gmane.network.quagga.devel

[quagga-dev 6268] Re: [PATCH 19/21] Manage system routes withzebra

Jeremy Jackson — 2008-12-04T02:45:56

On Wed, 2008-12-03 at 14:23 -0800, Stephen Hemminger wrote: There is also the issue of storing the statements in the config files with "wr m", if the route exists when quagga starts, the config file would need a "no ip route ..." to delete it, does a "no ip route" get saved?

[quagga-dev 6267] Re: [PATCH 19/21] Manage system routes with zebra

Joakim Tjernlund — 2008-12-04T00:25:22

So it seems. Would it be acceptable to add this to ip route? Perhaps one can extend the new -S option include kernel routes too, that is, kernel and static routes are the same? Jocke

[quagga-dev 6266] Re: [PATCH 19/21] Manage system routes with zebra

Stephen Hemminger — 2008-12-03T22:23:46

On Wed, 3 Dec 2008 22:19:34 +0100 "Joakim Tjernlund" transmode.se> wrote: the problem is that vtysh version of 'ip route' doesn't have enough options to specify the full state of the imported routes. It needs scope, protocol and metric to match.

[quagga-dev 6265] Re: [PATCH 19/21] Manage system routes with zebra

Joakim Tjernlund — 2008-12-03T21:19:34

So I tested it a bit and I can't change/delete kernel routes(not with ip route anyway) such as default gateway. I think it would be an useful addition if one could do that too. Jocke

[quagga-dev 6264] Re: [PATCH 20/21] Only announce connected routesif link is detected

Joakim Tjernlund — 2008-12-03T19:36:27

Exactly, OSPF has to follow up and propagate a /32 address in its LSA. It might not solve all routing problems, but I can't see how it could make things worse. Jocke

[quagga-dev 6263] Re: [PATCH 20/21] Only announce connected routes if link is detected

2008-12-03T18:55:20

Things like OSPF go by the interface list, not connected routes. So the change wouldn't affect that immediately. Indeed, I think OSPF might continue to list the connected subnet in its router-LSA - which would continue to cause the problems Stephen is trying to avoid methinks. I think what you want is for routing protocols to treat inoperative interfaces as effectively loopback, and advertise only the address/32. regards,

[quagga-dev 6262] Re: [PATCH 20/21] Only announce connected routesif link is detected

Joakim Tjernlund — 2008-12-03T18:31:29

This sounds like a workaround, still don't understand why this is needed. Can anyone explain what the problem is with my suggestion? Jocke

[quagga-dev 6261] Re: [PATCH 20/21] Only announce connected routesif link is detected

Joakim Tjernlund — 2008-12-03T18:27:55

But the link is up, it is just not operative(=RUNNING). So you could propagate a /32 route so that anyone logged into the node on that particular IP address stays connected. Perhaps this is the wrong place to do this, I haven't looked into the code. Jocke

[quagga-dev 6260] Re: [PATCH 20/21] Only announce connected routes if link is detected

Vincent JARDIN — 2008-12-03T18:23:32

right, it should not be announced. If you want to reach the IP address from any interfaces, you should configure an 1.2.3.4/32 on a loopback or distribute a /32 Best regards, Vincent

[quagga-dev 6259] Re: [PATCH 20/21] Only announce connected routes if link is detected

Stephen Hemminger — 2008-12-03T17:59:05

On Wed, 03 Dec 2008 14:33:18 +0100 Joakim Tjernlund transmode.se> wrote: No. if link detect is enabled, then you don't want the connected route to be propogated unless link is up.

[quagga-dev 6258] Re: [PATCH 20/21] Only announce connectedroutesif link is detected

Joakim Tjernlund — 2008-12-03T13:33:18

hmm, should you not announce a host route iff if_is_up(ifp) && !if_is_operative(ifp) is true? The IP address on the interface is still reachable from other interfaces Jocke

[quagga-dev 6257] Re: quagga & mpls

James R. Leu — 2008-12-02T20:45:48

[quagga-dev 6256] Re: [PATCH 19/21] Manage system routes with zebra

Joakim Tjernlund — 2008-12-02T18:45:52

Oh, maybe I misread you. I want to change/delete the default route added at boot time by the kernel. Is that possible? jocke

[quagga-dev 6255] Re: [PATCH 19/21] Manage system routes with zebra

Stephen Hemminger — 2008-12-02T18:35:14

On Tue, 2 Dec 2008 19:20:15 +0100 "Joakim Tjernlund" transmode.se> wrote: You can delete and add routes like always. You might want to add more options to set scope and protocol

[quagga-dev 6254] Re: [PATCH 19/21] Manage system routes with zebra

Joakim Tjernlund — 2008-12-02T18:20:15

hmm, I had hoped I could change/delete routes too. Would that be difficult to add? Jocke

[quagga-dev 6253] quagga & mpls

Pali — 2008-12-02T17:06:48

[quagga-dev 6252] Re: [PATCH 19/21] Manage system routes with zebra

Stephen Hemminger — 2008-12-02T16:58:05

On Tue, 02 Dec 2008 11:16:33 -0500 Jeremy Jackson coplanar.net> wrote: Zebra just acts as a routing store. It collects all the routing changes on the system and if link detect is enabled, removes those routes, then restores them when link comes back.

[quagga-dev 6251] Re: [PATCH 04/21] Mark assert() tests as unlikely

Andrew J. Schorr — 2008-12-02T16:36:33

Glad to hear I'm not crazy. This still looks the same to me... We need to make that LIKELY, right? Regards, Andy

[quagga-dev 6250] Re: [PATCH 04/21] Mark assert() tests as unlikely

Stephen Hemminger — 2008-12-02T16:17:46

Thanks Andrew, your correct LIKELY test was backwards. Here is an corrected patch with one addition as well. ----------------------------------------------------- This forces compiler to generate the more common code on the faster path. Also don't bother with the old __ASSERT_FUNCTION since many other places in Quagga now depend on __func__ being correct.. --- lib/zassert.h | 14 +++++++++++--- 1 files changed, 11 insertions(+), 3 deletions(-) --- a/lib/zassert.h2008-12-02 08:01:47.000000000 -0800 +++ b/lib/zassert.h2008-12-02 08:13:41.000000000 -0800 < at >< at > -9,17 +9,16 < at >< at > extern void _zlog_assert_failed (const c unsigned int line, const char *function) __attribute__ ((noreturn)); -#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L -#define __ASSERT_FUNCTION __func__ -#elif defined(__GNUC__) -#define __ASSERT_FUNCTION __FUNCTION__ +#ifdef __GNUC__ +#define UNLIKELY(EX)__builtin_expect(!!(EX), 0) +#define LIKELY(EX)__builtin_expect(!!(EX), 1) #else -#define __ASSERT_FUNCTI

[quagga-dev 6249] Re: [PATCH 19/21] Manage system routeswithzebra

Jeremy Jackson — 2008-12-02T16:16:33

On Tue, 2008-12-02 at 14:46 +0100, Joakim Tjernlund wrote: How about different routing tables?

[quagga-dev 6248] Re: [PATCH 04/21] Mark assert() tests as unlikely

Andrew J. Schorr — 2008-12-02T14:26:30

Hi, On Fri, Aug 08, 2008 at 03:33:16PM -0700, Stephen Hemminger wrote: I'm probably confused, but shouldn't that be LIKELY instead of UNLIKELY? In the usual case, we do not expect to call _zlog_assert_failed. So we expect the test to be true and execute "0" instead of _zlog_assert_failed. Or is my head twisted on backwards? Regards, Andy