Bug 48669
| Summary: | ifup setting IFF_* flags improperly, /sbin/ip not handling them | ||
|---|---|---|---|
| Product: | [Retired] Red Hat Linux | Reporter: | Pekka Savola <pekkas> |
| Component: | dhcpcd | Assignee: | Elliot Lee <sopwith> |
| Status: | CLOSED RAWHIDE | QA Contact: | Brock Organ <borgan> |
| Severity: | medium | Docs Contact: | |
| Priority: | high | ||
| Version: | 7.3 | CC: | msw, notting |
| Target Milestone: | --- | ||
| Target Release: | --- | ||
| Hardware: | i386 | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | Bug Fix | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2001-12-28 20:34:58 UTC | Type: | --- |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
pump has issues ATM. Looking into the other stuff. substitute however, of these, /sbin/ifup only supports arp. with however, of these, /sbin/ip only supports arp. if 1) is not making any sense. ;-) This defect considered MUST-FIX for Fairfax gold-release. 4) is worked around in beta3. 3): this only happens when DHCP is used. Statically configured, the interface is ok. I am not sure if this related or not but when using beta3 I get the following error message when bringing up an dhcp configured interface Determining IP information for eth0... failed; no link present. Check cable? This worked fine before upgrading to Beta3. Using dhcpcd (manually) instead of pump, I am able to get an address properly. Not directly related, see #49949 for your problem.. the NIC driver doesn't support MII properly. (The robustness of the check is a valid issue nonetheless). A good question is which interface flags you get when using dhcp manually. If you get the same as me with DHCP, this problem is probably with dhcpcd. Support for changing PROMISC & ALLMULTI flags added in iproute-2.2.4-14. Sorry for putting too many bugs in one report... 3), which is IMO most important, is still unsolved AFAIS. (also 2) but I don't think that's a show-stopper, especially if 3) is "fixed") Also, should initscripts Requires: iproute >= 2.2.4-14 ? (this is a marginal feature, so I'm not sure if the policy dictates this...) 3) is not an initscripts problem. According to the development staff, the original problem is fixed; there's a secondary problem that warrants Should-fix status. Thus, the tag for this defect is being changed. The legacy behaviour has been used for a very, very long time. I'm not sure what disabling trailers might do, but not having multicast enabled by default would definitely be a major drawback for many sites that do use it. So I'd like this to be a very strong SHOULD if not MUST. :-) FWIW, in regards to /sbin/ip: its docs say:
The {\tt ip} utility does not change the \verb|PROMISC|
or \verb|ALLMULTI| flags. These flags are considered
obsolete and should not be changed administratively.
According to the author, this should be enabled using packet socket, not a random
tool.
If only the world was perfect... For example xircom pcmcia cards have a bug (related to hardware I think) with handling of multicast, so that you must be in promisc or use allmulti for that to work... (e.g. ipv6 stuff doesn't work at all without allmulti flag on it). I don't see much need to set promisc administratively at network config scripts, but allmulti might be useful in scenarios like above. allmulti, AFAIK, doesn't even do much of anything. Well, for some drivers/hardware having problems with multicast, it apperently does :-) (I think it's like promiscuous mode but only for all multicast traffic, but I'm not sure...) Hmm, WRT #3, is it possible that the pump failures are caused by having the different interface flags set? Does anaconda use ifconfig and initscripts /sbin/ip for setting up the interface? That would explain why pump works in the installer but not on an installed system, maybe. I can't find out where exactly anaconda does its "turn on network for install" thing. IIRC, 3) happens because dhcpcd is used. With pump, I think it works all right (when pump works in the first place, that is). A problem here was, that the default changed from pump to dhcpcd due to pump failing mysteriously. I wonder if this would work now (post-roswell2), as initscripts has been fixed to assing 'scope host' to loopback interface. OK, I've totally forgotten the reasons for this bug still being open - can someone who remembers the problem describe what was remaining to be fixed, and whether the problem even exists in 7.2/rawhide? If dhcpcd is installed and BOOTPROTO=dhcp (so that it's used), interface flags after 'ifup eth0' will be: UP BROADCAST NOTRAILERS RUNNING Not: UP BROADCAST RUNNING MULTICAST The latter is how they're set with pump, and AFAIR how they've been set before RHL 7.2 with dhcpcd too (this might have something to do with switching from ifconfig to iproute). IMO, unless there are strong reason not to, the latter flags make much more sense. Well, as far as I can find, NOTRAILERS is a completely unused flag from BSD, so I'm leaving it there, but dhcpcd-1.3.21pl1-3 does now turn on the MULTICAST flag by default. |
initscripts 6.02. 1) /sbin/ifup performs: [ -n "${ARP}${PROMISC}${ALLMULTI}" ] && \ ip link set ${DEVICE} $(toggle_value arp $ARP) \ $(toggle_value promisc $PROMISC) \ $(toggle_value allmulti $ALLMULTI) however, of these, /sbin/ifup only supports arp. So, the fix is to hack /sbin/ip to include support for setting promisc,allmulti,trailers (not too difficult) or reverting back to the good ol' method using ifconfig and removing toggle_value. 2) There could be an interface hook also for: $MULTICAST $TRAILERS 3) Also, the default value for ethernet (at least) now seems to be: UP BROADCAST NOTRAILERS RUNNING MTU:1500 Metric:1 Where previously in RHL71 this was: UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 This is caused by /sbin ip link set eth0 up rather than ifconfig eth0 up. The latter is more correct and it's what other OS's (e.g. BSD) use too; at least multicast must be enabled. (Note: at least ifconfig has been tested to work rather ok on all kinds of mediums, like link layers that don't support multicast). 4) In DHCP environment at least, doing 'ifdown eth0' always gives 'Operation failed'. The interface has been disabled though. Doing 'ifdown eth0' does not fail anymore. Strace shows this is probably caused by pump, and happens after connect(3) to /var/run/pump.sock. Apparently releasing the lease of the interface didn't work properly.