Bug 8320 - replacing route in kernel doesn't send netlink message
Summary: replacing route in kernel doesn't send netlink message
Status: CLOSED CODE_FIX
Alias: None
Product: Networking
Classification: Unclassified
Component: IPV4 (show other bugs)
Hardware: i386 Linux
: P2 low
Assignee: Stephen Hemminger
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2007-04-11 02:36 UTC by Milan Kocian
Modified: 2007-07-18 05:45 UTC (History)
1 user (show)

See Also:
Kernel Version: 2.6.20.6
Subsystem:
Regression: ---
Bisected commit-id:


Attachments

Description Milan Kocian 2007-04-11 02:36:44 UTC
Most recent kernel where this bug did *NOT* occur: I think all 2.6 kernels
Distribution: Debian (with vanilla kernel)
Hardware Environment: PC
Software Environment: Debian
Problem Description:
When you replace route (via ip r r ), no netlink message is sent. Or is it feature?

Steps to reproduce:
1. run 'ip monitor all' on one console
2. do 'ip r r EXISTING_ROUTE via DST' on second console
3. no message on console one

Small patch for fib_hash (tested) but use carefully I am newbie :-) :

--- fib_hash.c.old      2007-04-11 10:39:34.895667672 +0200
+++ fib_hash.c  2007-04-11 10:41:34.623466280 +0200
@@ -457,6 +457,8 @@
                        fib_release_info(fi_drop);
                        if (state & FA_S_ACCESSED)
                                rt_cache_flush(-1);
+                       rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
+                                 &cfg->fc_nlinfo);
                        return 0;
                }

And for fib_trie (not tested):

--- fib_trie.c.old      2007-04-11 10:39:22.728517360 +0200
+++ fib_trie.c  2007-04-11 10:40:40.778651936 +0200
@@ -1205,6 +1205,8 @@
                        fib_release_info(fi_drop);
                        if (state & FA_S_ACCESSED)
                                rt_cache_flush(-1);
+                       rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
+                                 &cfg->fc_nlinfo);

                        goto succeeded;
                }
Comment 1 Anonymous Emailer 2007-04-11 09:47:44 UTC
Reply-To: akpm@linux-foundation.org

On Wed, 11 Apr 2007 02:37:01 -0700 bugme-daemon@bugzilla.kernel.org wrote:

> 
> http://bugzilla.kernel.org/show_bug.cgi?id=8320
> 
>            Summary: replacing route in kernel doesn't send netlink message
>     Kernel Version: 2.6.20.6
>             Status: NEW
>           Severity: low
>              Owner: shemminger@osdl.org
>          Submitter: milan.kocian@wq.cz
> 
> 
> Most recent kernel where this bug did *NOT* occur: I think all 2.6 kernels
> Distribution: Debian (with vanilla kernel)
> Hardware Environment: PC
> Software Environment: Debian
> Problem Description:
> When you replace route (via ip r r ), no netlink message is sent. Or is it feature?
> 
> Steps to reproduce:
> 1. run 'ip monitor all' on one console
> 2. do 'ip r r EXISTING_ROUTE via DST' on second console
> 3. no message on console one
> 
> Small patch for fib_hash (tested) but use carefully I am newbie :-) :
> 
> --- fib_hash.c.old      2007-04-11 10:39:34.895667672 +0200
> +++ fib_hash.c  2007-04-11 10:41:34.623466280 +0200
> @@ -457,6 +457,8 @@
>                         fib_release_info(fi_drop);
>                         if (state & FA_S_ACCESSED)
>                                 rt_cache_flush(-1);
> +                       rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
> +                                 &cfg->fc_nlinfo);
>                         return 0;
>                 }
> 
> And for fib_trie (not tested):
> 
> --- fib_trie.c.old      2007-04-11 10:39:22.728517360 +0200
> +++ fib_trie.c  2007-04-11 10:40:40.778651936 +0200
> @@ -1205,6 +1205,8 @@
>                         fib_release_info(fi_drop);
>                         if (state & FA_S_ACCESSED)
>                                 rt_cache_flush(-1);
> +                       rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
> +                                 &cfg->fc_nlinfo);
> 
>                         goto succeeded;
>                 }

Thanks.  We prefer to receive patches via email rather than via bugzilla. 
But that's a relatively minor matter - let's see what the net guys think
about the change first ;)

Comment 2 Patrick McHardy 2007-04-11 11:20:16 UTC
Andrew Morton wrote:
> On Wed, 11 Apr 2007 02:37:01 -0700 bugme-daemon@bugzilla.kernel.org wrote:
> 
> 
>>http://bugzilla.kernel.org/show_bug.cgi?id=8320
>>
>>           Summary: replacing route in kernel doesn't send netlink message
>>    Kernel Version: 2.6.20.6
>>            Status: NEW
>>          Severity: low
>>             Owner: shemminger@osdl.org
>>         Submitter: milan.kocian@wq.cz
>>
>>When you replace route (via ip r r ), no netlink message is sent. Or is it feature?
>>
>>Steps to reproduce:
>>1. run 'ip monitor all' on one console
>>2. do 'ip r r EXISTING_ROUTE via DST' on second console
>>3. no message on console one
>>
>>Small patch for fib_hash (tested) but use carefully I am newbie :-) :
>>
>>--- fib_hash.c.old      2007-04-11 10:39:34.895667672 +0200
>>+++ fib_hash.c  2007-04-11 10:41:34.623466280 +0200
>>@@ -457,6 +457,8 @@
>>                        fib_release_info(fi_drop);
>>                        if (state & FA_S_ACCESSED)
>>                                rt_cache_flush(-1);
>>+                       rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
>>+                                 &cfg->fc_nlinfo);
>>                        return 0;
>>                }


I think having notifications for this case makes sense (IIRC I used
to use a similar patch some time ago, but can't find it right now).
But we need to indicate somehow that it is a replacement and not a
completely new route, either by sending a RTM_DELROUTE for the old
route first (which would match what devinet does for addresses)
or by echoing the NLM_F_REPLACE flag. The former would probably be
easier for userspace to understand since it wouldn't need to
replicate the replacement logic just to find out which rule got
replaced.

Comment 3 Anonymous Emailer 2007-04-12 10:54:13 UTC
Reply-To: milon@wq.cz

On Wed, 2007-04-11 at 20:19 +0200, Patrick McHardy wrote:

> 
> I think having notifications for this case makes sense (IIRC I used
> to use a similar patch some time ago, but can't find it right now).
> But we need to indicate somehow that it is a replacement and not a
> completely new route, either by sending a RTM_DELROUTE for the old
> route first (which would match what devinet does for addresses)
> or by echoing the NLM_F_REPLACE flag. The former would probably be
> easier for userspace to understand since it wouldn't need to
> replicate the replacement logic just to find out which rule got
> replaced.
> 
> 

Hard to tell what is better. I slightly tried to test my patch with
quagga routing daemon. And then I tested second case: send RTM_DELROUTE
before RTM_NEWROUTE. Quagga updates internal rib in both cases (as I saw
in debug logs). I was in fear that quagga will try to install sefl route
but it doesn't catch.
So from my point of view is all the same :).

Comment 4 Patrick McHardy 2007-04-15 22:01:14 UTC
Milan Koci
Comment 5 Anonymous Emailer 2007-04-16 17:10:27 UTC
Reply-To: davem@davemloft.net

From: Patrick McHardy <kaber@trash.net>
Date: Mon, 16 Apr 2007 06:59:06 +0200

> RTM_DELROUTE + RTM_NEWROUTE seem to be safer, although you're correct
> that it might cause userspace to perform some action upon receiving
> the DELROUTE message since the update is non-atomic. So I really don't
> know, I'm in favour of having notifications for replacements, but I
> fear we might break something.

We can cry foul about a broken application if an application following
the API correctly would interpret the new messages correctly.

I think it doesn't make sense to do a delete then a newroute for
the atomicity issues, and therefore the replace makes the most
sense as long as existing correct uses of the API would not
explode on this.

Comment 6 Patrick McHardy 2007-04-17 05:58:54 UTC
David Miller wrote:
> From: Patrick McHardy <kaber@trash.net>
> Date: Mon, 16 Apr 2007 06:59:06 +0200
> 
> 
>>RTM_DELROUTE + RTM_NEWROUTE seem to be safer, although you're correct
>>that it might cause userspace to perform some action upon receiving
>>the DELROUTE message since the update is non-atomic. So I really don't
>>know, I'm in favour of having notifications for replacements, but I
>>fear we might break something.
> 
> 
> We can cry foul about a broken application if an application following
> the API correctly would interpret the new messages correctly.
> 
> I think it doesn't make sense to do a delete then a newroute for
> the atomicity issues, and therefore the replace makes the most
> sense as long as existing correct uses of the API would not
> explode on this.


They shouldn't, worst case is that they ignore NLM_F_REPLACE and treat
it as a completely new route, which is at least half way correct and
not really worse than today.

Milan, could you cook up another patch which uses NLM_F_REPLACE?

Comment 7 Anonymous Emailer 2007-04-18 05:48:35 UTC
Reply-To: milon@wq.cz

On Tue, 2007-04-17 at 14:58 +0200, Patrick McHardy wrote:
> David Miller wrote:
> > From: Patrick McHardy <kaber@trash.net>
> > Date: Mon, 16 Apr 2007 06:59:06 +0200
> > 
> > 
> >>RTM_DELROUTE + RTM_NEWROUTE seem to be safer, although you're correct
> >>that it might cause userspace to perform some action upon receiving
> >>the DELROUTE message since the update is non-atomic. So I really don't
> >>know, I'm in favour of having notifications for replacements, but I
> >>fear we might break something.
> > 
> > 
> > We can cry foul about a broken application if an application following
> > the API correctly would interpret the new messages correctly.
> > 
> > I think it doesn't make sense to do a delete then a newroute for
> > the atomicity issues, and therefore the replace makes the most
> > sense as long as existing correct uses of the API would not
> > explode on this.
>
> They shouldn't, worst case is that they ignore NLM_F_REPLACE and treat
> it as a completely new route, which is at least half way correct and
> not really worse than today.
> 
> Milan, could you cook up another patch which uses NLM_F_REPLACE?
> 

I can try it. Output is in patch below. Review carefully. I don't know
if it's best approach. It's tested and working without problem
(probably :-))

--- net/ipv4.old/fib_hash.c	2007-04-18 12:50:11.000000000 +0200
+++ net/ipv4/fib_hash.c	2007-04-18 12:39:49.081369320 +0200
@@ -443,7 +443,6 @@
 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
 			struct fib_info *fi_drop;
 			u8 state;
-
 			write_lock_bh(&fib_hash_lock);
 			fi_drop = fa->fa_info;
 			fa->fa_info = fi;
@@ -457,6 +456,8 @@
 			fib_release_info(fi_drop);
 			if (state & FA_S_ACCESSED)
 				rt_cache_flush(-1);
+			rtmsg_fib(RTM_NEWROUTE, key, fa, cfg->fc_dst_len, tb->tb_id,
+				  &cfg->fc_nlinfo, NLM_F_REPLACE);
 			return 0;
 		}
 
@@ -524,7 +525,7 @@
 	rt_cache_flush(-1);
 
 	rtmsg_fib(RTM_NEWROUTE, key, new_fa, cfg->fc_dst_len, tb->tb_id,
-		  &cfg->fc_nlinfo);
+		  &cfg->fc_nlinfo, 0);
 	return 0;
 
 out_free_new_fa:
@@ -590,7 +591,7 @@
 
 		fa = fa_to_delete;
 		rtmsg_fib(RTM_DELROUTE, key, fa, cfg->fc_dst_len,
-			  tb->tb_id, &cfg->fc_nlinfo);
+			  tb->tb_id, &cfg->fc_nlinfo, 0);
 
 		kill_fn = 0;
 		write_lock_bh(&fib_hash_lock);
--- net/ipv4.old/fib_trie.c	2007-04-18 12:50:11.000000000 +0200
+++ net/ipv4/fib_trie.c	2007-04-18 12:42:29.423993536 +0200
@@ -1205,6 +1205,9 @@
 			fib_release_info(fi_drop);
 			if (state & FA_S_ACCESSED)
 				rt_cache_flush(-1);
+			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
+				  &cfg->fc_nlinfo, NLM_F_REPLACE);
+
 			goto succeeded;
 		}
 		/* Error if we find a perfect match which
@@ -1256,7 +1259,7 @@
 
 	rt_cache_flush(-1);
 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
-		  &cfg->fc_nlinfo);
+		  &cfg->fc_nlinfo, 0);
 succeeded:
 	return 0;
 
@@ -1599,7 +1602,7 @@
 
 	fa = fa_to_delete;
 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
-		  &cfg->fc_nlinfo);
+		  &cfg->fc_nlinfo, 0);
 
 	l = fib_find_node(t, key);
 	li = find_leaf_info(l, plen);
--- net/ipv4.old/fib_semantics.c	2007-04-18 12:50:11.000000000 +0200
+++ net/ipv4/fib_semantics.c	2007-04-18 12:40:54.807377448 +0200
@@ -301,7 +301,7 @@
 }
 
 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
-	       int dst_len, u32 tb_id, struct nl_info *info)
+	       int dst_len, u32 tb_id, struct nl_info *info, unsigned int
nlm_flags)
 {
 	struct sk_buff *skb;
 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
@@ -313,7 +313,7 @@
 
 	err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 			    fa->fa_type, fa->fa_scope, key, dst_len,
-			    fa->fa_tos, fa->fa_info, 0);
+			    fa->fa_tos, fa->fa_info, nlm_flags);
 	/* failure implies BUG in fib_nlmsg_size() */
 	BUG_ON(err < 0);
 
--- net/ipv4.old/fib_lookup.h	2007-04-18 12:50:11.000000000 +0200
+++ net/ipv4/fib_lookup.h	2007-04-18 12:43:42.377902856 +0200
@@ -30,7 +30,7 @@
 			 int dst_len, u8 tos, struct fib_info *fi,
 			 unsigned int);
 extern void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
-		      int dst_len, u32 tb_id, struct nl_info *info);
+		      int dst_len, u32 tb_id, struct nl_info *info, unsigned int
nlm_flags);
 extern struct fib_alias *fib_find_alias(struct list_head *fah,
 					u8 tos, u32 prio);
 extern int fib_detect_death(struct fib_info *fi, int order,


Comment 8 Patrick McHardy 2007-04-18 07:07:25 UTC
Milan Koci
Comment 9 Adrian Bunk 2007-07-18 05:45:16 UTC
Patch was included in 2.6.22.

Note You need to log in before you can comment on or make changes to this bug.