DragonFly On-Line Manual Pages

NATM(4) 	      DragonFly Kernel Interfaces Manual	       NATM(4)


natm -- Native Mode ATM protocol layer


options NATM


The BSD ATM software comes with a native mode ATM protocol layer which provides socket level access to AAL0 and AAL5 virtual circuits.


The NATM layer uses a struct sockaddr_natm to specify a virtual circuit: struct sockaddr_natm { u_int8_t snatm_len; /* length */ u_int8_t snatm_family; /* AF_NATM */ char snatm_if[IFNAMSIZ]; /* interface name */ u_int16_t snatm_vci; /* vci */ u_int8_t snatm_vpi; /* vpi */ }; To create an AAL5 connection to a virtual circuit with VPI 0, VCI 201 one would use the following: struct sockaddr_natm snatm; int s, r; s = socket(AF_NATM, SOCK_STREAM, PROTO_NATMAAL5); /* note: PROTO_NATMAAL0 is AAL0 */ if (s < 0) { perror("socket"); exit(1); } bzero(&snatm, sizeof(snatm)); snatm.snatm_len = sizeof(snatm); snatm.snatm_family = AF_NATM; sprintf(snatm.snatm_if, "en0"); snatm.snatm_vci = 201; snatm.snatm_vpi = 0; r = connect(s, (struct sockaddr *)&snatm, sizeof(snatm)); if (r < 0) { perror("connect"); exit(1); } /* s now connected to ATM! */ The socket() call simply creates an unconnected NATM socket. The connect() call associates an unconnected NATM socket with a virtual cir- cuit and tells the driver to enable that virtual circuit for receiving data. After the connect() call one can read() or write() to the socket to perform ATM I/O. Internal NATM operation Internally, the NATM protocol layer keeps a list of all active virtual circuits on the system in natm_pcbs. This includes circuits currently being used for IP to prevent NATM and IP from clashing over virtual cir- cuit usage. When a virtual circuit is enabled for receiving data, the NATM protocol layer passes the address of the protocol control block down to the driver as a receive ``handle''. When inbound data arrives, the driver passes the data back with the appropriate receive handle. The NATM layer uses this to avoid the overhead of a protocol control block lookup. This allows us to take advantage of the fact that ATM has already demulti- plexed the data for us. Other NATM issues We are currently involved with a video server project and are using this driver as part of it. We have a device we build called an MMX. You can connect a video camera to an MMX and have it send you a stream of AAL0 cells with the video output in it. Of course this stream is pretty rapid (in fact, it is massive!), and the normal AAL0 handling of the driver is unable to handle it (you end up with a cell per small mbuf trying to make it to the application ... it turns out the socket layer can't keep up with that sort of data stream). To solve this problem we have imple- mented a ``raw'' mode which batches unprocessed AAL0 info from the card into larger data chunks blocks. We can save this data to disk in real- time without the socket layer freaking out. Unfortunately, the data has RBD (receive buffer descriptors) and cells headers in it, and this has to be filtered out after capture. To enable ``raw'' mode one does the following ioctl: int size = 4000; /* bytes */ ret = ioctl(s, SIOCRAWATM, (caddr_t)&size); This tells the driver to batch AAL0 data into 4000 bytes chunks, rather than the usual 48 bytes chunks. Admittedly this is somewhat gross, but our current application requires it. In the future we hope that video sources send data in nice large AAL5 frames.


The NATM protocol support is subject to change as the ATM protocols develop. Users should not depend on details of the current implementa- tion, but rather the services exported.




Chuck Cranor of Washington University implemented the NATM protocol layer along with the EN ATM driver in 1996 for NetBSD. DragonFly 3.7 December 29, 1997 DragonFly 3.7