1
0
mirror of https://git.FreeBSD.org/src.git synced 2024-12-26 11:47:31 +00:00
freebsd/usr.sbin/mrouted/route.c
Peter Wemm 2b026b8f2b Resync our mainline to mrouted release 3.8.
This will make FreeBSD boxes better behaved 'MBONE Citizens', based on
a couple of the comments about the severity of fixes..

Agreed to by: wollman, fenner@parc.xerox.com
1996-01-06 21:10:30 +00:00

1163 lines
30 KiB
C

/*
* The mrouted program is covered by the license in the accompanying file
* named "LICENSE". Use of the mrouted program represents acceptance of
* the terms and conditions listed in that file.
*
* The mrouted program is COPYRIGHT 1989 by The Board of Trustees of
* Leland Stanford Junior University.
*
*
* $Id: route.c,v 3.8 1995/11/29 22:36:57 fenner Rel $
*/
#include "defs.h"
/*
* This define statement saves a lot of space later
*/
#define RT_ADDR (struct rtentry *)&routing_table
/*
* Exported variables.
*/
int routes_changed; /* 1=>some routes have changed */
int delay_change_reports; /* 1=>postpone change reports */
/*
* The routing table is shared with prune.c , so must not be static.
*/
struct rtentry *routing_table; /* pointer to list of route entries */
/*
* Private variables.
*/
static struct rtentry *rtp; /* pointer to a route entry */
static struct rtentry *rt_end; /* pointer to last route entry */
unsigned int nroutes; /* current number of route entries */
/*
* Private functions.
*/
static int init_children_and_leaves __P((struct rtentry *r,
vifi_t parent));
static int find_route __P((u_int32 origin, u_int32 mask));
static void create_route __P((u_int32 origin, u_int32 mask));
static void discard_route __P((struct rtentry *prev_r));
static int compare_rts __P((const void *rt1, const void *rt2));
static int report_chunk __P((struct rtentry *start_rt, vifi_t vifi,
u_int32 dst));
/*
* Initialize the routing table and associated variables.
*/
void
init_routes()
{
routing_table = NULL;
rt_end = RT_ADDR;
nroutes = 0;
routes_changed = FALSE;
delay_change_reports = FALSE;
}
/*
* Initialize the children and leaf bits for route 'r', along with the
* associated dominant, subordinate, and leaf timing data structures.
* Return TRUE if this changes the value of either the children or
* leaf bitmaps for 'r'.
*/
static int
init_children_and_leaves(r, parent)
register struct rtentry *r;
register vifi_t parent;
{
register vifi_t vifi;
register struct uvif *v;
vifbitmap_t old_children, old_leaves;
VIFM_COPY(r->rt_children, old_children);
VIFM_COPY(r->rt_leaves, old_leaves );
VIFM_CLRALL(r->rt_children);
VIFM_CLRALL(r->rt_leaves);
r->rt_flags &= ~RTF_LEAF_TIMING;
for (vifi = 0, v = uvifs; vifi < numvifs; ++vifi, ++v) {
r->rt_dominants [vifi] = 0;
r->rt_subordinates[vifi] = 0;
if (vifi != parent && !(v->uv_flags & (VIFF_DOWN|VIFF_DISABLED))) {
VIFM_SET(vifi, r->rt_children);
if (v->uv_neighbors == NULL) {
VIFM_SET(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = 0;
}
else {
r->rt_leaf_timers[vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
}
else {
r->rt_leaf_timers[vifi] = 0;
}
}
return (!VIFM_SAME(r->rt_children, old_children) ||
!VIFM_SAME(r->rt_leaves, old_leaves));
}
/*
* A new vif has come up -- update the children and leaf bitmaps in all route
* entries to take that into account.
*/
void
add_vif_to_routes(vifi)
register vifi_t vifi;
{
register struct rtentry *r;
register struct uvif *v;
v = &uvifs[vifi];
for (r = routing_table; r != NULL; r = r->rt_next) {
if (r->rt_metric != UNREACHABLE &&
!VIFM_ISSET(vifi, r->rt_children)) {
VIFM_SET(vifi, r->rt_children);
r->rt_dominants [vifi] = 0;
r->rt_subordinates[vifi] = 0;
if (v->uv_neighbors == NULL) {
VIFM_SET(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = 0;
}
else {
VIFM_CLR(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
update_table_entry(r);
}
}
}
/*
* A vif has gone down -- expire all routes that have that vif as parent,
* and update the children bitmaps in all other route entries to take into
* account the failed vif.
*/
void
delete_vif_from_routes(vifi)
register vifi_t vifi;
{
register struct rtentry *r;
for (r = routing_table; r != NULL; r = r->rt_next) {
if (r->rt_metric != UNREACHABLE) {
if (vifi == r->rt_parent) {
del_table_entry(r, 0, DEL_ALL_ROUTES);
r->rt_timer = ROUTE_EXPIRE_TIME;
r->rt_metric = UNREACHABLE;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
}
else if (VIFM_ISSET(vifi, r->rt_children)) {
VIFM_CLR(vifi, r->rt_children);
VIFM_CLR(vifi, r->rt_leaves);
r->rt_subordinates[vifi] = 0;
r->rt_leaf_timers [vifi] = 0;
update_table_entry(r);
}
else {
r->rt_dominants[vifi] = 0;
}
}
}
}
/*
* A neighbor has failed or become unreachable. If that neighbor was
* considered a dominant or subordinate router in any route entries,
* take appropriate action.
*/
void
delete_neighbor_from_routes(addr, vifi)
register u_int32 addr;
register vifi_t vifi;
{
register struct rtentry *r;
register struct uvif *v;
v = &uvifs[vifi];
for (r = routing_table; r != NULL; r = r->rt_next) {
if (r->rt_metric != UNREACHABLE) {
if (r->rt_dominants[vifi] == addr) {
VIFM_SET(vifi, r->rt_children);
r->rt_dominants [vifi] = 0;
r->rt_subordinates[vifi] = 0;
if (v->uv_neighbors == NULL) {
VIFM_SET(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = 0;
}
else {
VIFM_CLR(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
update_table_entry(r);
}
else if (r->rt_subordinates[vifi] == addr) {
r->rt_subordinates[vifi] = 0;
if (v->uv_neighbors == NULL) {
VIFM_SET(vifi, r->rt_leaves);
update_table_entry(r);
}
else {
r->rt_leaf_timers[vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
}
else if (v->uv_neighbors == NULL &&
r->rt_leaf_timers[vifi] != 0) {
VIFM_SET(vifi, r->rt_leaves);
r->rt_leaf_timers[vifi] = 0;
update_table_entry(r);
}
}
}
}
/*
* Prepare for a sequence of ordered route updates by initializing a pointer
* to the start of the routing table. The pointer is used to remember our
* position in the routing table in order to avoid searching from the
* beginning for each update; this relies on having the route reports in
* a single message be in the same order as the route entries in the routing
* table.
*/
void
start_route_updates()
{
rtp = RT_ADDR;
}
/*
* Starting at the route entry following the one to which 'rtp' points,
* look for a route entry matching the specified origin and mask. If a
* match is found, return TRUE and leave 'rtp' pointing at the found entry.
* If no match is found, return FALSE and leave 'rtp' pointing to the route
* entry preceding the point at which the new origin should be inserted.
* This code is optimized for the normal case in which the first entry to
* be examined is the matching entry.
*/
static int
find_route(origin, mask)
register u_int32 origin, mask;
{
register struct rtentry *r;
r = rtp->rt_next;
while (r != NULL) {
if (origin == r->rt_origin && mask == r->rt_originmask) {
rtp = r;
return (TRUE);
}
if (ntohl(mask) < ntohl(r->rt_originmask) ||
(mask == r->rt_originmask &&
ntohl(origin) < ntohl(r->rt_origin))) {
rtp = r;
r = r->rt_next;
}
else break;
}
return (FALSE);
}
/*
* Create a new routing table entry for the specified origin and link it into
* the routing table. The shared variable 'rtp' is assumed to point to the
* routing entry after which the new one should be inserted. It is left
* pointing to the new entry.
*
* Only the origin, originmask, originwidth and flags fields are initialized
* in the new route entry; the caller is responsible for filling in the the
* rest.
*/
static void
create_route(origin, mask)
u_int32 origin, mask;
{
register struct rtentry *r;
if ((r = (struct rtentry *) malloc(sizeof(struct rtentry) +
(2 * numvifs * sizeof(u_int32)) +
(numvifs * sizeof(u_int)))) == NULL) {
log(LOG_ERR, 0, "ran out of memory"); /* fatal */
}
r->rt_origin = origin;
r->rt_originmask = mask;
if (((char *)&mask)[3] != 0) r->rt_originwidth = 4;
else if (((char *)&mask)[2] != 0) r->rt_originwidth = 3;
else if (((char *)&mask)[1] != 0) r->rt_originwidth = 2;
else r->rt_originwidth = 1;
r->rt_flags = 0;
r->rt_dominants = (u_int32 *)(r + 1);
r->rt_subordinates = (u_int32 *)(r->rt_dominants + numvifs);
r->rt_leaf_timers = (u_int *)(r->rt_subordinates + numvifs);
r->rt_groups = NULL;
r->rt_next = rtp->rt_next;
rtp->rt_next = r;
r->rt_prev = rtp;
if (r->rt_next != NULL)
(r->rt_next)->rt_prev = r;
else
rt_end = r;
rtp = r;
++nroutes;
}
/*
* Discard the routing table entry following the one to which 'prev_r' points.
*/
static void
discard_route(prev_r)
register struct rtentry *prev_r;
{
register struct rtentry *r;
r = prev_r->rt_next;
prev_r->rt_next = r->rt_next;
if (prev_r->rt_next != NULL)
(prev_r->rt_next)->rt_prev = prev_r;
else
rt_end = prev_r;
free((char *)r);
--nroutes;
}
/*
* Process a route report for a single origin, creating or updating the
* corresponding routing table entry if necessary. 'src' is either the
* address of a neighboring router from which the report arrived, or zero
* to indicate a change of status of one of our own interfaces.
*/
void
update_route(origin, mask, metric, src, vifi)
u_int32 origin, mask;
u_int metric;
u_int32 src;
vifi_t vifi;
{
register struct rtentry *r;
u_int adj_metric;
/*
* Compute an adjusted metric, taking into account the cost of the
* subnet or tunnel over which the report arrived, and normalizing
* all unreachable/poisoned metrics into a single value.
*/
if (src != 0 && (metric < 1 || metric >= 2*UNREACHABLE)) {
log(LOG_WARNING, 0,
"%s reports out-of-range metric %u for origin %s",
inet_fmt(src, s1), metric, inet_fmts(origin, mask, s2));
return;
}
adj_metric = metric + uvifs[vifi].uv_metric;
if (adj_metric > UNREACHABLE) adj_metric = UNREACHABLE;
/*
* Look up the reported origin in the routing table.
*/
if (!find_route(origin, mask)) {
/*
* Not found.
* Don't create a new entry if the report says it's unreachable,
* or if the reported origin and mask are invalid.
*/
if (adj_metric == UNREACHABLE) {
return;
}
if (src != 0 && !inet_valid_subnet(origin, mask)) {
log(LOG_WARNING, 0,
"%s reports an invalid origin (%s) and/or mask (%08x)",
inet_fmt(src, s1), inet_fmt(origin, s2), ntohl(mask));
return;
}
/*
* OK, create the new routing entry. 'rtp' will be left pointing
* to the new entry.
*/
create_route(origin, mask);
/*
* Now "steal away" any sources that belong under this route
* by deleting any cache entries they might have created
* and allowing the kernel to re-request them.
*/
steal_sources(rtp);
rtp->rt_metric = UNREACHABLE; /* temporary; updated below */
}
/*
* We now have a routing entry for the reported origin. Update it?
*/
r = rtp;
if (r->rt_metric == UNREACHABLE) {
/*
* The routing entry is for a formerly-unreachable or new origin.
* If the report claims reachability, update the entry to use
* the reported route.
*/
if (adj_metric == UNREACHABLE)
return;
r->rt_parent = vifi;
init_children_and_leaves(r, vifi);
r->rt_gateway = src;
r->rt_timer = 0;
r->rt_metric = adj_metric;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
update_table_entry(r);
}
else if (src == r->rt_gateway) {
/*
* The report has come either from the interface directly-connected
* to the origin subnet (src and r->rt_gateway both equal zero) or
* from the gateway we have chosen as the best first-hop gateway back
* towards the origin (src and r->rt_gateway not equal zero). Reset
* the route timer and, if the reported metric has changed, update
* our entry accordingly.
*/
r->rt_timer = 0;
if (adj_metric == r->rt_metric)
return;
if (adj_metric == UNREACHABLE) {
del_table_entry(r, 0, DEL_ALL_ROUTES);
r->rt_timer = ROUTE_EXPIRE_TIME;
}
else if (adj_metric < r->rt_metric) {
if (init_children_and_leaves(r, vifi)) {
update_table_entry(r);
}
}
r->rt_metric = adj_metric;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
}
else if (src == 0 ||
(r->rt_gateway != 0 &&
(adj_metric < r->rt_metric ||
(adj_metric == r->rt_metric &&
(ntohl(src) < ntohl(r->rt_gateway) ||
r->rt_timer >= ROUTE_SWITCH_TIME))))) {
/*
* The report is for an origin we consider reachable; the report
* comes either from one of our own interfaces or from a gateway
* other than the one we have chosen as the best first-hop gateway
* back towards the origin. If the source of the update is one of
* our own interfaces, or if the origin is not a directly-connected
* subnet and the reported metric for that origin is better than
* what our routing entry says, update the entry to use the new
* gateway and metric. We also switch gateways if the reported
* metric is the same as the one in the route entry and the gateway
* associated with the route entry has not been heard from recently,
* or if the metric is the same but the reporting gateway has a lower
* IP address than the gateway associated with the route entry.
* Did you get all that?
*/
if (r->rt_parent != vifi || adj_metric < r->rt_metric) {
/*
* XXX Why do we do this if we are just changing the metric?
*/
r->rt_parent = vifi;
if (init_children_and_leaves(r, vifi)) {
update_table_entry(r);
}
}
r->rt_gateway = src;
r->rt_timer = 0;
r->rt_metric = adj_metric;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
}
else if (vifi != r->rt_parent) {
/*
* The report came from a vif other than the route's parent vif.
* Update the children and leaf info, if necessary.
*/
if (VIFM_ISSET(vifi, r->rt_children)) {
/*
* Vif is a child vif for this route.
*/
if (metric < r->rt_metric ||
(metric == r->rt_metric &&
ntohl(src) < ntohl(uvifs[vifi].uv_lcl_addr))) {
/*
* Neighbor has lower metric to origin (or has same metric
* and lower IP address) -- it becomes the dominant router,
* and vif is no longer a child for me.
*/
VIFM_CLR(vifi, r->rt_children);
VIFM_CLR(vifi, r->rt_leaves);
r->rt_dominants [vifi] = src;
r->rt_subordinates[vifi] = 0;
r->rt_leaf_timers [vifi] = 0;
update_table_entry(r);
}
else if (metric > UNREACHABLE) { /* "poisoned reverse" */
/*
* Neighbor considers this vif to be on path to route's
* origin; if no subordinate recorded, record this neighbor
* as subordinate and clear the leaf flag.
*/
if (r->rt_subordinates[vifi] == 0) {
VIFM_CLR(vifi, r->rt_leaves);
r->rt_subordinates[vifi] = src;
r->rt_leaf_timers [vifi] = 0;
update_table_entry(r);
}
}
else if (src == r->rt_subordinates[vifi]) {
/*
* Current subordinate no longer considers this vif to be on
* path to route's origin; it is no longer a subordinate
* router, and we set the leaf confirmation timer to give
* us time to hear from other subordinates.
*/
r->rt_subordinates[vifi] = 0;
if (uvifs[vifi].uv_neighbors == NULL ||
uvifs[vifi].uv_neighbors->al_next == NULL) {
VIFM_SET(vifi, r->rt_leaves);
update_table_entry(r);
}
else {
r->rt_leaf_timers [vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
}
}
else if (src == r->rt_dominants[vifi] &&
(metric > r->rt_metric ||
(metric == r->rt_metric &&
ntohl(src) > ntohl(uvifs[vifi].uv_lcl_addr)))) {
/*
* Current dominant no longer has a lower metric to origin
* (or same metric and lower IP address); we adopt the vif
* as our own child.
*/
VIFM_SET(vifi, r->rt_children);
r->rt_dominants [vifi] = 0;
if (metric > UNREACHABLE) {
r->rt_subordinates[vifi] = src;
}
else if (uvifs[vifi].uv_neighbors == NULL ||
uvifs[vifi].uv_neighbors->al_next == NULL) {
VIFM_SET(vifi, r->rt_leaves);
}
else {
r->rt_leaf_timers[vifi] = LEAF_CONFIRMATION_TIME;
r->rt_flags |= RTF_LEAF_TIMING;
}
update_table_entry(r);
}
}
}
/*
* On every timer interrupt, advance the timer in each routing entry.
*/
void
age_routes()
{
register struct rtentry *r;
register struct rtentry *prev_r;
register vifi_t vifi;
for (prev_r = RT_ADDR, r = routing_table;
r != NULL;
prev_r = r, r = r->rt_next) {
if ((r->rt_timer += TIMER_INTERVAL) < ROUTE_EXPIRE_TIME) {
/*
* Route is still good; see if any leaf timers need to be
* advanced.
*/
if (r->rt_flags & RTF_LEAF_TIMING) {
r->rt_flags &= ~RTF_LEAF_TIMING;
for (vifi = 0; vifi < numvifs; ++vifi) {
if (r->rt_leaf_timers[vifi] != 0) {
/*
* Unlike other timers, leaf timers decrement.
*/
if ((r->rt_leaf_timers[vifi] -= TIMER_INTERVAL) == 0){
#ifdef NOTYET
/* If the vif is a physical leaf but has neighbors,
* it is not a tree leaf. If I am a leaf, then no
* interface with neighbors is a tree leaf. */
if (!(((uvifs[vifi].uv_flags & VIFF_LEAF) ||
(vifs_with_neighbors == 1)) &&
(uvifs[vifi].uv_neighbors != NULL))) {
#endif
VIFM_SET(vifi, r->rt_leaves);
update_table_entry(r);
#ifdef NOTYET
}
#endif
}
else {
r->rt_flags |= RTF_LEAF_TIMING;
}
}
}
}
}
else if (r->rt_timer >= ROUTE_DISCARD_TIME) {
/*
* Time to garbage-collect the route entry.
*/
del_table_entry(r, 0, DEL_ALL_ROUTES);
discard_route(prev_r);
r = prev_r;
}
else if (r->rt_metric != UNREACHABLE) {
/*
* Time to expire the route entry. If the gateway is zero,
* i.e., it is a route to a directly-connected subnet, just
* set the timer back to zero; such routes expire only when
* the interface to the subnet goes down.
*/
if (r->rt_gateway == 0) {
r->rt_timer = 0;
}
else {
del_table_entry(r, 0, DEL_ALL_ROUTES);
r->rt_metric = UNREACHABLE;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
}
}
}
}
/*
* Mark all routes as unreachable. This function is called only from
* hup() in preparation for informing all neighbors that we are going
* off the air. For consistency, we ought also to delete all reachable
* route entries from the kernel, but since we are about to exit we rely
* on the kernel to do its own cleanup -- no point in making all those
* expensive kernel calls now.
*/
void
expire_all_routes()
{
register struct rtentry *r;
for (r = routing_table; r != NULL; r = r->rt_next) {
r->rt_metric = UNREACHABLE;
r->rt_flags |= RTF_CHANGED;
routes_changed = TRUE;
}
}
/*
* Delete all the routes in the routing table.
*/
void
free_all_routes()
{
register struct rtentry *r;
r = RT_ADDR;
while (r->rt_next)
discard_route(r);
}
/*
* Process an incoming neighbor probe message.
*/
void
accept_probe(src, dst, p, datalen, level)
u_int32 src;
u_int32 dst;
char *p;
int datalen;
u_int32 level;
{
vifi_t vifi;
if ((vifi = find_vif(src, dst)) == NO_VIF) {
log(LOG_INFO, 0,
"ignoring probe from non-neighbor %s", inet_fmt(src, s1));
return;
}
update_neighbor(vifi, src, DVMRP_PROBE, p, datalen, level);
}
struct newrt {
u_int32 mask;
u_int32 origin;
int metric;
int pad;
};
static int
compare_rts(rt1, rt2)
const void *rt1;
const void *rt2;
{
register struct newrt *r1 = (struct newrt *)rt1;
register struct newrt *r2 = (struct newrt *)rt2;
register u_int32 m1 = ntohl(r1->mask);
register u_int32 m2 = ntohl(r2->mask);
register u_int32 o1, o2;
if (m1 > m2)
return (-1);
if (m1 < m2)
return (1);
/* masks are equal */
o1 = ntohl(r1->origin);
o2 = ntohl(r2->origin);
if (o1 > o2)
return (-1);
if (o1 < o2)
return (1);
return (0);
}
/*
* Process an incoming route report message.
*/
void
accept_report(src, dst, p, datalen, level)
u_int32 src, dst, level;
register char *p;
register int datalen;
{
vifi_t vifi;
register int width, i, nrt = 0;
int metric;
u_int32 mask;
u_int32 origin;
struct newrt rt[4096];
if ((vifi = find_vif(src, dst)) == NO_VIF) {
log(LOG_INFO, 0,
"ignoring route report from non-neighbor %s", inet_fmt(src, s1));
return;
}
if (!update_neighbor(vifi, src, DVMRP_REPORT, NULL, 0, level))
return;
if (datalen > 2*4096) {
log(LOG_INFO, 0,
"ignoring oversize (%d bytes) route report from %s",
datalen, inet_fmt(src, s1));
return;
}
while (datalen > 0) { /* Loop through per-mask lists. */
if (datalen < 3) {
log(LOG_WARNING, 0,
"received truncated route report from %s",
inet_fmt(src, s1));
return;
}
((u_char *)&mask)[0] = 0xff; width = 1;
if ((((u_char *)&mask)[1] = *p++) != 0) width = 2;
if ((((u_char *)&mask)[2] = *p++) != 0) width = 3;
if ((((u_char *)&mask)[3] = *p++) != 0) width = 4;
if (!inet_valid_mask(ntohl(mask))) {
log(LOG_WARNING, 0,
"%s reports bogus netmask 0x%08x (%s)",
inet_fmt(src, s1), ntohl(mask), inet_fmt(mask, s2));
return;
}
datalen -= 3;
do { /* Loop through (origin, metric) pairs */
if (datalen < width + 1) {
log(LOG_WARNING, 0,
"received truncated route report from %s",
inet_fmt(src, s1));
return;
}
origin = 0;
for (i = 0; i < width; ++i)
((char *)&origin)[i] = *p++;
metric = *p++;
datalen -= width + 1;
rt[nrt].mask = mask;
rt[nrt].origin = origin;
rt[nrt].metric = (metric & 0x7f);
++nrt;
} while (!(metric & 0x80));
}
qsort((char*)rt, nrt, sizeof(rt[0]), compare_rts);
start_route_updates();
/*
* If the last entry is default, change mask from 0xff000000 to 0
*/
if (rt[nrt-1].origin == 0)
rt[nrt-1].mask = 0;
log(LOG_DEBUG, 0, "Updating %d routes from %s to %s", nrt,
inet_fmt(src, s1), inet_fmt(dst, s2));
for (i = 0; i < nrt; ++i) {
if (i != 0 && rt[i].origin == rt[i-1].origin &&
rt[i].mask == rt[i-1].mask) {
log(LOG_WARNING, 0, "%s reports duplicate route for %s",
inet_fmt(src, s1), inet_fmts(rt[i].origin, rt[i].mask, s2));
continue;
}
update_route(rt[i].origin, rt[i].mask, rt[i].metric,
src, vifi);
}
if (routes_changed && !delay_change_reports)
report_to_all_neighbors(CHANGED_ROUTES);
}
/*
* Send a route report message to destination 'dst', via virtual interface
* 'vifi'. 'which_routes' specifies ALL_ROUTES or CHANGED_ROUTES.
*/
void
report(which_routes, vifi, dst)
int which_routes;
vifi_t vifi;
u_int32 dst;
{
register struct rtentry *r;
register char *p;
register int i;
int datalen = 0;
int width = 0;
u_int32 mask = 0;
u_int32 src;
u_int32 nflags;
src = uvifs[vifi].uv_lcl_addr;
p = send_buf + MIN_IP_HEADER_LEN + IGMP_MINLEN;
#ifdef NOTYET
/* If I'm not a leaf, but the neighbor is a leaf, only advertise default */
if ((vifs_with_neighbors != 1) && (uvifs[vifi].uv_flags & VIFF_LEAF)) {
*p++ = 0; /* 0xff000000 mask */
*p++ = 0;
*p++ = 0;
*p++ = 0; /* class A net 0.0.0.0 == default */
*p++ = 0x81; /*XXX metric 1, is this safe? */
datalen += 5;
send_igmp(src, dst, IGMP_DVMRP, DVMRP_REPORT,
htonl(MROUTED_LEVEL), datalen);
return;
}
#endif
nflags = (uvifs[vifi].uv_flags & VIFF_LEAF) ? 0 : LEAF_FLAGS;
for (r = rt_end; r != RT_ADDR; r = r->rt_prev) {
if (which_routes == CHANGED_ROUTES && !(r->rt_flags & RTF_CHANGED))
continue;
/*
* If there is no room for this route in the current message,
* send the message and start a new one.
*/
if (datalen + ((r->rt_originmask == mask) ?
(width + 1) :
(r->rt_originwidth + 4)) > MAX_DVMRP_DATA_LEN) {
*(p-1) |= 0x80;
send_igmp(src, dst, IGMP_DVMRP, DVMRP_REPORT,
htonl(MROUTED_LEVEL | nflags), datalen);
p = send_buf + MIN_IP_HEADER_LEN + IGMP_MINLEN;
datalen = 0;
mask = 0;
}
if (r->rt_originmask != mask || datalen == 0) {
mask = r->rt_originmask;
width = r->rt_originwidth;
if (datalen != 0) *(p-1) |= 0x80;
*p++ = ((char *)&mask)[1];
*p++ = ((char *)&mask)[2];
*p++ = ((char *)&mask)[3];
datalen += 3;
}
for (i = 0; i < width; ++i)
*p++ = ((char *)&(r->rt_origin))[i];
*p++ = (r->rt_parent == vifi && r->rt_metric != UNREACHABLE) ?
(char)(r->rt_metric + UNREACHABLE) : /* "poisoned reverse" */
(char)(r->rt_metric);
datalen += width + 1;
}
if (datalen != 0) {
*(p-1) |= 0x80;
send_igmp(src, dst, IGMP_DVMRP, DVMRP_REPORT,
htonl(MROUTED_LEVEL | nflags), datalen);
}
}
/*
* Send a route report message to all neighboring routers.
* 'which_routes' specifies ALL_ROUTES or CHANGED_ROUTES.
*/
void
report_to_all_neighbors(which_routes)
int which_routes;
{
register vifi_t vifi;
register struct uvif *v;
register struct rtentry *r;
int routes_changed_before;
/*
* Remember the state of the global routes_changed flag before
* generating the reports, and clear the flag.
*/
routes_changed_before = routes_changed;
routes_changed = FALSE;
for (vifi = 0, v = uvifs; vifi < numvifs; ++vifi, ++v) {
if (v->uv_neighbors != NULL) {
report(which_routes, vifi,
(v->uv_flags & VIFF_TUNNEL) ? v->uv_rmt_addr
: dvmrp_group);
}
}
/*
* If there were changed routes before we sent the reports AND
* if no new changes occurred while sending the reports, clear
* the change flags in the individual route entries. If changes
* did occur while sending the reports, new reports will be
* generated at the next timer interrupt.
*/
if (routes_changed_before && !routes_changed) {
for (r = routing_table; r != NULL; r = r->rt_next) {
r->rt_flags &= ~RTF_CHANGED;
}
}
/*
* Set a flag to inhibit further reports of changed routes until the
* next timer interrupt. This is to alleviate update storms.
*/
delay_change_reports = TRUE;
}
/*
* Send a route report message to destination 'dst', via virtual interface
* 'vifi'. 'which_routes' specifies ALL_ROUTES or CHANGED_ROUTES.
*/
static int
report_chunk(start_rt, vifi, dst)
register struct rtentry *start_rt;
vifi_t vifi;
u_int32 dst;
{
register struct rtentry *r;
register char *p;
register int i;
register int nrt = 0;
int datalen = 0;
int width = 0;
u_int32 mask = 0;
u_int32 src;
u_int32 nflags;
src = uvifs[vifi].uv_lcl_addr;
p = send_buf + MIN_IP_HEADER_LEN + IGMP_MINLEN;
nflags = (uvifs[vifi].uv_flags & VIFF_LEAF) ? 0 : LEAF_FLAGS;
for (r = start_rt; r != RT_ADDR; r = r->rt_prev) {
#ifdef NOTYET
/* Don't send poisoned routes back to parents if I am a leaf */
if ((vifs_with_neighbors == 1) && (r->rt_parent == vifi)
&& (r->rt_metric > 1)) {
++nrt;
continue;
}
#endif
/*
* If there is no room for this route in the current message,
* send it & return how many routes we sent.
*/
if (datalen + ((r->rt_originmask == mask) ?
(width + 1) :
(r->rt_originwidth + 4)) > MAX_DVMRP_DATA_LEN) {
*(p-1) |= 0x80;
send_igmp(src, dst, IGMP_DVMRP, DVMRP_REPORT,
htonl(MROUTED_LEVEL | nflags), datalen);
return (nrt);
}
if (r->rt_originmask != mask || datalen == 0) {
mask = r->rt_originmask;
width = r->rt_originwidth;
if (datalen != 0) *(p-1) |= 0x80;
*p++ = ((char *)&mask)[1];
*p++ = ((char *)&mask)[2];
*p++ = ((char *)&mask)[3];
datalen += 3;
}
for (i = 0; i < width; ++i)
*p++ = ((char *)&(r->rt_origin))[i];
*p++ = (r->rt_parent == vifi && r->rt_metric != UNREACHABLE) ?
(char)(r->rt_metric + UNREACHABLE) : /* "poisoned reverse" */
(char)(r->rt_metric);
++nrt;
datalen += width + 1;
}
if (datalen != 0) {
*(p-1) |= 0x80;
send_igmp(src, dst, IGMP_DVMRP, DVMRP_REPORT,
htonl(MROUTED_LEVEL | nflags), datalen);
}
return (nrt);
}
/*
* send the next chunk of our routing table to all neighbors.
* return the length of the smallest chunk we sent out.
*/
int
report_next_chunk()
{
register vifi_t vifi;
register struct uvif *v;
register struct rtentry *sr;
register int i, n = 0, min = 20000;
static int start_rt;
if (nroutes <= 0)
return (0);
/*
* find this round's starting route.
*/
for (sr = rt_end, i = start_rt; --i >= 0; ) {
sr = sr->rt_prev;
if (sr == RT_ADDR)
sr = rt_end;
}
/*
* send one chunk of routes starting at this round's start to
* all our neighbors.
*/
for (vifi = 0, v = uvifs; vifi < numvifs; ++vifi, ++v) {
if ((v->uv_neighbors != NULL)
#ifdef NOTYET
&& !(v->uv_flags & VIFF_LEAF)
#endif
) {
n = report_chunk(sr, vifi,
(v->uv_flags & VIFF_TUNNEL) ? v->uv_rmt_addr
: dvmrp_group);
if (n < min)
min = n;
}
}
if (min == 20000)
min = 0; /* Neighborless router didn't send any routes */
n = min;
log(LOG_INFO, 0, "update %d starting at %d of %d",
n, (nroutes - start_rt), nroutes);
start_rt = (start_rt + n) % nroutes;
return (n);
}
/*
* Print the contents of the routing table on file 'fp'.
*/
void
dump_routes(fp)
FILE *fp;
{
register struct rtentry *r;
register vifi_t i;
fprintf(fp,
"Multicast Routing Table (%u %s)\n%s\n",
nroutes, (nroutes == 1) ? "entry" : "entries",
" Origin-Subnet From-Gateway Metric Tmr In-Vif Out-Vifs");
for (r = routing_table; r != NULL; r = r->rt_next) {
fprintf(fp, " %-18s %-15s ",
inet_fmts(r->rt_origin, r->rt_originmask, s1),
(r->rt_gateway == 0) ? "" : inet_fmt(r->rt_gateway, s2));
fprintf(fp, (r->rt_metric == UNREACHABLE) ? " NR " : "%4u ",
r->rt_metric);
fprintf(fp, " %3u %3u ", r->rt_timer, r->rt_parent);
for (i = 0; i < numvifs; ++i) {
if (VIFM_ISSET(i, r->rt_children)) {
fprintf(fp, " %u%c",
i, VIFM_ISSET(i, r->rt_leaves) ? '*' : ' ');
}
}
fprintf(fp, "\n");
}
fprintf(fp, "\n");
}
struct rtentry *
determine_route(src)
u_int32 src;
{
struct rtentry *rt;
for (rt = routing_table; rt != NULL; rt = rt->rt_next) {
if (rt->rt_origin == (src & rt->rt_originmask))
break;
}
return rt;
}