On 10/02/2017 11:05 AM, Michal Privoznik wrote:
https://bugzilla.redhat.com/show_bug.cgi?id=1497410
The comment in virNetDevTapInterfaceStats() implementation for
Linux states that packets transmitted by domain are received by
the host and vice versa. Well, this is true but not for all types
of interfaces. For instance, for macvtaps when TAP device is
hooked right onto a physical device any packet that domain sends
looks also like a packet sent to the host. Therefore, we should
have caller chose if the stats returned should be straight copy
s/have/allow/
s/chose/to choose
BTW: My first read of this thought "caller" is user configurable :-)
or swapped.
Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com>
---
src/conf/domain_conf.c | 2 +-
src/conf/domain_conf.h | 37 ++++++++++++++++-
src/libxl/libxl_driver.c | 3 +-
src/lxc/lxc_driver.c | 3 +-
src/openvz/openvz_driver.c | 3 +-
src/qemu/qemu_driver.c | 18 +++++----
src/util/virnetdevtap.c | 99 +++++++++++++++++++++++++++++++---------------
src/util/virnetdevtap.h | 3 +-
src/xen/xen_hypervisor.c | 2 +-
9 files changed, 123 insertions(+), 47 deletions(-)
diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c
index 2289399cd..6569b94fa 100644
--- a/src/conf/domain_conf.c
+++ b/src/conf/domain_conf.c
@@ -26729,7 +26729,7 @@ virDomainStateReasonFromString(virDomainState state, const char
*reason)
*/
virDomainNetType
-virDomainNetGetActualType(virDomainNetDefPtr iface)
+virDomainNetGetActualType(const virDomainNetDef *iface)
Sigh, separable along w/ domain_conf.h change...
{
if (iface->type != VIR_DOMAIN_NET_TYPE_NETWORK)
return iface->type;
diff --git a/src/conf/domain_conf.h b/src/conf/domain_conf.h
index 9ba84a94d..6eda21ef7 100644
--- a/src/conf/domain_conf.h
+++ b/src/conf/domain_conf.h
@@ -3021,7 +3021,7 @@ int virDomainGraphicsListenAppendSocket(virDomainGraphicsDefPtr
def,
const char *socket)
ATTRIBUTE_NONNULL(1);
-virDomainNetType virDomainNetGetActualType(virDomainNetDefPtr iface);
+virDomainNetType virDomainNetGetActualType(const virDomainNetDef *iface);
e.g. this is separate and unrelated needs it's own already R-B patch ;-)
The rest is "this" patch specific...
const char *virDomainNetGetActualBridgeName(virDomainNetDefPtr
iface);
int virDomainNetGetActualBridgeMACTableManager(virDomainNetDefPtr iface);
const char *virDomainNetGetActualDirectDev(virDomainNetDefPtr iface);
@@ -3393,4 +3393,39 @@ virDomainGenerateMachineName(const char *drivername,
int id,
const char *name,
bool privileged);
+/**
+ * virDomainNetTypeSharesHostView:
+ * @net: interface
+ *
+ * Some types of interfaces "share" the host view. For instance,
+ * for macvtap interface, every domain RX is the host RX too. And
+ * every domain TX is host TX too. IOW, for some types of
+ * interfaces guest and host are on the same side of RX/TX
+ * barrier. This is important so that we set up QoS correctly and
+ * report proper stats.
+ */
+static inline bool
+virDomainNetTypeSharesHostView(const virDomainNetDef *net)
+{
+ virDomainNetType actualType = virDomainNetGetActualType(net);
+ switch (actualType) {
+ case VIR_DOMAIN_NET_TYPE_DIRECT:
+ case VIR_DOMAIN_NET_TYPE_ETHERNET:
+ return true;
+ case VIR_DOMAIN_NET_TYPE_USER:
+ case VIR_DOMAIN_NET_TYPE_VHOSTUSER:
+ case VIR_DOMAIN_NET_TYPE_SERVER:
+ case VIR_DOMAIN_NET_TYPE_CLIENT:
+ case VIR_DOMAIN_NET_TYPE_MCAST:
+ case VIR_DOMAIN_NET_TYPE_NETWORK:
+ case VIR_DOMAIN_NET_TYPE_BRIDGE:
+ case VIR_DOMAIN_NET_TYPE_INTERNAL:
+ case VIR_DOMAIN_NET_TYPE_HOSTDEV:
+ case VIR_DOMAIN_NET_TYPE_UDP:
+ case VIR_DOMAIN_NET_TYPE_LAST:
+ break;
+ }
+ return false;
+}
+
#endif /* __DOMAIN_CONF_H */
[...]
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c
index 1ab16e57c..a46409d70 100644
--- a/src/qemu/qemu_driver.c
+++ b/src/qemu/qemu_driver.c
@@ -11040,7 +11040,8 @@ qemuDomainInterfaceStats(virDomainPtr dom,
if (virNetDevOpenvswitchInterfaceStats(path, stats) < 0)
goto cleanup;
} else {
- if (virNetDevTapInterfaceStats(path, stats) < 0)
+ if (virNetDevTapInterfaceStats(path, stats,
+ !virDomainNetTypeSharesHostView(net)) < 0)
goto cleanup;
}
@@ -19559,29 +19560,30 @@ qemuDomainGetStatsInterface(virQEMUDriverPtr driver
ATTRIBUTE_UNUSED,
/* Check the path is one of the domain's network interfaces. */
for (i = 0; i < dom->def->nnets; i++) {
+ virDomainNetDefPtr net = dom->def->nets[i];
virDomainNetType actualType;
- if (!dom->def->nets[i]->ifname)
+ if (!net->ifname)
continue;
memset(&tmp, 0, sizeof(tmp));
- actualType = virDomainNetGetActualType(dom->def->nets[i]);
+ actualType = virDomainNetGetActualType(net);
QEMU_ADD_NAME_PARAM(record, maxparams,
- "net", "name", i,
dom->def->nets[i]->ifname);
+ "net", "name", i, net->ifname);
if (actualType == VIR_DOMAIN_NET_TYPE_VHOSTUSER) {
- if (virNetDevOpenvswitchInterfaceStats(dom->def->nets[i]->ifname,
+ if (virNetDevOpenvswitchInterfaceStats(net->ifname,
&tmp) < 0) {
This line may as well move to previous line as well for just one line.
virResetLastError();
continue;
}
} else {
- if (virNetDevTapInterfaceStats(dom->def->nets[i]->ifname, &tmp)
< 0) {
+ if (virNetDevTapInterfaceStats(net->ifname, &tmp,
+ !virDomainNetTypeSharesHostView(net)) <
0)
virResetLastError();
- continue;
- }
+ continue;
Is this one right? We won't QEMU_ADD_NET_PARAM when actual_type !=
VIR_DOMAIN_NET_TYPE_VHOSTUSER *and* virNetDevTapInterfaceStats succeeds?
Looks like a bad transposition of extracted patches.
}
QEMU_ADD_NET_PARAM(record, maxparams, i,
diff --git a/src/util/virnetdevtap.c b/src/util/virnetdevtap.c
index 175dc2bfa..a3ed59da8 100644
--- a/src/util/virnetdevtap.c
+++ b/src/util/virnetdevtap.c
@@ -676,14 +676,27 @@ int virNetDevTapCreateInBridgePort(const char *brname,
}
/*-------------------- interface stats --------------------*/
-/* Just reads the named interface, so not Xen or QEMU-specific.
- * NB. Caller must check that libvirt user is trying to query
- * the interface of a domain they own. We do no such checking.
+
+/**
+ * virNetDevTapInterfaceStats:
+ * @ifname: interface
+ * @stats: where to store statistics
+ * @swapped: whether to swap RX/TX fields
+ *
+ * Fetch RX/TX statistics for given named interface (@ifname) and
+ * store them at @stats. The returned statistics are always from
+ * domain POV. Because in some cases this means swapping RX/TX in
+ * the stats and in others this means no swapping (consider TAP
+ * vs macvtap) caller might choose if the returned stats should
+ * be @swapped or not.
+ *
+ * Returns 0 on success, -1 otherwise (with error reported).
*/
#ifdef __linux__
int
virNetDevTapInterfaceStats(const char *ifname,
- virDomainInterfaceStatsPtr stats)
+ virDomainInterfaceStatsPtr stats,
+ bool swapped)
{
int ifname_len;
FILE *fp;
@@ -718,30 +731,35 @@ virNetDevTapInterfaceStats(const char *ifname,
*colon = '\0';
if (colon-ifname_len >= line &&
STREQ(colon-ifname_len, ifname)) {
- /* IMPORTANT NOTE!
- * /proc/net/dev vif<domid>.nn sees the network from the point
- * of view of dom0 / hypervisor. So bytes TRANSMITTED by dom0
- * are bytes RECEIVED by the domain. That's why the TX/RX fields
- * appear to be swapped here.
Maybe it's just a wording thing, but "dom0 / hypervisor" from the above
note don't directly correspond to "are always from domain POV" described
above - at least as how domain gets generally described... The command
is run on the host right? So let's just be clear.
Otherwise, looks fine to me with some adjustments as described above,
Reviewed-by: John Ferlan <jferlan(a)redhat.com>
John
[...]