Added some comments.

This commit is contained in:
Ondrej Zajicek 2010-03-26 16:21:29 +01:00
parent af157fa3db
commit 48cff379a7
3 changed files with 48 additions and 1 deletions

View file

@ -97,6 +97,27 @@ ospf_sk_open(struct ospf_iface *ifa)
goto err; goto err;
#endif #endif
/*
* For OSPFv2: When sending a packet, it is important to have a
* proper source address. We expect that when we send one-hop
* unicast packets, OS chooses a source address according to the
* destination address (to be in the same prefix). We also expect
* that when we send multicast packets, OS uses the source address
* from sk->saddr registered to OS by sk_setup_multicast(). This
* behavior is needed to implement multiple virtual ifaces (struct
* ospf_iface) on one physical iface and is signalized by
* CONFIG_MC_PROPER_SRC.
*
* If this behavior is not available (for example on BSD), we create
* non-stub iface just for the primary IP address (see
* ospf_iface_stubby()) and we expect OS to use primary IP address
* as a source address for both unicast and multicast packets.
*
* FIXME: the primary IP address is currently just the
* lexicographically smallest address on an interface, it should be
* signalized by sysdep code which one is really the primary.
*/
sk->saddr = ifa->addr->ip; sk->saddr = ifa->addr->ip;
if (sk_setup_multicast(sk) < 0) if (sk_setup_multicast(sk) < 0)
goto err; goto err;

View file

@ -42,6 +42,32 @@
* and deletion. Each LSA is kept in two pieces: header and body. Both of them are * and deletion. Each LSA is kept in two pieces: header and body. Both of them are
* kept in the endianity of the CPU. * kept in the endianity of the CPU.
* *
* In OSPFv2 specification, it is implied that there is one IP prefix
* for each physical network/interface (unless it is an ptp link). But
* in modern systems, there might be more independent IP prefixes
* associated with an interface. To handle this situation, we have
* one &ospf_iface for each active IP prefix (instead for each active
* iface); This behaves like virtual interface for the purpose of OSPF.
* If we receive packet, we associate it with a proper virtual interface
* mainly according to its source address.
*
* OSPF keeps one socket per &ospf_iface. This allows us (compared to
* one socket approach) to evade problems with a limit of multicast
* groups per socket and with sending multicast packets to appropriate
* interface in a portable way. The socket is associated with
* underlying physical iface and should not receive packets received
* on other ifaces (unfortunately, this is not true on
* BSD). Generally, one packet can be received by more sockets (for
* example, if there are more &ospf_iface on one physical iface),
* therefore we explicitly filter received packets according to
* src/dst IP address and received iface.
*
* Vlinks are implemented using particularly degenerate form of
* &ospf_iface, which has several exceptions: it does not have its
* iface or socket (it copies these from 'parent' &ospf_iface) and it
* is present in iface list even when down (it is not freed in
* ospf_iface_down()).
*
* The heart beat of ospf is ospf_disp(). It is called at regular intervals * The heart beat of ospf is ospf_disp(). It is called at regular intervals
* (&proto_ospf->tick). It is responsible for aging and flushing of LSAs in * (&proto_ospf->tick). It is responsible for aging and flushing of LSAs in
* the database, for routing table calculaction and it call area_disp() of every * the database, for routing table calculaction and it call area_disp() of every

View file

@ -159,7 +159,7 @@ struct ospf_iface
{ {
node n; node n;
struct iface *iface; /* Nest's iface */ struct iface *iface; /* Nest's iface */
struct ifa *addr; struct ifa *addr; /* IP prefix associated with that OSPF iface */
struct ospf_area *oa; struct ospf_area *oa;
struct object_lock *lock; struct object_lock *lock;
sock *sk; /* IP socket (for DD ...) */ sock *sk; /* IP socket (for DD ...) */