Hi!
I observe an IP aliasing anomaly that occurs when I try to delete an IP
alias from an interface. That is, when I delete the first address in a
set of IP aliased addresses assigned according to a particular subnet,
rest of the aliases get deleted as well. Check out the below snippet.
$ *for I in `seq 1 6`; do sudo ip addr add 192.168.2.$I/29 dev eth0;
done*
$ ip addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
state UP qlen 1000
link/ether 00:24:54:b9:1c:f8 brd ff:ff:ff:ff:ff:ff
*inet 192.168.1.200/24 brd 192.168.1.255 scope global eth0**
inet 192.168.2.1/29 scope global eth0
inet 192.168.2.2/29 scope global secondary eth0
inet 192.168.2.3/29 scope global secondary eth0
inet 192.168.2.4/29 scope global secondary eth0
inet 192.168.2.5/29 scope global secondary eth0
inet 192.168.2.6/29 scope global secondary eth0*
inet6 fe80::224:54ff:feb9:1cf8/64 scope link
valid_lft forever preferred_lft forever
3: wlan0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN qlen 1000
link/ether e8:39:df:6a:21:2a brd ff:ff:ff:ff:ff:ff
$ *sudo ip addr del 192.168.2.1/29 dev eth0*
$ ip addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
state UP qlen 1000
link/ether 00:24:54:b9:1c:f8 brd ff:ff:ff:ff:ff:ff
*inet 192.168.1.200/24 brd 192.168.1.255 scope global eth0*
inet6 fe80::224:54ff:feb9:1cf8/64 scope link
valid_lft forever preferred_lft forever
3: wlan0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN qlen 1000
link/ether e8:39:df:6a:21:2a brd ff:ff:ff:ff:ff:ff
Per see, deleting 192.168.2.1/29 causes the rest of the aliased
interfaces get deleted as well. This is something that is slightly
documented in the ifconfig manual: /for every scope (i.e. same net with
address/netmask combination) all aliases are deleted, if you delete the
first (primary)/. So what is the right way of just deleting the first
(primary) alias without affecting the rest? If this is a scoping issue,
is it possible to assign each alias as primary within its own dedicated
scope?
As a side note, when I first asked this question to Stephen Hemminger
(he forwarded me to this mailing list) he also told me that "/In Linux
the interface aliases are really a legacy from the BSD style addressing,
and don't act the same. It is not common practice to use them./" Is that
really the case? Because, as you know, IP aliasing is the heart of a
majority of the high-availability and clustering solutions in Linux. Is
IP aliasing a really deprecated technology in Linux? Should we avoid
using it? If so, what do you recommend as an alternative?
Best.
On Mon, 2012-07-02 at 22:54 +0300, Volkan Yazıcı wrote:
Hi!
I observe an IP aliasing anomaly that occurs when I try to delete an IP
alias from an interface. That is, when I delete the first address in a
set of IP aliased addresses assigned according to a particular subnet,
rest of the aliases get deleted as well. Check out the below snippet.
[...]
As a side note, when I first asked this question to Stephen Hemminger
(he forwarded me to this mailing list) he also told me that "/In Linux
the interface aliases are really a legacy from the BSD style addressing,
and don't act the same. It is not common practice to use them./" Is that
really the case?
[...]
If you didn't give him the full details shown above, it's possible he
thought you meant alias interfaces such as 'eth0:0'.
Ben.
--
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
On Tue, 3 Jul 2012 02:12:17 +0100
Ben Hutchings [off-list ref] wrote:
On Mon, 2012-07-02 at 22:54 +0300, Volkan Yazıcı wrote:
quoted
Hi!
I observe an IP aliasing anomaly that occurs when I try to delete an IP
alias from an interface. That is, when I delete the first address in a
set of IP aliased addresses assigned according to a particular subnet,
rest of the aliases get deleted as well. Check out the below snippet.
[...]
quoted
As a side note, when I first asked this question to Stephen Hemminger
(he forwarded me to this mailing list) he also told me that "/In Linux
the interface aliases are really a legacy from the BSD style addressing,
and don't act the same. It is not common practice to use them./" Is that
really the case?
[...]
If you didn't give him the full details shown above, it's possible he
thought you meant alias interfaces such as 'eth0:0'.
Yes, that is what I was assuming by the word 'alias'
Hello,
On Mon, 2 Jul 2012, Volkan Yazıcı wrote:
Hi!
I observe an IP aliasing anomaly that occurs when I try to delete an IP alias
from an interface. That is, when I delete the first address in a set of IP
aliased addresses assigned according to a particular subnet, rest of the
aliases get deleted as well. Check out the below snippet.
This is in Linux may be from the 2.3/2.4 times
$ *for I in `seq 1 6`; do sudo ip addr add 192.168.2.$I/29 dev eth0;
done*
$ ip addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
state UP qlen 1000
link/ether 00:24:54:b9:1c:f8 brd ff:ff:ff:ff:ff:ff
*inet 192.168.1.200/24 brd 192.168.1.255 scope global eth0**
inet 192.168.2.1/29 scope global eth0
inet 192.168.2.2/29 scope global secondary eth0
inet 192.168.2.3/29 scope global secondary eth0
inet 192.168.2.4/29 scope global secondary eth0
inet 192.168.2.5/29 scope global secondary eth0
inet 192.168.2.6/29 scope global secondary eth0*
inet6 fe80::224:54ff:feb9:1cf8/64 scope link
valid_lft forever preferred_lft forever
3: wlan0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN qlen 1000
link/ether e8:39:df:6a:21:2a brd ff:ff:ff:ff:ff:ff
$ *sudo ip addr del 192.168.2.1/29 dev eth0*
$ ip addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue state UNKNOWN
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
state UP qlen 1000
link/ether 00:24:54:b9:1c:f8 brd ff:ff:ff:ff:ff:ff
*inet 192.168.1.200/24 brd 192.168.1.255 scope global eth0*
inet6 fe80::224:54ff:feb9:1cf8/64 scope link
valid_lft forever preferred_lft forever
3: wlan0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN qlen 1000
link/ether e8:39:df:6a:21:2a brd ff:ff:ff:ff:ff:ff
Per see, deleting 192.168.2.1/29 causes the rest of the aliased interfaces get
deleted as well. This is something that is slightly documented in the ifconfig
manual: /for every scope (i.e. same net with address/netmask combination) all
aliases are deleted, if you delete the first (primary)/. So what is the right
way of just deleting the first (primary) alias without affecting the rest? If
this is a scoping issue, is it possible to assign each alias as primary within
its own dedicated scope?
There is (yet) undocumented feature for the
interfaces:
/proc/sys/net/ipv4/conf/*/promote_secondaries
You set it for specific interface _or_ for "all".
It defaults to 0. When you enable it, deleting a primary
address will not delete all secondary addresses but will
change the next secondary as primary. The term alias may
refer for addresses on same interface while here the
problem is for a subset of addresses - from same subnet.
Primary is the first address added for the configured
subnet, all next addresses in subnet are added as
secondaries as shown by ip addr list.
As a side note, when I first asked this question to Stephen Hemminger (he
forwarded me to this mailing list) he also told me that "/In Linux the
interface aliases are really a legacy from the BSD style addressing, and don't
act the same. It is not common practice to use them./" Is that really the
case? Because, as you know, IP aliasing is the heart of a majority of the
high-availability and clustering solutions in Linux. Is IP aliasing a really
deprecated technology in Linux? Should we avoid using it? If so, what do you
recommend as an alternative?
Please do not stop using aliases!
Regards
--
Julian Anastasov [off-list ref]