[PATCH net-next v3 0/8] gtp: misc improvements
Hi Pablo, This is v3 of the GTP improvements. This series lays the groundwork for removing the socket references from the GTP netdevice by removing duplicate code and simplifying the logic on some code paths. It slighly changes the GTP genl API by making the socket parameters optional (though one of them is still required). The removal of the socket references will break the 1:1 releation between GTP netdevice and GTP socket that prevents us to support multiple VRFs with overlaping IP addresse spaces attached to the same GTP-U entity (needed for multi APN support). Pablo found a socket hold problem in v2. In order to solve that I had to switch the socket references from the struct socket to the internal struct sock. This should have no functionl impact, but we can now hang on to the reference without blocking user space from closing the GTP socket. There was also some length off-list conversation about why the netdevice to socket relation needs to be changed at all. Harald did already agree to this but Pablo had some questions. I've tried to address that with a bit of dokumentation for the GTP module. Hopefully this will explain the need for the change sufficiently. This series does have conflicts with the SGSN side tunnel work done by Jonas Bonn. The unification of the tunnel Rx code touches the same places he changed. v2->v3: * add documentation to explain the goal of all these changes * incorporate review comments * switch from struct socket to struct sock Regards Andreas -- Andreas Schultz (8): gtp: add documentation gtp: switch from struct socket to struct sock for the GTP sockets gtp: make GTP sockets in gtp_newlink optional gtp: merge gtp_get_net and gtp_genl_find_dev gtp: consolidate gtp socket rx path gtp: unify genl_find_pdp and prepare for per socket lookup gtp: consolidate pdp context destruction into helper gtp: add socket to pdp context Documentation/networking/gtp.txt | 112 ++++++++ drivers/net/gtp.c | 548 +++++++++++++++++++-------------------- 2 files changed, 386 insertions(+), 274 deletions(-) create mode 100644 Documentation/networking/gtp.txt -- 2.10.2
The GTP-U implementation in not only driven by 3GPP TS 29.281, but also by the requirements of the 3GPP network entities that use it. This document tries to explain and clarify some design decision and their origins. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- Documentation/networking/gtp.txt | 112 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 112 insertions(+) create mode 100644 Documentation/networking/gtp.txt diff --git a/Documentation/networking/gtp.txt b/Documentation/networking/gtp.txt new file mode 100644 index 0000000..cefd983 --- /dev/null +++ b/Documentation/networking/gtp.txt @@ -0,0 +1,112 @@ +Note: this document contain some forward looking statements and does + not (yet) reflect the implementation. This is done to motivate + and explain some of the changes to come. + + This note will be removed once the implementation matches this + document. + +General Packet Radio System (GPRS) Tunnelling Protocol User Plane (GTP-U) +========================================================================= + +The GTP-U protocol is a tunnelling protocol used in public land mobile networks +[PLMN]. It is always use together with a user space control instance implementing +a 3GPP network entity (e.g. GGSN, SGSN, PDN-GW, S-GW). + +The protocol is specified for version 0 in [GSM 09.60] and for version 1 in +[3GPP TS 29.281]. However, the functionality defined in those documents has +always to be taken in relation of the functional 3GPP entity that is using it. + +The rest of document is focusing on version 1 of the protocol. GTPv1 to GTPv0 +interworking and the use of GTPv0-U in general has be depreciated from 3GPP Rel8 +onward [3GPP TS 29.060]. + +Local GTP-U entity and tunnel identification +-------------------------------------------- + +GTP-U uses UDP for transporting PDU's. The receiving UDP port is 2152 for +GTPv1-U and 3386 for GTPv0-U. + +There is only one GTP-U entity (and therefor SGSN/GGSN/S-GW/PDN-GW instance) +per IP address. Tunnel Endpoint Identifier (TEID) are unique per GTP-U entity. + +A specific tunnel is only defined by the destination entity. Since the +destination port is constant, only the destination IP and TEID define +a tunnel. The source IP and Port have no meaning for the tunnel. + +Therefore: + + * when sending, the remote entity is defined by the remote IP and the tunnel + endpoint id. The source IP and port have no meaning and can be changed + at any time. + + * when receiving the local entity is defined by the local destination IP + and the tunnel endpoint id. The source IP and port have no meaning and can + change at any time. + +[3GPP TS 29.281] Section 4.3.0 defines this so: + +> The TEID in the GTP-U header is used to de-multiplex traffic incoming from +> remote tunnel endpoints so that it is delivered to the User plane entities +> in a way that allows multiplexing of different users, different packet +> protocols and different QoS levels. Therefore no two remote GTP-U endpoints +> shall send traffic to a GTP-U protocol entity using the same TEID value except +> for data forwarding as part of mobility procedures. + +The definition above only defines that two remote GTP-U endpoints *should not* +send to the same TEID, it *does not* forbid or exclude such a scenario. In +fact, the mentioned mobility procedures make it necessary that the GTP-U entity +accepts traffic for TEID's from multiple or unknown peers. + +Therefor the receiving side only identifies tunnels based on TEID's, not based +on the source IP. + +APN vs. Network Device +---------------------- + +The GTP-U driver creates a Linux network device for each Gi/SGi interface. + +[3GPP TS 29.281] calls the Gi/SGi reference point an interface. This may lead +to the impression that the GGSN/P-GW can have only one such interface. + +Correct is that the Gi/SGi reference point defines the interworking between +the 3GPP packet domain (PDN) based on GTP-U tunnel and IP based networks. + +There is no provision in any of the 3GPP documents that limits the number of +Gi/SGi interfaces implemented by a GGSN/P-GW. + +[3GPP TS 29.061] Section 11.3 makes it clear that the selection of a specific +Gi/SGi interfaces is made through the Access Point Name (APN): + +> 2. each private network manages its own addressing. In general this will +> result in different private networks having overlapping address ranges. +> A logically separate connection (e.g. an IP in IP tunnel or layer 2 +> virtual circuit) is used between the GGSN/P-GW and each private network. +> In this case the IP address alone is not necessarily unique. The pair +> of values, Access Point Name (APN) and IPv4 address and/or IPv6 +> prefixes, is unique. + +In order to support the overlapping address range use case, each APN is mapped +to a separate Gi/SGi interface (network device). + +NOTE: The Access Point Name is purely a control plane (GTP-C) concept. At the + GTP-U level, only Tunnel Endpoint Identifiers are present in GTP-U packets + and network devices are known. Therefore for a given UE the mapping in + IP to PDN network is: + + * network device + MS IP -> Peer IP + Peer TEID, + + and from PDN to IP network: + + * local GTP-U IP + TEID -> network device + + Furthermore, before a received T-PDU is injected into the network device + the MS IP is checked against the IP recorded in PDP context. + +References: +----------- + +[PLMN] https://en.wikipedia.org/wiki/Public_land_mobile_network +[3GPP TS 29.060] http://www.etsi.org/deliver/etsi_ts/129000_129099/129060/13.06.00_60/ts_1290... +[3GPP TS 29.061] http://www.etsi.org/deliver/etsi_ts/129000_129099/129061/13.06.00_60/ts_1290... +[3GPP TS 29.281] http://www.etsi.org/deliver/etsi_ts/129200_129299/129281/13.02.00_60/ts_1292... +[GSM 09.60] http://www.etsi.org/deliver/etsi_ts/101300_101399/101347/06.13.00_60/ts_1013... -- 2.10.2
Hi Andreas, this is not really about the documentation, but it still makes sense to discuss here as the issue is best described: On Mon, Feb 13, 2017 at 04:36:17PM +0100, Andreas Schultz wrote:
+Local GTP-U entity and tunnel identification +-------------------------------------------- + +GTP-U uses UDP for transporting PDU's. The receiving UDP port is 2152 for +GTPv1-U and 3386 for GTPv0-U. + +There is only one GTP-U entity (and therefor SGSN/GGSN/S-GW/PDN-GW instance) +per IP address. Tunnel Endpoint Identifier (TEID) are unique per GTP-U entity. + +A specific tunnel is only defined by the destination entity. Since the +destination port is constant, only the destination IP and TEID define +a tunnel. The source IP and Port have no meaning for the tunnel.
Are you absolutely sure about this?
+[3GPP TS 29.281] Section 4.3.0 defines this so: + +> The TEID in the GTP-U header is used to de-multiplex traffic incoming from +> remote tunnel endpoints so that it is delivered to the User plane entities +> in a way that allows multiplexing of different users, different packet +> protocols and different QoS levels. Therefore no two remote GTP-U endpoints +> shall send traffic to a GTP-U protocol entity using the same TEID value except +> for data forwarding as part of mobility procedures. + +The definition above only defines that two remote GTP-U endpoints *should not* +send to the same TEID, it *does not* forbid or exclude such a scenario. In +fact, the mentioned mobility procedures make it necessary that the GTP-U entity +accepts traffic for TEID's from multiple or unknown peers.
I so far always assumed that you use GTP-C "MODIFY PDP CONTEXT" in case of such mobility situations, i.e. the control plane explicitly notifies the GTP-U entity of a change in the SGSN/S-GW address. At least for 3G this seems to be the case, and my assumption appears to be confirmed with sources such as e.g. page 15 of http://www.nwadmin.de/presentation_mobility_manag ement_in_UMTS.pdf Also, for LTE, it seems that there's a GTPv2-C "Modify Bearer Request/Response" involved in relocation from one S-GW to another S-GW: http://www.lteandbeyond.com/2012/03/x2-based-handover-with-sgw-relocation.ht...
Step 3. The target Serving GW assigns addresses and TEIDs (one per bearer) for downlink traffic from the PDN GW. The Serving GW allocates DL TEIDs on S5/S8 even for non-accepted bearers. It sends a Modify Bearer Request (Serving GW addresses for user plane and TEID(s)) message per PDN connection to the PDN GW(s). The SGW also includes User Location Information IE and/or UE Time Zone IE if it is present in step 2. The PDN GW updates its context field and returns a Modify Bearer Response (Charging Id, MSISDN, etc.) message to the Serving GW. The MSISDN is included if the PDN GW has it stored in its UE context. The PDN GW starts sending downlink packets to the target GW using the newly received address and TEIDs. These downlink packets will use the new downlink path via the target Serving GW to the target eNodeB. The Serving GW shall allocate TEIDs for the failed bearers and inform to the MME.
Section 5.5.1.1.3 of TS 23.401 agrees with the GTPv2C signalling during relocation, AFAICT.
+Therefor the receiving side only identifies tunnels based on TEID's, not based +on the source IP.
I'm wondering if this really is neccesary. Do you have actual protocol traces, specs or any other literature confirming this? I find it somewhat surprising, given how much this opens the door for arbitrary spoofing from anyone (with access to the respective private network such as GRX). So in which situations specifically will thre be a S-GW side Address change without associated GTP-C signaling informing the P-GW about the new S-GW side Address + TEID? Regards, Harald -- - Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)
Hi, ----- On Feb 20, 2017, at 5:15 PM, laforge laforge@gnumonks.org wrote:
Hi Andreas,
this is not really about the documentation, but it still makes sense to discuss here as the issue is best described:
On Mon, Feb 13, 2017 at 04:36:17PM +0100, Andreas Schultz wrote:
+Local GTP-U entity and tunnel identification +-------------------------------------------- + +GTP-U uses UDP for transporting PDU's. The receiving UDP port is 2152 for +GTPv1-U and 3386 for GTPv0-U. + +There is only one GTP-U entity (and therefor SGSN/GGSN/S-GW/PDN-GW instance) +per IP address. Tunnel Endpoint Identifier (TEID) are unique per GTP-U entity. + +A specific tunnel is only defined by the destination entity. Since the +destination port is constant, only the destination IP and TEID define +a tunnel. The source IP and Port have no meaning for the tunnel.
Are you absolutely sure about this?
Yes. It took me a while to realize this. The crux is this statement in TS 29.060, Section 7.3.1:
The SGSN shall include an SGSN Address for control plane and an SGSN address for user traffic, which may differ from that provided by the underlying network service (e.g. IP). The GGSN shall store these SGSN Addresses and use them when sending control plane on this GTP tunnel or G-PDUs to the SGSN for the MS.
IMHO, this implies that the source IP of GTP-U and GTP-C frames does not have to match the GSN address specified by a Create PDP Context Request. For GTPv2-C there is no such clear statement. However, GTPv2-C makes it clear that a TEID describes the local tunnel ENDPOINT, there nothing that defines the tunnel *source* (the same is true for GTPv1-C, except that is not so explicit). TS 29.274, Section 4.2.2.1, Initial Messages:
During the establishment of the GTP tunnel, the GTPv2 entity selects and communicates to the peer GTPv2 entity the IP Destination Address at which it expects to receive subsequent control plane Initial messages related to that GTP tunnel via .....
TS 23.401, Annex D, Sect. D.3.3. makes is clear beyond doubt that we have to accept packet for a valid TEID from virtually any IP. If you look at figure D.3.3-1, step 16, you will see that only the new SGSN is contacting the P-GW. There is no prior advertisement of the change from the old SGSN. Step 16 would therefor not work if we limited the tunnel to a specific source IP. The same applies to figure D.3.4-1, step 18.
+[3GPP TS 29.281] Section 4.3.0 defines this so: + +> The TEID in the GTP-U header is used to de-multiplex traffic incoming from +> remote tunnel endpoints so that it is delivered to the User plane entities +> in a way that allows multiplexing of different users, different packet +> protocols and different QoS levels. Therefore no two remote GTP-U endpoints +> shall send traffic to a GTP-U protocol entity using the same TEID value except +> for data forwarding as part of mobility procedures. + +The definition above only defines that two remote GTP-U endpoints *should not* +send to the same TEID, it *does not* forbid or exclude such a scenario. In +fact, the mentioned mobility procedures make it necessary that the GTP-U entity +accepts traffic for TEID's from multiple or unknown peers.
I so far always assumed that you use GTP-C "MODIFY PDP CONTEXT" in case of such mobility situations, i.e. the control plane explicitly notifies the GTP-U entity of a change in the SGSN/S-GW address.
Yes, it does. But that notification only applies to the tunnel endpoint at the SGSN/S-GW in the GGSN/P-GW to SGSN/S-GW direction.
At least for 3G this seems to be the case, and my assumption appears to be confirmed with sources such as e.g. page 15 of http://www.nwadmin.de/presentation_mobility_manag ement_in_UMTS.pdf
Also, for LTE, it seems that there's a GTPv2-C "Modify Bearer Request/Response" involved in relocation from one S-GW to another S-GW: http://www.lteandbeyond.com/2012/03/x2-based-handover-with-sgw-relocation.ht...
I think this affirms my argument, Step 3a is sending GTP packets from a source IP that is know to the P-GW before. The fact that this is only used for GTP-C does IMHO not mean that the same should not apply to GTP-U.
Step 3. The target Serving GW assigns addresses and TEIDs (one per bearer) for downlink traffic from the PDN GW. The Serving GW allocates DL TEIDs on S5/S8 even for non-accepted bearers. It sends a Modify Bearer Request (Serving GW addresses for user plane and TEID(s)) message per PDN connection to the PDN GW(s). The SGW also includes User Location Information IE and/or UE Time Zone IE if it is present in step 2. The PDN GW updates its context field and returns a Modify Bearer Response (Charging Id, MSISDN, etc.) message to the Serving GW. The MSISDN is included if the PDN GW has it stored in its UE context. The PDN GW starts sending downlink packets to the target GW using the newly received address and TEIDs. These downlink packets will use the new downlink path via the target Serving GW to the target eNodeB. The Serving GW shall allocate TEIDs for the failed bearers and inform to the MME.
Section 5.5.1.1.3 of TS 23.401 agrees with the GTPv2C signalling during relocation, AFAICT.
Again, there is nothing in there that defines a IP *source*, a F-TEID is a tunnel ENDPOINT id not a tunnel SOURCE id.
+Therefor the receiving side only identifies tunnels based on TEID's, not based +on the source IP.
I'm wondering if this really is neccesary. Do you have actual protocol traces, specs or any other literature confirming this?
So far I have only seen symmetric GTP-U tunnels. However I believe there is nothing that would stop a SGSN/S-GW from switching GPT-U tunnel source transparently across IP's (for example a system with multiple shelves might use different shelve in DL/UL direction and have therefore multiple IP's.
I find it somewhat surprising, given how much this opens the door for arbitrary spoofing from anyone (with access to the respective private network such as GRX).
Yes it is, but it is mandated without a doubt by the specification for GTP-C. For the reasons outline before, I think that this applies to GTP-U as well.
So in which situations specifically will thre be a S-GW side Address change without associated GTP-C signaling informing the P-GW about the new S-GW side Address + TEID?
As above, multi shelve systems with asymmetric UL/DL path. Regards, Andreas
Regards, Harald -- - Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)
On Mon, Feb 20, 2017 at 06:28:07PM +0100, Andreas Schultz wrote:
+A specific tunnel is only defined by the destination entity. Since the +destination port is constant, only the destination IP and TEID define +a tunnel. The source IP and Port have no meaning for the tunnel.
Are you absolutely sure about this?
Yes. It took me a while to realize this. The crux is this statement in TS 29.060, Section 7.3.1:
Also sounds right to me. When the tunnel is established, each side of a GTP tunnel stores the other side's IP and TEID; port number is constant (unfortunately, which is one of the ideas/reasons behind gtphub). ~N
Hi Andreas, thanks for your feedback. On Mon, Feb 20, 2017 at 06:28:07PM +0100, Andreas Schultz wrote:
Are you absolutely sure about this?
Yes. It took me a while to realize this. The crux is this statement in TS 29.060, Section 7.3.1:
The SGSN shall include an SGSN Address for control plane and an SGSN address for user traffic, which may differ from that provided by the underlying network service (e.g. IP). The GGSN shall store these SGSN Addresses and use them when sending control plane on this GTP tunnel or G-PDUs to the SGSN for the MS.
IMHO, this implies that the source IP of GTP-U and GTP-C frames does not have to match the GSN address specified by a Create PDP Context Request.
Yes, I follow that reasoning for GTP-C (i.e. the transport-layer S-GW source IP of the CREATE PDP CTX REQ may differ from the one signalled inside GTP-C. For GTP-U however, this just means that GTP-U in general can use different IP addresses from GTP-C. This is well-known and I also have no doubts about this. But what I still doubt is whether within an ongoing GTP-U tunnel, the source IP address can change at any time without any signaling announcing that change.
TS 29.274, Section 4.2.2.1, Initial Messages:
During the establishment of the GTP tunnel, the GTPv2 entity selects and communicates to the peer GTPv2 entity the IP Destination Address at which it expects to receive subsequent control plane Initial messages related to that GTP tunnel via .....
TS 23.401, Annex D, Sect. D.3.3. makes is clear beyond doubt that we have to accept packet for a valid TEID from virtually any IP. If you look at figure D.3.3-1, step 16, you will see that only the new SGSN is contacting the P-GW. There is no prior advertisement of the change from the old SGSN.
Yes, but that is again GTP-C, and not GTP-U.
Step 16 would therefor not work if we limited the tunnel to a specific source IP.
correct. But it is a GTP-C transaction.
I so far always assumed that you use GTP-C "MODIFY PDP CONTEXT" in case of such mobility situations, i.e. the control plane explicitly notifies the GTP-U entity of a change in the SGSN/S-GW address.
Yes, it does. But that notification only applies to the tunnel endpoint at the SGSN/S-GW in the GGSN/P-GW to SGSN/S-GW direction.
At least for 3G this seems to be the case, and my assumption appears to be confirmed with sources such as e.g. page 15 of http://www.nwadmin.de/presentation_mobility_manag ement_in_UMTS.pdf
Also, for LTE, it seems that there's a GTPv2-C "Modify Bearer Request/Response" involved in relocation from one S-GW to another S-GW: http://www.lteandbeyond.com/2012/03/x2-based-handover-with-sgw-relocation.ht...
I think this affirms my argument, Step 3a is sending GTP packets from a source IP that is know to the P-GW before. The fact that this is only used for GTP-C does IMHO not mean that the same should not apply to GTP-U.
Step 3. The target Serving GW assigns addresses and TEIDs (one per bearer) for downlink traffic from the PDN GW. The Serving GW allocates DL TEIDs on S5/S8 even for non-accepted bearers. It sends a Modify Bearer Request (Serving GW addresses for user plane and TEID(s)) message per PDN connection to the PDN GW(s). The SGW also includes User Location Information IE and/or UE Time Zone IE if it is present in step 2. The PDN GW updates its context field and returns a Modify Bearer Response (Charging Id, MSISDN, etc.) message to the Serving GW. The MSISDN is included if the PDN GW has it stored in its UE context. The PDN GW starts sending downlink packets to the target GW using the newly received address and TEIDs. These downlink packets will use the new downlink path via the target Serving GW to the target eNodeB. The Serving GW shall allocate TEIDs for the failed bearers and inform to the MME.
Section 5.5.1.1.3 of TS 23.401 agrees with the GTPv2C signalling during relocation, AFAICT.
Again, there is nothing in there that defines a IP *source*, a F-TEID is a tunnel ENDPOINT id not a tunnel SOURCE id.
+Therefor the receiving side only identifies tunnels based on TEID's, not based +on the source IP.
I'm wondering if this really is neccesary. Do you have actual protocol traces, specs or any other literature confirming this?
So far I have only seen symmetric GTP-U tunnels. However I believe there is nothing that would stop a SGSN/S-GW from switching GPT-U tunnel source transparently across IP's (for example a system with multiple shelves might use different shelve in DL/UL direction and have therefore multiple IP's.
yes, it might. But then, on the other hand - at least for what I know about 2G/3G - it is typically very cumbersome to announce IP addresses to roaming partners. It involves manual updating of related GSMA [paper?] forms, exchanging them with all roaming partners, and coordinating a change with them. AFAIK, it is not possible to announce 'this is my range of S-GW IP addreses', but you have to individually manually announce each of them in the related documents. There are projects out there where people are building additional layers of address translation just to work around those long and error-prone manual procedures, making sure that the externally visible IP addresses of a given SGSN/GGSN are always the same ones. So yes, all of the above are practical/procedural concerns, but they would be explanations why this scenario - even if it might be supported by the specs - is unlikely in practise, at least within current/past procedures and practises. Finally, from the point of view of packet filtering (which operators typically implement, otherwise the manual announcement of SGSN IP addresses by the above procedures wouldn't be required), it would be impossible to do state based filtering if there were use cases where an random IP address colud send valid traffic for a given tunnel.
I find it somewhat surprising, given how much this opens the door for arbitrary spoofing from anyone (with access to the respective private network such as GRX).
Yes it is, but it is mandated without a doubt by the specification for GTP-C. For the reasons outline before, I think that this applies to GTP-U as well.
For GTP-U it may be the case. I'm not entirely convinced, though. Also, even if the specifications wanted to support such scenarios, I think it is doubtful that this is actually implemented in practise. If it was my choice, and I had to support the "loose matching", I would make it a configuration option (sysctl? netlink attribute?) and default to the more strict matching, including the source address. It just seems to make much more sense and be more safe/sane. For those people who really need the loose matching (and who are willing to pay the price for it in terms of opening up to spoofing and making packet filtering harder), they can set the option. But then, I'm not the one doing that implementation, so it is up to you. My opinion is just an opinion. Regards, Harald -- - Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)
Hi Harald, ----- On Feb 22, 2017, at 9:36 AM, laforge laforge@gnumonks.org wrote:
Hi Andreas,
thanks for your feedback.
On Mon, Feb 20, 2017 at 06:28:07PM +0100, Andreas Schultz wrote:
Are you absolutely sure about this?
Yes. It took me a while to realize this. The crux is this statement in TS 29.060, Section 7.3.1:
The SGSN shall include an SGSN Address for control plane and an SGSN address for user traffic, which may differ from that provided by the underlying network service (e.g. IP). The GGSN shall store these SGSN Addresses and use them when sending control plane on this GTP tunnel or G-PDUs to the SGSN for the MS.
IMHO, this implies that the source IP of GTP-U and GTP-C frames does not have to match the GSN address specified by a Create PDP Context Request.
Yes, I follow that reasoning for GTP-C (i.e. the transport-layer S-GW source IP of the CREATE PDP CTX REQ may differ from the one signalled inside GTP-C.
For GTP-U however, this just means that GTP-U in general can use different IP addresses from GTP-C. This is well-known and I also have no doubts about this.
But what I still doubt is whether within an ongoing GTP-U tunnel, the source IP address can change at any time without any signaling announcing that change.
TS 29.274, Section 4.2.2.1, Initial Messages:
During the establishment of the GTP tunnel, the GTPv2 entity selects and communicates to the peer GTPv2 entity the IP Destination Address at which it expects to receive subsequent control plane Initial messages related to that GTP tunnel via .....
TS 23.401, Annex D, Sect. D.3.3. makes is clear beyond doubt that we have to accept packet for a valid TEID from virtually any IP. If you look at figure D.3.3-1, step 16, you will see that only the new SGSN is contacting the P-GW. There is no prior advertisement of the change from the old SGSN.
Yes, but that is again GTP-C, and not GTP-U.
Step 16 would therefor not work if we limited the tunnel to a specific source IP.
correct. But it is a GTP-C transaction.
I so far always assumed that you use GTP-C "MODIFY PDP CONTEXT" in case of such mobility situations, i.e. the control plane explicitly notifies the GTP-U entity of a change in the SGSN/S-GW address.
Yes, it does. But that notification only applies to the tunnel endpoint at the SGSN/S-GW in the GGSN/P-GW to SGSN/S-GW direction.
At least for 3G this seems to be the case, and my assumption appears to be confirmed with sources such as e.g. page 15 of http://www.nwadmin.de/presentation_mobility_manag ement_in_UMTS.pdf
Also, for LTE, it seems that there's a GTPv2-C "Modify Bearer Request/Response" involved in relocation from one S-GW to another S-GW: http://www.lteandbeyond.com/2012/03/x2-based-handover-with-sgw-relocation.ht...
I think this affirms my argument, Step 3a is sending GTP packets from a source IP that is know to the P-GW before. The fact that this is only used for GTP-C does IMHO not mean that the same should not apply to GTP-U.
Step 3. The target Serving GW assigns addresses and TEIDs (one per bearer) for downlink traffic from the PDN GW. The Serving GW allocates DL TEIDs on S5/S8 even for non-accepted bearers. It sends a Modify Bearer Request (Serving GW addresses for user plane and TEID(s)) message per PDN connection to the PDN GW(s). The SGW also includes User Location Information IE and/or UE Time Zone IE if it is present in step 2. The PDN GW updates its context field and returns a Modify Bearer Response (Charging Id, MSISDN, etc.) message to the Serving GW. The MSISDN is included if the PDN GW has it stored in its UE context. The PDN GW starts sending downlink packets to the target GW using the newly received address and TEIDs. These downlink packets will use the new downlink path via the target Serving GW to the target eNodeB. The Serving GW shall allocate TEIDs for the failed bearers and inform to the MME.
Section 5.5.1.1.3 of TS 23.401 agrees with the GTPv2C signalling during relocation, AFAICT.
Again, there is nothing in there that defines a IP *source*, a F-TEID is a tunnel ENDPOINT id not a tunnel SOURCE id.
+Therefor the receiving side only identifies tunnels based on TEID's, not based +on the source IP.
I'm wondering if this really is neccesary. Do you have actual protocol traces, specs or any other literature confirming this?
So far I have only seen symmetric GTP-U tunnels. However I believe there is nothing that would stop a SGSN/S-GW from switching GPT-U tunnel source transparently across IP's (for example a system with multiple shelves might use different shelve in DL/UL direction and have therefore multiple IP's.
yes, it might. But then, on the other hand - at least for what I know about 2G/3G - it is typically very cumbersome to announce IP addresses to roaming partners. It involves manual updating of related GSMA [paper?] forms, exchanging them with all roaming partners, and coordinating a change with them. AFAIK, it is not possible to announce 'this is my range of S-GW IP addreses', but you have to individually manually announce each of them in the related documents. There are projects out there where people are building additional layers of address translation just to work around those long and error-prone manual procedures, making sure that the externally visible IP addresses of a given SGSN/GGSN are always the same ones.
So yes, all of the above are practical/procedural concerns, but they would be explanations why this scenario - even if it might be supported by the specs - is unlikely in practise, at least within current/past procedures and practises.
Finally, from the point of view of packet filtering (which operators typically implement, otherwise the manual announcement of SGSN IP addresses by the above procedures wouldn't be required), it would be impossible to do state based filtering if there were use cases where an random IP address colud send valid traffic for a given tunnel.
All of above applies to roaming, however provider internal traffic is usually not filtered and bound to GSMA IR.21 process. The use cases and procedures supported in roaming might therefore be different from the ones when connected to the home PLMN. If a firewall is deployed, then it's a decision made at that point to block or allow certain scenarios, but that should not have an impact on what we would like to support at the GTP-U instance. I believe we should attempt to support all defined and legal use-cases and leave the filtering to other layers (iptables, nftables or external firewalls). The GSMA has also something to say about firewall and GTP IP source addresses (although I think this mostly applies to GTP-C), GSMA IR.33 v8.01, Section A.2:
Unlike GTP version 0, in GTP version 1 the GGSN is allowed to send GTP response messages back to an SGSN with the source IP address set to an IP address different to that which was in the destination address of the associated GTP request message. The change was made in 3GPP to optimize internal processing of GGSNs.
I then goes on to describe how state-aware firewalls will have problems with that. Anyhow, this clearly indicates that using different source IP's is a valid use case (whether or not that might break in roaming scenarios).
I find it somewhat surprising, given how much this opens the door for arbitrary spoofing from anyone (with access to the respective private network such as GRX).
Yes it is, but it is mandated without a doubt by the specification for GTP-C. For the reasons outline before, I think that this applies to GTP-U as well.
For GTP-U it may be the case. I'm not entirely convinced, though. Also, even if the specifications wanted to support such scenarios, I think it is doubtful that this is actually implemented in practise.
I've done a test with a GTP implementation in a router from a big vendor (an ePDG, not a GGSN/PGW). That implementation did not like asymmetric GTP-U endpoint IP's. However, this seems to be a limitation of some kind of receive path optimization and not a deliberate filter (at least when I interpret the error messages correctly).
If it was my choice, and I had to support the "loose matching", I would make it a configuration option (sysctl? netlink attribute?) and default to the more strict matching, including the source address. It just seems to make much more sense and be more safe/sane. For those people who really need the loose matching (and who are willing to pay the price for it in terms of opening up to spoofing and making packet filtering harder), they can set the option.
That might be an idea, not in the sense of matching but as an additional filter. After the all the TEID is supposed the unique identifier for the tunnel in the GTP-U entity. We currently do not match or filter on the peer GSN source IP. That should IMHO remain the default. Adding a per PDP context flag to filter on the peer GSN address should be relatively simple.
But then, I'm not the one doing that implementation, so it is up to you. My opinion is just an opinion.
Regards, Harald
Regards, Andreas
-- - Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)
After enabling the UDP encapsulation, only the sk member is used. Holding the socket would prevent user space from closing the socket, but holding a reference to the sk member does not have the same effect. This change will make it simpler to later detach the sockets from the netdevice. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 42 +++++++++++++++++++++++------------------- 1 file changed, 23 insertions(+), 19 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index bda0c64..a8ce8c7 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -66,8 +66,8 @@ struct pdp_ctx { struct gtp_dev { struct list_head list; - struct socket *sock0; - struct socket *sock1u; + struct sock *sk0; + struct sock *sk1u; struct net_device *dev; @@ -261,17 +261,19 @@ static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, static void gtp_encap_disable(struct gtp_dev *gtp) { - if (gtp->sock0 && gtp->sock0->sk) { - udp_sk(gtp->sock0->sk)->encap_type = 0; - rcu_assign_sk_user_data(gtp->sock0->sk, NULL); + if (gtp->sk0) { + udp_sk(gtp->sk0)->encap_type = 0; + rcu_assign_sk_user_data(gtp->sk0, NULL); + sock_put(gtp->sk0); } - if (gtp->sock1u && gtp->sock1u->sk) { - udp_sk(gtp->sock1u->sk)->encap_type = 0; - rcu_assign_sk_user_data(gtp->sock1u->sk, NULL); + if (gtp->sk1u) { + udp_sk(gtp->sk1u)->encap_type = 0; + rcu_assign_sk_user_data(gtp->sk1u, NULL); + sock_put(gtp->sk1u); } - gtp->sock0 = NULL; - gtp->sock1u = NULL; + gtp->sk0 = NULL; + gtp->sk1u = NULL; } static void gtp_encap_destroy(struct sock *sk) @@ -484,14 +486,14 @@ static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev, switch (pctx->gtp_version) { case GTP_V0: - if (gtp->sock0) - sk = gtp->sock0->sk; + if (gtp->sk0) + sk = gtp->sk0; else sk = NULL; break; case GTP_V1: - if (gtp->sock1u) - sk = gtp->sock1u->sk; + if (gtp->sk1u) + sk = gtp->sk1u; else sk = NULL; break; @@ -504,7 +506,7 @@ static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev, return -ENOENT; } - rt = ip4_route_output_gtp(sock_net(sk), &fl4, gtp->sock0->sk, + rt = ip4_route_output_gtp(sock_net(sk), &fl4, gtp->sk0, pctx->sgsn_addr_ip4.s_addr); if (IS_ERR(rt)) { netdev_dbg(dev, "no route to SSGN %pI4\n", @@ -839,18 +841,20 @@ static int gtp_encap_enable(struct net_device *dev, struct gtp_dev *gtp, netdev_dbg(dev, "enable gtp on %p, %p\n", sock0, sock1u); - gtp->sock0 = sock0; - gtp->sock1u = sock1u; + sock_hold(sock0->sk); + gtp->sk0 = sock0->sk; + sock_hold(sock1u->sk); + gtp->sk1u = sock1u->sk; tuncfg.sk_user_data = gtp; tuncfg.encap_rcv = gtp_encap_recv; tuncfg.encap_destroy = gtp_encap_destroy; tuncfg.encap_type = UDP_ENCAP_GTP0; - setup_udp_tunnel_sock(sock_net(gtp->sock0->sk), gtp->sock0, &tuncfg); + setup_udp_tunnel_sock(sock_net(gtp->sk0), sock0, &tuncfg); tuncfg.encap_type = UDP_ENCAP_GTP1U; - setup_udp_tunnel_sock(sock_net(gtp->sock1u->sk), gtp->sock1u, &tuncfg); + setup_udp_tunnel_sock(sock_net(gtp->sk1u), sock1u, &tuncfg); err = 0; err2: -- 2.10.2
From: Andreas Schultz <aschultz@tpip.net> Date: Mon, 13 Feb 2017 16:36:18 +0100
+ if (gtp->sk0) { + udp_sk(gtp->sk0)->encap_type = 0; + rcu_assign_sk_user_data(gtp->sk0, NULL); + sock_put(gtp->sk0); }
This does "sock_put(NULL);" because you are assigning gtp->sk0 to NULL before the sock_put() call. So you are leaking the socket, at best. You need to load the socket pointer into a local variable in order to do this correctly.
----- On Feb 14, 2017, at 6:48 PM, David S. Miller davem@davemloft.net wrote:
From: Andreas Schultz <aschultz@tpip.net> Date: Mon, 13 Feb 2017 16:36:18 +0100
+ if (gtp->sk0) { + udp_sk(gtp->sk0)->encap_type = 0; + rcu_assign_sk_user_data(gtp->sk0, NULL); + sock_put(gtp->sk0); }
This does "sock_put(NULL);" because you are assigning gtp->sk0 to NULL before the sock_put() call. So you are leaking the socket, at best.
I don't understand how this should happen. If I where to use rcu_assign_pointer, then yes, but rcu_assign_sk_user_data does assign to the sk_user_data member of struct sock and not to the argument itself. Andreas
From: Andreas Schultz <aschultz@tpip.net> Date: Wed, 15 Feb 2017 08:04:56 +0100 (CET)
----- On Feb 14, 2017, at 6:48 PM, David S. Miller davem@davemloft.net wrote:
From: Andreas Schultz <aschultz@tpip.net> Date: Mon, 13 Feb 2017 16:36:18 +0100
+ if (gtp->sk0) { + udp_sk(gtp->sk0)->encap_type = 0; + rcu_assign_sk_user_data(gtp->sk0, NULL); + sock_put(gtp->sk0); }
This does "sock_put(NULL);" because you are assigning gtp->sk0 to NULL before the sock_put() call. So you are leaking the socket, at best.
I don't understand how this should happen. If I where to use rcu_assign_pointer, then yes, but rcu_assign_sk_user_data does assign to the sk_user_data member of struct sock and not to the argument itself.
You are right, I misread the assignment.
Having both GTPv0-U and GTPv1-U is not always desirable. Fallback from GTPv1-U to GTPv0-U was depreciated from 3GPP Rel-8 onwards. Post Rel-8 implementation are discuraged from listening on the v0 port (see 3GPP TS 29.281, Sect. 1). A future change will completely decouple the sockets from the network device. Till then, at least one of the sockets needs to be specified (either v0 or v1), the other is optional. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 149 +++++++++++++++++++++++++++++------------------------- 1 file changed, 79 insertions(+), 70 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index a8ce8c7..fe96d21 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -259,30 +259,30 @@ static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, return iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet); } -static void gtp_encap_disable(struct gtp_dev *gtp) -{ - if (gtp->sk0) { - udp_sk(gtp->sk0)->encap_type = 0; - rcu_assign_sk_user_data(gtp->sk0, NULL); - sock_put(gtp->sk0); - } - if (gtp->sk1u) { - udp_sk(gtp->sk1u)->encap_type = 0; - rcu_assign_sk_user_data(gtp->sk1u, NULL); - sock_put(gtp->sk1u); - } - - gtp->sk0 = NULL; - gtp->sk1u = NULL; -} - static void gtp_encap_destroy(struct sock *sk) { struct gtp_dev *gtp; gtp = rcu_dereference_sk_user_data(sk); - if (gtp) - gtp_encap_disable(gtp); + if (gtp) { + udp_sk(sk)->encap_type = 0; + rcu_assign_sk_user_data(sk, NULL); + sock_put(sk); + } +} + +static void gtp_encap_disable_sock(struct sock *sk) +{ + if (!sk) + return; + + gtp_encap_destroy(sk); +} + +static void gtp_encap_disable(struct gtp_dev *gtp) +{ + gtp_encap_disable_sock(gtp->sk0); + gtp_encap_disable_sock(gtp->sk1u); } /* UDP encapsulation receive handler. See net/ipv4/udp.c. @@ -642,27 +642,24 @@ static void gtp_link_setup(struct net_device *dev) static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize); static void gtp_hashtable_free(struct gtp_dev *gtp); -static int gtp_encap_enable(struct net_device *dev, struct gtp_dev *gtp, - int fd_gtp0, int fd_gtp1); +static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]); +static void gtp_encap_disable(struct gtp_dev *gtp); static int gtp_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[], struct nlattr *data[]) { - int hashsize, err, fd0, fd1; struct gtp_dev *gtp; struct gtp_net *gn; + int hashsize, err; - if (!data[IFLA_GTP_FD0] || !data[IFLA_GTP_FD1]) + if (!data[IFLA_GTP_FD0] && !data[IFLA_GTP_FD1]) return -EINVAL; gtp = netdev_priv(dev); - fd0 = nla_get_u32(data[IFLA_GTP_FD0]); - fd1 = nla_get_u32(data[IFLA_GTP_FD1]); - - err = gtp_encap_enable(dev, gtp, fd0, fd1); + err = gtp_encap_enable(gtp, data); if (err < 0) - goto out_err; + return err; if (!data[IFLA_GTP_PDP_HASHSIZE]) hashsize = 1024; @@ -690,7 +687,6 @@ static int gtp_newlink(struct net *src_net, struct net_device *dev, gtp_hashtable_free(gtp); out_encap: gtp_encap_disable(gtp); -out_err: return err; } @@ -805,63 +801,76 @@ static void gtp_hashtable_free(struct gtp_dev *gtp) kfree(gtp->tid_hash); } -static int gtp_encap_enable(struct net_device *dev, struct gtp_dev *gtp, - int fd_gtp0, int fd_gtp1) +static struct sock *gtp_encap_enable_socket(int fd, int type, + struct gtp_dev *gtp) { struct udp_tunnel_sock_cfg tuncfg = {NULL}; - struct socket *sock0, *sock1u; + struct socket *sock; + struct sock *sk; int err; - netdev_dbg(dev, "enable gtp on %d, %d\n", fd_gtp0, fd_gtp1); + pr_debug("enable gtp on %d, %d\n", fd, type); - sock0 = sockfd_lookup(fd_gtp0, &err); - if (sock0 == NULL) { - netdev_dbg(dev, "socket fd=%d not found (gtp0)\n", fd_gtp0); - return -ENOENT; + sock = sockfd_lookup(fd, &err); + if (!sock) { + pr_debug("gtp socket fd=%d not found\n", fd); + return NULL; } - if (sock0->sk->sk_protocol != IPPROTO_UDP) { - netdev_dbg(dev, "socket fd=%d not UDP\n", fd_gtp0); - err = -EINVAL; - goto err1; + if (sock->sk->sk_protocol != IPPROTO_UDP) { + pr_debug("socket fd=%d not UDP\n", fd); + sk = ERR_PTR(-EINVAL); + goto out_sock; } - sock1u = sockfd_lookup(fd_gtp1, &err); - if (sock1u == NULL) { - netdev_dbg(dev, "socket fd=%d not found (gtp1u)\n", fd_gtp1); - err = -ENOENT; - goto err1; + if (rcu_dereference_sk_user_data(sock->sk)) { + sk = ERR_PTR(-EBUSY); + goto out_sock; } - if (sock1u->sk->sk_protocol != IPPROTO_UDP) { - netdev_dbg(dev, "socket fd=%d not UDP\n", fd_gtp1); - err = -EINVAL; - goto err2; - } - - netdev_dbg(dev, "enable gtp on %p, %p\n", sock0, sock1u); - - sock_hold(sock0->sk); - gtp->sk0 = sock0->sk; - sock_hold(sock1u->sk); - gtp->sk1u = sock1u->sk; + sk = sock->sk; + sock_hold(sk); tuncfg.sk_user_data = gtp; + tuncfg.encap_type = type; tuncfg.encap_rcv = gtp_encap_recv; tuncfg.encap_destroy = gtp_encap_destroy; - tuncfg.encap_type = UDP_ENCAP_GTP0; - setup_udp_tunnel_sock(sock_net(gtp->sk0), sock0, &tuncfg); - - tuncfg.encap_type = UDP_ENCAP_GTP1U; - setup_udp_tunnel_sock(sock_net(gtp->sk1u), sock1u, &tuncfg); - - err = 0; -err2: - sockfd_put(sock1u); -err1: - sockfd_put(sock0); - return err; + setup_udp_tunnel_sock(sock_net(sock->sk), sock, &tuncfg); + +out_sock: + sockfd_put(sock); + return sk; +} + +static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]) +{ + struct sock *sk1u = NULL; + struct sock *sk0 = NULL; + + if (data[IFLA_GTP_FD0]) { + u32 fd0 = nla_get_u32(data[IFLA_GTP_FD0]); + + sk0 = gtp_encap_enable_socket(fd0, UDP_ENCAP_GTP0, gtp); + if (IS_ERR(sk0)) + return PTR_ERR(sk0); + } + + if (data[IFLA_GTP_FD1]) { + u32 fd1 = nla_get_u32(data[IFLA_GTP_FD1]); + + sk1u = gtp_encap_enable_socket(fd1, UDP_ENCAP_GTP1U, gtp); + if (IS_ERR(sk1u)) { + if (sk0) + gtp_encap_disable_sock(sk0); + return PTR_ERR(sk1u); + } + } + + gtp->sk0 = sk0; + gtp->sk1u = sk1u; + + return 0; } static struct net_device *gtp_find_dev(struct net *net, int ifindex) -- 2.10.2
Both function are always used together with the final goal to get the gtp_dev. This simplifies the code by merging them together. The netdevice lookup is changed to use the regular dev_get_by_index. The gtp netdevice list is now only used to find the PDP contexts for imcomming packets. It can be completely eliminated Once the TEID hash is moved into the GTP socket. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 147 ++++++++++++++++++++++++++---------------------------- 1 file changed, 70 insertions(+), 77 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index fe96d21..f340e20 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -745,21 +745,6 @@ static struct rtnl_link_ops gtp_link_ops __read_mostly = { .fill_info = gtp_fill_info, }; -static struct net *gtp_genl_get_net(struct net *src_net, struct nlattr *tb[]) -{ - struct net *net; - - /* Examine the link attributes and figure out which network namespace - * we are talking about. - */ - if (tb[GTPA_NET_NS_FD]) - net = get_net_ns_by_fd(nla_get_u32(tb[GTPA_NET_NS_FD])); - else - net = get_net(src_net); - - return net; -} - static int gtp_hashtable_new(struct gtp_dev *gtp, int hsize) { int i; @@ -873,16 +858,31 @@ static int gtp_encap_enable(struct gtp_dev *gtp, struct nlattr *data[]) return 0; } -static struct net_device *gtp_find_dev(struct net *net, int ifindex) +static struct gtp_dev *gtp_genl_find_dev(struct net *src_net, + struct nlattr *tb[]) { - struct gtp_net *gn = net_generic(net, gtp_net_id); - struct gtp_dev *gtp; - - list_for_each_entry_rcu(gtp, &gn->gtp_dev_list, list) { - if (ifindex == gtp->dev->ifindex) - return gtp->dev; - } - return NULL; + struct gtp_dev *gtp = NULL; + struct net_device *dev; + struct net *net; + + /* Examine the link attributes and figure out which network namespace + * we are talking about. + */ + if (tb[GTPA_NET_NS_FD]) + net = get_net_ns_by_fd(nla_get_u32(tb[GTPA_NET_NS_FD])); + else + net = get_net(src_net); + + if (IS_ERR(net)) + return NULL; + + /* Check if there's an existing gtpX device to configure */ + dev = dev_get_by_index_rcu(net, nla_get_u32(tb[GTPA_LINK])); + if (dev->netdev_ops == >p_netdev_ops) + gtp = netdev_priv(dev); + + put_net(net); + return gtp; } static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info) @@ -912,9 +912,9 @@ static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info) } } -static int ipv4_pdp_add(struct net_device *dev, struct genl_info *info) +static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) { - struct gtp_dev *gtp = netdev_priv(dev); + struct net_device *dev = gtp->dev; u32 hash_ms, hash_tid = 0; struct pdp_ctx *pctx; bool found = false; @@ -991,8 +991,8 @@ static int ipv4_pdp_add(struct net_device *dev, struct genl_info *info) static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) { - struct net_device *dev; - struct net *net; + struct gtp_dev *gtp; + int err; if (!info->attrs[GTPA_VERSION] || !info->attrs[GTPA_LINK] || @@ -1016,77 +1016,79 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) return -EINVAL; } - net = gtp_genl_get_net(sock_net(skb->sk), info->attrs); - if (IS_ERR(net)) - return PTR_ERR(net); + rcu_read_lock(); - /* Check if there's an existing gtpX device to configure */ - dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK])); - if (dev == NULL) { - put_net(net); - return -ENODEV; + gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); + if (!gtp) { + err = -ENODEV; + goto out_unlock; } - put_net(net); - return ipv4_pdp_add(dev, info); + err = ipv4_pdp_add(gtp, info); + +out_unlock: + rcu_read_unlock(); + return err; } static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info) { - struct net_device *dev; struct pdp_ctx *pctx; struct gtp_dev *gtp; - struct net *net; + int err = 0; if (!info->attrs[GTPA_VERSION] || !info->attrs[GTPA_LINK]) return -EINVAL; - net = gtp_genl_get_net(sock_net(skb->sk), info->attrs); - if (IS_ERR(net)) - return PTR_ERR(net); + rcu_read_lock(); - /* Check if there's an existing gtpX device to configure */ - dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK])); - if (dev == NULL) { - put_net(net); - return -ENODEV; + gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); + if (!gtp) { + err = -ENODEV; + goto out_unlock; } - put_net(net); - - gtp = netdev_priv(dev); switch (nla_get_u32(info->attrs[GTPA_VERSION])) { case GTP_V0: - if (!info->attrs[GTPA_TID]) - return -EINVAL; + if (!info->attrs[GTPA_TID]) { + err = -EINVAL; + goto out_unlock; + } pctx = gtp0_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_TID])); break; case GTP_V1: - if (!info->attrs[GTPA_I_TEI]) - return -EINVAL; + if (!info->attrs[GTPA_I_TEI]) { + err = -EINVAL; + goto out_unlock; + } pctx = gtp1_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_I_TEI])); break; default: - return -EINVAL; + err = -EINVAL; + goto out_unlock; } - if (pctx == NULL) - return -ENOENT; + if (!pctx) { + err = -ENOENT; + goto out_unlock; + } if (pctx->gtp_version == GTP_V0) - netdev_dbg(dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n", + netdev_dbg(gtp->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n", pctx->u.v0.tid, pctx); else if (pctx->gtp_version == GTP_V1) - netdev_dbg(dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n", + netdev_dbg(gtp->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n", pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx); hlist_del_rcu(&pctx->hlist_tid); hlist_del_rcu(&pctx->hlist_addr); kfree_rcu(pctx, rcu_head); - return 0; +out_unlock: + rcu_read_unlock(); + return err; } static struct genl_family gtp_genl_family; @@ -1130,11 +1132,9 @@ static int gtp_genl_fill_info(struct sk_buff *skb, u32 snd_portid, u32 snd_seq, static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info) { struct pdp_ctx *pctx = NULL; - struct net_device *dev; struct sk_buff *skb2; struct gtp_dev *gtp; u32 gtp_version; - struct net *net; int err; if (!info->attrs[GTPA_VERSION] || @@ -1150,21 +1150,14 @@ static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info) return -EINVAL; } - net = gtp_genl_get_net(sock_net(skb->sk), info->attrs); - if (IS_ERR(net)) - return PTR_ERR(net); - - /* Check if there's an existing gtpX device to configure */ - dev = gtp_find_dev(net, nla_get_u32(info->attrs[GTPA_LINK])); - if (dev == NULL) { - put_net(net); - return -ENODEV; - } - put_net(net); - - gtp = netdev_priv(dev); - rcu_read_lock(); + + gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); + if (!gtp) { + err = -ENODEV; + goto err_unlock; + } + if (gtp_version == GTP_V0 && info->attrs[GTPA_TID]) { u64 tid = nla_get_u64(info->attrs[GTPA_TID]); -- 2.10.2
Add network device to gtp context in preparation for splitting the TEID from the network device. Use this to rework the socker rx path. Move the common RX part of v0 and v1 into a helper. Also move the final rx part into that helper as well. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 80 ++++++++++++++++++++++++++++++------------------------- 1 file changed, 44 insertions(+), 36 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index f340e20..e21b663 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -58,6 +58,8 @@ struct pdp_ctx { struct in_addr ms_addr_ip4; struct in_addr sgsn_addr_ip4; + struct net_device *dev; + atomic_t tx_seq; struct rcu_head rcu_head; }; @@ -175,6 +177,40 @@ static bool gtp_check_src_ms(struct sk_buff *skb, struct pdp_ctx *pctx, return false; } +static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb, unsigned int hdrlen, + bool xnet) +{ + struct pcpu_sw_netstats *stats; + + if (!gtp_check_src_ms(skb, pctx, hdrlen)) { + pr_debug("No PDP ctx for this MS\n"); + return 1; + } + + /* Get rid of the GTP + UDP headers. */ + if (iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet)) + return -1; + + pr_debug("forwarding packet from GGSN to uplink\n"); + + /* Now that the UDP and the GTP header have been removed, set up the + * new network header. This is required by the upper layer to + * calculate the transport header. + */ + skb_reset_network_header(skb); + + skb->dev = pctx->dev; + + stats = this_cpu_ptr(pctx->dev->tstats); + u64_stats_update_begin(&stats->syncp); + stats->rx_packets++; + stats->rx_bytes += skb->len; + u64_stats_update_end(&stats->syncp); + + netif_rx(skb); + return 0; +} + /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */ static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, bool xnet) @@ -201,13 +237,7 @@ static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, return 1; } - if (!gtp_check_src_ms(skb, pctx, hdrlen)) { - netdev_dbg(gtp->dev, "No PDP ctx for this MS\n"); - return 1; - } - - /* Get rid of the GTP + UDP headers. */ - return iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet); + return gtp_rx(pctx, skb, hdrlen, xnet); } static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, @@ -250,13 +280,7 @@ static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, return 1; } - if (!gtp_check_src_ms(skb, pctx, hdrlen)) { - netdev_dbg(gtp->dev, "No PDP ctx for this MS\n"); - return 1; - } - - /* Get rid of the GTP + UDP headers. */ - return iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet); + return gtp_rx(pctx, skb, hdrlen, xnet); } static void gtp_encap_destroy(struct sock *sk) @@ -290,10 +314,9 @@ static void gtp_encap_disable(struct gtp_dev *gtp) */ static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb) { - struct pcpu_sw_netstats *stats; struct gtp_dev *gtp; + int ret = 0; bool xnet; - int ret; gtp = rcu_dereference_sk_user_data(sk); if (!gtp) @@ -319,33 +342,17 @@ static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb) switch (ret) { case 1: netdev_dbg(gtp->dev, "pass up to the process\n"); - return 1; + break; case 0: - netdev_dbg(gtp->dev, "forwarding packet from GGSN to uplink\n"); break; case -1: netdev_dbg(gtp->dev, "GTP packet has been dropped\n"); kfree_skb(skb); - return 0; + ret = 0; + break; } - /* Now that the UDP and the GTP header have been removed, set up the - * new network header. This is required by the upper layer to - * calculate the transport header. - */ - skb_reset_network_header(skb); - - skb->dev = gtp->dev; - - stats = this_cpu_ptr(gtp->dev->tstats); - u64_stats_update_begin(&stats->syncp); - stats->rx_packets++; - stats->rx_bytes += skb->len; - u64_stats_update_end(&stats->syncp); - - netif_rx(skb); - - return 0; + return ret; } static int gtp_dev_init(struct net_device *dev) @@ -953,6 +960,7 @@ static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) if (pctx == NULL) return -ENOMEM; + pctx->dev = gtp->dev; ipv4_pdp_fill(pctx, info); atomic_set(&pctx->tx_seq, 0); -- 2.10.2
This unifies duplicate code into a helper. It also prepares the groundwork to add a lookup version that uses the socket to find attache pdp contexts. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 124 +++++++++++++++++++++++------------------------------- 1 file changed, 53 insertions(+), 71 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index e21b663..d3c384e 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -1039,55 +1039,70 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) return err; } +static struct pdp_ctx *gtp_genl_find_pdp_by_link(struct sk_buff *skb, + struct genl_info *info) +{ + struct gtp_dev *gtp; + + gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); + if (!gtp) + return ERR_PTR(-ENODEV); + + if (info->attrs[GTPA_MS_ADDRESS]) { + __be32 ip = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]); + + return ipv4_pdp_find(gtp, ip); + } else if (info->attrs[GTPA_VERSION]) { + u32 gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]); + + if (gtp_version == GTP_V0 && info->attrs[GTPA_TID]) + return gtp0_pdp_find(gtp, nla_get_u64( + info->attrs[GTPA_TID])); + else if (gtp_version == GTP_V1 && info->attrs[GTPA_I_TEI]) + return gtp1_pdp_find(gtp, nla_get_u32( + info->attrs[GTPA_I_TEI])); + } + + return ERR_PTR(-EINVAL); +} + +static struct pdp_ctx *gtp_genl_find_pdp(struct sk_buff *skb, + struct genl_info *info) +{ + struct pdp_ctx *pctx; + + if (info->attrs[GTPA_LINK]) + pctx = gtp_genl_find_pdp_by_link(skb, info); + else + pctx = ERR_PTR(-EINVAL); + + if (!pctx) + pctx = ERR_PTR(-ENOENT); + + return pctx; +} + static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info) { struct pdp_ctx *pctx; - struct gtp_dev *gtp; int err = 0; - if (!info->attrs[GTPA_VERSION] || - !info->attrs[GTPA_LINK]) + if (!info->attrs[GTPA_VERSION]) return -EINVAL; rcu_read_lock(); - gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); - if (!gtp) { - err = -ENODEV; - goto out_unlock; - } - - switch (nla_get_u32(info->attrs[GTPA_VERSION])) { - case GTP_V0: - if (!info->attrs[GTPA_TID]) { - err = -EINVAL; - goto out_unlock; - } - pctx = gtp0_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_TID])); - break; - case GTP_V1: - if (!info->attrs[GTPA_I_TEI]) { - err = -EINVAL; - goto out_unlock; - } - pctx = gtp1_pdp_find(gtp, nla_get_u64(info->attrs[GTPA_I_TEI])); - break; - - default: - err = -EINVAL; - goto out_unlock; - } - - if (!pctx) { - err = -ENOENT; + pctx = gtp_genl_find_pdp(skb, info); + if (IS_ERR(pctx)) { + err = PTR_ERR(pctx); goto out_unlock; } if (pctx->gtp_version == GTP_V0) - netdev_dbg(gtp->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n", + netdev_dbg(pctx->dev, "GTPv0-U: deleting tunnel id = %llx (pdp %p)\n", pctx->u.v0.tid, pctx); else if (pctx->gtp_version == GTP_V1) - netdev_dbg(gtp->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n", + netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n", pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx); hlist_del_rcu(&pctx->hlist_tid); @@ -1141,49 +1156,16 @@ static int gtp_genl_get_pdp(struct sk_buff *skb, struct genl_info *info) { struct pdp_ctx *pctx = NULL; struct sk_buff *skb2; - struct gtp_dev *gtp; - u32 gtp_version; int err; - if (!info->attrs[GTPA_VERSION] || - !info->attrs[GTPA_LINK]) + if (!info->attrs[GTPA_VERSION]) return -EINVAL; - gtp_version = nla_get_u32(info->attrs[GTPA_VERSION]); - switch (gtp_version) { - case GTP_V0: - case GTP_V1: - break; - default: - return -EINVAL; - } - rcu_read_lock(); - gtp = gtp_genl_find_dev(sock_net(skb->sk), info->attrs); - if (!gtp) { - err = -ENODEV; - goto err_unlock; - } - - if (gtp_version == GTP_V0 && - info->attrs[GTPA_TID]) { - u64 tid = nla_get_u64(info->attrs[GTPA_TID]); - - pctx = gtp0_pdp_find(gtp, tid); - } else if (gtp_version == GTP_V1 && - info->attrs[GTPA_I_TEI]) { - u32 tid = nla_get_u32(info->attrs[GTPA_I_TEI]); - - pctx = gtp1_pdp_find(gtp, tid); - } else if (info->attrs[GTPA_MS_ADDRESS]) { - __be32 ip = nla_get_be32(info->attrs[GTPA_MS_ADDRESS]); - - pctx = ipv4_pdp_find(gtp, ip); - } - - if (pctx == NULL) { - err = -ENOENT; + pctx = gtp_genl_find_pdp(skb, info); + if (IS_ERR(pctx)) { + err = PTR_ERR(pctx); goto err_unlock; } -- 2.10.2
Consolidate duplicate code into helper. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index d3c384e..62f598b 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -86,6 +86,8 @@ struct gtp_net { static u32 gtp_h_initval; +static void pdp_context_delete(struct pdp_ctx *pctx); + static inline u32 gtp0_hashfn(u64 tid) { u32 *tid32 = (u32 *) &tid; @@ -781,13 +783,10 @@ static void gtp_hashtable_free(struct gtp_dev *gtp) struct pdp_ctx *pctx; int i; - for (i = 0; i < gtp->hash_size; i++) { - hlist_for_each_entry_rcu(pctx, >p->tid_hash[i], hlist_tid) { - hlist_del_rcu(&pctx->hlist_tid); - hlist_del_rcu(&pctx->hlist_addr); - kfree_rcu(pctx, rcu_head); - } - } + for (i = 0; i < gtp->hash_size; i++) + hlist_for_each_entry_rcu(pctx, >p->tid_hash[i], hlist_tid) + pdp_context_delete(pctx); + synchronize_rcu(); kfree(gtp->addr_hash); kfree(gtp->tid_hash); @@ -997,6 +996,13 @@ static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) return 0; } +static void pdp_context_delete(struct pdp_ctx *pctx) +{ + hlist_del_rcu(&pctx->hlist_tid); + hlist_del_rcu(&pctx->hlist_addr); + kfree_rcu(pctx, rcu_head); +} + static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) { struct gtp_dev *gtp; @@ -1105,9 +1111,7 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info) netdev_dbg(pctx->dev, "GTPv1-U: deleting tunnel id = %x/%x (pdp %p)\n", pctx->u.v1.i_tei, pctx->u.v1.o_tei, pctx); - hlist_del_rcu(&pctx->hlist_tid); - hlist_del_rcu(&pctx->hlist_addr); - kfree_rcu(pctx, rcu_head); + pdp_context_delete(pctx); out_unlock: rcu_read_unlock(); -- 2.10.2
Having the socket present in context simplifies the sending logic. It also fixes the invalid assumption that we have to use the same sending socket for all client IP's on a specific gtp interface. Signed-off-by: Andreas Schultz <aschultz@tpip.net> --- drivers/net/gtp.c | 94 +++++++++++++++++++++++++++---------------------------- 1 file changed, 47 insertions(+), 47 deletions(-) diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c index 62f598b..45d6f7e 100644 --- a/drivers/net/gtp.c +++ b/drivers/net/gtp.c @@ -58,6 +58,7 @@ struct pdp_ctx { struct in_addr ms_addr_ip4; struct in_addr sgsn_addr_ip4; + struct sock *sk; struct net_device *dev; atomic_t tx_seq; @@ -179,8 +180,7 @@ static bool gtp_check_src_ms(struct sk_buff *skb, struct pdp_ctx *pctx, return false; } -static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb, unsigned int hdrlen, - bool xnet) +static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb, unsigned int hdrlen) { struct pcpu_sw_netstats *stats; @@ -190,7 +190,8 @@ static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb, unsigned int hdrlen } /* Get rid of the GTP + UDP headers. */ - if (iptunnel_pull_header(skb, hdrlen, skb->protocol, xnet)) + if (iptunnel_pull_header(skb, hdrlen, skb->protocol, + !net_eq(sock_net(pctx->sk), dev_net(pctx->dev)))) return -1; pr_debug("forwarding packet from GGSN to uplink\n"); @@ -214,8 +215,7 @@ static int gtp_rx(struct pdp_ctx *pctx, struct sk_buff *skb, unsigned int hdrlen } /* 1 means pass up to the stack, -1 means drop and 0 means decapsulated. */ -static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, - bool xnet) +static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb) { unsigned int hdrlen = sizeof(struct udphdr) + sizeof(struct gtp0_header); @@ -239,11 +239,10 @@ static int gtp0_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, return 1; } - return gtp_rx(pctx, skb, hdrlen, xnet); + return gtp_rx(pctx, skb, hdrlen); } -static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, - bool xnet) +static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb) { unsigned int hdrlen = sizeof(struct udphdr) + sizeof(struct gtp1_header); @@ -282,7 +281,7 @@ static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb, return 1; } - return gtp_rx(pctx, skb, hdrlen, xnet); + return gtp_rx(pctx, skb, hdrlen); } static void gtp_encap_destroy(struct sock *sk) @@ -318,7 +317,6 @@ static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb) { struct gtp_dev *gtp; int ret = 0; - bool xnet; gtp = rcu_dereference_sk_user_data(sk); if (!gtp) @@ -326,16 +324,14 @@ static int gtp_encap_recv(struct sock *sk, struct sk_buff *skb) netdev_dbg(gtp->dev, "encap_recv sk=%p\n", sk); - xnet = !net_eq(sock_net(sk), dev_net(gtp->dev)); - switch (udp_sk(sk)->encap_type) { case UDP_ENCAP_GTP0: netdev_dbg(gtp->dev, "received GTP0 packet\n"); - ret = gtp0_udp_encap_recv(gtp, skb, xnet); + ret = gtp0_udp_encap_recv(gtp, skb); break; case UDP_ENCAP_GTP1U: netdev_dbg(gtp->dev, "received GTP1U packet\n"); - ret = gtp1u_udp_encap_recv(gtp, skb, xnet); + ret = gtp1u_udp_encap_recv(gtp, skb); break; default: ret = -1; /* Shouldn't happen. */ @@ -378,8 +374,9 @@ static void gtp_dev_uninit(struct net_device *dev) free_percpu(dev->tstats); } -static struct rtable *ip4_route_output_gtp(struct net *net, struct flowi4 *fl4, - const struct sock *sk, __be32 daddr) +static struct rtable *ip4_route_output_gtp(struct flowi4 *fl4, + const struct sock *sk, + __be32 daddr) { memset(fl4, 0, sizeof(*fl4)); fl4->flowi4_oif = sk->sk_bound_dev_if; @@ -388,7 +385,7 @@ static struct rtable *ip4_route_output_gtp(struct net *net, struct flowi4 *fl4, fl4->flowi4_tos = RT_CONN_FLAGS(sk); fl4->flowi4_proto = sk->sk_protocol; - return ip_route_output_key(net, fl4); + return ip_route_output_key(sock_net(sk), fl4); } static inline void gtp0_push_header(struct sk_buff *skb, struct pdp_ctx *pctx) @@ -477,7 +474,6 @@ static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev, struct rtable *rt; struct flowi4 fl4; struct iphdr *iph; - struct sock *sk; __be16 df; int mtu; @@ -493,30 +489,7 @@ static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev, } netdev_dbg(dev, "found PDP context %p\n", pctx); - switch (pctx->gtp_version) { - case GTP_V0: - if (gtp->sk0) - sk = gtp->sk0; - else - sk = NULL; - break; - case GTP_V1: - if (gtp->sk1u) - sk = gtp->sk1u; - else - sk = NULL; - break; - default: - return -ENOENT; - } - - if (!sk) { - netdev_dbg(dev, "no userspace socket is available, skip\n"); - return -ENOENT; - } - - rt = ip4_route_output_gtp(sock_net(sk), &fl4, gtp->sk0, - pctx->sgsn_addr_ip4.s_addr); + rt = ip4_route_output_gtp(&fl4, pctx->sk, pctx->sgsn_addr_ip4.s_addr); if (IS_ERR(rt)) { netdev_dbg(dev, "no route to SSGN %pI4\n", &pctx->sgsn_addr_ip4.s_addr); @@ -561,7 +534,7 @@ static int gtp_build_skb_ip4(struct sk_buff *skb, struct net_device *dev, goto err_rt; } - gtp_set_pktinfo_ipv4(pktinfo, sk, iph, pctx, rt, &fl4, dev); + gtp_set_pktinfo_ipv4(pktinfo, pctx->sk, iph, pctx, rt, &fl4, dev); gtp_push_header(skb, pktinfo); return 0; @@ -918,7 +891,8 @@ static void ipv4_pdp_fill(struct pdp_ctx *pctx, struct genl_info *info) } } -static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) +static int ipv4_pdp_add(struct gtp_dev *gtp, struct sock *sk, + struct genl_info *info) { struct net_device *dev = gtp->dev; u32 hash_ms, hash_tid = 0; @@ -959,6 +933,8 @@ static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) if (pctx == NULL) return -ENOMEM; + sock_hold(sk); + pctx->sk = sk; pctx->dev = gtp->dev; ipv4_pdp_fill(pctx, info); atomic_set(&pctx->tx_seq, 0); @@ -996,16 +972,26 @@ static int ipv4_pdp_add(struct gtp_dev *gtp, struct genl_info *info) return 0; } +static void pdp_context_free(struct rcu_head *head) +{ + struct pdp_ctx *pctx = container_of(head, struct pdp_ctx, rcu_head); + + sock_put(pctx->sk); + kfree(pctx); +} + static void pdp_context_delete(struct pdp_ctx *pctx) { hlist_del_rcu(&pctx->hlist_tid); hlist_del_rcu(&pctx->hlist_addr); - kfree_rcu(pctx, rcu_head); + call_rcu(&pctx->rcu_head, pdp_context_free); } static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) { + unsigned int version; struct gtp_dev *gtp; + struct sock *sk; int err; if (!info->attrs[GTPA_VERSION] || @@ -1014,7 +1000,9 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) !info->attrs[GTPA_MS_ADDRESS]) return -EINVAL; - switch (nla_get_u32(info->attrs[GTPA_VERSION])) { + version = nla_get_u32(info->attrs[GTPA_VERSION]); + + switch (version) { case GTP_V0: if (!info->attrs[GTPA_TID] || !info->attrs[GTPA_FLOW]) @@ -1038,7 +1026,19 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info) goto out_unlock; } - err = ipv4_pdp_add(gtp, info); + if (version == GTP_V0) + sk = gtp->sk0; + else if (version == GTP_V1) + sk = gtp->sk1u; + else + sk = NULL; + + if (!sk) { + err = -ENODEV; + goto out_unlock; + } + + err = ipv4_pdp_add(gtp, sk, info); out_unlock: rcu_read_unlock(); -- 2.10.2
Hi Pablo, Harald, Do you have some time to look at the patches? Andreas ----- On Feb 13, 2017, at 4:36 PM, Andreas Schultz aschultz@tpip.net wrote:
Hi Pablo,
This is v3 of the GTP improvements.
This series lays the groundwork for removing the socket references from the GTP netdevice by removing duplicate code and simplifying the logic on some code paths.
It slighly changes the GTP genl API by making the socket parameters optional (though one of them is still required).
The removal of the socket references will break the 1:1 releation between GTP netdevice and GTP socket that prevents us to support multiple VRFs with overlaping IP addresse spaces attached to the same GTP-U entity (needed for multi APN support).
Pablo found a socket hold problem in v2. In order to solve that I had to switch the socket references from the struct socket to the internal struct sock. This should have no functionl impact, but we can now hang on to the reference without blocking user space from closing the GTP socket.
There was also some length off-list conversation about why the netdevice to socket relation needs to be changed at all. Harald did already agree to this but Pablo had some questions. I've tried to address that with a bit of dokumentation for the GTP module. Hopefully this will explain the need for the change sufficiently.
This series does have conflicts with the SGSN side tunnel work done by Jonas Bonn. The unification of the tunnel Rx code touches the same places he changed.
v2->v3: * add documentation to explain the goal of all these changes * incorporate review comments * switch from struct socket to struct sock
Regards Andreas
-- Andreas Schultz (8): gtp: add documentation gtp: switch from struct socket to struct sock for the GTP sockets gtp: make GTP sockets in gtp_newlink optional gtp: merge gtp_get_net and gtp_genl_find_dev gtp: consolidate gtp socket rx path gtp: unify genl_find_pdp and prepare for per socket lookup gtp: consolidate pdp context destruction into helper gtp: add socket to pdp context
Documentation/networking/gtp.txt | 112 ++++++++ drivers/net/gtp.c | 548 +++++++++++++++++++-------------------- 2 files changed, 386 insertions(+), 274 deletions(-) create mode 100644 Documentation/networking/gtp.txt
-- 2.10.2
participants (4)
-
Andreas Schultz -
David Miller -
Harald Welte -
Neels Hofmeyr