Broken? 2.6.6 + IP_ADD_SOURCE_MEMBERSHIP + SO_REUSEADDR
Subsystems (3)
Hi!
I am using the new IGMPv3 support in the 2.6.* series for Single Source Multicast (SSM). However, the kernel appears to (incorrectly) drop packets in some situations.
What I do is this: open a UDP port, set it SO_REUSEADDR, bind it to port 6767, and then use IP_ADD_SOURCE_MEMBERSHIP to listen to multicast group 232.65.43.21 and with a command-line controlled sender.
If I launch the same program twice with different senders, the first program ceases to receive multicast packets. (Neither from its own sender, nor the second program's sender) The second program receives packets from its designated sender only (as expected).
I know from tcpdump that the switch is delivering messages from the first designated sender. The kernel is simply not giving them to the application.
Some more observations:
If both programs specify the same sender, then both programs receive the message (as expected).
This is not an issue with subscribing to multiple senders in general. A single program listening to two senders does receive messages from both.
This seems like a bug to me.
PS. I am not subscribed to this list, so please CC me.
Signature · 1 line
Wesley W. Terpstra
In article <20040604155423.GA5656@muffin> (at Fri, 4 Jun 2004 17:54:23 +0200), "Wesley W. Terpstra" <[email protected]> says:
Quoted 4 lines
If I launch the same program twice with different senders, the first program ceases to receive multicast packets. (Neither from its own sender, nor the second program's sender) The second program receives packets from its designated sender only (as expected).
:
Quoted 2 lines
If both programs specify the same sender, then both programs receive the message (as expected).
Thanks for the report.
The following patch should fix the issue. Please try. Thanks again.
===== net/ipv4/udp.c 1.60 vs edited =====
net/ipv4/udp.c +1 -1
@@ -294,7 +294,7 @@ipv6_only_sock (s) || (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) continue; -if (!ip_mc_sf_allow (sk, loc_addr, rmt_addr, dif)) +if (!ip_mc_sf_allow (s, loc_addr, rmt_addr, dif))continue ; goto found; }
Signature · 2 lines
Hideaki YOSHIFUJI @ USAGI Project <[email protected]> GPG FP: 9022 65EB 1ECF 3AD1 0BDF 80D8 4807 F894 E062 0EEA
In article <[email protected]> (at Sat, 05 Jun 2004 01:55:44 +0900 (JST)), YOSHIFUJI Hideaki / 吉藤英明 <[email protected]> says:
Quoted 1 line
The following patch should fix the issue. Please try.
And, here's a variant for 2.4.
===== net/ipv4/udp.c 1.13 vs edited =====
net/ipv4/udp.c +1 -1
@@ -290,7 +290,7 @@ipv6_only_sock (s) || (s->bound_dev_if && s->bound_dev_if != dif)) continue; -if (!ip_mc_sf_allow (sk, loc_addr, rmt_addr, dif)) +if (!ip_mc_sf_allow (s, loc_addr, rmt_addr, dif))continue ;break ; }
-- yoshfuji @ time to sleep...
Quoted 5 lines · On Sat, 05 Jun 2004 02:14:11 +0900 (JST) YOSHIFUJI Hideaki / $B5HF#1QL@(B…
In article <[email protected]> (at Sat, 05 Jun 2004 01:55:44 +0900 (JST)), YOSHIFUJI Hideaki / $B5HF#1QL@(B <[email protected]> says:
Quoted 1 line
The following patch should fix the issue. Please try.
And, here's a variant for 2.4.
Both patches applied, thanks.
Quoted 17 lines · YOSHIFUJI Hideaki / 吉藤英明
In article <20040604155423.GA5656@muffin> (at Fri, 4 Jun 2004 17:54:23 +0200), "Wesley W. Terpstra" <[email protected]> says:
Quoted 7 lines
If I launch the same program twice with different senders, the first program ceases to receive multicast packets. (Neither from its own sender, nor the second program's sender) The second program receives packets from its designated sender only (as expected).
If both programs specify the same sender, then both programs receive the message (as expected).
Thanks for the report.
The following patch should fix the issue. Please try. Thanks again.=====
net/ipv4/udp.c1.60 vs edited =====
net/ipv4/udp.c line 300
Quoted 8 lines
net/ipv4/udp.c +1
-1@@ -294,7 +294,7 @@ipv6_only_sock (s) || (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) continue; -if (!ip_mc_sf_allow (sk, loc_addr, rmt_addr, dif)) +if (!ip_mc_sf_allow (s, loc_addr, rmt_addr, dif))continue ; goto found; }
That works.
I have confirmed that multiple senders for multiple processes is fixed.
I am amazed you fixed it so fast!
Will this make it into 2.6.7?
Signature · 1 line
Wesley W. Terpstra
Quoted 2 lines · On Fri, 4 Jun 2004 23:29:16 +0200 "Wesley W. Terpstra"…
I am amazed you fixed it so fast!
Will this make it into 2.6.7?
Yes, and 2.4.27 hopefully as well.