The kernel's more broken than one would think. Various drivers report various (usually spurious) values if the interface is down. While on some we experience -EINVAL when read()-ing the speed sysfs file, with other drivers we might get anything from 0 to UINT_MAX. If that's the case it's better to not report link speed. Well, the interface is down anyway. Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com&gt; --- src/util/virnetdev.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) diff --git a/src/util/virnetdev.c b/src/util/virnetdev.c index 6f3a202..80ef572 100644 --- a/src/util/virnetdev.c +++ b/src/util/virnetdev.c @@ -1843,7 +1843,7 @@ virNetDevGetLinkInfo(const char *ifname, char *buf = NULL; char *tmp; int tmp_state; - unsigned int tmp_speed; + unsigned int tmp_speed; /* virInterfaceState */

if (virNetDevSysfsFile(&path, ifname, "operstate") < 0) goto cleanup; @@ -1875,6 +1875,16 @@ virNetDevGetLinkInfo(const char *ifname, lnk->state = tmp_state; + /* Shortcut to avoid some kernel issues. If link is down (and possibly in + * other states too) several drivers report several values. While igb + * reports 65535, realtek goes with 10. To avoid muddying XML with insane + * values, don't report link speed */ + if (lnk->state == VIR_INTERFACE_STATE_DOWN) { + lnk->speed = 0; + ret = 0; + goto cleanup; + } +

VIR_FREE(path); VIR_FREE(buf); @@ -1884,6 +1894,7 @@ virNetDevGetLinkInfo(const char *ifname, if (virFileReadAll(path, 1024, &buf) < 0) { /* Some devices doesn't report speed, in which case we get EINVAL */ if (errno == EINVAL) { + lnk->speed = 0; ret = 0; goto cleanup; } -- 1.8.5.5

On 06/13/2014 10:10 AM, Martin Kletzander wrote: > On Fri, Jun 13, 2014 at 08:46:59AM +0200, Michal Privoznik wrote: >> The kernel's more broken than one would think. Various drivers report >> various (usually spurious) values if the interface is down. While on >> some we experience -EINVAL when read()-ing the speed sysfs file, with >> other drivers we might get anything from 0 to UINT_MAX. If that's the >> case it's better to not report link speed. Well, the interface is down >> anyway. >> >> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com&gt; >> --- >> src/util/virnetdev.c | 13 ++++++++++++- >> 1 file changed, 12 insertions(+), 1 deletion(-) >> >> diff --git a/src/util/virnetdev.c b/src/util/virnetdev.c >> index 6f3a202..80ef572 100644 >> --- a/src/util/virnetdev.c >> +++ b/src/util/virnetdev.c >> @@ -1843,7 +1843,7 @@ virNetDevGetLinkInfo(const char *ifname, >> char *buf = NULL; >> char *tmp; >> int tmp_state; >> - unsigned int tmp_speed; >> + unsigned int tmp_speed; /* virInterfaceState */ >> > > You probably wanted to put this comment next to the line with > tmp_state and not tmp_speed. > >> if (virNetDevSysfsFile(&path, ifname, "operstate") < 0) >> goto cleanup; >> @@ -1875,6 +1875,16 @@ virNetDevGetLinkInfo(const char *ifname, >> >> lnk->state = tmp_state; >> >> + /* Shortcut to avoid some kernel issues. If link is down (and >> possibly in >> + * other states too) several drivers report several values. >> While igb >> + * reports 65535, realtek goes with 10. To avoid muddying XML >> with insane >> + * values, don't report link speed */ >> + if (lnk->state == VIR_INTERFACE_STATE_DOWN) { Also for VIR_INTERFACE_LOWER_LAYER_DOWN (verified by looking at the speed reported by a macvtap device when its physdev is down). And I'm not sure how to get an interface into "NOT_PRESENT" or "DORMANT" state, but I would imagine that the speed should be 0 in those cases too.

ACK with LOWER_LAYER_DOWN added (I won't insist on the others until/unless I see experimental evidence that they need it). BTW, thinking more about bridge devices - maybe they should be given state "up" if the device has been ifup'ed. In other words, in their case you could call the functional equivalent of if_is_active() in netcf (which does an SIOCGIFFLAGS ioctl and checks for the IFF_UP flag). (in any case, bridges should probably just always report a speed of 0).

On 06/13/2014 10:50 AM, Martin Kletzander wrote: > On Fri, Jun 13, 2014 at 10:31:53AM +0300, Laine Stump wrote: >> On 06/13/2014 10:10 AM, Martin Kletzander wrote: >>> On Fri, Jun 13, 2014 at 08:46:59AM +0200, Michal Privoznik wrote: >>>> The kernel's more broken than one would think. Various drivers report >>>> various (usually spurious) values if the interface is down. While on >>>> some we experience -EINVAL when read()-ing the speed sysfs file, with >>>> other drivers we might get anything from 0 to UINT_MAX. If that's the >>>> case it's better to not report link speed. Well, the interface is down >>>> anyway. >>>> >>>> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com&gt; >>>> --- >>>> src/util/virnetdev.c | 13 ++++++++++++- >>>> 1 file changed, 12 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/src/util/virnetdev.c b/src/util/virnetdev.c >>>> index 6f3a202..80ef572 100644 >>>> --- a/src/util/virnetdev.c >>>> +++ b/src/util/virnetdev.c >>>> @@ -1843,7 +1843,7 @@ virNetDevGetLinkInfo(const char *ifname, >>>> char *buf = NULL; >>>> char *tmp; >>>> int tmp_state; >>>> - unsigned int tmp_speed; >>>> + unsigned int tmp_speed; /* virInterfaceState */ >>>> >>> >>> You probably wanted to put this comment next to the line with >>> tmp_state and not tmp_speed. >>> >>>> if (virNetDevSysfsFile(&path, ifname, "operstate") < 0) >>>> goto cleanup; >>>> @@ -1875,6 +1875,16 @@ virNetDevGetLinkInfo(const char *ifname, >>>> >>>> lnk->state = tmp_state; >>>> >>>> + /* Shortcut to avoid some kernel issues. If link is down (and >>>> possibly in >>>> + * other states too) several drivers report several values. >>>> While igb >>>> + * reports 65535, realtek goes with 10. To avoid muddying XML >>>> with insane >>>> + * values, don't report link speed */ >>>> + if (lnk->state == VIR_INTERFACE_STATE_DOWN) { >> >> Also for VIR_INTERFACE_LOWER_LAYER_DOWN (verified by looking at the >> speed reported by a macvtap device when its physdev is down). And I'm >> not sure how to get an interface into "NOT_PRESENT" or "DORMANT" state, >> but I would imagine that the speed should be 0 in those cases too. >> > > I've seen many other states I have no idea how to achieve. Wouldn't > it make more sense to report the speed only if the state is UP? That makes enough sense to me that I changed my netcf patch to do just that - it sets speed to 0 unless operstate is "up". Still open for debate though :-) I just sent the patch to netcf-devel(a)lists.fedorahosted.org