Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Andrew Morton]

On Sat, 14 Jul 2007 15:21:21 -0700 (PDT) bugme-daemon@bugzilla.kernel.org wrote:

> http://bugzilla.kernel.org/show_bug.cgi?id=8755
> 
>            Summary: "ip -6 route change " behaves like "ip -6 route add"
>            Product: Networking
>            Version: 2.5
>           Platform: All
>         OS/Version: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: IPV6
>         AssignedTo: yoshfuji@linux-ipv6.org
>         ReportedBy: bugzilla@fire.lp0.eu
> 
> 
> The "ip -6 route change" command behaves as though "ip -6 route add" was used.
> The source of iproute2 shows that "change" uses the "NLM_F_REPLACE" netlink
> flag only.
> 
> ip utility, iproute2-ss070313
> 
> Changing an existing route:
> # ip -6 r show 2002::/16
> 2002::/16 dev sit0  metric 1024  expires 4482618sec mtu 1480 advmss 7140
> hoplimit 4294967295
> # ip -6 r change 2002::/16 dev sit0 mtu 1280
> RTNETLINK answers: File exists
> 
> Adding a route using "change":
> # ip -6 r change 2002::/17 dev sit0 mtu 1280
> # ip -6 r show 2002::/17
> 2002::/17 dev sit0  metric 1024  expires 21334368sec mtu 1280 advmss 1220
> hoplimit 4294967295
> 

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Patrick McHardy]

Andrew Morton wrote:
> On Sat, 14 Jul 2007 15:21:21 -0700 (PDT) bugme-daemon@bugzilla.kernel.org wrote:
> 
>>The "ip -6 route change" command behaves as though "ip -6 route add" was used.
>>The source of iproute2 shows that "change" uses the "NLM_F_REPLACE" netlink
>>flag only.
>>
>>ip utility, iproute2-ss070313
>>
>>Changing an existing route:
>># ip -6 r show 2002::/16
>>2002::/16 dev sit0  metric 1024  expires 4482618sec mtu 1480 advmss 7140
>>hoplimit 4294967295
>># ip -6 r change 2002::/16 dev sit0 mtu 1280
>>RTNETLINK answers: File exists
>>
>>Adding a route using "change":
>># ip -6 r change 2002::/17 dev sit0 mtu 1280
>># ip -6 r show 2002::/17
>>2002::/17 dev sit0  metric 1024  expires 21334368sec mtu 1280 advmss 1220
>>hoplimit 4294967295


There is a difference between "add" and "change". "add" will only add a
new address if there isn't already one with the same identity, "change"
will allow to change attributes of an existing address, like flags,
lifetime, ..., or behave similar to "add" in case the address doesn't
already exist.

What kind of behaviour are you expecting?

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Simon Arlott]

On 15/07/07 16:07, Patrick McHardy wrote:
>>>Changing an existing route:
>>># ip -6 r show 2002::/16
>>>2002::/16 dev sit0  metric 1024  expires 4482618sec mtu 1480 advmss 7140
>>>hoplimit 4294967295
>>># ip -6 r change 2002::/16 dev sit0 mtu 1280
>>>RTNETLINK answers: File exists
^ This is clearly a bug, since I'm trying to change an existing route
and it gives an error as if it tried to add it.

>>>Adding a route using "change":
>>># ip -6 r change 2002::/17 dev sit0 mtu 1280
>>># ip -6 r show 2002::/17
>>>2002::/17 dev sit0  metric 1024  expires 21334368sec mtu 1280 advmss 1220
>>>hoplimit 4294967295
^ This shouldn't be possible!

> There is a difference between "add" and "change". "add" will only add a
> new address if there isn't already one with the same identity, "change"
> will allow to change attributes of an existing address, like flags,
> lifetime, ..., or behave similar to "add" in case the address doesn't
> already exist.
> 
> What kind of behaviour are you expecting?

That "change" would actually change not simply add. It won't let me
change anything and in fact can only add instead.

Compare it to ipv4 where "change" never adds - "replace" is "change, or
add". (Also, "replace" doesn't work for v6 either).

-- 
Simon Arlott

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Patrick McHardy]

Simon Arlott wrote:
> On 15/07/07 16:07, Patrick McHardy wrote:
> 
>>>>Changing an existing route:
>>>># ip -6 r show 2002::/16
>>>>2002::/16 dev sit0  metric 1024  expires 4482618sec mtu 1480 advmss 7140
>>>>hoplimit 4294967295
>>>># ip -6 r change 2002::/16 dev sit0 mtu 1280
>>>>RTNETLINK answers: File exists
> 
> ^ This is clearly a bug, since I'm trying to change an existing route
> and it gives an error as if it tried to add it.


Agreed.

>>>>Adding a route using "change":
>>>># ip -6 r change 2002::/17 dev sit0 mtu 1280
>>>># ip -6 r show 2002::/17
>>>>2002::/17 dev sit0  metric 1024  expires 21334368sec mtu 1280 advmss 1220
>>>>hoplimit 4294967295
> 
> ^ This shouldn't be possible!


That one will probably be impossible to fix since IPv6 has never checked
for NLM_F_EXCL, doing to now might break things.

>>There is a difference between "add" and "change". "add" will only add a
>>new address if there isn't already one with the same identity, "change"
>>will allow to change attributes of an existing address, like flags,
>>lifetime, ..., or behave similar to "add" in case the address doesn't
>>already exist.
>>
>>What kind of behaviour are you expecting?
> 
> 
> That "change" would actually change not simply add. It won't let me
> change anything and in fact can only add instead.


The code looks like it would support it properly. Please add a
few printks to inet6_rtm_newaddr to find out what goes wrong.

> Compare it to ipv4 where "change" never adds - "replace" is "change, or
> add". (Also, "replace" doesn't work for v6 either).


IPv4 doesn't check any netlink flags, so I don't think that is correct.

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Simon Arlott]

[-- Attachment #1: Type: text/plain, Size: 1635 bytes --]

On 16/07/07 14:01, Patrick McHardy wrote:
> Simon Arlott wrote:
>> On 15/07/07 16:07, Patrick McHardy wrote:
>>>>>Adding a route using "change":
>>>>># ip -6 r change 2002::/17 dev sit0 mtu 1280
>>>>># ip -6 r show 2002::/17
>>>>>2002::/17 dev sit0  metric 1024  expires 21334368sec mtu 1280 advmss 1220
>>>>>hoplimit 4294967295
>> 
>> ^ This shouldn't be possible!
> 
> 
> That one will probably be impossible to fix since IPv6 has never checked
> for NLM_F_EXCL, doing to now might break things.

NLM_F_EXCL isn't used:
   if (matches(*argv, "add") == 0)
      return iproute_modify(RTM_NEWROUTE, NLM_F_CREATE|NLM_F_EXCL,

   if (matches(*argv, "change") == 0 || strcmp(*argv, "chg") == 0)
      return iproute_modify(RTM_NEWROUTE, NLM_F_REPLACE,

   if (matches(*argv, "replace") == 0)
      return iproute_modify(RTM_NEWROUTE, NLM_F_CREATE|NLM_F_REPLACE,

"test" only uses NLM_F_EXCL... I didn't even know that existed and it's
not shown in help text.


> The code looks like it would support it properly. Please add a
> few printks to inet6_rtm_newaddr to find out what goes wrong.

Ok.


>> Compare it to ipv4 where "change" never adds - "replace" is "change, or
>> add". (Also, "replace" doesn't work for v6 either).
> 
> 
> IPv4 doesn't check any netlink flags, so I don't think that is correct.

It looks like it works perfectly for add, change, and replace:
# ip r add 2.0.0.0/32 dev lo
# ip r change 3.0.0.0/32 dev lo
RTNETLINK answers: No such file or directory
# ip r replace 3.0.0.0/32 dev lo
# ip r change 2.0.0.0/32 dev lo mtu 100
# ip r
3.0.0.0 dev lo  scope link
2.0.0.0 dev lo  scope link  mtu 100

-- 
Simon Arlott

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/x-pkcs7-signature, Size: 3623 bytes --]

Re: [Bugme-new] [Bug 8755] New: "ip -6 route change " behaves like "ip -6 route add"

[Simon Arlott]

[-- Attachment #1: Type: text/plain, Size: 1782 bytes --]

On 16/07/07 14:01, Patrick McHardy wrote:
> Simon Arlott wrote:
>>>>>Changing an existing route:
>>>>># ip -6 r show 2002::/16
>>>>>2002::/16 dev sit0  metric 1024  expires 4482618sec mtu 1480 advmss 7140
>>>>>hoplimit 4294967295
>>>>># ip -6 r change 2002::/16 dev sit0 mtu 1280
>>>>>RTNETLINK answers: File exists

> The code looks like it would support it properly. Please add a
> few printks to inet6_rtm_newaddr to find out what goes wrong.

Erm. It's routing configuration that's broken not address configuration
(ip doesn't even . Looking at route.c through to ip6_fib.c, it doesn't
support replacing routes at all.

$ grep NLM_F_ ipv4/fib_*.c ipv4/route.c ipv6/fib6_* ipv6/ip6_fib.c
ipv6/route.c

ipv4/fib_frontend.c:            cfg->fc_nlflags = NLM_F_CREATE;
ipv4/fib_frontend.c:            .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
ipv4/fib_hash.c:                if (cfg->fc_nlflags & NLM_F_EXCL)
ipv4/fib_hash.c:                if (cfg->fc_nlflags & NLM_F_REPLACE) {
ipv4/fib_hash.c:                                  &cfg->fc_nlinfo,
NLM_F_REPLACE);
ipv4/fib_hash.c:                if (!(cfg->fc_nlflags & NLM_F_APPEND))
ipv4/fib_hash.c:        if (!(cfg->fc_nlflags & NLM_F_CREATE))
ipv4/fib_hash.c:                                          NLM_F_MULTI) <
0) {
ipv4/fib_trie.c:                if (cfg->fc_nlflags & NLM_F_EXCL)
ipv4/fib_trie.c:                if (cfg->fc_nlflags & NLM_F_REPLACE) {
ipv4/fib_trie.c:                                tb->tb_id,
&cfg->fc_nlinfo, NLM_F_REPLACE);
ipv4/fib_trie.c:                if (!(cfg->fc_nlflags & NLM_F_APPEND))
ipv4/fib_trie.c:        if (!(cfg->fc_nlflags & NLM_F_CREATE))
ipv4/route.c:                                    1, NLM_F_MULTI) <= 0) {

ipv6/route.c:                prefix, NLM_F_MULTI);

-- 
Simon Arlott

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/x-pkcs7-signature, Size: 3623 bytes --]