Bug 116982

Summary:	ip address flush deadlocks doing netlink communication over and over on 2.6 kernel
Product:	[Fedora] Fedora	Reporter:	Arkadiusz Miskiewicz <arekm>
Component:	iproute	Assignee:	Phil Knirsch <pknirsch>
Status:	CLOSED UPSTREAM	QA Contact:	Brock Organ <borgan>
Severity:	medium	Docs Contact:
Priority:	medium
Version:	rawhide	CC:	alex.kiernan, rvokal
Target Milestone:	---
Target Release:	---
Hardware:	All
OS:	Linux
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2004-03-17 14:34:56 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:
Bug Depends On:
Bug Blocks:	114963

Description Arkadiusz Miskiewicz 2004-02-27 00:39:29 UTC

On 2.6 kernel iproute-2.4.7-11:

ip a a 192.168.0.1/24 dev eth0
ip link set eth0 down
ip a flush dev eth0

Here on my vanilla 2.6.2 it locks eating CPU - it does netlink 
communication over and over. This ,,hang'' doesn't happen when 
interface is in UP state. Also doesn't happen on 2.4 kernels.

This also could be kernel bug...

Comment 1 Bill Nottingham 2004-03-02 04:19:38 UTC

What adapter?

Comment 2 Arkadiusz Miskiewicz 2004-03-02 10:19:19 UTC

Doesn't really matter (but it's 3c905C-TX/TX-M using 3c59x and some 
RTL-8139/8139C/8139C+ using 8139too). It even hapens on lo device so 
should be easy to reproduce. ,,how to reproduce'' recipe doesn't work 
for you?

Comment 3 Arkadiusz Miskiewicz 2004-03-04 01:57:16 UTC

It seems that in ipaddr_list_or_flush() in for(;;) loop 
rtnl_dump_filter() function executes filter++ at each pass so it 
never leaves that for(;;) loop.

On 2.6 one thing happens different from 2.4 kernels.
rtnl_wilddump_request() sends request with rth->dump == 1078365203 
but gets answer with h->nlmsg_seq == 1078365202 and in next while 
(NLMSG_OK(h, status)) pass it gets the right one h->nlmsg_seq  == 
1078365203.

on 2.4 it always gets right reply. Maybe this has nothing to do with 
the problem or maybe it has.

Comment 4 Arkadiusz Miskiewicz 2004-03-04 22:26:33 UTC

For now I've just limited number of loop passed to 10k. (hack).

diff -urN iproute2.org/ip/ipaddress.c iproute2/ip/ipaddress.c
--- iproute2.org/ip/ipaddress.c 2004-03-04 23:00:41.050515248 +0100
+++ iproute2/ip/ipaddress.c     2004-03-04 23:08:08.810575433 +0100
@@ -603,7 +603,7 @@
                                fprintf(stderr, "Flush terminated\n")
;
                                exit(1);
                        }
-                       if (filter.flushed == 0) {
+                       if (filter.flushed == 0 || round > 10000) {
                                if (round == 0) {
                                        fprintf(stderr, "Nothing to 
flush.\n");
                                } else if (show_stats)

Comment 5 Arkadiusz Miskiewicz 2004-03-17 14:34:56 UTC

Fixed in kernel
http://oss.sgi.com/projects/netdev/archive/2004-03/msg00190.html