It doesn't look like the disable_ipv6 devconf setting works
as described in ip-sysctl.txt. I think it was originally
designed to only be set by addrconf_dad_failure(), but since
it's mode 0644 it can be controlled through sysctl -w.
Here's an example:
# sysctl -w net.ipv6.conf.eth0.disable_ipv6=1
net.ipv6.conf.eth0.disable_ipv6 = 1
# sysctl -w net.ipv6.conf.eth0.accept_dad=2
net.ipv6.conf.eth0.accept_dad = 2
# ifdown eth0
# ifup eth0
Mar 5 11:23:45 titan kernel: eth0: duplicate address detected!
# ip -6 a show dev eth0
5: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qlen 1000
inet6 fe80::21f:29ff:fe5a:ef04/64 scope link tentative
valid_lft forever preferred_lft forever
The address didn't actually fail DAD, but since disable_ipv6
was set, addrconf_dad_timer() called addrconf_dad_failure()
before checking if it succeeded.
The following patch fixes the behavior. The other option is
to make this value read-only, but disable_ipv6 can be useful.
---
Fix the behavior of allowing both sysctl and addrconf_dad_failure()
to set the disable_ipv6 parameter without any bad side-effects.
If DAD fails and accept_dad > 1, we will still set disable_ipv6=1,
but then instead of allowing an RA to add an address then
immediately fail DAD, we simply don't allow the address to be
added in the first place. This also lets the user set this flag
and disable all IPv6 addresses on the interface, or on the entire
system.
Signed-off-by: Brian Haley <brian.haley@hp.com>
---
Documentation/networking/ip-sysctl.txt | 4 +++-
net/ipv6/addrconf.c | 21 ++++++++++++++-------
2 files changed, 17 insertions(+), 8 deletions(-)
diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index 7185e4c..ec5de02 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -1043,7 +1043,9 @@ max_addresses - INTEGER
Default: 16
disable_ipv6 - BOOLEAN
- Disable IPv6 ...
