On 04/23/2013 05:30 AM, cngesaint@outlook.com wrote:

From: Xu Wang <cngesaint@outlook.com>

Original code will report xml text missing when a disk is not accessable, make user confuse. This patch will report the real error to tip user check

s/accessable/accessible/

its system health state on the server.

Can you provide an example test or command - so that it's "testable"? Whether that's by adding a new cimtest or some other means. There seems to be two errors serviced by DISK_UNKNOWN - the first one is a failure on a 'stat64()' and the second is the st_mode not being a BLK device, a REG (file), or a DIR (file system). How are they differentiated? Seems to me earlier checks should determine that the path doesn't exist while this check should be limited to invalid format. My other experience with CIM enum's is that there's supposed to be an "UNKNOWN" and "OTHER" values, where UNKNOWN was always 0 and OTHER was always 1. I may be the "OTHER" name space incorrect it's been a while...

Signed-off-by: Xu Wang <cngesaint@outlook.com> --- src/Virt_VirtualSystemManagementService.c | 65 +++++++++++++++++++++-------- 1 files changed, 48 insertions(+), 17 deletions(-)

diff --git a/src/Virt_VirtualSystemManagementService.c b/src/Virt_VirtualSystemManagementService.c index 4e93ef0..d252e12 100644 --- a/src/Virt_VirtualSystemManagementService.c +++ b/src/Virt_VirtualSystemManagementService.c @@ -964,11 +964,13 @@ static const char *net_rasd_to_vdev(CMPIInstance *inst, }

static const char *disk_rasd_to_vdev(CMPIInstance *inst, - struct virt_device *dev) + struct virt_device *dev, + char **p_error) { const char *val = NULL; uint16_t type; bool read = false; + int rc;

CU_DEBUG("Enter disk_rasd_to_vdev"); if (cu_get_str_prop(inst, "VirtualDevice", &val) != CMPI_RC_OK) @@ -984,6 +986,17 @@ static const char *disk_rasd_to_vdev(CMPIInstance *inst, dev->dev.disk.source = strdup(val); dev->dev.disk.disk_type = disk_type_from_file(val);

+ if (dev->dev.disk.disk_type == DISK_UNKNOWN) { + /* on success or fail caller should try free it */ + rc = asprintf(p_error, "Device %s, Address %s, " + "make sure Address can be accessed on host system.", + dev->dev.disk.virtual_dev, dev->dev.disk.source);

There's no error checking on whether the strdup()'s succeeded and thus this could cause problems with %s and NULL strings. For that matter there's very little error checking w/r/t strdup() failures so you're following existing potential issues...

+ if (rc == -1) { + CU_DEBUG("error during recording exception!");

Since asprintf() says the parameter 1 is "undefined" if rc == -1, so be safe and set p_error to NULL again...

+ } + return "Can't get a valid disk type, ";

Looks like a cut-n-paste - just snip the ", ". Other error returns don't use the ", " list marker...

+ } + if (cu_get_u16_prop(inst, "EmulatedType", &type) != CMPI_RC_OK) type = VIRT_DISK_TYPE_DISK;

@@ -1452,10 +1465,11 @@ static const char *input_rasd_to_vdev(CMPIInstance *inst, static const char *_sysvirt_rasd_to_vdev(CMPIInstance *inst, struct virt_device *dev, uint16_t type, - const char *ns) + const char *ns, + char **p_error) { if (type == CIM_RES_TYPE_DISK) { - return disk_rasd_to_vdev(inst, dev); + return disk_rasd_to_vdev(inst, dev, p_error); } else if (type == CIM_RES_TYPE_NET) { return net_rasd_to_vdev(inst, dev, ns); } else if (type == CIM_RES_TYPE_MEM) { @@ -1494,7 +1508,8 @@ static const char *_container_rasd_to_vdev(CMPIInstance *inst, static const char *rasd_to_vdev(CMPIInstance *inst, struct domain *domain, struct virt_device *dev, - const char *ns) + const char *ns, + char **p_error) { uint16_t type; CMPIObjectPath *op; @@ -1516,7 +1531,7 @@ static const char *rasd_to_vdev(CMPIInstance *inst, if (domain->type == DOMAIN_LXC) msg = _container_rasd_to_vdev(inst, dev, type, ns); else - msg = _sysvirt_rasd_to_vdev(inst, dev, type, ns); + msg = _sysvirt_rasd_to_vdev(inst, dev, type, ns, p_error); out: if (msg && op) CU_DEBUG("rasd_to_vdev(%s): %s", CLASSNAME(op), msg); @@ -1560,7 +1575,8 @@ static char *add_device_nodup(struct virt_device *dev,

static const char *classify_resources(CMPIArray *resources, const char *ns, - struct domain *domain) + struct domain *domain, + char **p_error) { int i; uint16_t type; @@ -1613,13 +1629,15 @@ static const char *classify_resources(CMPIArray *resources, msg = rasd_to_vdev(inst, domain, &domain->dev_vcpu[0], - veillard@redhat.com ns); + ns, + p_error); } else if (type == CIM_RES_TYPE_MEM) { domain->dev_mem_ct = 1; msg = rasd_to_vdevvirQEMUDriverCreateCapabilities(inst, domain, &domain->dev_mem[0], - ns); + ns, + p_error); } else if (type == CIM_RES_TYPE_DISK) { struct virt_device dev; int dcount = count + domain->dev_disk_ct; @@ -1628,7 +1646,8 @@ static const char *classify_resources(CMPIArray *resources, msg = rasd_to_vdev(inst, domain, &dev, - ns); + ns, + p_error); if (msg == NULL) msg = add_device_nodup(&dev, domain->dev_disk, @@ -1646,7 +1665,8 @@ static const char *classify_resources(CMPIArray *resources, msg = rasd_to_vdev(inst, domain, &dev, - ns); + ns, + p_error); if (msg == NULL) msg = add_device_nodup(&dev, domain->dev_net, @@ -1676,7 +1696,8 @@ static const char *classify_resources(CMPIArray *resources, msg = rasd_to_vdev(inst, domain, &dev, - ns); + ns, + p_error); if (msg == NULL) msg = add_device_nodup(&dev, domain->dev_graphics, @@ -1687,7 +1708,8 @@ static const char *classify_resources(CMPIArray *resources, msg = rasd_to_vdev(inst, domain, &domain->dev_input[0], - ns); + ns, + p_error); } if (msg != NULL) return msg; @@ -2083,6 +2105,7 @@ static CMPIInstance *create_system(const CMPIContext *context, struct inst_list list; const char *props[] = {NULL}; struct domain *domain = NULL; + char *error_msg = NULL;

inst_list_init(&list);

@@ -2113,12 +2136,13 @@ static CMPIInstance *create_system(const CMPIContext *context, if (s->rc != CMPI_RC_OK) goto out;

- msg = classify_resources(resources, NAMESPACE(ref), domain); + msg = classify_resources(resources, NAMESPACE(ref), domain, &error_msg); if (msg != NULL) { - CU_DEBUG("Failed to classify resources: %s", msg); + CU_DEBUG("Failed to classify resources: %s, %s", + msg, error_msg);

Since error_msg could be NULL - it should be handled...

cu_statusf(_BROKER, s, CMPI_RC_ERR_FAILED, - "ResourceSettings Error: %s", msg); + "ResourceSettings Error: %s, %s", msg, error_msg);

Same here...

goto out; }

@@ -2159,6 +2183,7 @@ static CMPIInstance *create_system(const CMPIContext *context,

out: + free(error_msg); cleanup_dominfo(&domain); free(xml); inst_list_free(&list); @@ -2638,6 +2663,7 @@ static CMPIStatus resource_add(struct domain *dominfo, struct virt_device *dev; int *count = NULL; const char *msg = NULL; + char *error_msg = NULL;

op = CMGetObjectPath(rasd, &s); if ((op == NULL) || (s.rc != CMPI_RC_OK)) @@ -2677,7 +2703,7 @@ static CMPIStatus resource_add(struct domain *dominfo, dev = &list[*count];

dev->type = type; - msg = rasd_to_vdev(rasd, dominfo, dev, ns); + msg = rasd_to_vdev(rasd, dominfo, dev, ns, &error_msg); if (msg != NULL) { cu_statusf(_BROKER, &s, CMPI_RC_ERR_FAILED,

Why not add the "error_msg" output here too like create_system?

@@ -2702,6 +2728,8 @@ static CMPIStatus resource_add(struct domain *dominfo, (*count)++;

out: + free(error_msg); +virQEMUDriverCreateCapabilities return s; }

@@ -2718,6 +2746,7 @@ static CMPIStatus resource_mod(struct domain *dominfo, int *count; int i; const char *msg = NULL; + char *error_msg = NULL;

CU_DEBUG("Enter resource_mod"); if (devid == NULL) { @@ -2749,7 +2778,7 @@ static CMPIStatus resource_mod(struct domain *dominfo, struct virt_device *dev = &list[i];

if (STREQ(dev->id, devid)) { - msg = rasd_to_vdev(rasd, dominfo, dev, ns); + msg = rasd_to_vdev(rasd, dominfo, dev, ns, &error_msg); if (msg != NULL) { cu_statusf(_BROKER, &s, CMPI_RC_ERR_FAILED,

Same comment John

@@ -2793,6 +2822,8 @@ static CMPIStatus resource_mod(struct domain *dominfo, }

out: + free(error_msg); + return s; }

于 2013-4-26 0:42, John Ferlan 写道:

On 04/23/2013 05:30 AM, cngesaint@outlook.com wrote:

...

From: Wenchao Xia <xiawenc@linux.vnet.ibm.com>

Since in nested KVM, libvirt-cim doesn't handle it well now, add this option to make it run well with qemu wchi help development and test.

I think your commit message more simply stated is:

Allow libvirt-cim to be supported within a nested KVM environment in order to more easily develop and test various configurations

What happens if someone sets this in a non nested environment? I would think you'd want to test if the 'nested' property is set...

Hi, John The commit message is not clear, actually this patch provide a way to manually disable KVM and fall back to qemu(without usage of hardware acceleration module, kvm.ko), Since using KVM have some problem in libvirt-cim in nested KVM env(which is a test env, most issues happens in the xml gen in libvirt-cim and parsing in libvirt). In non nested environment, it is same, kvm acceleration is disabled.

...

Signed-off-by: Xu Wang <cngesaint@outlook.com> --- libvirt-cim.conf | 8 ++++++++ libxkutil/misc_util.c | 8 ++++++++ libxkutil/misc_util.h | 1 + src/Virt_VirtualSystemManagementService.c | 7 +++++++ 4 files changed, 24 insertions(+), 0 deletions(-)

diff --git a/libvirt-cim.conf b/libvirt-cim.conf index 37d7b0f..3244ee3 100644 --- a/libvirt-cim.conf +++ b/libvirt-cim.conf @@ -30,3 +30,11 @@ # Default value: NULL, that is not set. # # migrate_ssh_temp_key = "/root/vm_migrate_tmp_id_rsa"; + +# force_use_qemu (bool) +# Since in nested KVM, libvirt-cim doesn't handler it well now, so add this +# option to make it run well with qemu which help development and test. +# Possible values: {true,false} +# Default value: false +# +# force_use_qemu = false;

I suggest using the comments from above here too...

...

diff --git a/libxkutil/misc_util.c b/libxkutil/misc_util.c index 00eb4b1..4c0b0a1 100644 --- a/libxkutil/misc_util.c +++ b/libxkutil/misc_util.c @@ -227,6 +227,14 @@ static int is_read_only(void) return prop.value_bool; }

+bool get_force_use_qemu(void) +{ + static LibvirtcimConfigProperty prop = { + "force_use_qemu", CONFIG_BOOL, {0}, 0}; + libvirt_cim_config_get(&prop); + return prop.value_bool; +} + const char *get_mig_ssh_tmp_key(void) { static LibvirtcimConfigProperty prop = { diff --git a/libxkutil/misc_util.h b/libxkutil/misc_util.h index 0f52290..9e6b419 100644 --- a/libxkutil/misc_util.h +++ b/libxkutil/misc_util.h @@ -154,6 +154,7 @@ int virt_set_status(const CMPIBroker *broker,

/* get libvirt-cim config */ const char *get_mig_ssh_tmp_key(void); +bool get_force_use_qemu(void);

/* * Local Variables: diff --git a/src/Virt_VirtualSystemManagementService.c b/src/Virt_VirtualSystemManagementService.c index cbb646d..4e93ef0 100644 --- a/src/Virt_VirtualSystemManagementService.c +++ b/src/Virt_VirtualSystemManagementService.c @@ -394,6 +394,13 @@ static bool system_has_kvm(const char *pfx) virConnectPtr conn; char *caps = NULL; bool kvm = false; + bool force_use_qemu = get_force_use_qemu(); + + /* hack for nested KVM */ + if (force_use_qemu) { + CU_DEBUG("Enter force use qemu mode!"); + return false; + }

The above check is being done on the "local" system right? While the following check is being done on the "BROKER" system, right? Which may They both happen on local system.

not be the same as the "local" system? So where should the check of whether the system in which is executing libvirt-cim code be really made? IOW, are you sure this is the right place to check? What's the differentiation in libvirt-cim between QEMU & KVM being made for? Being "new" I thought they were the same.

I'm not even convinced the routine is doing the right thing, but perhaps I just don't have enough history. I see a virsh capabilities on my system returns some kvm defs and some qemu defs, so a blind strstr is returning true...

The prime differentiator is that the calling routine will set domain->type to QEMU now rather than KVM and I'm not sure I understand how by using that setting we'll be able support nested KVM. Support nested KVM will need more efforts, I think you have found the right place, there are some issues in following path, what I saw is libvirt rejected some xml generated. To avoid the troubles, so disable it first, to test libvirt-cim for qemu, for that most part are the same with KVM case,

John

...

conn = connect_by_classname(_BROKER, pfx, &s); if ((conn == NULL) || (s.rc != CMPI_RC_OK)) {

_______________________________________________ Libvirt-cim mailing list Libvirt-cim@redhat.com https://www.redhat.com/mailman/listinfo/libvirt-cim

-- Best Regards Wenchao Xia