Commit c10421d3 authored by Martin Mareš's avatar Martin Mareš
Browse files

More changes to the kernel syncer.

o  Now compatible with filtering.
o  Learning of kernel routes supported only on CONFIG_SELF_CONSCIOUS
   systems (on the others it's impossible to get it semantically correct).
o  Learning now stores all of its routes in a separate fib and selects
   the ones the kernel really uses for forwarding packets.
o  Better treatment of CONFIG_AUTO_ROUTES ports.
o  Lots of internal changes.
parent 69ec9087
......@@ -140,18 +140,20 @@ typedef struct rte {
struct {
} bgp;
#endif
struct { /* Routes internally generated by krt sync */
int src; /* Alleged route source (see krt.h) */
} krt_sync;
struct { /* Routes generated by krt sync (both temporary and inherited ones) */
s8 src; /* Alleged route source (see krt.h) */
u8 proto; /* Kernel source protocol ID */
u8 type; /* Kernel route type */
u8 seen; /* Seen during last scan */
u32 metric; /* Kernel metric */
} krt;
} u;
} rte;
#define REF_CHOSEN 1 /* Currently chosen route */
extern rtable master_table;
void rt_init(void);
void rt_setup(rtable *, char *);
void rt_setup(pool *, rtable *, char *);
net *net_find(rtable *tab, unsigned tos, ip_addr addr, unsigned len);
net *net_get(rtable *tab, unsigned tos, ip_addr addr, unsigned len);
rte *rte_find(net *net, struct proto *p);
......
......@@ -38,10 +38,10 @@ rte_init(struct fib_node *N)
}
void
rt_setup(rtable *t, char *name)
rt_setup(pool *p, rtable *t, char *name)
{
bzero(t, sizeof(*t));
fib_init(&t->fib, &root_pool, sizeof(rte), 0, rte_init);
fib_init(&t->fib, p, sizeof(rte), 0, rte_init);
t->name = name;
}
......@@ -67,7 +67,7 @@ net_get(rtable *tab, unsigned tos, ip_addr mask, unsigned len)
while (tab->sibling)
tab = tab->sibling;
t = mb_alloc(&root_pool, sizeof(rtable));
rt_setup(t, NULL);
rt_setup(&root_pool, t, NULL); /* FIXME: Either delete all the TOS logic or use the right pool */
tab->sibling = t;
t->tos = tos;
}
......@@ -305,10 +305,10 @@ rte_dump(rte *e)
debug("%1I/%2d ", n->n.prefix, n->n.pxlen);
else
debug("??? ");
debug("PF=%02x pref=%d lm=%d ", e->pflags, e->pref, now-e->lastmod);
debug("KF=%02x PF=%02x pref=%d lm=%d ", n->n.flags, e->pflags, e->pref, now-e->lastmod);
rta_dump(e->attrs);
if (e->flags & REF_CHOSEN)
debug(" [*]");
if (e->attrs->proto->proto->dump_attrs)
e->attrs->proto->proto->dump_attrs(e);
debug("\n");
}
......@@ -357,7 +357,7 @@ rt_init(void)
{
rta_init();
rt_table_pool = rp_new(&root_pool, "Routing tables");
rt_setup(&master_table, "master");
rt_setup(rt_table_pool, &master_table, "master");
rte_slab = sl_new(rt_table_pool, sizeof(rte));
rt_last_gc = now;
rt_gc_event = ev_new(rt_table_pool);
......
......@@ -127,7 +127,7 @@ krt_parse_entry(byte *ent, struct krt_proto *p)
e = rte_get_temp(&a);
e->net = net;
e->u.krt_sync.src = KRT_SRC_UNKNOWN;
e->u.krt.src = KRT_SRC_UNKNOWN;
krt_got_route(p, e);
}
......
......@@ -276,7 +276,7 @@ nl_parse_link(struct nlmsghdr *h, int scan)
ifi = if_find_by_index(i->ifi_index);
if (!new)
{
DBG("KRT: IF%d(%s) goes down\n", i->ifi_index, name);
DBG("KIF: IF%d(%s) goes down\n", i->ifi_index, name);
if (ifi && !scan)
{
memcpy(&f, ifi, sizeof(struct iface));
......@@ -286,7 +286,7 @@ nl_parse_link(struct nlmsghdr *h, int scan)
}
else
{
DBG("KRT: IF%d(%s) goes up (mtu=%d,flg=%x)\n", i->ifi_index, name, mtu, i->ifi_flags);
DBG("KIF: IF%d(%s) goes up (mtu=%d,flg=%x)\n", i->ifi_index, name, mtu, i->ifi_flags);
if (ifi)
memcpy(&f, ifi, sizeof(f));
else
......@@ -332,14 +332,14 @@ nl_parse_addr(struct nlmsghdr *h)
}
if (i->ifa_flags & IFA_F_SECONDARY)
{
DBG("KRT: Received address message for secondary address which is not supported.\n"); /* FIXME */
DBG("KIF: Received address message for secondary address which is not supported.\n"); /* FIXME */
return;
}
ifi = if_find_by_index(i->ifa_index);
if (!ifi)
{
log(L_ERR "KRT: Received address message for unknown interface %d\n", i->ifa_index);
log(L_ERR "KIF: Received address message for unknown interface %d\n", i->ifa_index);
return;
}
memcpy(&f, ifi, sizeof(f));
......@@ -347,14 +347,14 @@ nl_parse_addr(struct nlmsghdr *h)
if (i->ifa_prefixlen > 32 || i->ifa_prefixlen == 31 ||
(f.flags & IF_UNNUMBERED) && i->ifa_prefixlen != 32)
{
log(L_ERR "KRT: Invalid prefix length for interface %s: %d\n", f.name, i->ifa_prefixlen);
log(L_ERR "KIF: Invalid prefix length for interface %s: %d\n", f.name, i->ifa_prefixlen);
new = 0;
}
f.ip = f.brd = f.opposite = IPA_NONE;
if (!new)
{
DBG("KRT: IF%d IP address deleted\n");
DBG("KIF: IF%d IP address deleted\n");
f.pxlen = 0;
}
else
......@@ -374,7 +374,7 @@ nl_parse_addr(struct nlmsghdr *h)
}
/* else a NBMA link */
f.prefix = ipa_and(f.ip, ipa_mkmask(f.pxlen));
DBG("KRT: IF%d IP address set to %I, net %I/%d, brd %I, opp %I\n", f.index, f.ip, f.prefix, f.pxlen, f.brd, f.opposite);
DBG("KIF: IF%d IP address set to %I, net %I/%d, brd %I, opp %I\n", f.index, f.ip, f.prefix, f.pxlen, f.brd, f.opposite);
}
if_update(&f);
}
......@@ -485,12 +485,8 @@ nl_send_route(rte *e, int new)
}
void
krt_set_notify(struct proto *p, net *n, rte *new, rte *old)
krt_set_notify(struct krt_proto *p, net *n, rte *new, rte *old)
{
if (old && !krt_capable(old))
old = NULL;
if (new && !krt_capable(new))
new = NULL;
if (old && new && old->attrs->tos == new->attrs->tos)
{
/* FIXME: Priorities should be identical as well, but we don't use them yet. */
......@@ -546,6 +542,7 @@ nl_parse_route(struct krt_proto *p, struct nlmsghdr *h, int scan)
return;
if ((a[RTA_DST] && RTA_PAYLOAD(a[RTA_DST]) != sizeof(ip_addr)) ||
(a[RTA_OIF] && RTA_PAYLOAD(a[RTA_OIF]) != 4) ||
(a[RTA_PRIORITY] && RTA_PAYLOAD(a[RTA_PRIORITY]) != 4) ||
(a[RTA_GATEWAY] && RTA_PAYLOAD(a[RTA_GATEWAY]) != sizeof(ip_addr)))
{
log(L_ERR "nl_parse_route: Malformed message received");
......@@ -649,9 +646,22 @@ nl_parse_route(struct krt_proto *p, struct nlmsghdr *h, int scan)
DBG("KRT: Ignoring route with type=%d\n", i->rtm_type);
return;
}
if (i->rtm_scope != RT_SCOPE_UNIVERSE) /* FIXME: Other scopes? */
{
DBG("KRT: Ignoring route with scope=%d\n", i->rtm_scope);
return;
}
e = rte_get_temp(&ra);
e->net = net;
e->u.krt_sync.src = src;
e->u.krt.src = src;
e->u.krt.proto = i->rtm_protocol;
e->u.krt.type = i->rtm_type;
if (a[RTA_PRIORITY])
memcpy(&e->u.krt.metric, RTA_DATA(a[RTA_PRIORITY]), sizeof(e->u.krt.metric));
else
e->u.krt.metric = 0;
if (scan)
krt_got_route(p, e);
else
......@@ -797,7 +807,7 @@ krt_scan_start(struct krt_proto *p)
{
init_list(&p->scan.temp_ifs);
nl_open();
if (KRT_CF->scan.async)
if (KRT_CF->scan.async) /* FIXME: Async is for debugging only. Get rid of it some day. */
nl_open_async(p);
}
......@@ -810,4 +820,5 @@ void
krt_if_start(struct kif_proto *p)
{
nl_open();
/* FIXME: nl_open_async() after scan.async is gone */
}
......@@ -77,41 +77,19 @@ krt_ioctl(int ioc, rte *e, char *name)
log(L_ERR "%s(%I/%d): %m", name, net->n.prefix, net->n.pxlen);
}
static inline void
krt_remove_route(rte *old)
void
krt_set_notify(struct krt_proto *p, net *net, rte *new, rte *old)
{
net *net = old->net;
if (!krt_capable(old))
if (old)
{
DBG("krt_remove_route(ignored %I/%d)\n", net->n.prefix, net->n.pxlen);
return;
DBG("krt_remove_route(%I/%d)\n", net->n.prefix, net->n.pxlen);
krt_ioctl(SIOCDELRT, old, "SIOCDELRT");
}
DBG("krt_remove_route(%I/%d)\n", net->n.prefix, net->n.pxlen);
krt_ioctl(SIOCDELRT, old, "SIOCDELRT");
}
static inline void
krt_add_route(rte *new)
{
net *net = new->net;
if (!krt_capable(new))
if (new)
{
DBG("krt_add_route(ignored %I/%d)\n", net->n.prefix, net->n.pxlen);
return;
DBG("krt_add_route(%I/%d)\n", net->n.prefix, net->n.pxlen);
krt_ioctl(SIOCADDRT, new, "SIOCADDRT");
}
DBG("krt_add_route(%I/%d)\n", net->n.prefix, net->n.pxlen);
krt_ioctl(SIOCADDRT, new, "SIOCADDRT");
}
void
krt_set_notify(struct proto *x, net *net, rte *new, rte *old)
{
if (old)
krt_remove_route(old);
if (new)
krt_add_route(new);
}
void
......
......@@ -45,7 +45,13 @@ kern_item:
/* Scan time of 0 means scan on startup only */
THIS_KRT->scan_time = $3;
}
| LEARN bool { THIS_KRT->learn = $2; }
| LEARN bool {
THIS_KRT->learn = $2;
#ifndef KRT_ALLOW_LEARN
if ($2)
cf_error("Learning of kernel routes not supported in this configuration");
#endif
}
;
/* Kernel interface protocol */
......
......@@ -17,6 +17,8 @@
#include "unix.h"
#include "krt.h"
static int krt_uptodate(rte *k, rte *e);
/*
* Global resources
*/
......@@ -106,6 +108,247 @@ struct protocol proto_unix_iface = {
shutdown: kif_shutdown,
};
/*
* Inherited Routes
*/
#ifdef KRT_ALLOW_LEARN
static inline int
krt_same_key(rte *a, rte *b)
{
return a->u.krt.proto == b->u.krt.proto &&
a->u.krt.metric == b->u.krt.metric &&
a->u.krt.type == b->u.krt.type;
}
static void
krt_learn_announce_update(struct krt_proto *p, rte *e)
{
net *n = e->net;
rta *aa = rta_clone(e->attrs);
rte *ee = rte_get_temp(aa);
net *nn = net_get(p->p.table, 0, n->n.prefix, n->n.pxlen); /* FIXME: TOS */
ee->net = nn;
ee->pflags = 0;
ee->u.krt = e->u.krt;
rte_update(nn, &p->p, ee);
}
static void
krt_learn_announce_delete(struct krt_proto *p, net *n)
{
n = net_find(p->p.table, 0, n->n.prefix, n->n.pxlen); /* FIXME: TOS */
if (n)
rte_update(n, &p->p, NULL);
}
static void
krt_learn_scan(struct krt_proto *p, rte *e)
{
net *n0 = e->net;
net *n = net_get(&p->krt_table, 0, n0->n.prefix, n0->n.pxlen); /* FIXME: TOS */
rte *m, **mm;
e->attrs->source = RTS_INHERIT;
for(mm=&n->routes; m = *mm; mm=&m->next)
if (krt_same_key(m, e))
break;
if (m)
{
if (krt_uptodate(m, e))
{
DBG("krt_learn_scan: SEEN\n");
rte_free(e);
m->u.krt.seen = 1;
}
else
{
DBG("krt_learn_scan: OVERRIDE\n");
*mm = m->next;
rte_free(m);
m = NULL;
}
}
else
DBG("krt_learn_scan: CREATE\n");
if (!m)
{
e->attrs = rta_lookup(e->attrs);
e->next = n->routes;
n->routes = e;
e->u.krt.seen = 1;
}
}
/* FIXME: Add dump function */
static void
krt_learn_prune(struct krt_proto *p)
{
struct fib *fib = &p->krt_table.fib;
struct fib_iterator fit;
DBG("Pruning inheritance data...\n");
FIB_ITERATE_INIT(&fit, fib);
again:
FIB_ITERATE_START(fib, &fit, f)
{
net *n = (net *) f;
rte *e, **ee, *best, **pbest, *old_best;
old_best = n->routes;
best = NULL;
pbest = NULL;
ee = &n->routes;
while (e = *ee)
{
if (!e->u.krt.seen)
{
*ee = e->next;
rte_free(e);
continue;
}
if (!best || best->u.krt.metric > e->u.krt.metric)
{
best = e;
pbest = ee;
}
e->u.krt.seen = 0;
ee = &e->next;
}
if (!n->routes)
{
DBG("%I/%d: deleting\n", n->n.prefix, n->n.pxlen);
if (old_best)
{
krt_learn_announce_delete(p, n);
n->n.flags &= ~KRF_INSTALLED;
}
FIB_ITERATE_PUT(&fit, f);
fib_delete(fib, f);
goto again;
}
*pbest = best->next;
best->next = n->routes;
n->routes = best;
if (best != old_best || !(n->n.flags & KRF_INSTALLED))
{
DBG("%I/%d: announcing (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric);
krt_learn_announce_update(p, best);
n->n.flags |= KRF_INSTALLED;
}
else
DBG("%I/%d: uptodate (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric);
}
FIB_ITERATE_END(f);
}
static void
krt_learn_async(struct krt_proto *p, rte *e, int new)
{
net *n0 = e->net;
net *n = net_get(&p->krt_table, 0, n0->n.prefix, n0->n.pxlen); /* FIXME: TOS */
rte *g, **gg, *best, **bestp, *old_best;
e->attrs->source = RTS_INHERIT;
old_best = n->routes;
for(gg=&n->routes; g = *gg; gg = &g->next)
if (krt_same_key(g, e))
break;
if (new)
{
if (g)
{
if (krt_uptodate(g, e))
{
DBG("krt_learn_async: same\n");
rte_free(e);
return;
}
DBG("krt_learn_async: update\n");
*gg = g->next;
rte_free(g);
}
else
DBG("krt_learn_async: create\n");
e->attrs = rta_lookup(e->attrs);
e->next = n->routes;
n->routes = e;
}
else if (!g)
{
DBG("krt_learn_async: not found\n");
rte_free(e);
return;
}
else
{
DBG("krt_learn_async: delete\n");
*gg = g->next;
rte_free(e);
rte_free(g);
}
best = n->routes;
bestp = &n->routes;
for(gg=&n->routes; g=*gg; gg=&g->next)
if (best->u.krt.metric > g->u.krt.metric)
{
best = g;
bestp = gg;
}
if (best)
{
*bestp = best->next;
best->next = n->routes;
n->routes = best;
}
if (best != old_best)
{
DBG("krt_learn_async: distributing change\n");
if (best)
{
krt_learn_announce_update(p, best);
n->n.flags |= KRF_INSTALLED;
}
else
{
n->routes = NULL;
krt_learn_announce_delete(p, n);
n->n.flags &= ~KRF_INSTALLED;
}
}
}
static void
krt_learn_init(struct krt_proto *p)
{
if (KRT_CF->learn)
rt_setup(p->p.pool, &p->krt_table, "Inherited");
}
static void
krt_dump(struct proto *P)
{
struct krt_proto *p = (struct krt_proto *) P;
if (!KRT_CF->learn)
return;
debug("KRT: Table of inheritable routes\n");
rt_dump(&p->krt_table);
}
static void
krt_dump_attrs(rte *e)
{
debug(" [m=%d,p=%d,t=%d]", e->u.krt.metric, e->u.krt.proto, e->u.krt.type);
}
#endif
/*
* Routes
*/
......@@ -128,13 +371,12 @@ krt_flush_routes(struct krt_proto *p)
{
rta *a = e->attrs;
if (a->source != RTS_DEVICE && a->source != RTS_INHERIT)
krt_set_notify(&p->p, e->net, NULL, e);
krt_set_notify(p, e->net, NULL, e);
}
}
FIB_WALK_END;
}
/* FIXME: Inbound/outbound route filtering? */
/* FIXME: Synchronization of multiple routing tables? */
static int
......@@ -165,41 +407,53 @@ krt_got_route(struct krt_proto *p, rte *e)
{
rte *old;
net *net = e->net;
int src = e->u.krt_sync.src;
int src = e->u.krt.src;
int verdict;
if (net->n.flags)
#ifdef CONFIG_AUTO_ROUTES
if (e->attrs->dest == RTD_DEVICE)
{
/* It's a device route. Probably a kernel-generated one. */
verdict = KRF_IGNORE;
goto sentenced;
}
#endif
#ifdef KRT_ALLOW_LEARN
if (src == KRT_SRC_ALIEN)
{
if (KRT_CF->learn)
krt_learn_scan(p, e);
else
DBG("krt_parse_entry: Alien route, ignoring\n");
return;
}
#endif
if (net->n.flags & KRF_VERDICT_MASK)
{
/* Route to this destination was already seen. Strange, but it happens... */
DBG("Already seen.\n");
return;
}
if (old = net->routes)
if (net->n.flags & KRF_INSTALLED)
{
if (!krt_capable(old))
{
#ifdef CONFIG_AUTO_ROUTES
if (old->attrs->source == RTS_DEVICE)
verdict = KRF_SEEN;
else
#endif
verdict = krt_capable(e) ? KRF_DELETE : KRF_SEEN;
}
else if (krt_uptodate(e, net->routes))
old = net->routes;
ASSERT(old);
if (krt_uptodate(e, old))
verdict = KRF_SEEN;
else
verdict = KRF_UPDATE;
}
else if (KRT_CF->learn && !net->routes && (src == KRT_SRC_ALIEN || src < 0))
verdict = KRF_LEARN;
else
verdict = KRF_DELETE;
DBG("krt_parse_entry: verdict=%s\n", ((char *[]) { "CREATE", "SEEN", "UPDATE", "DELETE", "LEARN" }) [verdict]);
sentenced:
DBG("krt_parse_entry: verdict=%s\n", ((char *[]) { "CREATE", "SEEN", "UPDATE", "DELETE", "IGNORE" }) [verdict]);
net->n.flags = verdict;
if (verdict != KRF_SEEN)
net->n.flags = (net->n.flags & ~KRF_VERDICT_MASK) | verdict;
if (verdict == KRF_UPDATE || verdict == KRF_DELETE)
{
/* Get a cached copy of attributes and link the route */
rta *a = e->attrs;
......@@ -227,10 +481,10 @@ krt_prune(struct krt_proto *p)
FIB_WALK(&t->fib, f)
{
net *n = (net *) f;
int verdict = f->flags;
int verdict = f->flags & KRF_VERDICT_MASK;
rte *new, *old;