Re: [OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
If you have UAs behind your NAT why not just have them register to a different IP address on the OpenSIPS box, and don't specify an advertised address for that binding? The registrar will automatically select this interface when you do a lookup() for one of those UAs. Double Record-Route headers are properly added by OpenSIPS when a message enters one interface and leaves another. That worked even before this patch, which is why you do not see it explicitly mentioned here. One of the things the patch does is make sure that the advertised address (if specified) winds up on the Record-Route header corresponding to the interface with the alias. Thanks, Phil Date: Sun, 23 Aug 2009 22:33:43 +0200 From: Emmanuel BUU emmanuel@ives.fr Subject: Re: [OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised To: OpenSIPS devel mailling list devel@lists.opensips.org Cc: nore...@sourceforge.net Message-ID: 4a91a7a7.4090...@ives.fr Content-Type: text/plain; charset=iso-8859-1 Hello, This is a good idea except that the real live situation is more complicated. Advertised addresses are used whenether OpenSIPS is running behind an external NAT (provided by a router). If fact, the opensips may have to deal with UAs that are also behind the NAT (e.g. inside a corporate network) where there is no need to advertise other address the bind address. So, we need to distinguish between: - messages remaining inside an internal network - messages going through the NAT To be able to make this distinction, we would need to have some description of what the Internal network is lile internal_network=10.0.0.0/8 If any outgoing message matches these networks, the advertised address would not be used. It would be great to be able to have more than one such internal network description. And for advertised address, there is another issue: loose routing. In case of routing a packet from an internal network to the external network, we should have a double record route (like if we had two interfaces) Record-Route: sip:bind_address:port Record-Route: sip:adv_address:port Tricky ... but asterisk does it nicely. SourceForge.net a ?crit : Patches item #2706135, was opened at 2009-03-23 15:50 Message generated for change (Settings changed) made by bogdan_iancu You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Closed Resolution: Accepted Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 15:50 Message generated for change (Comment added) made by bogdan_iancu You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-07-13 14:36 Message: Hi Phil, Is the listen_alias_port_fixed.patch the final version with all the bugs fixed ? If so, I will review the code and upload it on SVN. Thanks and regards, Bogdan -- Comment By: Phil D'Amore (ptdamore) Date: 2009-07-10 21:05 Message: Hi Bogdan: Any word if this will be included in the next release now that the bug is fixed? Thanks, Phil -- Comment By: Nergal (nergalex) Date: 2009-06-03 13:07 Message: Hello Phil, That's great, the bug is fixed! Results of no regression tests is OK too. Thank you for your help. Alexis -- Comment By: Phil D'Amore (ptdamore) Date: 2009-05-26 20:03 Message: Alexis: Thanks for the extra information. I think I found the cause of the issue, and I've uploaded a new version of the patch. Please try the listen_alias_port_fixed.patch. I have not had time to test the issue in my environment since I use port 5060 everywhere. Since I don't know if I'll be able to do it today I'm uploading the modified patch so you can test if you'd like. The issue was in the changes to msg_translator.c. I was incorrectly using the length of the original port to calculate an offset. It was not properly re-written with the rest of the code to use the alias port when it's available. Thanks, Phil -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-05-26 19:52 Message: Hi Guys, Is the bug fixed in the patch ? I just want to know to start evaluating the patch for upload.
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 09:50 Message generated for change (Comment added) made by ptdamore You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Phil D'Amore (ptdamore) Date: 2009-07-13 12:47 Message: Bogdan, Yes the listen_alias_port_fixed.patch file contains a complete patch with the known issues fixed. Looking forward to your assessment. Thanks, Phil -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-07-13 07:36 Message: Hi Phil, Is the listen_alias_port_fixed.patch the final version with all the bugs fixed ? If so, I will review the code and upload it on SVN. Thanks and regards, Bogdan -- Comment By: Phil D'Amore (ptdamore) Date: 2009-07-10 14:05 Message: Hi Bogdan: Any word if this will be included in the next release now that the bug is fixed? Thanks, Phil -- Comment By: Nergal (nergalex) Date: 2009-06-03 06:07 Message: Hello Phil, That's great, the bug is fixed! Results of no regression tests is OK too. Thank you for your help. Alexis -- Comment By: Phil D'Amore (ptdamore) Date: 2009-05-26 13:03 Message: Alexis: Thanks for the extra information. I think I found the cause of the issue, and I've uploaded a new version of the patch. Please try the listen_alias_port_fixed.patch. I have not had time to test the issue in my environment since I use port 5060 everywhere. Since I don't know if I'll be able to do it today I'm uploading the modified patch so you can test if you'd like. The issue was in the changes to msg_translator.c. I was incorrectly using the length of the original port to calculate an offset. It was not properly re-written with the rest of the code to use the alias port
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 09:50 Message generated for change (Comment added) made by ptdamore You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Phil D'Amore (ptdamore) Date: 2009-07-10 14:05 Message: Hi Bogdan: Any word if this will be included in the next release now that the bug is fixed? Thanks, Phil -- Comment By: Nergal (nergalex) Date: 2009-06-03 06:07 Message: Hello Phil, That's great, the bug is fixed! Results of no regression tests is OK too. Thank you for your help. Alexis -- Comment By: Phil D'Amore (ptdamore) Date: 2009-05-26 13:03 Message: Alexis: Thanks for the extra information. I think I found the cause of the issue, and I've uploaded a new version of the patch. Please try the listen_alias_port_fixed.patch. I have not had time to test the issue in my environment since I use port 5060 everywhere. Since I don't know if I'll be able to do it today I'm uploading the modified patch so you can test if you'd like. The issue was in the changes to msg_translator.c. I was incorrectly using the length of the original port to calculate an offset. It was not properly re-written with the rest of the code to use the alias port when it's available. Thanks, Phil -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-05-26 12:52 Message: Hi Guys, Is the bug fixed in the patch ? I just want to know to start evaluating the patch for upload. Thanks and regards, Bogdan -- Comment By: Nergal (nergalex) Date: 2009-05-26 05:44 Message: Hello Phil, After few tests, I identified the exact context when the problem appear. I answer to your questions before. 1) b) An extra random character, this morning
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 09:50 Message generated for change (Comment added) made by ptdamore You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Phil D'Amore (ptdamore) Date: 2009-05-26 13:03 Message: Alexis: Thanks for the extra information. I think I found the cause of the issue, and I've uploaded a new version of the patch. Please try the listen_alias_port_fixed.patch. I have not had time to test the issue in my environment since I use port 5060 everywhere. Since I don't know if I'll be able to do it today I'm uploading the modified patch so you can test if you'd like. The issue was in the changes to msg_translator.c. I was incorrectly using the length of the original port to calculate an offset. It was not properly re-written with the rest of the code to use the alias port when it's available. Thanks, Phil -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-05-26 12:52 Message: Hi Guys, Is the bug fixed in the patch ? I just want to know to start evaluating the patch for upload. Thanks and regards, Bogdan -- Comment By: Nergal (nergalex) Date: 2009-05-26 05:44 Message: Hello Phil, After few tests, I identified the exact context when the problem appear. I answer to your questions before. 1) b) An extra random character, this morning is '.' that appears in the first Record-Route header after host:port BUT always in a specific context, see #3. 2) I have listen=X.X.X.X:5060 AS X.X.X.Y:5060 in my configuration and also: listen=X.X.X.X:15060 AS X.X.X.Y:5060 3) [NEW] The problem appears when the SIP message match the X.X.X.X:15060 instance. The rewrite of RR is done by the line listen=X.X.X.X:15060 AS X.X.X.Y:5060. In this case, the extra caracter is suffixed to the first RR port (public inbound interface). The problem appears for
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 14:50 Message generated for change (Comment added) made by nergalex You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Nergal (nergalex) Date: 2009-05-25 10:20 Message: Hello, Sorry to post this comment so late, I was in vacancy. History: https://sourceforge.net/forum/message.php?msg_id=7379817 Best regards Alexis --- I have a bug on the patch Change to listen= syntax to support per-interface advertised - ID: 2706135. In the first Record-Route header, a random caracter ends the IP:Port entry (see below 2 tests with the same configuration and the extra caracter is ';'). Could you help me to fix it? I can do more test or give you more information if you need. Maybe a clue: the Advertised port is 4 digits long but the real port is 5 digits long. conf extract: - X.X.X.X is a public IP - listen=tcp:10.143.136.8:15060 AS 10.143.136.29:5060 - force_send_socket(10.143.136.8:15060); - t_relay(tcp:10.143.81.130:5060) --- Test OK / proto TCP --- 1. Before SIP proxy processing (Front interface capture) SUBSCRIBE sip:serv...@pilote-sip.my_domain:5060 SIP/2.0 Via: SIP/2.0/TCP 10.95.10.1:10005;branch=z9hG4bK-1338-1-0 From: us...@sfr.fr sip:s...@10.95.10.1:10005;tag=1338SIPpTag001 To: us...@sfr.fr sip:serv...@pilote-sip.my_domain:5060 Call-ID: 1-1...@10.95.10.1 CSeq: 1 SUBSCRIBE Contact: sip:s...@10.95.10.1:10005 Max-Forwards: 70 Subject: Subscribe Test Content-Length: 0 2. After SIP proxy processing (Back interface capture) SUBSCRIBE sip:serv...@pilote-sip.my_domain:5060 SIP/2.0 Record-Route: sip:10.143.136.29:5060;transport=tcp;r2=on;lr=on;ftag=1338SIPpTag001 Record-Route: sip:X.X.X.X:5060;transport=tcp;r2=on;lr=on;ftag=1338SIPpTag001 Via: SIP/2.0/TCP 10.143.136.29:5060;branch=z9hG4bKa4b3.1d7725a.0;i=9e8f Via: SIP/2.0/TCP 10.95.10.1:10005;rport=32769;received=10.95.10.1;branch=z9hG4bK-1338-1-0 From: us...@sfr.fr
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 09:50 Message generated for change (Comment added) made by ptdamore You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Phil D'Amore (ptdamore) Date: 2009-05-25 16:18 Message: Alexis : I have a few questions for you: 1) To make sure I understand the problem, I want to re-word it a bit. Which of these statements is true about the problem a) An extra ';' (always a ';') character appears in the first Record-Route header after host:port, at random times. b) An extra random character (not always ';') appears in the first Record-Route header after host:port, at random times 2) Since you are using double-rr can I assume that you also have a listen=X.X.X.X:5060 line in your configuration for your public inbound interface, but it was not included in your message because it does not specify the [AS host:port] portion? If the answer to 2 is yes, can you try to specify an alias on that listen= line like this: listen=X.X.X.X:5060 AS X.X.X.X:5060 The effect should be that no re-write appears to happen as the left and right side are the same, but the code used to get there would be slightly different. I'm curious to see if the extra ';' goes away in that case. Thanks, Phil -- Comment By: Nergal (nergalex) Date: 2009-05-25 04:20 Message: Hello, Sorry to post this comment so late, I was in vacancy. History: https://sourceforge.net/forum/message.php?msg_id=7379817 Best regards Alexis --- I have a bug on the patch Change to listen= syntax to support per-interface advertised - ID: 2706135. In the first Record-Route header, a random caracter ends the IP:Port entry (see below 2 tests with the same configuration and the extra caracter is ';'). Could you help me to fix it? I can do more test or give you more information if you need. Maybe a clue: the Advertised port is 4 digits long
[OpenSIPS-Devel] [ opensips-Patches-2706135 ] Change to listen= syntax to support per-interface advertised
Patches item #2706135, was opened at 2009-03-23 15:50 Message generated for change (Comment added) made by bogdan_iancu You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: core Group: trunk Status: Open Resolution: None Priority: 5 Private: No Submitted By: Phil D'Amore (ptdamore) Assigned to: Bogdan-Andrei Iancu (bogdan_iancu) Summary: Change to listen= syntax to support per-interface advertised Initial Comment: [This is the patch for an idea I posted several weeks ago to the devel list. Below is the original list posting as an explanation.] The patch extends the idea of the advertised_address and advertised_port parameters so that it is possible to bind such an address to each listen= statement. Syntax-wise, it works like this: listen=[proto:]host[:port][ AS host[:port]] This new AS host[:port] syntax binds the advertised address specifically to the corresponding socket_info structure. If such an address is specified, it will be used before the address specified as advertised_address, or set in the set_advertised_address function for packets being sent on that particular socket. If it is not specified, then the old methods work as before. The core has been modified so that address tuples using the advertised address work interchangeably with the real address where you'd expect them to. grep_sock_info and find_si have been updated to allow this. Therefore, it is not necessary to explicitly name your advertised address on an alias= line. The following are now equivalent: force_send_socket(udp:adv_address:adv_port); force_send_socket(udp:real_address:real_port); Note we don't specify protocol on the advertised address. The way I see things, it makes no sense to advertise a different protocol, so it inherits the protocol of the socket. Via construction is affected in the same way that the old global advertised_address does, except the Via is written specifically for the socket used to send the message. Lump substitution is also affected. The following lump substitutions are aware of this new facility: SUBST_RCV_IP SUBST_RCV_PORT SUBST_RCV_ALL SUBST_SND_IP SUBST_SND_PORT SUBST_SND_ALL As a result of this, I was able to stop using record_route_preset in my own config, and just use standard record_route. This winds up working much better, because double-rr works properly, and add_rr_param now works as expected for me, where it was a problem using record_route_preset. Additionally, registrar and usrloc had to be modified: registrar - update the add_sock_hdr function to put the advertised address in the message instead of the real address. This allows registration information to be replicated between multi-homed proxies, and the correct interface on the peer can be used. usrloc - update db_insert_ucontact to store the advertised address in the database instead of the real address. Same reasoning as the mods to registrar. -- Comment By: Bogdan-Andrei Iancu (bogdan_iancu) Date: 2009-05-09 11:15 Message: Hi Phil, I will start reviewing the patch in order to upload it on SVN. If I have some questions about it, I will post them here. Thanks and regards, Bogdan -- You can respond by visiting: https://sourceforge.net/tracker/?func=detailatid=1086412aid=2706135group_id=232389 ___ Devel mailing list Devel@lists.opensips.org http://lists.opensips.org/cgi-bin/mailman/listinfo/devel