diff options
Diffstat (limited to 'proto/ospf')
-rw-r--r-- | proto/ospf/iface.c | 2 | ||||
-rw-r--r-- | proto/ospf/lsalib.c | 4 | ||||
-rw-r--r-- | proto/ospf/lsreq.c | 2 | ||||
-rw-r--r-- | proto/ospf/lsupd.c | 545 | ||||
-rw-r--r-- | proto/ospf/lsupd.h | 16 | ||||
-rw-r--r-- | proto/ospf/neighbor.c | 2 | ||||
-rw-r--r-- | proto/ospf/ospf.c | 2 | ||||
-rw-r--r-- | proto/ospf/topology.c | 10 |
8 files changed, 307 insertions, 276 deletions
diff --git a/proto/ospf/iface.c b/proto/ospf/iface.c index d08fa8b6..0cc88cb0 100644 --- a/proto/ospf/iface.c +++ b/proto/ospf/iface.c @@ -102,7 +102,7 @@ iface_chstate(struct ospf_iface *ifa, u8 state) ifa->nlsa->lsa.age=LSA_MAXAGE; if(state>=OSPF_IS_WAITING) { - net_flush_lsa(ifa->nlsa,po,ifa->oa); + ospf_lsupd_flush_nlsa(ifa->nlsa,ifa->oa); } if(can_flush_lsa(ifa->oa)) flush_lsa(ifa->nlsa,ifa->oa); ifa->nlsa=NULL; diff --git a/proto/ospf/lsalib.c b/proto/ospf/lsalib.c index 1a9653cf..82a5938b 100644 --- a/proto/ospf/lsalib.c +++ b/proto/ospf/lsalib.c @@ -1,7 +1,7 @@ /* * BIRD -- OSPF * - * (c) 1999 - 2000 Ondrej Filip <feela@network.cz> + * (c) 1999--2004 Ondrej Filip <feela@network.cz> * * Can be freely distributed and used under the terms of the GNU GPL. */ @@ -70,7 +70,7 @@ ospf_age(struct ospf_area *oa) en->inst_t=now; en->ini_age=0; lsasum_calculate(&en->lsa,en->lsa_body,po); - flood_lsa(NULL,NULL,&en->lsa,po,NULL,oa,1); + ospf_lsupd_flood(NULL,NULL,&en->lsa,NULL,oa,1); continue; } if((en->lsa.age=(en->ini_age+(now-en->inst_t)))>=LSA_MAXAGE) diff --git a/proto/ospf/lsreq.c b/proto/ospf/lsreq.c index 4e85fad7..c862d73f 100644 --- a/proto/ospf/lsreq.c +++ b/proto/ospf/lsreq.c @@ -116,6 +116,6 @@ ospf_lsreq_receive(struct ospf_lsreq_packet *ps, return; } } - ospf_lsupd_tx_list(n, &uplist); + ospf_lsupd_send_list(n, &uplist); rfree(upslab); } diff --git a/proto/ospf/lsupd.c b/proto/ospf/lsupd.c index c04f16bc..8c82588b 100644 --- a/proto/ospf/lsupd.c +++ b/proto/ospf/lsupd.c @@ -1,7 +1,7 @@ /* * BIRD -- OSPF * - * (c) 2000-2004 Ondrej Filip <feela@network.cz> + * (c) 2000--2004 Ondrej Filip <feela@network.cz> * * Can be freely distributed and used under the terms of the GNU GPL. */ @@ -9,7 +9,7 @@ #include "ospf.h" /** - * flood_lsa - send received or generated lsa to the neighbors + * ospf_lsupd_flood - send received or generated lsa to the neighbors * @n: neighbor than sent this lsa (or NULL if generated) * @hn: LSA header followed by lsa body in network endianity (may be NULL) * @hh: LSA header in host endianity (must be filled) @@ -22,110 +22,127 @@ */ int -flood_lsa(struct ospf_neighbor *n, struct ospf_lsa_header *hn, - struct ospf_lsa_header *hh, struct proto_ospf *po, struct ospf_iface *iff, - struct ospf_area *oa, int rtl) +ospf_lsupd_flood(struct ospf_neighbor *n, struct ospf_lsa_header *hn, + struct ospf_lsa_header *hh, struct ospf_iface *iff, + struct ospf_area *oa, int rtl) { struct ospf_iface *ifa; struct ospf_neighbor *nn; struct top_hash_entry *en; - struct proto *p=&po->proto; + struct proto_ospf *po = oa->po; + struct proto *p = &po->proto; int ret, retval = 0; /* pg 148 */ - WALK_LIST(NODE ifa,po->iface_list) + WALK_LIST(NODE ifa, po->iface_list) { - if(ifa->stub) continue; + if (ifa->stub) + continue; - if(hh->type==LSA_T_EXT) + if (hh->type == LSA_T_EXT) { - if(ifa->type==OSPF_IT_VLINK) continue; - if(ifa->oa->stub) continue; + if (ifa->type == OSPF_IT_VLINK) + continue; + if (ifa->oa->stub) + continue; } else { - if(oa->areaid==BACKBONE) + if (oa->areaid == BACKBONE) { - if((ifa->type!=OSPF_IT_VLINK)&&(ifa->oa!=oa)) continue; + if ((ifa->type != OSPF_IT_VLINK) && (ifa->oa != oa)) + continue; } else { - if(ifa->oa!=oa) continue; + if (ifa->oa != oa) + continue; } } - ret=0; + ret = 0; WALK_LIST(NODE nn, ifa->neigh_list) { - if(nn->state<NEIGHBOR_EXCHANGE) continue; - if(nn->state<NEIGHBOR_FULL) + if (nn->state < NEIGHBOR_EXCHANGE) + continue; + if (nn->state < NEIGHBOR_FULL) { - if((en=ospf_hash_find_header(nn->lsrqh,hh))!=NULL) + if ((en = ospf_hash_find_header(nn->lsrqh, hh)) != NULL) { - switch(lsa_comp(hh,&en->lsa)) + switch (lsa_comp(hh, &en->lsa)) { - case CMP_OLDER: - continue; - break; - case CMP_SAME: - s_rem_node(SNODE en); - if(en->lsa_body!=NULL) mb_free(en->lsa_body); - en->lsa_body=NULL; - DBG("Removing from lsreq list for neigh %I\n", nn->rid); - ospf_hash_delete(nn->lsrqh,en); - if(EMPTY_SLIST(nn->lsrql)) ospf_neigh_sm(nn, INM_LOADDONE); - continue; - break; - case CMP_NEWER: - s_rem_node(SNODE en); - if(en->lsa_body!=NULL) mb_free(en->lsa_body); - en->lsa_body=NULL; - DBG("Removing from lsreq list for neigh %I\n", nn->rid); - ospf_hash_delete(nn->lsrqh,en); - if(EMPTY_SLIST(nn->lsrql)) ospf_neigh_sm(nn, INM_LOADDONE); - break; - default: bug("Bug in lsa_comp?"); + case CMP_OLDER: + continue; + break; + case CMP_SAME: + s_rem_node(SNODE en); + if (en->lsa_body != NULL) + mb_free(en->lsa_body); + en->lsa_body = NULL; + DBG("Removing from lsreq list for neigh %I\n", nn->rid); + ospf_hash_delete(nn->lsrqh, en); + if (EMPTY_SLIST(nn->lsrql)) + ospf_neigh_sm(nn, INM_LOADDONE); + continue; + break; + case CMP_NEWER: + s_rem_node(SNODE en); + if (en->lsa_body != NULL) + mb_free(en->lsa_body); + en->lsa_body = NULL; + DBG("Removing from lsreq list for neigh %I\n", nn->rid); + ospf_hash_delete(nn->lsrqh, en); + if (EMPTY_SLIST(nn->lsrql)) + ospf_neigh_sm(nn, INM_LOADDONE); + break; + default: + bug("Bug in lsa_comp?"); } } } - if(nn==n) continue; + if (nn == n) + continue; - if(rtl) + if (rtl) { - if((en=ospf_hash_find_header(nn->lsrth, hh))==NULL) - { - en=ospf_hash_get_header(nn->lsrth, hh); - } - else - { - s_rem_node(SNODE en); - } - s_add_tail(&nn->lsrtl, SNODE en); - memcpy(&en->lsa,hh,sizeof(struct ospf_lsa_header)); - DBG("Adding LSA lsrt RT: %I, Id: %I, Type: %u for n: %I\n", - en->lsa.rt,en->lsa.id, en->lsa.type, nn->ip); + if ((en = ospf_hash_find_header(nn->lsrth, hh)) == NULL) + { + en = ospf_hash_get_header(nn->lsrth, hh); + } + else + { + s_rem_node(SNODE en); + } + s_add_tail(&nn->lsrtl, SNODE en); + memcpy(&en->lsa, hh, sizeof(struct ospf_lsa_header)); + DBG("Adding LSA lsrt RT: %I, Id: %I, Type: %u for n: %I\n", + en->lsa.rt, en->lsa.id, en->lsa.type, nn->ip); } else { - if((en=ospf_hash_find_header(nn->lsrth, hh))!=NULL) - { - s_rem_node(SNODE en); - if(en->lsa_body!=NULL) mb_free(en->lsa_body); - en->lsa_body=NULL; - ospf_hash_delete(nn->lsrth, en); - } + if ((en = ospf_hash_find_header(nn->lsrth, hh)) != NULL) + { + s_rem_node(SNODE en); + if (en->lsa_body != NULL) + mb_free(en->lsa_body); + en->lsa_body = NULL; + ospf_hash_delete(nn->lsrth, en); + } } - ret=1; + ret = 1; } - if(ret==0) continue; /* pg 150 (2) */ + if (ret == 0) + continue; /* pg 150 (2) */ - if(ifa==iff) + if (ifa == iff) { - if((n->rid==iff->drid)||n->rid==iff->bdrid) continue; /* pg 150 (3) */ - if(iff->state==OSPF_IS_BACKUP) continue; /* pg 150 (4) */ - retval = 1; /* FIXME !!!!!!!!!!! */ + if ((n->rid == iff->drid) || n->rid == iff->bdrid) + continue; /* pg 150 (3) */ + if (iff->state == OSPF_IS_BACKUP) + continue; /* pg 150 (4) */ + retval = 1; } { @@ -136,65 +153,70 @@ flood_lsa(struct ospf_neighbor *n, struct ospf_lsa_header *hn, struct ospf_packet *op; struct ospf_lsa_header *lh; - if(ifa->type==OSPF_IT_NBMA) sk=ifa->ip_sk; - else sk=ifa->hello_sk; /* FIXME is this true for PTP? */ + if (ifa->type == OSPF_IT_NBMA) + sk = ifa->ip_sk; + else + sk = ifa->hello_sk; /* FIXME is this true for PTP? */ - pk=(struct ospf_lsupd_packet *)sk->tbuf; - op=(struct ospf_packet *)sk->tbuf; + pk = (struct ospf_lsupd_packet *) sk->tbuf; + op = (struct ospf_packet *) sk->tbuf; fill_ospf_pkt_hdr(ifa, pk, LSUPD_P); - pk->lsano=htonl(1); + pk->lsano = htonl(1); - lh = (struct ospf_lsa_header *)(pk+1); + lh = (struct ospf_lsa_header *) (pk + 1); /* Copy LSA into the packet */ - if(hn) + if (hn) { - memcpy(lh, hn, ntohs(hn->length)); + memcpy(lh, hn, ntohs(hn->length)); } else { - u8 *help; + u8 *help; struct top_hash_entry *en; - htonlsah(hh,lh); - help = (u8 *)(lh+1); - en = ospf_hash_find_header(oa->gr,hh); - htonlsab(en->lsa_body,help,hh->type,hh->length - -sizeof(struct ospf_lsa_header)); + htonlsah(hh, lh); + help = (u8 *) (lh + 1); + en = ospf_hash_find_header(oa->gr, hh); + htonlsab(en->lsa_body, help, hh->type, hh->length + - sizeof(struct ospf_lsa_header)); } - len = sizeof(struct ospf_lsupd_packet)+ntohs(lh->length); + len = sizeof(struct ospf_lsupd_packet) + ntohs(lh->length); age = ntohs(lh->age); age += ifa->inftransdelay; - if( age > LSA_MAXAGE ) age = LSA_MAXAGE; + if (age > LSA_MAXAGE) + age = LSA_MAXAGE; lh->age = htons(age); - op->length=htons(len); + op->length = htons(len); ospf_pkt_finalize(ifa, op); OSPF_TRACE(D_PACKETS, "LS upd flooded via %s", ifa->iface->name); - if(ifa->type==OSPF_IT_NBMA) + if (ifa->type == OSPF_IT_NBMA) { - if((ifa->state==OSPF_IS_BACKUP)||(ifa->state==OSPF_IS_DR)) - sk_send_to_agt(sk ,len, ifa, NEIGHBOR_EXCHANGE); - else sk_send_to_bdr(sk ,len, ifa); + if ((ifa->state == OSPF_IS_BACKUP) || (ifa->state == OSPF_IS_DR)) + sk_send_to_agt(sk, len, ifa, NEIGHBOR_EXCHANGE); + else + sk_send_to_bdr(sk, len, ifa); } else { - if((ifa->state==OSPF_IS_BACKUP)||(ifa->state==OSPF_IS_DR)|| - (ifa->type==OSPF_IT_PTP)) - sk_send_to(sk,len, AllSPFRouters, OSPF_PROTO); - else sk_send_to(sk,len, AllDRouters, OSPF_PROTO); + if ((ifa->state == OSPF_IS_BACKUP) || (ifa->state == OSPF_IS_DR) || + (ifa->type == OSPF_IT_PTP)) + sk_send_to(sk, len, AllSPFRouters, OSPF_PROTO); + else + sk_send_to(sk, len, AllDRouters, OSPF_PROTO); } } } return retval; } -void /* I send all I received in LSREQ */ -ospf_lsupd_tx_list(struct ospf_neighbor *n, list *l) +void /* I send all I received in LSREQ */ +ospf_lsupd_send_list(struct ospf_neighbor *n, list * l) { struct l_lsr_head *llsh; u16 len; @@ -202,95 +224,98 @@ ospf_lsupd_tx_list(struct ospf_neighbor *n, list *l) struct top_hash_entry *en; struct ospf_lsupd_packet *pk; struct ospf_packet *op; - struct proto *p=&n->ifa->oa->po->proto; + struct proto *p = &n->ifa->oa->po->proto; void *pktpos; u8 ii; - u8 *jj=n->ifa->ip_sk->tbuf; + u8 *jj = n->ifa->ip_sk->tbuf; - if(EMPTY_LIST(*l)) return; + if (EMPTY_LIST(*l)) + return; - pk=(struct ospf_lsupd_packet *)n->ifa->ip_sk->tbuf; - op=(struct ospf_packet *)n->ifa->ip_sk->tbuf; + pk = (struct ospf_lsupd_packet *) n->ifa->ip_sk->tbuf; + op = (struct ospf_packet *) n->ifa->ip_sk->tbuf; DBG("LSupd: 1st packet\n"); - + fill_ospf_pkt_hdr(n->ifa, pk, LSUPD_P); - len=SIPH+sizeof(struct ospf_lsupd_packet); - lsano=0; - pktpos=(pk+1); + len = SIPH + sizeof(struct ospf_lsupd_packet); + lsano = 0; + pktpos = (pk + 1); WALK_LIST(llsh, *l) { - if((en=ospf_hash_find(n->ifa->oa->gr,llsh->lsh.id,llsh->lsh.rt, - llsh->lsh.type))==NULL) continue; /* Probably flushed LSA */ + if ((en = ospf_hash_find(n->ifa->oa->gr, llsh->lsh.id, llsh->lsh.rt, + llsh->lsh.type)) == NULL) + continue; /* Probably flushed LSA */ DBG("Sending ID=%I, Type=%u, RT=%I Sn: 0x%x Age: %u\n", - llsh->lsh.id, llsh->lsh.type, llsh->lsh.rt, en->lsa.sn, en->lsa.age); - if(((u32)(len+en->lsa.length))>n->ifa->iface->mtu) + llsh->lsh.id, llsh->lsh.type, llsh->lsh.rt, en->lsa.sn, en->lsa.age); + if (((u32) (len + en->lsa.length)) > n->ifa->iface->mtu) { - pk->lsano=htonl(lsano); - op->length=htons(len-SIPH); + pk->lsano = htonl(lsano); + op->length = htons(len - SIPH); ospf_pkt_finalize(n->ifa, op); - - sk_send_to(n->ifa->ip_sk,len-SIPH, n->ip, OSPF_PROTO); + + sk_send_to(n->ifa->ip_sk, len - SIPH, n->ip, OSPF_PROTO); OSPF_TRACE(D_PACKETS, "LS upd sent to %I (%d LSAs)", n->ip, lsano); DBG("LSupd: next packet\n"); fill_ospf_pkt_hdr(n->ifa, pk, LSUPD_P); - len=SIPH+sizeof(struct ospf_lsupd_packet); - lsano=0; - pktpos=(pk+1); + len = SIPH + sizeof(struct ospf_lsupd_packet); + lsano = 0; + pktpos = (pk + 1); } htonlsah(&(en->lsa), pktpos); - pktpos=pktpos+sizeof(struct ospf_lsa_header); + pktpos = pktpos + sizeof(struct ospf_lsa_header); htonlsab(en->lsa_body, pktpos, en->lsa.type, en->lsa.length - -sizeof(struct ospf_lsa_header)); - pktpos=pktpos+en->lsa.length-sizeof(struct ospf_lsa_header); - len+=en->lsa.length; + - sizeof(struct ospf_lsa_header)); + pktpos = pktpos + en->lsa.length - sizeof(struct ospf_lsa_header); + len += en->lsa.length; lsano++; } - if(lsano>0) + if (lsano > 0) { - pk->lsano=htonl(lsano); - op->length=htons(len-SIPH); + pk->lsano = htonl(lsano); + op->length = htons(len - SIPH); ospf_pkt_finalize(n->ifa, op); OSPF_TRACE(D_PACKETS, "LS upd sent to %I (%d LSAs)", n->ip, lsano); - sk_send_to(n->ifa->ip_sk,len-SIPH, n->ip, OSPF_PROTO); + sk_send_to(n->ifa->ip_sk, len - SIPH, n->ip, OSPF_PROTO); } } void ospf_lsupd_receive(struct ospf_lsupd_packet *ps, - struct ospf_iface *ifa, u16 size) + struct ospf_iface *ifa, u16 size) { - u32 area,nrid; - struct ospf_neighbor *n,*ntmp; + u32 area, nrid; + struct ospf_neighbor *n, *ntmp; struct ospf_lsa_header *lsa; struct ospf_area *oa; - struct proto_ospf *po=ifa->proto; - struct proto *p = (struct proto *)po; + struct proto_ospf *po = ifa->proto; + struct proto *p = (struct proto *) po; u16 length; u8 i; int sendreq = 1; - nrid=ntohl(ps->ospf_packet.routerid); + nrid = ntohl(ps->ospf_packet.routerid); - if((n=find_neigh(ifa, nrid))==NULL) + if ((n = find_neigh(ifa, nrid)) == NULL) { - OSPF_TRACE(D_PACKETS, "Received lsupd from unknown neighbor! (%I)", - nrid); - return ; + OSPF_TRACE(D_PACKETS, "Received lsupd from unknown neighbor! (%I)", nrid); + return; } - if(n->state<NEIGHBOR_EXCHANGE) + if (n->state < NEIGHBOR_EXCHANGE) { - OSPF_TRACE(D_PACKETS, "Received lsupd in lesser state than EXCHANGE from (%I)", - n->ip); + OSPF_TRACE(D_PACKETS, + "Received lsupd in lesser state than EXCHANGE from (%I)", + n->ip); return; } - if(size<=(sizeof(struct ospf_lsupd_packet)+sizeof(struct ospf_lsa_header))) + if (size <= + (sizeof(struct ospf_lsupd_packet) + sizeof(struct ospf_lsa_header))) { log(L_WARN "Received lsupd from %I is too short!", n->ip); return; @@ -299,26 +324,28 @@ ospf_lsupd_receive(struct ospf_lsupd_packet *ps, OSPF_TRACE(D_PACKETS, "Received LS upd from %I", n->ip); ospf_neigh_sm(n, INM_HELLOREC); /* Questionable */ - lsa=(struct ospf_lsa_header *)(ps+1); - area=htonl(ps->ospf_packet.areaid); - oa=ospf_find_area((struct proto_ospf *)p,area); + lsa = (struct ospf_lsa_header *) (ps + 1); + area = htonl(ps->ospf_packet.areaid); + oa = ospf_find_area((struct proto_ospf *) p, area); - for(i=0; i<ntohl(ps->lsano); i++, - lsa=(struct ospf_lsa_header *)(((u8 *)lsa)+ntohs(lsa->length))) + for (i = 0; i < ntohl(ps->lsano); i++, + lsa = (struct ospf_lsa_header *) (((u8 *) lsa) + ntohs(lsa->length))) { struct ospf_lsa_header lsatmp; struct top_hash_entry *lsadb; - int diff=((u8 *)lsa)-((u8 *)ps); + int diff = ((u8 *) lsa) - ((u8 *) ps); u16 chsum, lenn = ntohs(lsa->length); - if(((diff+sizeof(struct ospf_lsa_header))>=size) || ((lenn+diff)>size)) + if (((diff + sizeof(struct ospf_lsa_header)) >= size) + || ((lenn + diff) > size)) { log(L_WARN "Received lsupd from %I is too short!", n->ip); ospf_neigh_sm(n, INM_BADLSREQ); break; } - if((lenn<=sizeof(struct ospf_lsa_header))||(lenn!=(4*(lenn/4)))) + if ((lenn <= sizeof(struct ospf_lsa_header)) + || (lenn != (4 * (lenn / 4)))) { log(L_WARN "Received LSA from %I with bad length", n->ip); ospf_neigh_sm(n, INM_BADLSREQ); @@ -326,53 +353,53 @@ ospf_lsupd_receive(struct ospf_lsupd_packet *ps, } /* pg 143 (1) */ - chsum=lsa->checksum; - if(chsum!=lsasum_check(lsa,NULL,po)) + chsum = lsa->checksum; + if (chsum != lsasum_check(lsa, NULL, po)) { log(L_WARN "Received bad lsa checksum from %I", n->ip); continue; } /* pg 143 (2) */ - if((lsa->type<LSA_T_RT)||(lsa->type>LSA_T_EXT)) + if ((lsa->type < LSA_T_RT) || (lsa->type > LSA_T_EXT)) { log(L_WARN "Unknown LSA type from %I", n->ip); continue; } /* pg 143 (3) */ - if((lsa->type==LSA_T_EXT)&&oa->stub) + if ((lsa->type == LSA_T_EXT) && oa->stub) { log(L_WARN "Received External LSA in stub area from %I", n->ip); continue; } - ntohlsah(lsa,&lsatmp); + ntohlsah(lsa, &lsatmp); DBG("Update Type: %u ID: %I RT: %I, Sn: 0x%08x Age: %u, Sum: %u\n", - lsatmp.type, lsatmp.id, lsatmp.rt, lsatmp.sn, lsatmp.age, - lsatmp.checksum); + lsatmp.type, lsatmp.id, lsatmp.rt, lsatmp.sn, lsatmp.age, + lsatmp.checksum); - lsadb=ospf_hash_find_header(oa->gr, &lsatmp); + lsadb = ospf_hash_find_header(oa->gr, &lsatmp); #ifdef LOCAL_DEBUG - if(lsadb) + if (lsadb) DBG("I have Type: %u ID: %I RT: %I, Sn: 0x%08x Age: %u, Sum: %u\n", - lsadb->lsa.type, lsadb->lsa.id, lsadb->lsa.rt, lsadb->lsa.sn, - lsadb->lsa.age, lsadb->lsa.checksum); + lsadb->lsa.type, lsadb->lsa.id, lsadb->lsa.rt, lsadb->lsa.sn, + lsadb->lsa.age, lsadb->lsa.checksum); #endif /* pg 143 (4) */ - if((lsatmp.age==LSA_MAXAGE)&&(lsadb==NULL)&&can_flush_lsa(oa)) + if ((lsatmp.age == LSA_MAXAGE) && (lsadb == NULL) && can_flush_lsa(oa)) { ospf_lsack_enqueue(n, lsa, ACKL_DIRECT); continue; } /* pg 144 (5) */ - if((lsadb==NULL)||(lsa_comp(&lsatmp,&lsadb->lsa)==CMP_NEWER)) + if ((lsadb == NULL) || (lsa_comp(&lsatmp, &lsadb->lsa) == CMP_NEWER)) { - struct ospf_iface *ift=NULL; + struct ospf_iface *ift = NULL; void *body; struct ospf_iface *nifa; int self = (lsatmp.rt == p->cf->global->router_id); @@ -380,93 +407,98 @@ ospf_lsupd_receive(struct ospf_lsupd_packet *ps, DBG("PG143(5): Received LSA is newer\n"); /* pg 145 (5f) - premature aging of self originated lsa */ - if((!self) && (lsatmp.type==LSA_T_NET)) + if ((!self) && (lsatmp.type == LSA_T_NET)) { - WALK_LIST(nifa,po->iface_list) - { - if(ipa_compare(nifa->iface->addr->ip,ipa_from_u32(lsatmp.id))==0) - { - self = 1; - break; - } - } + WALK_LIST(nifa, po->iface_list) + { + if (ipa_compare(nifa->iface->addr->ip, ipa_from_u32(lsatmp.id)) == + 0) + { + self = 1; + break; + } + } } - if(self) + if (self) { - struct top_hash_entry *en; - - if((lsatmp.age==LSA_MAXAGE)&&(lsatmp.sn==LSA_MAXSEQNO)) - { - ospf_lsack_enqueue(n, lsa, ACKL_DIRECT); - continue; - } - - lsatmp.age=LSA_MAXAGE; - lsatmp.sn=LSA_MAXSEQNO; - lsa->age=htons(LSA_MAXAGE); - lsa->sn=htonl(LSA_MAXSEQNO); - OSPF_TRACE(D_EVENTS, "Premature aging self originated lsa."); - OSPF_TRACE(D_EVENTS, "Type: %d, Id: %I, Rt: %I",lsatmp.type, - lsatmp.id, lsatmp.rt); - lsasum_check(lsa,(lsa+1),po); /* It also calculates chsum! */ - lsatmp.checksum=ntohs(lsa->checksum); - flood_lsa(NULL,lsa,&lsatmp,po,NULL,oa,0); - if(en=ospf_hash_find_header(oa->gr,&lsatmp)) - { - flood_lsa(NULL,NULL,&en->lsa,po,NULL,oa,1); - } - continue; + struct top_hash_entry *en; + + if ((lsatmp.age == LSA_MAXAGE) && (lsatmp.sn == LSA_MAXSEQNO)) + { + ospf_lsack_enqueue(n, lsa, ACKL_DIRECT); + continue; + } + + lsatmp.age = LSA_MAXAGE; + lsatmp.sn = LSA_MAXSEQNO; + lsa->age = htons(LSA_MAXAGE); + lsa->sn = htonl(LSA_MAXSEQNO); + OSPF_TRACE(D_EVENTS, "Premature aging self originated lsa."); + OSPF_TRACE(D_EVENTS, "Type: %d, Id: %I, Rt: %I", lsatmp.type, + lsatmp.id, lsatmp.rt); + lsasum_check(lsa, (lsa + 1), po); /* It also calculates chsum! */ + lsatmp.checksum = ntohs(lsa->checksum); + ospf_lsupd_flood(NULL, lsa, &lsatmp, NULL, oa, 0); + if (en = ospf_hash_find_header(oa->gr, &lsatmp)) + { + ospf_lsupd_flood(NULL, NULL, &en->lsa, NULL, oa, 1); + } + continue; } /* pg 144 (5a) */ - if(lsadb && ((now-lsadb->inst_t)<=MINLSARRIVAL)) /* FIXME: test for flooding? */ + if (lsadb && ((now - lsadb->inst_t) <= MINLSARRIVAL)) /* FIXME: test for flooding? */ { - DBG("I got it in less that MINLSARRIVAL\n"); - sendreq = 0; + DBG("I got it in less that MINLSARRIVAL\n"); + sendreq = 0; continue; } - - if(flood_lsa(n,lsa,&lsatmp,po,ifa,ifa->oa,1)==0) + + if (ospf_lsupd_flood(n, lsa, &lsatmp, ifa, ifa->oa, 1) == 0) { - DBG("Wasn't flooded back\n"); /* ps 144(5e), pg 153 */ - if(ifa->state==OSPF_IS_BACKUP) + DBG("Wasn't flooded back\n"); /* ps 144(5e), pg 153 */ + if (ifa->state == OSPF_IS_BACKUP) { - if(ifa->drid==n->rid) ospf_lsack_enqueue(n, lsa, ACKL_DELAY); + if (ifa->drid == n->rid) + ospf_lsack_enqueue(n, lsa, ACKL_DELAY); } - else ospf_lsack_enqueue(n, lsa, ACKL_DELAY); + else + ospf_lsack_enqueue(n, lsa, ACKL_DELAY); } /* Remove old from all ret lists */ /* pg 144 (5c) */ - if(lsadb) - WALK_LIST(NODE ift,po->iface_list) - WALK_LIST(NODE ntmp,ift->neigh_list) - { - struct top_hash_entry *en; - if(ntmp->state>NEIGHBOR_EXSTART) - if((en=ospf_hash_find_header(ntmp->lsrth,&lsadb->lsa))!=NULL) - { - s_rem_node(SNODE en); - if(en->lsa_body!=NULL) mb_free(en->lsa_body); - en->lsa_body=NULL; - ospf_hash_delete(ntmp->lsrth,en); - } - } - - if((lsatmp.age==LSA_MAXAGE)&&(lsatmp.sn==LSA_MAXSEQNO) - &&lsadb&&can_flush_lsa(oa)) + if (lsadb) + WALK_LIST(NODE ift, po->iface_list) + WALK_LIST(NODE ntmp, ift->neigh_list) { - flush_lsa(lsadb,oa); - schedule_rtcalc(oa); - continue; - } /* FIXME lsack? */ + struct top_hash_entry *en; + if (ntmp->state > NEIGHBOR_EXSTART) + if ((en = ospf_hash_find_header(ntmp->lsrth, &lsadb->lsa)) != NULL) + { + s_rem_node(SNODE en); + if (en->lsa_body != NULL) + mb_free(en->lsa_body); + en->lsa_body = NULL; + ospf_hash_delete(ntmp->lsrth, en); + } + } + + if ((lsatmp.age == LSA_MAXAGE) && (lsatmp.sn == LSA_MAXSEQNO) + && lsadb && can_flush_lsa(oa)) + { + flush_lsa(lsadb, oa); + schedule_rtcalc(oa); + continue; + } /* FIXME lsack? */ /* pg 144 (5d) */ - body=mb_alloc(p->pool,lsatmp.length-sizeof(struct ospf_lsa_header)); - ntohlsab(lsa+1,body,lsatmp.type, - lsatmp.length-sizeof(struct ospf_lsa_header)); - lsadb=lsa_install_new(&lsatmp,body, oa); + body = + mb_alloc(p->pool, lsatmp.length - sizeof(struct ospf_lsa_header)); + ntohlsab(lsa + 1, body, lsatmp.type, + lsatmp.length - sizeof(struct ospf_lsa_header)); + lsadb = lsa_install_new(&lsatmp, body, oa); DBG("New LSA installed in DB\n"); continue; @@ -475,33 +507,35 @@ ospf_lsupd_receive(struct ospf_lsupd_packet *ps, /* FIXME pg145 (6) */ /* pg145 (7) */ - if(lsa_comp(&lsatmp,&lsadb->lsa)==CMP_SAME) + if (lsa_comp(&lsatmp, &lsadb->lsa) == CMP_SAME) { struct top_hash_entry *en; DBG("PG145(7) Got the same LSA\n"); - if((en=ospf_hash_find_header(n->lsrth,&lsadb->lsa))!=NULL) + if ((en = ospf_hash_find_header(n->lsrth, &lsadb->lsa)) != NULL) { - /* pg145 (7a) */ - s_rem_node(SNODE en); - if(en->lsa_body!=NULL) mb_free(en->lsa_body); - en->lsa_body=NULL; - ospf_hash_delete(n->lsrth, en); - if(ifa->state==OSPF_IS_BACKUP) - { - if(n->rid==ifa->drid) ospf_lsack_enqueue(n, lsa, ACKL_DELAY); - } + /* pg145 (7a) */ + s_rem_node(SNODE en); + if (en->lsa_body != NULL) + mb_free(en->lsa_body); + en->lsa_body = NULL; + ospf_hash_delete(n->lsrth, en); + if (ifa->state == OSPF_IS_BACKUP) + { + if (n->rid == ifa->drid) + ospf_lsack_enqueue(n, lsa, ACKL_DELAY); + } } else { - /* pg145 (7b) */ - ospf_lsack_enqueue(n, lsa, ACKL_DIRECT); + /* pg145 (7b) */ + ospf_lsack_enqueue(n, lsa, ACKL_DIRECT); } sendreq = 0; continue; } /* pg145 (8) */ - if((lsadb->lsa.age==LSA_MAXAGE)&&(lsadb->lsa.sn==LSA_MAXSEQNO)) + if ((lsadb->lsa.age == LSA_MAXAGE) && (lsadb->lsa.sn == LSA_MAXSEQNO)) { continue; } @@ -510,36 +544,35 @@ ospf_lsupd_receive(struct ospf_lsupd_packet *ps, list l; struct l_lsr_head ll; init_list(&l); - ll.lsh.id=lsadb->lsa.id; - ll.lsh.rt=lsadb->lsa.rt; - ll.lsh.type=lsadb->lsa.type; - add_tail(&l, NODE &ll); - ospf_lsupd_tx_list(n, &l); + ll.lsh.id = lsadb->lsa.id; + ll.lsh.rt = lsadb->lsa.rt; + ll.lsh.type = lsadb->lsa.type; + add_tail(&l, NODE & ll); + ospf_lsupd_send_list(n, &l); } } /* Send direct LSAs */ ospf_lsack_send(n, ACKL_DIRECT); - if(sendreq && (n->state==NEIGHBOR_LOADING)) + if (sendreq && (n->state == NEIGHBOR_LOADING)) { - ospf_lsreq_send(n); /* Ask for another part of neighbor's database */ + ospf_lsreq_send(n); /* Ask for another part of neighbor's database */ } } void -net_flush_lsa(struct top_hash_entry *en, struct proto_ospf *po, - struct ospf_area *oa) +ospf_lsupd_flush_nlsa(struct top_hash_entry *en, struct ospf_area *oa) { - struct ospf_lsa_header *lsa=&en->lsa; - struct proto *p=&po->proto; + struct ospf_lsa_header *lsa = &en->lsa; + struct proto_ospf *po = oa->po; + struct proto *p = &po->proto; - lsa->age=LSA_MAXAGE; - lsa->sn=LSA_MAXSEQNO; - lsasum_calculate(lsa,en->lsa_body,po); + lsa->age = LSA_MAXAGE; + lsa->sn = LSA_MAXSEQNO; + lsasum_calculate(lsa, en->lsa_body, po); OSPF_TRACE(D_EVENTS, "Premature aging self originated lsa!"); OSPF_TRACE(D_EVENTS, "Type: %d, Id: %I, Rt: %I", lsa->type, - lsa->id, lsa->rt); - flood_lsa(NULL,NULL,lsa,po,NULL,oa,0); + lsa->id, lsa->rt); + ospf_lsupd_flood(NULL, NULL, lsa, NULL, oa, 0); } - diff --git a/proto/ospf/lsupd.h b/proto/ospf/lsupd.h index 5bbc97ab..8c046759 100644 --- a/proto/ospf/lsupd.h +++ b/proto/ospf/lsupd.h @@ -1,7 +1,7 @@ /* * BIRD -- OSPF * - * (c) 2000 Ondrej Filip <feela@network.cz> + * (c) 2000--2004 Ondrej Filip <feela@network.cz> * * Can be freely distributed and used under the terms of the GNU GPL. * @@ -10,14 +10,12 @@ #ifndef _BIRD_OSPF_LSUPD_H_ #define _BIRD_OSPF_LSUPD_H_ -void ospf_lsupd_tx_list(struct ospf_neighbor *n, list *l); +void ospf_lsupd_send_list(struct ospf_neighbor *n, list * l); void ospf_lsupd_receive(struct ospf_lsupd_packet *ps, - struct ospf_iface *ifa, u16 size); -int flood_lsa(struct ospf_neighbor *n, struct ospf_lsa_header *hn, - struct ospf_lsa_header *hh, struct proto_ospf *po, struct ospf_iface *iff, - struct ospf_area *oa, int rtl); -void net_flush_lsa(struct top_hash_entry *en, struct proto_ospf *po, - struct ospf_area *oa); - + struct ospf_iface *ifa, u16 size); +int ospf_lsupd_flood(struct ospf_neighbor *n, struct ospf_lsa_header *hn, + struct ospf_lsa_header *hh, struct ospf_iface *iff, + struct ospf_area *oa, int rtl); +void ospf_lsupd_flush_nlsa(struct top_hash_entry *en, struct ospf_area *oa); #endif /* _BIRD_OSPF_LSUPD_H_ */ diff --git a/proto/ospf/neighbor.c b/proto/ospf/neighbor.c index 9c9bd376..885b0b89 100644 --- a/proto/ospf/neighbor.c +++ b/proto/ospf/neighbor.c @@ -600,7 +600,7 @@ rxmt_timer_hook(timer *timer) en->lsa.id, en->lsa.rt, en->lsa.type); add_tail(&uplist, NODE llsh); } - ospf_lsupd_tx_list(n, &uplist); + ospf_lsupd_send_list(n, &uplist); rfree(upslab); } } diff --git a/proto/ospf/ospf.c b/proto/ospf/ospf.c index 498b978b..53ad210a 100644 --- a/proto/ospf/ospf.c +++ b/proto/ospf/ospf.c @@ -405,7 +405,7 @@ ospf_rt_notify(struct proto *p, net *n, rte *new, rte *old, ea_list *attrs) ext=en->lsa_body; if(ipa_compare(ext->netmask, ipa_mkmask(n->n.pxlen))==0) { - net_flush_lsa(en,po,oa); + ospf_lsupd_flush_nlsa(en,oa); break; } } diff --git a/proto/ospf/topology.c b/proto/ospf/topology.c index 29629e20..9d35bac8 100644 --- a/proto/ospf/topology.c +++ b/proto/ospf/topology.c @@ -2,7 +2,7 @@ * BIRD -- OSPF Topological Database * * (c) 1999 Martin Mares <mj@ucw.cz> - * (c) 1999 - 2000 Ondrej Filip <feela@network.cz> + * (c) 1999 - 2004 Ondrej Filip <feela@network.cz> * * Can be freely distributed and used under the terms of the GNU GPL. */ @@ -215,7 +215,7 @@ originate_rt_lsa(struct ospf_area *oa) lsasum_calculate(&lsa,body,po); en=lsa_install_new(&lsa, body, oa); oa->rt=en; - flood_lsa(NULL,NULL,&oa->rt->lsa,po,NULL,oa,1); + ospf_lsupd_flood(NULL,NULL,&oa->rt->lsa,NULL,oa,1); schedule_rtcalc(oa); oa->origrt=0; } @@ -282,7 +282,7 @@ originate_net_lsa(struct ospf_iface *ifa) ifa->iface->name); ifa->nlsa->lsa.sn+=1; ifa->nlsa->lsa.age=LSA_MAXAGE; - flood_lsa(NULL,NULL,&ifa->nlsa->lsa,po,NULL,ifa->oa,0); + ospf_lsupd_flood(NULL,NULL,&ifa->nlsa->lsa,NULL,ifa->oa,0); s_rem_node(SNODE ifa->nlsa); if(ifa->nlsa->lsa_body!=NULL) mb_free(ifa->nlsa->lsa_body); ifa->nlsa->lsa_body=NULL; @@ -312,7 +312,7 @@ originate_net_lsa(struct ospf_iface *ifa) body=originate_net_lsa_body(ifa, &lsa.length, po); lsasum_calculate(&lsa,body,po); ifa->nlsa=lsa_install_new(&lsa, body, ifa->oa); - flood_lsa(NULL,NULL,&ifa->nlsa->lsa,po,NULL,ifa->oa,1); + ospf_lsupd_flood(NULL,NULL,&ifa->nlsa->lsa,NULL,ifa->oa,1); ifa->orignet=0; } @@ -445,7 +445,7 @@ originate_ext_lsa(net *n, rte *e, struct proto_ospf *po, struct ea_list *attrs) WALK_LIST(oa, po->area_list) { en=lsa_install_new(&lsa, body, oa); - flood_lsa(NULL,NULL,&en->lsa,po,NULL,oa,1); + ospf_lsupd_flood(NULL,NULL,&en->lsa,NULL,oa,1); body=originate_ext_lsa_body(n, e, po, attrs); } mb_free(body); |