Bug 2097861 - iPXE causes an LACP flood
Summary: iPXE causes an LACP flood
Keywords:
Status: ASSIGNED
Alias: None
Product: Red Hat Enterprise Linux 8
Classification: Red Hat
Component: ipxe
Version: 8.4
Hardware: x86_64
OS: Linux
unspecified
unspecified
Target Milestone: rc
: ---
Assignee: Michal Schmidt
QA Contact: Oliver Gutiérrez
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2022-06-16 18:16 UTC by François Rigault
Modified: 2023-06-22 20:37 UTC (History)
9 users (show)

Fixed In Version:
Doc Type: No Doc Update
Doc Text:
Clone Of:
Environment:
Last Closed:
Type: Bug
Target Upstream Version:
Embargoed:


Attachments (Terms of Use)
Unknown_krgfs_crash (29.51 KB, text/plain)
2022-09-08 07:06 UTC, François Rigault
no flags Details


Links
System ID Private Priority Status Summary Last Updated
Github ipxe ipxe issues 157 0 None closed iPXE causes an LACP flood, DoS-ing the control plane of the switch 2022-06-16 18:16:24 UTC
Red Hat Issue Tracker RHELPLAN-125522 0 None None None 2022-06-16 18:24:27 UTC

Description François Rigault 2022-06-16 18:16:24 UTC
Description of problem:
https://github.com/ipxe/ipxe/issues/157
iPXE floods the physical switch with LACP frames, fails to configure network interface.

Version-Release number of selected component (if applicable):
from RHOSP16.2.2
$ sudo podman exec -ti ironic_pxe_tftp rpm -q ipxe-bootimgs
ipxe-bootimgs-20181214-8.git133f4c47.el8.noarch

How reproducible:
I need
- a Dell PowerEdge R830, configured for PXE and UEFI boot
- Mellanox ConnectX-3 cards
- connected to 2 switches with active LACP bonding


Steps to Reproduce:
1. openstack baremetal introspection start <node>
2. 
3.

Actual results:
ipxe.efi is properly downloaded, but then the interface fails to configure.
The interface on the switch goes to errdisabled because lacp-rate-limit
tcpdump on the switch reveals the same trace as in the github ticket: switch being flooded with LACP frames.

Expected results:
should work


Additional info:
it works replacing the ipxe.efi from ipxe-bootimgs-20181214-8.git133f4c47.el8.noarch with one I built myself that includes more recent sources + a patch to disable NET_PROTO_LACP 
(https://koji.fedoraproject.org/koji/taskinfo?taskID=88334805)

Comment 21 Oliver Gutiérrez 2022-09-06 15:54:23 UTC
Hi François. As the bug has been fixed in ipxe-20181214-10.git133f4c47.el8, could you check if the problem is fixed using this version?

If it is, please report us any information that could be useful to mark this bug as verified.

Thanks!

Comment 22 François Rigault 2022-09-07 11:49:40 UTC
Thanks Oliver, unfortunately I don't have a way to retrieve this artifact.

Comment 23 Oliver Gutiérrez 2022-09-07 12:12:08 UTC
Jarod: Could you give François a public link to download the fixed version package and do the testing for the package?

Comment 24 Jarod Wilson 2022-09-08 03:14:35 UTC
Packages temporarily available here: https://people.redhat.com/~jwilson/testing/ipxe/

Comment 25 François Rigault 2022-09-08 07:06:28 UTC
Created attachment 1910398 [details]
Unknown_krgfs_crash

used the ipxe-snponly-x86_64.efi to deploy 2 servers:

- 1 unrelated machine: deploys fine
- 1 machine with the ConnectX-3 pro -> crash, log attached.

Comment 26 Oliver Gutiérrez 2022-09-26 14:51:51 UTC
Hi @frigo . 

The crash you specify in this test is something that happened also with previous version or only with this new one?

Comment 28 François Rigault 2022-09-26 15:06:47 UTC
hi! no I have never seen this crash before, it is specific to this new version.

Comment 31 Jarod Wilson 2022-09-29 20:01:05 UTC
Does the package you built yourself (mentioned in the original report description) work on that host where this build failed, or does that fail similarly?

Comment 33 François Rigault 2022-09-29 20:42:55 UTC
the package I built myself works on that host flawlessly yes.

I understand this requires a specific type of hardware to reproduce. I can live with my hacks if fixing it is not trivial.

Comment 35 Jarod Wilson 2022-09-29 21:47:01 UTC
(In reply to François Rigault from comment #33)
> the package I built myself works on that host flawlessly yes.
> 
> I understand this requires a specific type of hardware to reproduce. I can
> live with my hacks if fixing it is not trivial.

It sounds like it would require tracking down the specific changes from upstream ipxe that make this work with cx-3 or a full rebase (which we generally frown upon doing at any point in a major release, let alone this late), and without a readily available reproducer and some time and effort (I'm lacking in time), I think we may well have to revert that one patch and revisit this for the next minor release.


Note You need to log in before you can comment on or make changes to this bug.