Bug 1463077
| Summary: | Network disconnected by NetworkManager when bonding "updelay=100" specified for network install | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | Taketo Kabe <rkabe> |
| Component: | NetworkManager | Assignee: | Beniamino Galvani <bgalvani> |
| Status: | CLOSED ERRATA | QA Contact: | Desktop QE <desktop-qa-list> |
| Severity: | medium | Docs Contact: | |
| Priority: | unspecified | ||
| Version: | 7.3 | CC: | atragler, bgalvani, fgiudici, lrintel, rkhan, sukulkar, thaller, vbenes |
| Target Milestone: | rc | ||
| Target Release: | --- | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | NetworkManager-1.10.2-3.el7 | Doc Type: | If docs needed, set a value |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2018-04-10 13:27:23 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
| Attachments: | |||
Additional diagnostics and patch available at: https://bugs.centos.org/view.php?id=13423 Created attachment 1290569 [details] [PATCH] bond: ignore miimon option only when it is zero (In reply to Taketo Kabe from comment #2) > Additional diagnostics and patch available at: > https://bugs.centos.org/view.php?id=13423 While this looks correct, I think it's not necessary to always add miimon,updelay,downdelay to the generated connection. Updelay and downdelay should be ignored if they are set to the default value, and miimon should only be set when different from 0. How about this patch instead? Created attachment 1290574 [details]
[CI PATCH] bond: add tests for miimon option
CI tests for this bz
(In reply to Beniamino Galvani from comment #3) > Created attachment 1290569 [details] > [PATCH] bond: ignore miimon option only when it is zero > > (In reply to Taketo Kabe from comment #2) > > Additional diagnostics and patch available at: > > https://bugs.centos.org/view.php?id=13423 > > While this looks correct, I think it's not necessary to always add > miimon,updelay,downdelay to the generated connection. Updelay and downdelay > should be ignored if they are set to the default value, and miimon should > only be set when different from 0. How about this patch instead? lgtm (In reply to Beniamino Galvani from comment #3) > Created attachment 1290569 [details] > [PATCH] bond: ignore miimon option only when it is zero > > While this looks correct, I think it's not necessary to always add > miimon,updelay,downdelay to the generated connection. Updelay and downdelay > should be ignored if they are set to the default value, and miimon should > only be set when different from 0. How about this patch instead? Confirmed that this patch works with both "updelay=100" and "arp_interval=1000,arp_ip_target=10.6.0.1" modes; network installer doesn't freeze and NFS isn't disconnected. I also had the feeling that only miimon needs be treated specially. The code to not ignore all arp_interval/miimon/downdelay/updelay had been in initial commit c90730fa 2013/08/28, so the exact rationale is unknown. Just-to-be-sure defensive programming? (In reply to Taketo Kabe from comment #6) > Confirmed that this patch works with both > "updelay=100" and "arp_interval=1000,arp_ip_target=10.6.0.1" modes; > network installer doesn't freeze and NFS isn't disconnected. Thanks for trying this. > The code to not ignore all arp_interval/miimon/downdelay/updelay had been in > initial commit c90730fa 2013/08/28, so the exact rationale is unknown. > Just-to-be-sure defensive programming? Yes, I guess so. Patch applied to master and nm-1-8. https://cgit.freedesktop.org/NetworkManager/NetworkManager/commit/?id=92fc10918337f45ef5eb80bb663a8916894110fe Thanks for the detailed bug report! Created attachment 1364944 [details]
this doesn't work as those tests are written
Not sure if the fix is present in NetworkManager-1.10.0-1.el7.x86_64 All 3 test cases pass with NetworkManager-1.10.2-3.el7. 3 tests running in CI for some time already, working well, no issues seen. Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2018:0778 |
Created attachment 1289398 [details] Patch for NetworkManager Description of problem: Network disconnected by NetworkManager when bonding "updelay=100" specified for network install Version-Release number of selected component (if applicable): NetworkManager-1.4.0-20.el7_3 dracut-network-033-463.el7 How reproducible: always Steps to Reproduce: - Prepare an NFS server [ip=10.6.0.30] exporting /export/rhel/rhel-*-7.3-x86_64-dvd.iso . - Boot a network installation with kernel command line: net.ifnames=0 inst.repo=nfs:vers=4:10.6.0.30:/export/rhel ip=bond0:dhcp bond=bond0:eth0:mode=1,fail_over_mac=1,miimon=100,updelay=100 Actual results: tmux starts (sometimes not), and stage2 of anaconda doesn't start and freezes. Expected results: Anaconda stage2 runs, and anaconda greeter output and shell on tty2 should be available. Workaround: Specify non-default miimon value like bond=bond0:eth0:miimon=101,updelay=101 This will properly inherit the bond0 interface settings setup by dracut and anaconda could continue. Additional info: The bond0 interface seems to have lost its IP address by NetworkManager. To diagnose the problem, you can disable NetworkManager on boot to have a shell available: - boot with kernel command line net.ifnames=0 inst.repo=nfs:vers=4:10.6.0.30:/export/rhel ip=bond0:dhcp bond=bond0:eth0:mode=1,miimon=100,updelay=100 init=/bin/sh - After switch-root, it drops to shell. Disable NetworkManager and continue boot by: # ln -s /dev/null /etc/systemd/system/NetworkManager.service # equivalent of "systemctl mask NetworkManager" # exec /usr/lib/systemd/systemd --system - Eventually anaconda stage2 runs, and shell is available at tty2. Manually invoking NetworkManager from the prompt will break the network: # /usr/sbin/NetworkManager -d --log-level TRACE & Now, commands other than bash builtins stop responding since NFS connection is broken.