* Can be freely distributed and used under the terms of the GNU GPL.
*/
-#define LOCAL_DEBUG
+/**
+ * DOC: Kernel synchronization
+ *
+ * This system dependent module implements the Kernel and Device protocol,
+ * that is synchronization of interface lists and routing tables with the
+ * OS kernel.
+ *
+ * The whole kernel synchronization is a bit messy and touches some internals
+ * of the routing table engine, because routing table maintenance is a typical
+ * example of the proverbial compatibility between different Unices and we want
+ * to keep the overhead of our KRT business as low as possible and avoid maintaining
+ * a local routing table copy.
+ *
+ * The kernel syncer can work in three different modes (according to system config header):
+ * Either with a single routing table and single KRT protocol [traditional UNIX]
+ * or with many routing tables and separate KRT protocols for all of them
+ * or with many routing tables, but every scan including all tables, so we start
+ * separate KRT protocols which cooperate with each other [Linux].
+ * In this case, we keep only a single scan timer.
+ *
+ * We use FIB node flags in the routing table to keep track of route
+ * synchronization status. We also attach temporary &rte's to the routing table,
+ * but it cannot do any harm to the rest of BIRD since table synchronization is
+ * an atomic process.
+ *
+ * When starting up, we cheat by looking if there is another
+ * KRT instance to be initialized later and performing table scan
+ * only once for all the instances.
+ *
+ * The code uses OS-dependent parts for kernel updates and scans. These parts are
+ * in more specific sysdep directories (e.g. sysdep/linux) in functions krt_sys_*
+ * and kif_sys_* (and some others like krt_replace_rte()) and krt-sys.h header file.
+ * This is also used for platform specific protocol options and route attributes.
+ *
+ * There was also an old code that used traditional UNIX ioctls for these tasks.
+ * It was unmaintained and later removed. For reference, see sysdep/krt-* files
+ * in commit 396dfa9042305f62da1f56589c4b98fac57fc2f6
+ */
+
+/*
+ * If you are brave enough, continue now. You cannot say you haven't been warned.
+ */
+
+#undef LOCAL_DEBUG
#include "nest/bird.h"
#include "nest/iface.h"
#include "nest/route.h"
#include "nest/protocol.h"
+#include "filter/filter.h"
#include "lib/timer.h"
#include "conf/conf.h"
+#include "lib/string.h"
#include "unix.h"
#include "krt.h"
-/*
- * The whole kernel synchronization is a bit messy and touches some internals
- * of the routing table engine, because routing table maintenance is a typical
- * example of the proverbial compatibility between different Unices and we want
- * to keep the overhead of our krt business as low as possible and avoid maintaining
- * a local routing table copy.
- *
- * The kernel syncer can work in three different modes (according to system config header):
- * o Single routing table, single krt protocol. [traditional Unix]
- * o Many routing tables, separate krt protocols for all of them.
- * o Many routing tables, but every scan includes all tables, so we start
- * separate krt protocols which cooperate with each other. [Linux 2.2]
- * In this case, we keep only a single scan timer.
- *
- * The hacky bits:
- * o We use FIB node flags to keep track of route synchronization status.
- * o When starting up, we cheat by looking if there is another kernel
- * krt instance to be initialized later and performing table scan
- * only once for all the instances.
- * o We attach temporary rte's to routing tables.
- *
- * If you are brave enough, continue now. You cannot say you haven't been warned.
- */
-
-static int krt_uptodate(rte *k, rte *e);
-
/*
* Global resources
*/
pool *krt_pool;
+static linpool *krt_filter_lp;
+static list krt_proto_list;
void
krt_io_init(void)
{
krt_pool = rp_new(&root_pool, "Kernel Syncer");
- krt_if_io_init();
+ krt_filter_lp = lp_new(krt_pool, 4080);
+ init_list(&krt_proto_list);
+ krt_sys_io_init();
}
/*
* Interfaces
*/
-struct proto_config *cf_kif;
-
-static struct kif_proto *kif_proto;
+struct kif_proto *kif_proto;
+static struct kif_config *kif_cf;
static timer *kif_scan_timer;
static bird_clock_t kif_last_shot;
-static void
-kif_preconfig(struct protocol *P, struct config *c)
-{
- cf_kif = NULL;
-}
-
static void
kif_scan(timer *t)
{
struct kif_proto *p = t->data;
- DBG("KIF: It's interface scan time...\n");
+ KRT_TRACE(p, D_EVENTS, "Scanning interfaces");
kif_last_shot = now;
- krt_if_scan(p);
+ kif_do_scan(p);
}
static void
}
}
+void
+kif_request_scan(void)
+{
+ if (kif_proto && kif_scan_timer->expires > now)
+ tm_start(kif_scan_timer, 1);
+}
+
+static inline int
+prefer_addr(struct ifa *a, struct ifa *b)
+{
+ int sa = a->scope > SCOPE_LINK;
+ int sb = b->scope > SCOPE_LINK;
+
+ if (sa < sb)
+ return 0;
+ else if (sa > sb)
+ return 1;
+ else
+ return ipa_compare(a->ip, b->ip) < 0;
+}
+
+static inline struct ifa *
+find_preferred_ifa(struct iface *i, ip_addr prefix, ip_addr mask)
+{
+ struct ifa *a, *b = NULL;
+
+ WALK_LIST(a, i->addrs)
+ {
+ if (!(a->flags & IA_SECONDARY) &&
+ ipa_equal(ipa_and(a->ip, mask), prefix) &&
+ (!b || prefer_addr(a, b)))
+ b = a;
+ }
+
+ return b;
+}
+
+struct ifa *
+kif_choose_primary(struct iface *i)
+{
+ struct kif_config *cf = (struct kif_config *) (kif_proto->p.cf);
+ struct kif_primary_item *it;
+ struct ifa *a;
+
+ WALK_LIST(it, cf->primary)
+ {
+ if (!it->pattern || patmatch(it->pattern, i->name))
+ if (a = find_preferred_ifa(i, it->prefix, ipa_mkmask(it->pxlen)))
+ return a;
+ }
+
+ if (a = kif_get_primary_ip(i))
+ return a;
+
+ return find_preferred_ifa(i, IPA_NONE, IPA_NONE);
+}
+
+
static struct proto *
kif_init(struct proto_config *c)
{
struct kif_proto *p = proto_new(c, sizeof(struct kif_proto));
+
+ kif_sys_init(p);
return &p->p;
}
struct kif_proto *p = (struct kif_proto *) P;
kif_proto = p;
- krt_if_start(p);
+ kif_sys_start(p);
/* Start periodic interface scanning */
kif_scan_timer = tm_new(P->pool);
struct kif_proto *p = (struct kif_proto *) P;
tm_stop(kif_scan_timer);
- krt_if_shutdown(p);
+ kif_sys_shutdown(p);
kif_proto = NULL;
- if_start_update(); /* Remove all interfaces */
- if_end_update();
- /*
- * FIXME: Is it really a good idea? It causes routes to be flushed,
- * but at the same time it avoids sending of these deletions to the kernel,
- * because krt thinks the kernel itself has already removed the route
- * when downing the interface. Sad.
- */
-
return PS_DOWN;
}
+static int
+kif_reconfigure(struct proto *p, struct proto_config *new)
+{
+ struct kif_config *o = (struct kif_config *) p->cf;
+ struct kif_config *n = (struct kif_config *) new;
+
+ if (!kif_sys_reconfigure((struct kif_proto *) p, n, o))
+ return 0;
+
+ if (o->scan_time != n->scan_time)
+ {
+ tm_stop(kif_scan_timer);
+ kif_scan_timer->recurrent = n->scan_time;
+ kif_scan(kif_scan_timer);
+ tm_start(kif_scan_timer, n->scan_time);
+ }
+
+ if (!EMPTY_LIST(o->primary) || !EMPTY_LIST(n->primary))
+ {
+ /* This is hack, we have to update a configuration
+ * to the new value just now, because it is used
+ * for recalculation of primary addresses.
+ */
+ p->cf = new;
+
+ ifa_recalc_all_primary_addresses();
+ }
+
+ return 1;
+}
+
+
+static void
+kif_preconfig(struct protocol *P UNUSED, struct config *c)
+{
+ kif_cf = NULL;
+ kif_sys_preconfig(c);
+}
+
+struct proto_config *
+kif_init_config(int class)
+{
+ if (kif_cf)
+ cf_error("Kernel device protocol already defined");
+
+ kif_cf = (struct kif_config *) proto_config_new(&proto_unix_iface, class);
+ kif_cf->scan_time = 60;
+ init_list(&kif_cf->primary);
+
+ kif_sys_init_config(kif_cf);
+ return (struct proto_config *) kif_cf;
+}
+
+static void
+kif_copy_config(struct proto_config *dest, struct proto_config *src)
+{
+ struct kif_config *d = (struct kif_config *) dest;
+ struct kif_config *s = (struct kif_config *) src;
+
+ /* Shallow copy of everything (just scan_time currently) */
+ proto_copy_rest(dest, src, sizeof(struct kif_config));
+
+ /* Copy primary addr list */
+ cfg_copy_list(&d->primary, &s->primary, sizeof(struct kif_primary_item));
+
+ /* Fix sysdep parts */
+ kif_sys_copy_config(d, s);
+}
+
+
struct protocol proto_unix_iface = {
- name: "Device",
- priority: 100,
- preconfig: kif_preconfig,
- init: kif_init,
- start: kif_start,
- shutdown: kif_shutdown,
+ .name = "Device",
+ .template = "device%d",
+ .preference = DEF_PREF_DIRECT,
+ .config_size = sizeof(struct kif_config),
+ .preconfig = kif_preconfig,
+ .init = kif_init,
+ .start = kif_start,
+ .shutdown = kif_shutdown,
+ .reconfigure = kif_reconfigure,
+ .copy_config = kif_copy_config
};
+/*
+ * Tracing of routes
+ */
+
+static inline void
+krt_trace_in(struct krt_proto *p, rte *e, char *msg)
+{
+ if (p->p.debug & D_PACKETS)
+ log(L_TRACE "%s: %I/%d: %s", p->p.name, e->net->n.prefix, e->net->n.pxlen, msg);
+}
+
+static inline void
+krt_trace_in_rl(struct tbf *f, struct krt_proto *p, rte *e, char *msg)
+{
+ if (p->p.debug & D_PACKETS)
+ log_rl(f, L_TRACE "%s: %I/%d: %s", p->p.name, e->net->n.prefix, e->net->n.pxlen, msg);
+}
+
/*
* Inherited Routes
*/
#ifdef KRT_ALLOW_LEARN
+static struct tbf rl_alien = TBF_DEFAULT_LOG_LIMITS;
+
+/*
+ * krt_same_key() specifies what (aside from the net) is the key in
+ * kernel routing tables. It should be OS-dependent, this is for
+ * Linux. It is important for asynchronous alien updates, because a
+ * positive update is implicitly a negative one for any old route with
+ * the same key.
+ */
+
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;
+ return a->u.krt.metric == b->u.krt.metric;
+}
+
+static inline int
+krt_uptodate(rte *a, rte *b)
+{
+ if (a->attrs != b->attrs)
+ return 0;
+
+ if (a->u.krt.proto != b->u.krt.proto)
+ return 0;
+
+ return 1;
}
static void
net *nn = net_get(p->p.table, n->n.prefix, n->n.pxlen);
ee->net = nn;
ee->pflags = 0;
+ ee->pref = p->p.preference;
ee->u.krt = e->u.krt;
- rte_update(p->p.table, nn, &p->p, ee);
+ rte_update(&p->p, nn, ee);
}
static void
krt_learn_announce_delete(struct krt_proto *p, net *n)
{
n = net_find(p->p.table, n->n.prefix, n->n.pxlen);
- if (n)
- rte_update(p->p.table, n, &p->p, NULL);
+ rte_update(&p->p, n, NULL);
}
+/* Called when alien route is discovered during scan */
static void
krt_learn_scan(struct krt_proto *p, rte *e)
{
net *n = net_get(&p->krt_table, n0->n.prefix, n0->n.pxlen);
rte *m, **mm;
- e->attrs->source = RTS_INHERIT;
+ e->attrs = rta_lookup(e->attrs);
for(mm=&n->routes; m = *mm; mm=&m->next)
if (krt_same_key(m, e))
{
if (krt_uptodate(m, e))
{
- DBG("krt_learn_scan: SEEN\n");
+ krt_trace_in_rl(&rl_alien, p, e, "[alien] seen");
rte_free(e);
m->u.krt.seen = 1;
}
else
{
- DBG("krt_learn_scan: OVERRIDE\n");
+ krt_trace_in(p, e, "[alien] updated");
*mm = m->next;
rte_free(m);
m = NULL;
}
}
else
- DBG("krt_learn_scan: CREATE\n");
+ krt_trace_in(p, e, "[alien] created");
if (!m)
{
- e->attrs = rta_lookup(e->attrs);
e->next = n->routes;
n->routes = e;
e->u.krt.seen = 1;
struct fib *fib = &p->krt_table.fib;
struct fib_iterator fit;
- DBG("Pruning inheritance data...\n");
+ KRT_TRACE(p, D_EVENTS, "Pruning inherited routes");
FIB_ITERATE_INIT(&fit, fib);
again:
net *n = (net *) f;
rte *e, **ee, *best, **pbest, *old_best;
- old_best = n->routes;
+ /*
+ * Note that old_best may be NULL even if there was an old best route in
+ * the previous step, because it might be replaced in krt_learn_scan().
+ * But in that case there is a new valid best route.
+ */
+
+ old_best = NULL;
best = NULL;
pbest = NULL;
ee = &n->routes;
while (e = *ee)
{
+ if (e->u.krt.best)
+ old_best = e;
+
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;
+ e->u.krt.best = 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;
- }
+ krt_learn_announce_delete(p, n);
+
FIB_ITERATE_PUT(&fit, f);
fib_delete(fib, f);
goto again;
}
+
+ best->u.krt.best = 1;
*pbest = best->next;
best->next = n->routes;
n->routes = best;
- if (best != old_best || !(n->n.flags & KRF_INSTALLED))
+
+ if ((best != old_best) || p->reload)
{
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);
+
+ p->reload = 0;
}
static void
net *n = net_get(&p->krt_table, n0->n.prefix, n0->n.pxlen);
rte *g, **gg, *best, **bestp, *old_best;
- e->attrs->source = RTS_INHERIT;
+ e->attrs = rta_lookup(e->attrs);
old_best = n->routes;
for(gg=&n->routes; g = *gg; gg = &g->next)
{
if (krt_uptodate(g, e))
{
- DBG("krt_learn_async: same\n");
+ krt_trace_in(p, e, "[alien async] same");
rte_free(e);
return;
}
- DBG("krt_learn_async: update\n");
+ krt_trace_in(p, e, "[alien async] updated");
*gg = g->next;
rte_free(g);
}
else
- DBG("krt_learn_async: create\n");
- e->attrs = rta_lookup(e->attrs);
+ krt_trace_in(p, e, "[alien async] created");
+
e->next = n->routes;
n->routes = e;
}
else if (!g)
{
- DBG("krt_learn_async: not found\n");
+ krt_trace_in(p, e, "[alien async] delete failed");
rte_free(e);
return;
}
else
{
- DBG("krt_learn_async: delete\n");
+ krt_trace_in(p, e, "[alien async] removed");
*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;
}
+
+ g->u.krt.best = 0;
+ }
+
if (best)
{
+ best->u.krt.best = 1;
*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;
- }
+ krt_learn_announce_update(p, best);
else
- {
- n->routes = NULL;
- krt_learn_announce_delete(p, n);
- n->n.flags &= ~KRF_INSTALLED;
- }
+ krt_learn_announce_delete(p, n);
}
}
krt_learn_init(struct krt_proto *p)
{
if (KRT_CF->learn)
- rt_setup(p->p.pool, &p->krt_table, "Inherited");
+ rt_setup(p->p.pool, &p->krt_table, "Inherited", NULL);
}
static void
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);
+ debug(" [m=%d,p=%d]", e->u.krt.metric, e->u.krt.proto);
}
#endif
* Routes
*/
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
-static timer *krt_scan_timer;
-static int krt_instance_count;
-static list krt_instance_list;
-#endif
-
static void
krt_flush_routes(struct krt_proto *p)
{
struct rtable *t = p->p.table;
- DBG("Flushing kernel routes...\n");
+ KRT_TRACE(p, D_EVENTS, "Flushing kernel routes");
FIB_WALK(&t->fib, f)
{
net *n = (net *) f;
rte *e = n->routes;
- if (e)
+ if (rte_is_valid(e) && (n->n.flags & KRF_INSTALLED))
{
- rta *a = e->attrs;
- if (a->source != RTS_DEVICE && a->source != RTS_INHERIT)
- krt_set_notify(p, e->net, NULL, e);
+ /* FIXME: this does not work if gw is changed in export filter */
+ krt_replace_rte(p, e->net, NULL, e, NULL);
+ n->n.flags &= ~KRF_INSTALLED;
}
}
FIB_WALK_END;
}
+static struct rte *
+krt_export_net(struct krt_proto *p, net *net, rte **rt_free, ea_list **tmpa)
+{
+ struct announce_hook *ah = p->p.main_ahook;
+ struct filter *filter = ah->out_filter;
+ rte *rt;
+
+ if (p->p.accept_ra_types == RA_MERGED)
+ return rt_export_merged(ah, net, rt_free, tmpa, krt_filter_lp, 1);
+
+ rt = net->routes;
+ *rt_free = NULL;
+
+ if (!rte_is_valid(rt))
+ return NULL;
+
+ if (filter == FILTER_REJECT)
+ return NULL;
+
+ struct proto *src = rt->attrs->src->proto;
+ *tmpa = src->make_tmp_attrs ? src->make_tmp_attrs(rt, krt_filter_lp) : NULL;
+
+ /* We could run krt_import_control() here, but it is already handled by KRF_INSTALLED */
+
+ if (filter == FILTER_ACCEPT)
+ goto accept;
+
+ if (f_run(filter, &rt, tmpa, krt_filter_lp, FF_FORCE_TMPATTR | FF_SILENT) > F_ACCEPT)
+ goto reject;
+
+
+accept:
+ if (rt != net->routes)
+ *rt_free = rt;
+ return rt;
+
+reject:
+ if (rt != net->routes)
+ rte_free(rt);
+ return NULL;
+}
+
static int
-krt_uptodate(rte *k, rte *e)
+krt_same_dest(rte *k, rte *e)
{
rta *ka = k->attrs, *ea = e->attrs;
return ipa_equal(ka->gw, ea->gw);
case RTD_DEVICE:
return !strcmp(ka->iface->name, ea->iface->name);
+ case RTD_MULTIPATH:
+ return mpnh_same(ka->nexthops, ea->nexthops);
default:
return 1;
}
void
krt_got_route(struct krt_proto *p, rte *e)
{
- rte *old;
net *net = e->net;
- int src = e->u.krt.src;
int verdict;
-#ifdef CONFIG_AUTO_ROUTES
- if (e->attrs->dest == RTD_DEVICE)
+#ifdef KRT_ALLOW_LEARN
+ switch (e->u.krt.src)
{
- /* It's a device route. Probably a kernel-generated one. */
+ case KRT_SRC_KERNEL:
verdict = KRF_IGNORE;
goto sentenced;
- }
-#endif
-#ifdef KRT_ALLOW_LEARN
- if (src == KRT_SRC_ALIEN)
- {
+ case KRT_SRC_REDIRECT:
+ verdict = KRF_DELETE;
+ goto sentenced;
+
+ case KRT_SRC_ALIEN:
if (KRT_CF->learn)
krt_learn_scan(p, e);
else
- DBG("krt_parse_entry: Alien route, ignoring\n");
+ {
+ krt_trace_in_rl(&rl_alien, p, e, "[alien] ignored");
+ rte_free(e);
+ }
return;
}
#endif
+ /* The rest is for KRT_SRC_BIRD (or KRT_SRC_UNKNOWN) */
if (net->n.flags & KRF_VERDICT_MASK)
{
/* Route to this destination was already seen. Strange, but it happens... */
- DBG("Already seen.\n");
+ krt_trace_in(p, e, "already seen");
+ rte_free(e);
return;
}
+ if (!p->ready)
+ {
+ /* We wait for the initial feed to have correct KRF_INSTALLED flag */
+ verdict = KRF_IGNORE;
+ goto sentenced;
+ }
+
if (net->n.flags & KRF_INSTALLED)
{
- old = net->routes;
- ASSERT(old);
- if (krt_uptodate(e, old))
- verdict = KRF_SEEN;
- else
+ rte *new, *rt_free;
+ ea_list *tmpa;
+
+ new = krt_export_net(p, net, &rt_free, &tmpa);
+
+ /* TODO: There also may be changes in route eattrs, we ignore that for now. */
+
+ if (!new)
+ verdict = KRF_DELETE;
+ else if ((net->n.flags & KRF_SYNC_ERROR) || !krt_same_dest(e, new))
verdict = KRF_UPDATE;
+ else
+ verdict = KRF_SEEN;
+
+ if (rt_free)
+ rte_free(rt_free);
+
+ lp_flush(krt_filter_lp);
}
else
verdict = KRF_DELETE;
-sentenced:
- DBG("krt_parse_entry: verdict=%s\n", ((char *[]) { "CREATE", "SEEN", "UPDATE", "DELETE", "IGNORE" }) [verdict]);
-
+ sentenced:
+ krt_trace_in(p, e, ((char *[]) { "?", "seen", "will be updated", "will be removed", "ignored" }) [verdict]);
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 */
+ /* Get a cached copy of attributes and temporarily link the route */
rta *a = e->attrs;
a->source = RTS_DUMMY;
e->attrs = rta_lookup(a);
static void
krt_prune(struct krt_proto *p)
{
- struct proto *pp = &p->p;
struct rtable *t = p->p.table;
- struct fib_node *f;
- DBG("Pruning routes in table %s...\n", t->name);
+ KRT_TRACE(p, D_EVENTS, "Pruning table %s", t->name);
FIB_WALK(&t->fib, f)
{
net *n = (net *) f;
int verdict = f->flags & KRF_VERDICT_MASK;
- rte *new, *old;
+ rte *new, *old, *rt_free = NULL;
+ ea_list *tmpa = NULL;
- if (verdict != KRF_CREATE && verdict != KRF_SEEN && verdict != KRF_IGNORE)
+ if (verdict == KRF_UPDATE || verdict == KRF_DELETE)
{
+ /* Get a dummy route from krt_got_route() */
old = n->routes;
n->routes = old->next;
}
else
old = NULL;
- new = n->routes;
+
+ if (verdict == KRF_CREATE || verdict == KRF_UPDATE)
+ {
+ /* We have to run export filter to get proper 'new' route */
+ new = krt_export_net(p, n, &rt_free, &tmpa);
+
+ if (!new)
+ verdict = (verdict == KRF_CREATE) ? KRF_IGNORE : KRF_DELETE;
+ else
+ tmpa = ea_append(tmpa, new->attrs->eattrs);
+ }
+ else
+ new = NULL;
switch (verdict)
{
case KRF_CREATE:
if (new && (f->flags & KRF_INSTALLED))
{
- DBG("krt_prune: reinstalling %I/%d\n", n->n.prefix, n->n.pxlen);
- krt_set_notify(p, n, new, NULL);
+ krt_trace_in(p, new, "reinstalling");
+ krt_replace_rte(p, n, new, NULL, tmpa);
}
break;
case KRF_SEEN:
/* Nothing happens */
break;
case KRF_UPDATE:
- DBG("krt_prune: updating %I/%d\n", n->n.prefix, n->n.pxlen);
- krt_set_notify(p, n, new, old);
+ krt_trace_in(p, new, "updating");
+ krt_replace_rte(p, n, new, old, tmpa);
break;
case KRF_DELETE:
- DBG("krt_prune: deleting %I/%d\n", n->n.prefix, n->n.pxlen);
- krt_set_notify(p, n, NULL, old);
+ krt_trace_in(p, old, "deleting");
+ krt_replace_rte(p, n, NULL, old, NULL);
break;
default:
bug("krt_prune: invalid route status");
}
+
if (old)
rte_free(old);
+ if (rt_free)
+ rte_free(rt_free);
+ lp_flush(krt_filter_lp);
f->flags &= ~KRF_VERDICT_MASK;
}
FIB_WALK_END;
if (KRT_CF->learn)
krt_learn_prune(p);
#endif
+
+ if (p->ready)
+ p->initialized = 1;
}
void
krt_got_route_async(struct krt_proto *p, rte *e, int new)
{
net *net = e->net;
- rte *old = net->routes;
- int src = e->u.krt.src;
- switch (src)
+ switch (e->u.krt.src)
{
case KRT_SRC_BIRD:
- ASSERT(0);
+ ASSERT(0); /* Should be filtered by the back end */
+
case KRT_SRC_REDIRECT:
- DBG("It's a redirect, kill him! Kill! Kill!\n");
- krt_set_notify(p, net, NULL, e);
+ if (new)
+ {
+ krt_trace_in(p, e, "[redirect] deleting");
+ krt_replace_rte(p, net, NULL, e, NULL);
+ }
+ /* If !new, it is probably echo of our deletion */
break;
- case KRT_SRC_ALIEN:
+
#ifdef KRT_ALLOW_LEARN
+ case KRT_SRC_ALIEN:
if (KRT_CF->learn)
{
krt_learn_async(p, e, new);
return;
}
#endif
- /* Fall-thru */
- default:
- DBG("Discarding\n");
- rte_update(p->p.table, net, &p->p, NULL);
}
rte_free(e);
}
* Periodic scanning
*/
+
+#ifdef CONFIG_ALL_TABLES_AT_ONCE
+
+static timer *krt_scan_timer;
+static int krt_scan_count;
+
static void
-krt_scan(timer *t)
+krt_scan(timer *t UNUSED)
{
struct krt_proto *p;
kif_force_scan();
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
+
+ /* We need some node to decide whether to print the debug messages or not */
+ p = SKIP_BACK(struct krt_proto, krt_node, HEAD(krt_proto_list));
+ KRT_TRACE(p, D_EVENTS, "Scanning routing table");
+
+ krt_do_scan(NULL);
+
+ void *q;
+ WALK_LIST(q, krt_proto_list)
{
- void *q;
- DBG("KRT: It's route scan time...\n");
- krt_scan_fire(NULL);
- WALK_LIST(q, krt_instance_list)
- {
- p = SKIP_BACK(struct krt_proto, instance_node, q);
- krt_prune(p);
- }
+ p = SKIP_BACK(struct krt_proto, krt_node, q);
+ krt_prune(p);
}
+}
+
+static void
+krt_scan_timer_start(struct krt_proto *p)
+{
+ if (!krt_scan_count)
+ krt_scan_timer = tm_new_set(krt_pool, krt_scan, NULL, 0, KRT_CF->scan_time);
+
+ krt_scan_count++;
+
+ tm_start(krt_scan_timer, 1);
+}
+
+static void
+krt_scan_timer_stop(struct krt_proto *p UNUSED)
+{
+ krt_scan_count--;
+
+ if (!krt_scan_count)
+ {
+ rfree(krt_scan_timer);
+ krt_scan_timer = NULL;
+ }
+}
+
+static void
+krt_scan_timer_kick(struct krt_proto *p UNUSED)
+{
+ tm_start(krt_scan_timer, 0);
+}
+
#else
- p = t->data;
- DBG("KRT: It's route scan time for %s...\n", p->p.name);
- krt_scan_fire(p);
+
+static void
+krt_scan(timer *t)
+{
+ struct krt_proto *p = t->data;
+
+ kif_force_scan();
+
+ KRT_TRACE(p, D_EVENTS, "Scanning routing table");
+ krt_do_scan(p);
krt_prune(p);
-#endif
}
+static void
+krt_scan_timer_start(struct krt_proto *p)
+{
+ p->scan_timer = tm_new_set(p->p.pool, krt_scan, p, 0, KRT_CF->scan_time);
+ tm_start(p->scan_timer, 1);
+}
+
+static void
+krt_scan_timer_stop(struct krt_proto *p)
+{
+ tm_stop(p->scan_timer);
+}
+
+static void
+krt_scan_timer_kick(struct krt_proto *p)
+{
+ tm_start(p->scan_timer, 0);
+}
+
+#endif
+
+
+
+
/*
* Updates
*/
+static struct ea_list *
+krt_make_tmp_attrs(rte *rt, struct linpool *pool)
+{
+ struct ea_list *l = lp_alloc(pool, sizeof(struct ea_list) + 2 * sizeof(eattr));
+
+ l->next = NULL;
+ l->flags = EALF_SORTED;
+ l->count = 2;
+
+ l->attrs[0].id = EA_KRT_SOURCE;
+ l->attrs[0].flags = 0;
+ l->attrs[0].type = EAF_TYPE_INT | EAF_TEMP;
+ l->attrs[0].u.data = rt->u.krt.proto;
+
+ l->attrs[1].id = EA_KRT_METRIC;
+ l->attrs[1].flags = 0;
+ l->attrs[1].type = EAF_TYPE_INT | EAF_TEMP;
+ l->attrs[1].u.data = rt->u.krt.metric;
+
+ return l;
+}
+
static void
-krt_notify(struct proto *P, net *net, rte *new, rte *old, struct ea_list *tmpa)
+krt_store_tmp_attrs(rte *rt, struct ea_list *attrs)
+{
+ /* EA_KRT_SOURCE is read-only */
+ rt->u.krt.metric = ea_get_int(attrs, EA_KRT_METRIC, 0);
+}
+
+static int
+krt_import_control(struct proto *P, rte **new, ea_list **attrs UNUSED, struct linpool *pool UNUSED)
{
struct krt_proto *p = (struct krt_proto *) P;
+ rte *e = *new;
+
+ if (e->attrs->src->proto == P)
+ {
+#ifdef CONFIG_SINGLE_ROUTE
+ /*
+ * Implicit withdraw - when the imported kernel route becomes the best one,
+ * we know that the previous one exported to the kernel was already removed,
+ * but if we processed the update as usual, we would send withdraw to the
+ * kernel, which would remove the new imported route instead.
+ *
+ * We will remove KRT_INSTALLED flag, which stops such withdraw to be
+ * processed in krt_rt_notify() and krt_replace_rte().
+ */
+ if (e == e->net->routes)
+ e->net->n.flags &= ~KRF_INSTALLED;
+#endif
+ return -1;
+ }
+
+ if (!KRT_CF->devroutes &&
+ (e->attrs->dest == RTD_DEVICE) &&
+ (e->attrs->source != RTS_STATIC_DEVICE))
+ return -1;
+
+ if (!krt_capable(e))
+ return -1;
- if (new && (!krt_capable(new) || new->attrs->source == RTS_INHERIT))
- new = NULL;
+ return 0;
+}
+
+static void
+krt_rt_notify(struct proto *P, struct rtable *table UNUSED, net *net,
+ rte *new, rte *old, struct ea_list *eattrs)
+{
+ struct krt_proto *p = (struct krt_proto *) P;
+
+ if (config->shutdown)
+ return;
if (!(net->n.flags & KRF_INSTALLED))
old = NULL;
if (new)
net->n.flags |= KRF_INSTALLED;
else
net->n.flags &= ~KRF_INSTALLED;
- krt_set_notify(p, net, new, old);
+ if (p->initialized) /* Before first scan we don't touch the routes */
+ krt_replace_rte(p, net, new, old, eattrs);
}
-/*
- * Protocol glue
- */
+static void
+krt_if_notify(struct proto *P, uint flags, struct iface *iface UNUSED)
+{
+ struct krt_proto *p = (struct krt_proto *) P;
-struct proto_config *cf_krt;
+ /*
+ * When interface went down, we should remove routes to it. In the ideal world,
+ * OS kernel would send us route removal notifications in such cases, but we
+ * cannot rely on it as it is often not true. E.g. Linux kernel removes related
+ * routes when an interface went down, but it does not notify userspace about
+ * that. To be sure, we just schedule a scan to ensure synchronization.
+ */
-static void
-krt_preconfig(struct protocol *P, struct config *c)
+ if ((flags & IF_CHANGE_DOWN) && KRT_CF->learn)
+ krt_scan_timer_kick(p);
+}
+
+static int
+krt_reload_routes(struct proto *P)
{
- cf_krt = NULL;
- krt_scan_preconfig(c);
+ struct krt_proto *p = (struct krt_proto *) P;
+
+ /* Although we keep learned routes in krt_table, we rather schedule a scan */
+
+ if (KRT_CF->learn)
+ {
+ p->reload = 1;
+ krt_scan_timer_kick(p);
+ }
+
+ return 1;
}
static void
-krt_postconfig(struct proto_config *C)
+krt_feed_end(struct proto *P)
{
- struct krt_config *c = (struct krt_config *) C;
+ struct krt_proto *p = (struct krt_proto *) P;
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
- struct krt_config *first = (struct krt_config *) cf_krt;
- if (first->scan_time != c->scan_time)
- cf_error("All kernel syncers must use the same table scan interval");
-#endif
+ p->ready = 1;
+ krt_scan_timer_kick(p);
+}
- if (C->table->krt_attached)
- cf_error("Kernel syncer (%s) already attached to table %s", C->table->krt_attached->name, C->table->name);
- C->table->krt_attached = C;
- krt_scan_postconfig(c);
+
+static int
+krt_rte_same(rte *a, rte *b)
+{
+ /* src is always KRT_SRC_ALIEN and type is irrelevant */
+ return (a->u.krt.proto == b->u.krt.proto) && (a->u.krt.metric == b->u.krt.metric);
}
-static timer *
-krt_start_timer(struct krt_proto *p)
+
+/*
+ * Protocol glue
+ */
+
+struct krt_config *krt_cf;
+
+static struct proto *
+krt_init(struct proto_config *C)
{
- timer *t;
+ struct krt_proto *p = proto_new(C, sizeof(struct krt_proto));
+ struct krt_config *c = (struct krt_config *) C;
- t = tm_new(p->krt_pool);
- t->hook = krt_scan;
- t->data = p;
- t->recurrent = KRT_CF->scan_time;
- tm_start(t, KRT_CF->scan_time);
- return t;
+ p->p.accept_ra_types = c->merge_paths ? RA_MERGED : RA_OPTIMAL;
+ p->p.merge_limit = c->merge_paths;
+ p->p.import_control = krt_import_control;
+ p->p.rt_notify = krt_rt_notify;
+ p->p.if_notify = krt_if_notify;
+ p->p.reload_routes = krt_reload_routes;
+ p->p.feed_end = krt_feed_end;
+ p->p.make_tmp_attrs = krt_make_tmp_attrs;
+ p->p.store_tmp_attrs = krt_store_tmp_attrs;
+ p->p.rte_same = krt_rte_same;
+
+ krt_sys_init(p);
+ return &p->p;
}
static int
krt_start(struct proto *P)
{
struct krt_proto *p = (struct krt_proto *) P;
- int first = 1;
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
- if (!krt_instance_count++)
- init_list(&krt_instance_list);
- else
- first = 0;
- p->krt_pool = krt_pool;
- add_tail(&krt_instance_list, &p->instance_node);
-#else
- p->krt_pool = P->pool;
-#endif
+ add_tail(&krt_proto_list, &p->krt_node);
#ifdef KRT_ALLOW_LEARN
krt_learn_init(p);
#endif
- krt_scan_start(p, first);
- krt_set_start(p, first);
+ if (!krt_sys_start(p))
+ {
+ rem_node(&p->krt_node);
+ return PS_START;
+ }
- /* Start periodic routing table scanning */
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
- if (first)
- krt_scan_timer = krt_start_timer(p);
- p->scan_timer = krt_scan_timer;
- /* If this is the last instance to be initialized, kick the timer */
- if (!P->proto->startup_counter)
- krt_scan(p->scan_timer);
-#else
- p->scan_timer = krt_start_timer(p);
- krt_scan(p->scan_timer);
-#endif
+ krt_scan_timer_start(p);
+
+ if (P->gr_recovery && KRT_CF->graceful_restart)
+ P->gr_wait = 1;
return PS_UP;
}
krt_shutdown(struct proto *P)
{
struct krt_proto *p = (struct krt_proto *) P;
- int last = 1;
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
- rem_node(&p->instance_node);
- if (--krt_instance_count)
- last = 0;
- else
-#endif
- tm_stop(p->scan_timer);
+ krt_scan_timer_stop(p);
- if (!KRT_CF->persist)
+ /* FIXME we should flush routes even when persist during reconfiguration */
+ if (p->initialized && !KRT_CF->persist)
krt_flush_routes(p);
- krt_set_shutdown(p, last);
- krt_scan_shutdown(p, last);
+ p->ready = 0;
+ p->initialized = 0;
+
+ if (p->p.proto_state == PS_START)
+ return PS_DOWN;
+
+ krt_sys_shutdown(p);
+ rem_node(&p->krt_node);
+
+ return PS_DOWN;
+}
+
+static int
+krt_reconfigure(struct proto *p, struct proto_config *new)
+{
+ struct krt_config *o = (struct krt_config *) p->cf;
+ struct krt_config *n = (struct krt_config *) new;
+
+ if (!krt_sys_reconfigure((struct krt_proto *) p, n, o))
+ return 0;
+
+ /* persist, graceful restart need not be the same */
+ return o->scan_time == n->scan_time && o->learn == n->learn &&
+ o->devroutes == n->devroutes && o->merge_paths == n->merge_paths;
+}
+
+static void
+krt_preconfig(struct protocol *P UNUSED, struct config *c)
+{
+ krt_cf = NULL;
+ krt_sys_preconfig(c);
+}
+
+static void
+krt_postconfig(struct proto_config *C)
+{
+ struct krt_config *c = (struct krt_config *) C;
#ifdef CONFIG_ALL_TABLES_AT_ONCE
- if (last)
- rfree(krt_scan_timer);
+ if (krt_cf->scan_time != c->scan_time)
+ cf_error("All kernel syncers must use the same table scan interval");
#endif
- return PS_DOWN;
+ if (C->table->krt_attached)
+ cf_error("Kernel syncer (%s) already attached to table %s", C->table->krt_attached->name, C->table->name);
+ C->table->krt_attached = C;
+ krt_sys_postconfig(c);
}
-static struct proto *
-krt_init(struct proto_config *c)
+struct proto_config *
+krt_init_config(int class)
{
- struct krt_proto *p = proto_new(c, sizeof(struct krt_proto));
+#ifndef CONFIG_MULTIPLE_TABLES
+ if (krt_cf)
+ cf_error("Kernel protocol already defined");
+#endif
- p->p.rt_notify = krt_notify;
- p->p.min_scope = SCOPE_HOST;
- return &p->p;
+ krt_cf = (struct krt_config *) proto_config_new(&proto_unix_kernel, class);
+ krt_cf->scan_time = 60;
+
+ krt_sys_init_config(krt_cf);
+ return (struct proto_config *) krt_cf;
}
+static void
+krt_copy_config(struct proto_config *dest, struct proto_config *src)
+{
+ struct krt_config *d = (struct krt_config *) dest;
+ struct krt_config *s = (struct krt_config *) src;
+
+ /* Shallow copy of everything */
+ proto_copy_rest(dest, src, sizeof(struct krt_config));
+
+ /* Fix sysdep parts */
+ krt_sys_copy_config(d, s);
+}
+
+static int
+krt_get_attr(eattr *a, byte *buf, int buflen)
+{
+ switch (a->id)
+ {
+ case EA_KRT_SOURCE:
+ bsprintf(buf, "source");
+ return GA_NAME;
+
+ case EA_KRT_METRIC:
+ bsprintf(buf, "metric");
+ return GA_NAME;
+
+ default:
+ return krt_sys_get_attr(a, buf, buflen);
+ }
+}
+
+
struct protocol proto_unix_kernel = {
- name: "Kernel",
- priority: 80,
- preconfig: krt_preconfig,
- postconfig: krt_postconfig,
- init: krt_init,
- start: krt_start,
- shutdown: krt_shutdown,
+ .name = "Kernel",
+ .template = "kernel%d",
+ .attr_class = EAP_KRT,
+ .preference = DEF_PREF_INHERITED,
+ .config_size = sizeof(struct krt_config),
+ .preconfig = krt_preconfig,
+ .postconfig = krt_postconfig,
+ .init = krt_init,
+ .start = krt_start,
+ .shutdown = krt_shutdown,
+ .reconfigure = krt_reconfigure,
+ .copy_config = krt_copy_config,
+ .get_attr = krt_get_attr,
#ifdef KRT_ALLOW_LEARN
- dump: krt_dump,
- dump_attrs: krt_dump_attrs,
+ .dump = krt_dump,
+ .dump_attrs = krt_dump_attrs,
#endif
};