 4 of 5] Fix KVM network xml to use virtual network type
JG> Is making the Xen/KVM distinction on network type safe? (That's
JG> not a rhetorical way of saying I don't think it is, this is just
JG> something I have basically no info on.)
I'm not sure what you're getting at. I don't think that right now
anything but a KVM guest can have a network type of "network", so it's
a reasonable key for the moment. However, as Kaitlin pointed out (and
I eluded to in the updated patch with a comment), I'm only claiming
support for "bridge" interfaces on Xen and "[virtual] network"
interfaces on KVM.
That call really needs an additional bit of information to know what
kind of domain it is to be 100% safe, but I didn't want to refactor it
all right now. Perhaps just adding a more specific type enum to the
network struct would be a good idea.
--
Dan Smith
IBM Linux Technology Center
Open Hypervisor Team
email: danms(a)us.ibm.com
10 p.m.
New subject: [PATCH 4 of 5] Fix KVM network xml to use virtual network type
Dan Smith wrote:
JG> Is making the Xen/KVM distinction on network type safe?
(That's
JG> not a rhetorical way of saying I don't think it is, this is just
JG> something I have basically no info on.)
I'm not sure what you're getting at. I don't think that right now
anything but a KVM guest can have a network type of "network", so it's
a reasonable key for the moment. However, as Kaitlin pointed out (and
I eluded to in the updated patch with a comment), I'm only claiming
support for "bridge" interfaces on Xen and "[virtual] network"
interfaces on KVM.
That call really needs an additional bit of information to know what
kind of domain it is to be 100% safe, but I didn't want to refactor it
all right now. Perhaps just adding a more specific type enum to the
network struct would be a good idea.
Yea that's what I meant. If the only thing that can reasonably have
type "network" then it's cool by me. We've got bigger fish to fry than
100% safety on this one.
--
-Jay
5:58 p.m.
New subject: [PATCH 5 of 5] FullVirt changes to VSMS
# HG changeset patch
# User Dan Smith <danms(a)us.ibm.com>
# Date 1200506277 28800
# Node ID 2f141015fd40a4766dd9057dc567159d446da6b8
# Parent 81149ab841550817357e38e7d6c88c3049a0fd32
FullVirt changes to VSMS
This includes checking the isFullVirt flag in Xen VSSDs, and assuming it
for KVM. Also, support creating Xen-type bridge interfaces and KVM-type
virtual network interfaces.
Signed-off-by: Dan Smith <danms(a)us.ibm.com>
diff -r 81149ab84155 -r 2f141015fd40 src/Virt_VirtualSystemManagementService.c
--- a/src/Virt_VirtualSystemManagementService.c Wed Jan 16 09:36:15 2008 -0800
+++ b/src/Virt_VirtualSystemManagementService.c Wed Jan 16 09:57:57 2008 -0800
@@ -127,12 +127,73 @@ static CMPIStatus define_system_parse_ar
return s;
}
+static int xenpv_vssd_to_domain(CMPIInstance *inst,
+ struct domain *domain)
+{
+ int ret;
+ const char *val;
+
+ domain->type = DOMAIN_XENPV;
+
+ ret = cu_get_str_prop(inst, "Bootloader", &val);
+ if (ret != CMPI_RC_OK)
+ val = "";
+
+ free(domain->bootloader);
+ domain->bootloader = strdup(val);
+
+ ret = cu_get_str_prop(inst, "BootloaderArgs", &val);
+ if (ret != CMPI_RC_OK)
+ val = "";
+
+ free(domain->bootloader_args);
+ domain->bootloader_args = strdup(val);
+
+ return 1;
+}
+
+static int fv_vssd_to_domain(CMPIInstance *inst,
+ struct domain *domain,
+ const char *pfx)
+{
+ int ret;
+ const char *val;
+
+ if (STREQC(pfx, "KVM"))
+ domain->type = DOMAIN_KVM;
+ else if (STREQC(pfx, "Xen"))
+ domain->type = DOMAIN_XENFV;
+ else {
+ CU_DEBUG("Unknown fullvirt domain type: %s", pfx);
+ return 0;
+ }
+
+ ret = cu_get_str_prop(inst, "BootDevice", &val);
+ if (ret != CMPI_RC_OK)
+ val = "hd";
+
+ free(domain->os_info.fv.boot);
+ domain->os_info.fv.boot = strdup(val);
+
+ return 1;
+}
+
static int vssd_to_domain(CMPIInstance *inst,
struct domain *domain)
{
uint16_t tmp;
int ret = 0;
const char *val;
+ const char *cn;
+ char *pfx = NULL;
+ bool fullvirt;
+
+ cn = CLASSNAME(CMGetObjectPath(inst, NULL));
+ pfx = class_prefix_name(cn);
+ if (pfx == NULL) {
+ CU_DEBUG("Unknown prefix for class: %s", cn);
+ return 0;
+ }
ret = cu_get_str_prop(inst, "VirtualSystemIdentifier", &val);
if (ret != CMPI_RC_OK)
@@ -153,23 +214,41 @@ static int vssd_to_domain(CMPIInstance *
domain->on_crash = (int)tmp;
- ret = cu_get_str_prop(inst, "Bootloader", &val);
- if (ret != CMPI_RC_OK)
- val = "";
-
- free(domain->bootloader);
- domain->bootloader = strdup(val);
-
- ret = cu_get_str_prop(inst, "BootloaderArgs", &val);
- if (ret != CMPI_RC_OK)
- val = "";
-
- free(domain->bootloader_args);
- domain->bootloader_args = strdup(val);
-
- ret = 1;
- out:
+ if (STREQC(pfx, "KVM"))
+ fullvirt = true;
+ else if (cu_get_bool_prop(inst, "isFullVirt", &fullvirt) !=
CMPI_RC_OK)
+ fullvirt = false;
+
+ if (fullvirt)
+ ret = fv_vssd_to_domain(inst, domain, pfx);
+ else
+ ret = xenpv_vssd_to_domain(inst, domain);
+
+ out:
+ free(pfx);
+
return ret;
+}
+
+static int xen_net_rasd_to_vdev(CMPIInstance *inst,
+ struct virt_device *dev)
+{
+ free(dev->dev.net.type);
+ dev->dev.net.type = strdup("bridge");
+
+ return 1;
+}
+
+static int kvm_net_rasd_to_vdev(CMPIInstance *inst,
+ struct virt_device *dev)
+{
+ free(dev->dev.net.type);
+ dev->dev.net.type = strdup("network");
+
+ free(dev->dev.net.source);
+ dev->dev.net.source = strdup("default");
+
+ return 1;
}
static int rasd_to_vdev(CMPIInstance *inst,
@@ -211,11 +290,14 @@ static int rasd_to_vdev(CMPIInstance *in
free(dev->dev.net.mac);
dev->dev.net.mac = devid;
- if (cu_get_str_prop(inst, "NetworkType", &val) !=
CMPI_RC_OK)
- val = "bridge";
-
- free(dev->dev.net.type);
- dev->dev.net.type = strdup(val);
+ if (STARTS_WITH(CLASSNAME(op), "Xen"))
+ xen_net_rasd_to_vdev(inst, dev);
+ else if (STARTS_WITH(CLASSNAME(op), "KVM"))
+ kvm_net_rasd_to_vdev(inst, dev);
+ else
+ CU_DEBUG("Unknown class type for net device: %s",
+ CLASSNAME(op));
+
} else if (type == VIRT_DEV_MEM) {
cu_get_u64_prop(inst, "VirtualQuantity",
&dev->dev.mem.size);
cu_get_u64_prop(inst, "Reservation",
&dev->dev.mem.size);
6091
days inactive
6091
days old
10 comments
3 participants
participants (3)
-
Dan Smith -
Jay Gagnon -
Kaitlin Rupert