Hi Lars, Thanks for picking this up! On Tue, Jan 06, 2026 at 04:43:41AM +0100, Lars Gierth via Bird-users wrote:
Hi all,
happy new year and thanks for the fresh releases!
We have been successfully running the Babel route selection patch [1] with Bird 2.17 in our community mesh network with ~200 Babel nodes since Q3/2024. It has allowed us to really fine-tune a mix of all kinds of wire types connecting the locations: wifi at 60 / 5 / 2 GHz, tunnels over LTE or DSL, fiber in a datacenter, etc. etc. With Babel being a critical part of our network, the patch is running on a broad range of architectures: mips, ppc64, x86_64, and various arm and aarch64 types.
With dxld's consent I took his patch and rebased it on Bird 3.1, and we've been testing that for 2 months on the live network. Today I rebased it on 3.2 (thread-next branch), it applied cleanly but I haven't tested it yet. Nevertheless I decided to share it now, see attachment.
Kind regards, Lars Freifunk Berlin
[1] https://bird.network.cz/pipermail/bird-users/2023-February/016705.html
From d2ee215ea7927fe5677990781bffe682b7e53719 Mon Sep 17 00:00:00 2001 From: Packet Please <pktpls@systemli.org> Date: Sat, 8 Nov 2025 18:57:18 +0100 Subject: [PATCH] Babel: Replace internal route selection by bird's nest MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit
This introduces the ability to filter routes from specific interfaces and neighbours. With the current internal route selection proto babel exports only up to one route and an admin cannot do fine-grained filtering.
To fix this we rip out the internal route selection entirely and export evey babel route as a distinct route into the bird's nest instead.
In the presence of esoteric route filters this change may be observable by users however we feel the additional filtering power is more important. A release note entry for this change is recommended.
Signed-off-by: Daniel GrM-CM-6ber <dxld@darkboxed.org>
Also add a Co-authored-by/Co-developed-by please. Also Is the encoding broken? https://www.kernel.org/doc/html/latest/process/submitting-patches.html#when-...
Signed-off-by: Lars Gierth <larsg@systemli.org> --- proto/babel/babel.c | 306 ++++++++++++++++++++++---------------------- proto/babel/babel.h | 3 + 2 files changed, 154 insertions(+), 155 deletions(-)
diff --git a/proto/babel/babel.c b/proto/babel/babel.c index c7d47c01f..5f3e94fe1 100644
The patch doesn't apply. On thread-next (0b91feeda): $ git am < babel.patch Applying: Babel: Replace internal route selection by bird's nest error: patch failed: proto/babel/babel.c:251 error: proto/babel/babel.c: patch does not apply error: patch failed: proto/babel/babel.h:25 error: proto/babel/babel.h: patch does not apply Patch failed at 0001 Babel: Replace internal route selection by bird's nest hint: Use 'git am --show-current-patch=diff' to see the failed patch hint: When you have resolved this problem, run "git am --continue". hint: If you prefer to skip this patch, run "git am --skip" instead. hint: To restore the original branch and stop patching, run "git am --abort". hint: Disable this message with "git config advice.mergeConflict false" Could you use git-send-email(1) please :> https://git-send-email.io
--- a/proto/babel/babel.c +++ b/proto/babel/babel.c @@ -28,10 +28,12 @@ * possible routes for the prefix are tracked as &babel_route entries and the * feasibility distance is maintained through &babel_source structures. * - * The main route selection is done in babel_select_route(). This is called when - * an entry is updated by receiving updates from the network or when modified by - * internal timers. The function selects from feasible and reachable routes the - * one with the lowest metric to be announced to the core. + * The main route selection is done by bird's nest (heh). For each prefix we + * export all feasible routes to the core with a distinct source (rte_src) per + * neihbour so bird handles them as distinct routes. Nest will then notify us of + * one best route for each prefix, either our own (internal) or from another + * protocol, by calling babel_rt_notify. For internal best routes we remember + * which babel_route it selected in babel_entry.selected. * * Supported standards: * RFC 8966 - The Babel Routing Protocol @@ -60,8 +62,8 @@ static inline int gt_mod64k(uint a, uint b) { return ge_mod64k(a, b) && a != b; }
static void babel_expire_requests(struct babel_proto *p, struct babel_entry *e); -static void babel_select_route(struct babel_proto *p, struct babel_entry *e, struct babel_route *mod); -static inline void babel_announce_retraction(struct babel_proto *p, struct babel_entry *e); +static void babel_announce_rte(struct babel_proto *p, struct babel_entry *e, struct babel_route *r); +static void babel_rte_update_unreachable(struct babel_proto *p, struct babel_entry *e, u8 announce); static void babel_send_route_request(struct babel_proto *p, struct babel_entry *e, struct babel_neighbor *n); static void babel_send_seqno_request(struct babel_proto *p, struct babel_entry *e, struct babel_seqno_request *sr, struct babel_neighbor *n); static void babel_update_cost(struct babel_neighbor *n); @@ -181,9 +183,7 @@ static inline void babel_retract_route(struct babel_proto *p, struct babel_route *r) { r->metric = r->advert_metric = BABEL_INFINITY; - - if (r == r->e->selected) - babel_select_route(p, r->e, r); + babel_announce_rte(p, r->e, r); }
static void @@ -195,9 +195,6 @@ babel_flush_route(struct babel_proto *p UNUSED, struct babel_route *r) rem_node(NODE r); rem_node(&r->neigh_route);
- if (r->e->selected == r) - r->e->selected = NULL; - sl_free(r); }
@@ -213,6 +210,7 @@ babel_expire_route(struct babel_proto *p, struct babel_route *r) { r->metric = r->advert_metric = BABEL_INFINITY; r->expires = current_time() + cf->hold_time; + babel_announce_rte(p, r->e, r); } else { @@ -242,8 +240,6 @@ babel_expire_routes_(struct babel_proto *p, struct fib *rtable) loop: FIB_ITERATE_START(rtable, &fit, struct babel_entry, e) { - int changed = 0; - WALK_LIST_DELSAFE(r, rx, e->routes) { if (r->refresh_time && r->refresh_time <= now_) @@ -251,23 +247,12 @@ babel_expire_routes_(struct babel_proto *p, struct fib *rtable)
if (r->expires && r->expires <= now_) { - changed = changed || (r == e->selected); + FIB_ITERATE_PUT(&fit); babel_expire_route(p, r); + goto loop; } }
- if (changed) - { - /* - * We have to restart the iteration because there may be a cascade of - * synchronous events babel_select_route() -> nest table change -> - * babel_rt_notify() -> rtable change, invalidating hidden variables. - */ - FIB_ITERATE_PUT(&fit); - babel_select_route(p, e, NULL); - goto loop; - } - /* Clean up stale entries */ if ((e->valid == BABEL_ENTRY_STALE) && ((e->updated + cf->hold_time) <= now_)) e->valid = BABEL_ENTRY_DUMMY; @@ -276,7 +261,7 @@ babel_expire_routes_(struct babel_proto *p, struct fib *rtable) if (e->unreachable && (!e->valid || (e->router_id == p->router_id))) { FIB_ITERATE_PUT(&fit); - babel_announce_retraction(p, e); + babel_rte_update_unreachable(p, e, 0); goto loop; }
@@ -461,6 +446,8 @@ babel_get_neighbor(struct babel_iface *ifa, ip_addr addr)
nbr = mb_allocz(ifa->pool, sizeof(struct babel_neighbor)); nbr->ifa = ifa; + nbr->src = rt_get_source(&p->p, idm_alloc(&p->src_ids)); + rt_lock_source(nbr->src); nbr->addr = addr; nbr->rxcost = BABEL_INFINITY; nbr->txcost = BABEL_INFINITY; @@ -488,6 +475,9 @@ babel_flush_neighbor(struct babel_proto *p, struct babel_neighbor *nbr) }
nbr->ifa = NULL; + + rt_unlock_source(nbr->src); + idm_free(&p->src_ids, nbr->src->private_id); rem_node(NODE nbr); mb_free(nbr); } @@ -667,11 +657,60 @@ babel_update_cost(struct babel_neighbor *nbr) WALK_LIST2(r, n, nbr->routes, neigh_route) { r->metric = babel_compute_metric(nbr, r->advert_metric); - babel_select_route(p, r->e, r); + babel_announce_rte(p, r->e, r); } } }
+/** + * This function handles announcing the special unreachable route we insert for + * a prefix whenever we have no more feasible routes available as per RFC8966 + * section 3.5.4 as well as retracting it when such routes are available + * again. + * + * We also remember if we inserted an unreachable route in e->unreachable in + * order to clean it up later in babel_expire_routes_. + */ +static void +babel_rte_update_unreachable(struct babel_proto *p, struct babel_entry *e, u8 announce) +{ + struct channel *c = (e->n.addr->type == NET_IP4) ? p->ip4_channel : p->ip6_channel; + + if (announce) { + struct nexthop_adata nhad = NEXTHOP_DEST_LITERAL(RTD_UNREACHABLE); + struct { + ea_list l; + eattr a[4]; + } eattrs = { + .l.count = ARRAY_SIZE(eattrs.a), + .a = { + EA_LITERAL_EMBEDDED(&ea_gen_preference, 0, 1), + EA_LITERAL_EMBEDDED(&ea_gen_source, 0, RTS_BABEL), + EA_LITERAL_STORE_ADATA(&ea_gen_nexthop, 0, nhad.ad.data, nhad.ad.length), + EA_LITERAL_EMBEDDED(&ea_babel_metric, 0, 1), // XXX this one is just for debugging + } + }; + rte e0 = { + .attrs = &eattrs.l, + .src = p->p.main_source, + }; + + e->unreachable = 1; + rte_update(c, e->n.addr, &e0, p->p.main_source); + } + else + { + e->unreachable = 0; + rte_update(c, e->n.addr, NULL, p->p.main_source); + } + + /* Unlike the regular per-neighbour routes we only want one unreachable route + * for each prefix. This is mainly due to the lifetime of the unreachable rte + * exceeding that of the neighbour's rte_src ID as the babal neighbour may be + * flushed before the unreachable route is retracted. */ + // rte_update(c, e->n.addr, rtee, p->p.main_source); +} + /** * babel_announce_rte - announce selected route to the core * @p: Babel protocol instance @@ -682,12 +721,11 @@ babel_update_cost(struct babel_neighbor *nbr) * the entry is valid and ours, the unreachable route is announced instead. */ static void -babel_announce_rte(struct babel_proto *p, struct babel_entry *e) +babel_announce_rte(struct babel_proto *p, struct babel_entry *e, struct babel_route *r) { - struct babel_route *r = e->selected; struct channel *c = (e->n.addr->type == NET_IP4) ? p->ip4_channel : p->ip6_channel;
- if (r) + if (r->metric != BABEL_INFINITY && r->feasible) { struct nexthop_adata nhad = { .nh = { @@ -706,7 +744,7 @@ babel_announce_rte(struct babel_proto *p, struct babel_entry *e) */ if (!neigh_find(&p->p, r->next_hop, r->neigh->ifa->iface, 0)) nhad.nh.flags = RNF_ONLINK; - + struct { ea_list l; eattr a[7]; @@ -725,124 +763,25 @@ babel_announce_rte(struct babel_proto *p, struct babel_entry *e)
rte e0 = { .attrs = &eattrs.l, - .src = p->p.main_source, + .src = r->neigh->src, };
- e->unreachable = 0; - rte_update(c, e->n.addr, &e0, p->p.main_source); - } - else if (e->valid && (e->router_id != p->router_id)) - { - /* Unreachable */ - ea_list *ea = NULL; - - ea_set_attr_u32(&ea, &ea_gen_preference, 0, 1); - ea_set_attr_u32(&ea, &ea_gen_source, 0, RTS_BABEL); - ea_set_dest(&ea, 0, RTD_UNREACHABLE); - - rte e0 = { - .attrs = ea, - .src = p->p.main_source, - }; - - e->unreachable = 1; - rte_update(c, e->n.addr, &e0, p->p.main_source); - } - else - { - /* Retraction */ - e->unreachable = 0; - rte_update(c, e->n.addr, NULL, p->p.main_source); - } -} - -/* Special case of babel_announce_rte() just for retraction */ -static inline void -babel_announce_retraction(struct babel_proto *p, struct babel_entry *e) -{ - struct channel *c = (e->n.addr->type == NET_IP4) ? p->ip4_channel : p->ip6_channel; - e->unreachable = 0; - rte_update(c, e->n.addr, NULL, p->p.main_source); -} - - -/** - * babel_select_route - select best route for given route entry - * @p: Babel protocol instance - * @e: Babel entry to select the best route for - * @mod: Babel route that was modified or NULL if unspecified - * - * Select the best reachable and feasible route for a given prefix among the - * routes received from peers, and propagate it to the nest. This just selects - * the reachable and feasible route with the lowest metric, but keeps selected - * the old one in case of tie. - * - * If no feasible route is available for a prefix that previously had a route - * selected, a seqno request is sent to try to get a valid route. If the entry - * is valid and not owned by us, the unreachable route is announced to the nest - * (to blackhole packets going to it, as per section 2.8). It is later removed - * by babel_expire_routes(). Otherwise, the route is just removed from the nest. - * - * Argument @mod is used to optimize best route calculation. When specified, the - * function can assume that only the @mod route was modified to avoid full best - * route selection and announcement when non-best route was modified in minor - * way. The caller is advised to not call babel_select_route() when no change is - * done (e.g. periodic route updates) to avoid unnecessary announcements of the - * same best route. The caller is not required to call the function in case of a - * retraction of a non-best route. - * - * Note that the function does not active triggered updates. That is done by - * babel_rt_notify() when the change is propagated back to Babel. - */ -static void -babel_select_route(struct babel_proto *p, struct babel_entry *e, struct babel_route *mod) -{ - struct babel_route *r, *best = e->selected; - - /* Shortcut if only non-best was modified */ - if (mod && (mod != best)) - { - /* Either select modified route, or keep old best route */ - if ((mod->metric < (best ? best->metric : BABEL_INFINITY)) && mod->feasible) - best = mod; - else - return; + rte_update(c, e->n.addr, &e0, r->neigh->src); + if (e->unreachable) + babel_rte_update_unreachable(p, e, 0); } else { - /* Selected route may be modified and no longer admissible */ - if (!best || (best->metric == BABEL_INFINITY) || !best->feasible) - best = NULL; + if (e->selected == r) + /* We NULL e->selected here rather than wait for babel_rt_notify in order + * to stay in a sync context. This is to be prepared for async rt_notify + * in BIRD 3. This is critical as RFC8966 demands unfeasible or infinite + * metric routes never being selected (see section 3.6 Route + * Selection). */ + e->selected = NULL;
- /* Find the best feasible route from all routes */ - WALK_LIST(r, e->routes) - if ((r->metric < (best ? best->metric : BABEL_INFINITY)) && r->feasible) - best = r; + rte_update(c, e->n.addr, NULL, r->neigh->src); } - - if (best) - { - if (best != e->selected) - TRACE(D_EVENTS, "Picked new route for prefix %N: router-id %lR metric %d", - e->n.addr, best->router_id, best->metric); - } - else if (e->selected) - { - /* - * We have lost all feasible routes. We have to broadcast seqno request - * (Section 3.8.2.1) and keep unreachable route for a while (section 2.8). - * The later is done automatically by babel_announce_rte(). - */ - - TRACE(D_EVENTS, "Lost feasible route for prefix %N", e->n.addr); - if (e->valid && (e->selected->router_id == e->router_id)) - babel_add_seqno_request(p, e, e->selected->router_id, e->selected->seqno + 1, 0, NULL); - } - else - return; - - e->selected = best; - babel_announce_rte(p, e); }
/* @@ -1414,7 +1353,12 @@ babel_handle_update(union babel_msg *m, struct babel_iface *ifa) /* * RFC 8966 3.8.2.2 - dealing with unfeasible updates. Generate a one-off * (not retransmitted) unicast seqno request to the originator of this update. - * Note: !feasible -> s exists, check for 's' is just for clarity / safety. + * + * Notes: + * - (!feasible) implies (s) is non-NULL, check for (s) is just for clarity + * and safety. + * - (!best) condition is not explicitly recommeded by RFCC8966 text but was + * in the old version, RFC6126. */ if (!feasible && s && (metric != BABEL_INFINITY) && (!best || (r == best) || (metric < best->metric))) @@ -1448,7 +1392,7 @@ babel_handle_update(union babel_msg *m, struct babel_iface *ifa) e->updated = current_time(); }
- babel_select_route(p, e, r); + babel_announce_rte(p, e, r); }
void @@ -2136,7 +2080,7 @@ static void babel_dump_entry(struct dump_request *dreq, struct babel_entry *e) { struct babel_source *s; - struct babel_route *r; + struct babel_route *r, *best = e->selected;
RDUMP("Babel: Entry %N:\n", e->n.addr);
@@ -2146,7 +2090,7 @@ babel_dump_entry(struct dump_request *dreq, struct babel_entry *e) WALK_LIST(r,e->routes) { RDUMP(" "); - if (r == e->selected) RDUMP("*"); + if (r == best) RDUMP("*"); babel_dump_route(dreq, r); } } @@ -2324,7 +2268,7 @@ babel_show_entries_(struct babel_proto *p, struct fib *rtable)
FIB_WALK(rtable, struct babel_entry, e) { - struct babel_route *r = NULL; + struct babel_route *r = NULL, *best = e->selected; uint rts = 0, srcs = 0; node *n;
@@ -2337,7 +2281,7 @@ babel_show_entries_(struct babel_proto *p, struct fib *rtable) if (e->valid) cli_msg(-1025, "%-*N %-23lR %6u %5u %7u %7u", width, e->n.addr, e->router_id, e->metric, e->seqno, rts, srcs); - else if (r = e->selected) + else if (r = best) cli_msg(-1025, "%-*N %-23lR %6u %5u %7u %7u", width, e->n.addr, r->router_id, r->metric, r->seqno, rts, srcs); else @@ -2374,10 +2318,10 @@ babel_show_routes_(struct babel_proto *p, struct fib *rtable)
FIB_WALK(rtable, struct babel_entry, e) { - struct babel_route *r; + struct babel_route *r, *best = e->selected; WALK_LIST(r, e->routes) { - char c = (r == e->selected) ? '*' : (r->feasible ? '+' : ' '); + char c = (r == best) ? '*' : (r->feasible ? '+' : ' '); btime time = r->expires ? r->expires - current_time() : 0; cli_msg(-1025, "%-*N %-25I %-10s %5u %c %5u %7t", width, e->n.addr, r->next_hop, r->neigh->ifa->ifname, @@ -2466,8 +2410,9 @@ babel_entry_invalidate(struct babel_entry *e)
/* - * babel_rt_notify - core tells us about new route (possibly our own), - * so store it into our data structures. + * babel_rt_notify - BIRD core notifis us of changes in the selected optimal + * route. We can either get one of the routes back we exported or one from a + * different protocol. */ static void babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, @@ -2475,6 +2420,8 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, { struct babel_proto *p = (void *) P; struct babel_entry *e; + struct babel_iface *ifa; + struct babel_neighbor *n;
/* Ignore everything on shutdown */ if (SHUTTING_DOWN) @@ -2483,11 +2430,13 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, if (new) { /* Update */ + uint internal = new->src->owner == &P->sources; uint rt_seqno; uint rt_metric = ea_get_int(new->attrs, &ea_babel_metric, 0); u64 rt_router_id = 0; + struct babel_route *best;
- if (new->src->owner == &P->sources) + if (internal) { rt_seqno = ea_get_int(new->attrs, &ea_babel_seqno, 0); eattr *e = ea_find(new->attrs, &ea_babel_router_id); @@ -2496,6 +2445,8 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, } else { + /* If the selected route isn't one of ours we inject a new route into the + * babel domain as per RFC8966 section 3.7 */ rt_seqno = p->update_seqno; rt_router_id = p->router_id; } @@ -2509,6 +2460,29 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net,
e = babel_get_entry(p, net);
+ best = e->selected; + if (internal) { + eattr *ea = ea_find(new->attrs, &ea_gen_nexthop); + struct nexthop_adata *nhad = (void *) ea->u.ptr; + ifa = babel_find_iface(p, nhad->nh.iface); + n = ifa ? babel_find_neighbor(ifa, ea_get_ip(new->attrs, &ea_gen_from, IPA_NONE)) : NULL; + best = n ? babel_get_route(p, e, n) : NULL; + ASSERT(best); /* Note: We think this can't happen but the lifetimes of the + * involved objects are complicated so we're not yet + * completely convinced. */ + } + else + { + best = NULL; + } + + if (best && best != e->selected) + TRACE(D_EVENTS, + "Picked new route for prefix %N: router-id %lR metric %d", + e->n.addr, best->router_id, best->metric); + + e->selected = best; + /* Activate triggered updates */ if (!(e->valid & BABEL_ENTRY_VALID) || (e->router_id != rt_router_id)) @@ -2530,6 +2504,27 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, if (!e || e->valid != BABEL_ENTRY_VALID) return;
+ if (e->selected) { + TRACE(D_EVENTS, "Lost all feasible routes for prefix %N", e->n.addr); + /* We had a feasible route, but now it's gone. We must send a seqno + * request (Section 3.8.2.1). As allowed by RFC8966 we choose to always do + * so even if we don't have any more unfeasible routes and we use the + * simple unconditional multicast strategy. + */ + if (e->valid && (e->selected->router_id == e->router_id)) + babel_add_seqno_request( + p, e, e->selected->router_id, e->selected->seqno + 1, 0, NULL); + } + + + e->valid = BABEL_ENTRY_STALE; + e->metric = BABEL_INFINITY; + e->selected = NULL; + + /* Install an unreachable route for prefix hold time as per RFC8966 section + * 3.5.4. */ + babel_rte_update_unreachable(p, e, 1); + babel_entry_invalidate(e); babel_trigger_update(p); } @@ -2626,6 +2621,7 @@ babel_start(struct proto *P) p->source_slab = sl_new(P->pool, sizeof(struct babel_source)); p->msg_slab = sl_new(P->pool, sizeof(struct babel_msg_node)); p->seqno_slab = sl_new(P->pool, sizeof(struct babel_seqno_request)); + idm_init(&p->src_ids, P->pool, 8);
p->log_pkt_tbf = (struct tbf){ .rate = 1, .burst = 5 };
diff --git a/proto/babel/babel.h b/proto/babel/babel.h index 146179cda..0699a2974 100644 --- a/proto/babel/babel.h +++ b/proto/babel/babel.h @@ -25,6 +25,7 @@ #include "lib/socket.h" #include "lib/string.h" #include "lib/timer.h" +#include "lib/idm.h"
#define BABEL_MAGIC 42 #define BABEL_VERSION 2 @@ -193,6 +194,7 @@ struct babel_proto { slab *source_slab; slab *msg_slab; slab *seqno_slab; + struct idm src_ids;
struct tbf log_pkt_tbf; /* TBF for packet messages */ }; @@ -235,6 +237,7 @@ struct babel_iface { struct babel_neighbor { node n; struct babel_iface *ifa; + struct rte_src *src;
ip_addr addr; u16 rxcost; /* Sent in last IHU */
--Daniel