bird/proto/ospf/topology.c

398 lines
9.3 KiB
C
Raw Normal View History

/*
* BIRD -- OSPF Topological Database
*
2000-02-26 03:19:41 +08:00
* (c) 1999 Martin Mares <mj@ucw.cz>
* (c) 1999 - 2000 Ondrej Filip <feela@network.cz>
*
* Can be freely distributed and used under the terms of the GNU GPL.
*/
#define LOCAL_DEBUG
#include "nest/bird.h"
#include "lib/string.h"
#include "ospf.h"
#define HASH_DEF_ORDER 6 /* FIXME: Increase */
#define HASH_HI_MARK *4
#define HASH_HI_STEP 2
#define HASH_HI_MAX 16
#define HASH_LO_MARK /5
#define HASH_LO_STEP 2
#define HASH_LO_MIN 8
void *
originate_rt_lsa_body(struct ospf_area *oa, u16 *length, struct proto_ospf *p)
{
struct ospf_iface *ifa;
int j=0,k=0,v=0,e=0,b=0;
u16 i=0;
struct ospf_lsa_rt *rt;
struct ospf_lsa_rt_link *ln;
struct ospf_neighbor *neigh;
struct top_hash_entry *old;
struct proto_ospf *po=(struct proto_ospf *)p;
2000-04-19 03:22:49 +08:00
DBG("%s: Originating RT_lsa body for area \"%I\".\n", po->proto.name, oa->areaid);
WALK_LIST (ifa, p->iface_list) i++;
{
if((ifa->an==oa->areaid) && (ifa->state!=OSPF_IS_DOWN))
{
i++;
if(ifa->type==OSPF_IT_VLINK) v=1;
}
}
rt=mb_allocz(p->proto.pool, sizeof(struct ospf_lsa_rt)+
i*sizeof(struct ospf_lsa_rt_link));
if((p->areano>1) && (!oa->stub)) e=1;
rt->VEB=(v>>LSA_RT_V)+(e>>LSA_RT_E)+(b>>LSA_RT_B);
ln=(struct ospf_lsa_rt_link *)(rt+1);
WALK_LIST (ifa, p->iface_list)
{
if((ifa->an==oa->areaid) && (ifa->state!=OSPF_IS_DOWN))
{
if(ifa->state==OSPF_IS_LOOP)
{
ln->type=3;
ln->id=ipa_to_u32(ifa->iface->addr->ip);
ln->data=0xffffffff;
ln->metric=0;
ln->notos=0;
}
else
{
switch(ifa->type)
{
case OSPF_IT_PTP: /* rfc2328 - pg126 */
neigh=(struct ospf_neighbor *)HEAD(ifa->neigh_list);
if((neigh!=NULL) || (neigh->state==NEIGHBOR_FULL))
{
ln->type=LSART_PTP;
ln->id=neigh->rid;
ln->metric=ifa->cost;
ln->notos=0;
if(ifa->iface->flags && IA_UNNUMBERED)
{
ln->data=ifa->iface->index;
}
else
{
ln->id=ipa_to_u32(ifa->iface->addr->ip);
}
}
else
{
if(ifa->state==OSPF_IS_PTP)
{
ln->type=LSART_STUB;
ln->id=ln->id=ipa_to_u32(ifa->iface->addr->opposite);
ln->metric=ifa->cost;
ln->notos=0;
ln->data=0xffffffff;
}
else
{
i--; /* No link added */
}
}
break;
case OSPF_IT_BCAST: /*FIXME Go on */
case OSPF_IT_NBMA:
if(ifa->state==OSPF_IS_WAITING)
{
ln->type=LSART_STUB;
ln->id=ipa_to_u32(ifa->iface->addr->prefix);
ln->data=ipa_to_u32(ipa_mkmask(ifa->iface->addr->pxlen));
ln->metric=ifa->cost;
ln->notos=0;
}
else
{
j=0,k=0;
WALK_LIST(neigh, ifa->neigh_list)
{
if((neigh->rid==ifa->drid) &&
(neigh->state==NEIGHBOR_FULL)) k=1;
if(neigh->state==NEIGHBOR_FULL) j=1;
}
if(((ifa->state=OSPF_IS_DR) && (j==1)) || (k==1))
{
ln->type=LSART_NET;
ln->id=ipa_to_u32(ifa->drip);
ln->data=ipa_to_u32(ifa->iface->addr->ip);
ln->metric=ifa->cost;
ln->notos=0;
}
else
{
ln->type=LSART_STUB;
ln->id=ipa_to_u32(ifa->iface->addr->prefix);
ln->data=ipa_to_u32(ipa_mkmask(ifa->iface->addr->pxlen));
ln->metric=ifa->cost;
ln->notos=0;
}
}
break;
case OSPF_IT_VLINK: /* FIXME Add virtual links! */
i--;
break;
}
}
if(ifa->type==OSPF_IT_VLINK) v=1;
}
ln=(ln+1);
}
rt->links=i;
*length=i*sizeof(struct ospf_lsa_rt_link)+sizeof(struct ospf_lsa_rt)+
sizeof(struct ospf_lsa_header);
return rt;
}
void
addifa_rtlsa(struct ospf_iface *ifa)
{
struct ospf_area *oa;
struct proto_ospf *po;
2000-02-26 03:19:41 +08:00
u32 rtid;
struct top_graph_rtlsa_link *li, *lih;
po=ifa->proto;
2000-02-26 03:19:41 +08:00
rtid=po->proto.cf->global->router_id;
DBG("%s: New OSPF area \"%d\" adding.\n", po->proto.name, ifa->an);
oa=NULL;
WALK_LIST(NODE oa,po->area_list)
{
if(oa->areaid==ifa->an) break;
}
if(EMPTY_LIST(po->area_list) || (oa->areaid!=ifa->an)) /* New area */
{
2000-03-29 08:34:28 +08:00
struct ospf_lsa_header *lsa;
oa=mb_allocz(po->proto.pool, sizeof(struct ospf_area));
add_tail(&po->area_list,NODE oa);
oa->areaid=ifa->an;
2000-02-26 03:19:41 +08:00
oa->gr=ospf_top_new(po);
s_init_list(&(oa->lsal));
oa->rt=NULL;
po->areano++;
DBG("%s: New OSPF area \"%d\" added.\n", po->proto.name, ifa->an);
2000-02-26 03:19:41 +08:00
}
ifa->oa=oa;
oa->rt=originate_rt_lsa(oa,po);
2000-04-18 04:42:42 +08:00
flood_lsa(NULL,NULL,&oa->rt->lsa,po,NULL,oa);
}
struct top_hash_entry *
originate_rt_lsa(struct ospf_area *oa, struct proto_ospf *po)
{
struct ospf_lsa_header lsa;
u32 rtid=po->proto.cf->global->router_id;
struct top_hash_entry *en;
void *body;
DBG("%s: Originating RT_lsa for area \"%d\".\n", po->proto.name, oa->areaid);
lsa.age=0;
lsa.id=rtid;
lsa.type=LSA_T_RT;
lsa.rt=rtid;
if(oa->rt==NULL)
{
lsa.sn=LSA_INITSEQNO;
}
else
{
lsa.sn=oa->rt->lsa.sn+1;
}
body=originate_rt_lsa_body(oa, &lsa.length, po);
lsasum_calculate(&lsa,body,po);
en=lsa_install_new(&lsa, body, oa);
return en;
}
static void
ospf_top_ht_alloc(struct top_graph *f)
{
f->hash_size = 1 << f->hash_order;
f->hash_mask = f->hash_size - 1;
if (f->hash_order > HASH_HI_MAX - HASH_HI_STEP)
f->hash_entries_max = ~0;
else
f->hash_entries_max = f->hash_size HASH_HI_MARK;
if (f->hash_order < HASH_LO_MIN + HASH_LO_STEP)
f->hash_entries_min = 0;
else
f->hash_entries_min = f->hash_size HASH_LO_MARK;
DBG("Allocating OSPF hash of order %d: %d hash_entries, %d low, %d high\n",
f->hash_order, f->hash_size, f->hash_entries_min, f->hash_entries_max);
f->hash_table = mb_alloc(f->pool, f->hash_size * sizeof(struct top_hash_entry *));
bzero(f->hash_table, f->hash_size * sizeof(struct top_hash_entry *));
}
static inline void
ospf_top_ht_free(struct top_hash_entry **h)
{
mb_free(h);
}
static inline u32
ospf_top_hash_u32(u32 a)
{
/* Shamelessly stolen from IP address hashing in ipv4.h */
a ^= a >> 16;
a ^= a << 10;
return a;
}
static inline unsigned
ospf_top_hash(struct top_graph *f, u32 lsaid, u32 rtrid, u32 type)
{
return (ospf_top_hash_u32(lsaid) + ospf_top_hash_u32(rtrid) + type) & f->hash_mask;
}
struct top_graph *
ospf_top_new(struct proto_ospf *p)
{
struct top_graph *f;
f = mb_allocz(p->proto.pool, sizeof(struct top_graph));
f->pool = p->proto.pool;
f->hash_slab = sl_new(f->pool, sizeof(struct top_hash_entry));
f->hash_order = HASH_DEF_ORDER;
ospf_top_ht_alloc(f);
f->hash_entries = 0;
f->hash_entries_min = 0;
return f;
}
void
ospf_top_free(struct top_graph *f)
{
rfree(f->hash_slab);
ospf_top_ht_free(f->hash_table);
mb_free(f);
}
static void
ospf_top_rehash(struct top_graph *f, int step)
{
unsigned int oldn, oldh;
struct top_hash_entry **n, **oldt, **newt, *e, *x;
oldn = f->hash_size;
oldt = f->hash_table;
DBG("Re-hashing topology hash from order %d to %d\n", f->hash_order, f->hash_order+step);
f->hash_order += step;
ospf_top_ht_alloc(f);
newt = f->hash_table;
for(oldh=0; oldh < oldn; oldh++)
{
e = oldt[oldh];
while (e)
{
x = e->next;
n = newt + ospf_top_hash(f, e->lsa.id, e->lsa.rt, e->lsa.type);
e->next = *n;
*n = e;
e = x;
}
}
ospf_top_ht_free(oldt);
}
2000-04-04 08:32:17 +08:00
struct top_hash_entry *
ospf_hash_find_header(struct top_graph *f, struct ospf_lsa_header *h)
{
return ospf_hash_find(f,h->id,h->rt,h->type);
}
struct top_hash_entry *
ospf_hash_get_header(struct top_graph *f, struct ospf_lsa_header *h)
{
return ospf_hash_get(f,h->id,h->rt,h->type);
}
struct top_hash_entry *
ospf_hash_find(struct top_graph *f, u32 lsa, u32 rtr, u32 type)
{
struct top_hash_entry *e = f->hash_table[ospf_top_hash(f, lsa, rtr, type)];
while (e && (e->lsa.id != lsa || e->lsa.rt != rtr || e->lsa.type != type))
e = e->next;
return e;
}
struct top_hash_entry *
ospf_hash_get(struct top_graph *f, u32 lsa, u32 rtr, u32 type)
{
struct top_hash_entry **ee = f->hash_table + ospf_top_hash(f, lsa, rtr, type);
struct top_hash_entry *e = *ee;
while (e && (e->lsa.id != lsa || e->lsa.rt != rtr || e->lsa.type != type))
e = e->next;
if (e)
return e;
e = sl_alloc(f->hash_slab);
e->lsa.id = lsa;
e->lsa.rt = rtr;
e->lsa.type = type;
e->lsa_body = NULL;
2000-02-26 03:19:41 +08:00
e->next=*ee; /* MJ you forgot this :-) */
*ee=e;
if (f->hash_entries++ > f->hash_entries_max)
ospf_top_rehash(f, HASH_HI_STEP);
return e;
}
void
ospf_hash_delete(struct top_graph *f, struct top_hash_entry *e)
{
unsigned int h = ospf_top_hash(f, e->lsa.id, e->lsa.rt, e->lsa.type);
struct top_hash_entry **ee = f->hash_table + h;
while (*ee)
{
if (*ee == e)
{
*ee = e->next;
sl_free(f->hash_slab, e);
if (f->hash_entries-- < f->hash_entries_min)
ospf_top_rehash(f, -HASH_LO_STEP);
return;
}
ee = &((*ee)->next);
}
bug("ospf_hash_delete() called for invalid node");
}
void
ospf_top_dump(struct top_graph *f)
{
unsigned int i;
2000-02-26 03:19:41 +08:00
debug("Hash entries: %d\n", f->hash_entries);
for(i=0; i<f->hash_size; i++)
{
struct top_hash_entry *e = f->hash_table[i];
while (e)
{
debug("\t%04x %08x %08x %p\n", e->lsa.type, e->lsa.id,
e->lsa.rt, e->lsa_body);
e = e->next;
}
}
}
2000-02-26 03:19:41 +08:00