diff options
Diffstat (limited to 'proto/bgp')
| -rw-r--r-- | proto/bgp/attrs.c | 1121 | ||||
| -rw-r--r-- | proto/bgp/bgp.c | 541 | ||||
| -rw-r--r-- | proto/bgp/bgp.h | 187 | ||||
| -rw-r--r-- | proto/bgp/config.Y | 50 | ||||
| -rw-r--r-- | proto/bgp/packets.c | 340 |
5 files changed, 1381 insertions, 858 deletions
diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c index 1e234b16..039fdff5 100644 --- a/proto/bgp/attrs.c +++ b/proto/bgp/attrs.c @@ -15,12 +15,13 @@ #include "nest/bird.h" #include "nest/iface.h" #include "nest/protocol.h" -#include "nest/route.h" -#include "nest/attrs.h" +#include "nest/rt.h" +#include "lib/attrs.h" #include "conf/conf.h" #include "lib/resource.h" #include "lib/string.h" #include "lib/unaligned.h" +#include "lib/macro.h" #include "bgp.h" @@ -64,37 +65,72 @@ * format - Optional hook that converts eattr to textual representation. */ - -struct bgp_attr_desc { - const char *name; - uint type; - uint flags; - void (*export)(struct bgp_export_state *s, eattr *a); - int (*encode)(struct bgp_write_state *s, eattr *a, byte *buf, uint size); - void (*decode)(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to); - void (*format)(const eattr *ea, byte *buf, uint size); +union bgp_attr_desc { + struct ea_class class; + struct { + EA_CLASS_INSIDE; + uint flags; + void (*export)(struct bgp_export_state *s, eattr *a); + int (*encode)(struct bgp_write_state *s, eattr *a, byte *buf, uint size); + void (*decode)(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to); + }; }; -static const struct bgp_attr_desc bgp_attr_table[]; +static union bgp_attr_desc bgp_attr_table[]; +static inline const union bgp_attr_desc *bgp_find_attr_desc(eattr *a) +{ + const struct ea_class *class = ea_class_find(a->id); -static inline int bgp_attr_known(uint code); + if ((class < &bgp_attr_table[0].class) || (class >= &bgp_attr_table[BGP_ATTR_MAX].class)) + return NULL; -eattr * -bgp_set_attr(ea_list **attrs, struct linpool *pool, uint code, uint flags, uintptr_t val) + return (const union bgp_attr_desc *) class; +} + +#define BGP_EA_ID(code) (bgp_attr_table[code].id) +#define EA_BGP_ID(code) (((union bgp_attr_desc *) ea_class_find(code)) - bgp_attr_table) + +void bgp_set_attr_u32(ea_list **to, uint code, uint flags, u32 val) +{ + const union bgp_attr_desc *desc = &bgp_attr_table[code]; + + ea_set_attr(to, EA_LITERAL_EMBEDDED( + &desc->class, + flags & ~BAF_EXT_LEN, + val + )); +} + +void bgp_set_attr_ptr(ea_list **to, uint code, uint flags, const struct adata *ad) { - ASSERT(bgp_attr_known(code)); + const union bgp_attr_desc *desc = &bgp_attr_table[code]; - return ea_set_attr( - attrs, - pool, - EA_CODE(PROTOCOL_BGP, code), - flags & ~BAF_EXT_LEN, - bgp_attr_table[code].type, - val - ); + ea_set_attr(to, EA_LITERAL_DIRECT_ADATA( + &desc->class, + flags & ~BAF_EXT_LEN, + ad + )); } +void +bgp_set_attr_data(ea_list **to, uint code, uint flags, void *data, uint len) +{ + const union bgp_attr_desc *desc = &bgp_attr_table[code]; + + ea_set_attr(to, EA_LITERAL_STORE_ADATA( + &desc->class, + flags & ~BAF_EXT_LEN, + data, + len + )); +} +void +bgp_unset_attr(ea_list **to, uint code) +{ + const union bgp_attr_desc *desc = &bgp_attr_table[code]; + ea_unset_attr(to, 0, &desc->class); +} #define REPORT(msg, args...) \ ({ log(L_REMOTE "%s: " msg, s->proto->p.name, ## args); }) @@ -151,7 +187,7 @@ bgp_encode_u8(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size) if (size < (3+1)) return -1; - bgp_put_attr_hdr3(buf, EA_ID(a->id), a->flags, 1); + bgp_put_attr_hdr3(buf, EA_BGP_ID(a->id), a->flags, 1); buf[3] = a->u.data; return 3+1; @@ -163,7 +199,7 @@ bgp_encode_u32(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size) if (size < (3+4)) return -1; - bgp_put_attr_hdr3(buf, EA_ID(a->id), a->flags, 4); + bgp_put_attr_hdr3(buf, EA_BGP_ID(a->id), a->flags, 4); put_u32(buf+3, a->u.data); return 3+4; @@ -177,7 +213,7 @@ bgp_encode_u32s(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size if (size < (4+len)) return -1; - uint hdr = bgp_put_attr_hdr(buf, EA_ID(a->id), a->flags, len); + uint hdr = bgp_put_attr_hdr(buf, EA_BGP_ID(a->id), a->flags, len); put_u32s(buf + hdr, (u32 *) a->u.ptr->data, len / 4); return hdr + len; @@ -198,7 +234,7 @@ bgp_put_attr(byte *buf, uint size, uint code, uint flags, const byte *data, uint static int bgp_encode_raw(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size) { - return bgp_put_attr(buf, size, EA_ID(a->id), a->flags, a->u.ptr->data, a->u.ptr->length); + return bgp_put_attr(buf, size, EA_BGP_ID(a->id), a->flags, a->u.ptr->data, a->u.ptr->length); } @@ -336,9 +372,9 @@ bgp_aigp_set_metric(struct linpool *pool, const struct adata *ad, u64 metric) } int -bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad) +bgp_total_aigp_metric_(const rte *e, u64 *metric, const struct adata **ad) { - eattr *a = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AIGP)); + eattr *a = ea_find(e->attrs, BGP_EA_ID(BA_AIGP)); if (!a) return 0; @@ -347,7 +383,7 @@ bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad) return 0; u64 aigp = get_u64(b + 3); - u64 step = e->attrs->igp_metric; + u64 step = rt_get_igp_metric(e); if (!rte_resolvable(e) || (step >= IGP_METRIC_UNKNOWN)) step = BGP_AIGP_MAX; @@ -366,7 +402,7 @@ bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad) static inline int bgp_init_aigp_metric(rte *e, u64 *metric, const struct adata **ad) { - if (e->attrs->source == RTS_BGP) + if (rt_get_source_attr(e) == RTS_BGP) return 0; *metric = rt_get_igp_metric(e); @@ -375,7 +411,7 @@ bgp_init_aigp_metric(rte *e, u64 *metric, const struct adata **ad) } u32 -bgp_rte_igp_metric(struct rte *rt) +bgp_rte_igp_metric(const rte *rt) { u64 metric = bgp_total_aigp_metric(rt); return (u32) MIN(metric, (u64) IGP_METRIC_UNKNOWN); @@ -402,7 +438,7 @@ bgp_decode_origin(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte if (data[0] > 2) WITHDRAW(BAD_VALUE, "ORIGIN", data[0]); - bgp_set_attr_u32(to, s->pool, BA_ORIGIN, flags, data[0]); + bgp_set_attr_u32(to, BA_ORIGIN, flags, data[0]); } static void @@ -470,7 +506,7 @@ bgp_decode_as_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte !bgp_as_path_first_as_equal(data, len, p->remote_as)) WITHDRAW("Malformed AS_PATH attribute - %s", "First AS differs from neigbor AS"); - bgp_set_attr_data(to, s->pool, BA_AS_PATH, flags, data, len); + bgp_set_attr_data(to, BA_AS_PATH, flags, data, len); } @@ -542,7 +578,7 @@ bgp_decode_med(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *da WITHDRAW(BAD_LENGTH, "MULTI_EXIT_DISC", len); u32 val = get_u32(data); - bgp_set_attr_u32(to, s->pool, BA_MULTI_EXIT_DISC, flags, val); + bgp_set_attr_u32(to, BA_MULTI_EXIT_DISC, flags, val); } @@ -563,7 +599,7 @@ bgp_decode_local_pref(struct bgp_parse_state *s, uint code UNUSED, uint flags, b WITHDRAW(BAD_LENGTH, "LOCAL_PREF", len); u32 val = get_u32(data); - bgp_set_attr_u32(to, s->pool, BA_LOCAL_PREF, flags, val); + bgp_set_attr_u32(to, BA_LOCAL_PREF, flags, val); } @@ -573,7 +609,7 @@ bgp_decode_atomic_aggr(struct bgp_parse_state *s, uint code UNUSED, uint flags, if (len != 0) DISCARD(BAD_LENGTH, "ATOMIC_AGGR", len); - bgp_set_attr_data(to, s->pool, BA_ATOMIC_AGGR, flags, NULL, 0); + bgp_set_attr_data(to, BA_ATOMIC_AGGR, flags, NULL, 0); } static int @@ -607,7 +643,7 @@ bgp_decode_aggregator(struct bgp_parse_state *s, uint code UNUSED, uint flags, b len = aggregator_16to32(data, src); } - bgp_set_attr_data(to, s->pool, BA_AGGREGATOR, flags, data, len); + bgp_set_attr_data(to, BA_AGGREGATOR, flags, data, len); } static void @@ -636,7 +672,7 @@ bgp_decode_community(struct bgp_parse_state *s, uint code UNUSED, uint flags, by struct adata *ad = lp_alloc_adata(s->pool, len); get_u32s(data, (u32 *) ad->data, len / 4); - bgp_set_attr_ptr(to, s->pool, BA_COMMUNITY, flags, ad); + bgp_set_attr_ptr(to, BA_COMMUNITY, flags, ad); } @@ -657,7 +693,7 @@ bgp_decode_originator_id(struct bgp_parse_state *s, uint code UNUSED, uint flags WITHDRAW(BAD_LENGTH, "ORIGINATOR_ID", len); u32 val = get_u32(data); - bgp_set_attr_u32(to, s->pool, BA_ORIGINATOR_ID, flags, val); + bgp_set_attr_u32(to, BA_ORIGINATOR_ID, flags, val); } @@ -682,7 +718,7 @@ bgp_decode_cluster_list(struct bgp_parse_state *s, uint code UNUSED, uint flags, struct adata *ad = lp_alloc_adata(s->pool, len); get_u32s(data, (u32 *) ad->data, len / 4); - bgp_set_attr_ptr(to, s->pool, BA_CLUSTER_LIST, flags, ad); + bgp_set_attr_ptr(to, BA_CLUSTER_LIST, flags, ad); } static void @@ -801,7 +837,7 @@ bgp_decode_ext_community(struct bgp_parse_state *s, uint code UNUSED, uint flags struct adata *ad = lp_alloc_adata(s->pool, len); get_u32s(data, (u32 *) ad->data, len / 4); - bgp_set_attr_ptr(to, s->pool, BA_EXT_COMMUNITY, flags, ad); + bgp_set_attr_ptr(to, BA_EXT_COMMUNITY, flags, ad); } @@ -814,7 +850,7 @@ bgp_decode_as4_aggregator(struct bgp_parse_state *s, uint code UNUSED, uint flag if (len != 8) DISCARD(BAD_LENGTH, "AS4_AGGREGATOR", len); - bgp_set_attr_data(to, s->pool, BA_AS4_AGGREGATOR, flags, data, len); + bgp_set_attr_data(to, BA_AS4_AGGREGATOR, flags, data, len); } static void @@ -844,7 +880,7 @@ bgp_decode_as4_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byt a = as_path_strip_confed(s->pool, a); } - bgp_set_attr_ptr(to, s->pool, BA_AS4_PATH, flags, a); + bgp_set_attr_ptr(to, BA_AS4_PATH, flags, a); } @@ -868,7 +904,7 @@ bgp_decode_aigp(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *d if (!bgp_aigp_valid(data, len, err, sizeof(err))) DISCARD("Malformed AIGP attribute - %s", err); - bgp_set_attr_data(to, s->pool, BA_AIGP, flags, data, len); + bgp_set_attr_data(to, BA_AIGP, flags, data, len); } static void @@ -900,7 +936,7 @@ bgp_decode_large_community(struct bgp_parse_state *s, uint code UNUSED, uint fla struct adata *ad = lp_alloc_adata(s->pool, len); get_u32s(data, (u32 *) ad->data, len / 4); - bgp_set_attr_ptr(to, s->pool, BA_LARGE_COMMUNITY, flags, ad); + bgp_set_attr_ptr(to, BA_LARGE_COMMUNITY, flags, ad); } @@ -911,14 +947,14 @@ bgp_decode_otc(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *da WITHDRAW(BAD_LENGTH, "OTC", len); u32 val = get_u32(data); - bgp_set_attr_u32(to, s->pool, BA_ONLY_TO_CUSTOMER, flags, val); + bgp_set_attr_u32(to, BA_ONLY_TO_CUSTOMER, flags, val); } static void bgp_export_mpls_label_stack(struct bgp_export_state *s, eattr *a) { - net_addr *n = s->route->net->n.addr; + const net_addr *n = s->route->net; u32 *labels = (u32 *) a->u.ptr->data; uint lnum = a->u.ptr->length / 4; @@ -985,10 +1021,29 @@ bgp_format_mpls_label_stack(const eattr *a, byte *buf, uint size) } static inline void -bgp_decode_unknown(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to) +bgp_export_unknown(struct bgp_export_state *s UNUSED, eattr *a) +{ + if (!(a->flags & BAF_TRANSITIVE)) + UNSET(a); + + a->flags |= BAF_PARTIAL; +} + +static inline void +bgp_decode_unknown(struct bgp_parse_state *s UNUSED, uint code, uint flags, byte *data, uint len, ea_list **to) { + if (!(flags & BAF_OPTIONAL)) + WITHDRAW("Unknown attribute (code %u) - conflicting flags (%02x)", code, flags); + /* Cannot use bgp_set_attr_data() as it works on known attributes only */ - ea_set_attr_data(to, s->pool, EA_CODE(PROTOCOL_BGP, code), flags, EAF_TYPE_OPAQUE, data, len); + ea_set_attr_data(to, &bgp_attr_table[code].class, flags, data, len); +} + +static inline void +bgp_format_unknown(const eattr *a, byte *buf, uint size) +{ + if (a->flags & BAF_TRANSITIVE) + bsnprintf(buf, size, "(transitive)"); } @@ -996,10 +1051,10 @@ bgp_decode_unknown(struct bgp_parse_state *s, uint code, uint flags, byte *data, * Attribute table */ -static const struct bgp_attr_desc bgp_attr_table[] = { +static union bgp_attr_desc bgp_attr_table[BGP_ATTR_MAX] = { [BA_ORIGIN] = { - .name = "origin", - .type = EAF_TYPE_INT, + .name = "bgp_origin", + .type = T_ENUM_BGP_ORIGIN, .flags = BAF_TRANSITIVE, .export = bgp_export_origin, .encode = bgp_encode_u8, @@ -1007,69 +1062,69 @@ static const struct bgp_attr_desc bgp_attr_table[] = { .format = bgp_format_origin, }, [BA_AS_PATH] = { - .name = "as_path", - .type = EAF_TYPE_AS_PATH, + .name = "bgp_path", + .type = T_PATH, .flags = BAF_TRANSITIVE, .encode = bgp_encode_as_path, .decode = bgp_decode_as_path, }, [BA_NEXT_HOP] = { - .name = "next_hop", - .type = EAF_TYPE_IP_ADDRESS, + .name = "bgp_next_hop", + .type = T_IP, .flags = BAF_TRANSITIVE, .encode = bgp_encode_next_hop, .decode = bgp_decode_next_hop, .format = bgp_format_next_hop, }, [BA_MULTI_EXIT_DISC] = { - .name = "med", - .type = EAF_TYPE_INT, + .name = "bgp_med", + .type = T_INT, .flags = BAF_OPTIONAL, .encode = bgp_encode_u32, .decode = bgp_decode_med, }, [BA_LOCAL_PREF] = { - .name = "local_pref", - .type = EAF_TYPE_INT, + .name = "bgp_local_pref", + .type = T_INT, .flags = BAF_TRANSITIVE, .export = bgp_export_local_pref, .encode = bgp_encode_u32, .decode = bgp_decode_local_pref, }, [BA_ATOMIC_AGGR] = { - .name = "atomic_aggr", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_atomic_aggr", + .type = T_OPAQUE, .flags = BAF_TRANSITIVE, .encode = bgp_encode_raw, .decode = bgp_decode_atomic_aggr, }, [BA_AGGREGATOR] = { - .name = "aggregator", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_aggregator", + .type = T_OPAQUE, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .encode = bgp_encode_aggregator, .decode = bgp_decode_aggregator, .format = bgp_format_aggregator, }, [BA_COMMUNITY] = { - .name = "community", - .type = EAF_TYPE_INT_SET, + .name = "bgp_community", + .type = T_CLIST, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .export = bgp_export_community, .encode = bgp_encode_u32s, .decode = bgp_decode_community, }, [BA_ORIGINATOR_ID] = { - .name = "originator_id", - .type = EAF_TYPE_ROUTER_ID, + .name = "bgp_originator_id", + .type = T_QUAD, .flags = BAF_OPTIONAL, .export = bgp_export_originator_id, .encode = bgp_encode_u32, .decode = bgp_decode_originator_id, }, [BA_CLUSTER_LIST] = { - .name = "cluster_list", - .type = EAF_TYPE_INT_SET, + .name = "bgp_cluster_list", + .type = T_CLIST, .flags = BAF_OPTIONAL, .export = bgp_export_cluster_list, .encode = bgp_encode_u32s, @@ -1077,43 +1132,47 @@ static const struct bgp_attr_desc bgp_attr_table[] = { .format = bgp_format_cluster_list, }, [BA_MP_REACH_NLRI] = { - .name = "mp_reach_nlri", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_mp_reach_nlri", + .type = T_OPAQUE, + .hidden = 1, .flags = BAF_OPTIONAL, .decode = bgp_decode_mp_reach_nlri, }, [BA_MP_UNREACH_NLRI] = { - .name = "mp_unreach_nlri", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_mp_unreach_nlri", + .type = T_OPAQUE, + .hidden = 1, .flags = BAF_OPTIONAL, .decode = bgp_decode_mp_unreach_nlri, }, [BA_EXT_COMMUNITY] = { - .name = "ext_community", - .type = EAF_TYPE_EC_SET, + .name = "bgp_ext_community", + .type = T_ECLIST, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .export = bgp_export_ext_community, .encode = bgp_encode_u32s, .decode = bgp_decode_ext_community, }, [BA_AS4_PATH] = { - .name = "as4_path", - .type = EAF_TYPE_AS_PATH, + .name = "bgp_as4_path", + .type = T_PATH, + .hidden = 1, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .encode = bgp_encode_raw, .decode = bgp_decode_as4_path, }, [BA_AS4_AGGREGATOR] = { - .name = "as4_aggregator", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_as4_aggregator", + .type = T_OPAQUE, + .hidden = 1, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .encode = bgp_encode_raw, .decode = bgp_decode_as4_aggregator, .format = bgp_format_aggregator, }, [BA_AIGP] = { - .name = "aigp", - .type = EAF_TYPE_OPAQUE, + .name = "bgp_aigp", + .type = T_OPAQUE, .flags = BAF_OPTIONAL | BAF_DECODE_FLAGS, .export = bgp_export_aigp, .encode = bgp_encode_raw, @@ -1121,8 +1180,8 @@ static const struct bgp_attr_desc bgp_attr_table[] = { .format = bgp_format_aigp, }, [BA_LARGE_COMMUNITY] = { - .name = "large_community", - .type = EAF_TYPE_LC_SET, + .name = "bgp_large_community", + .type = T_LCLIST, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .export = bgp_export_large_community, .encode = bgp_encode_u32s, @@ -1130,14 +1189,15 @@ static const struct bgp_attr_desc bgp_attr_table[] = { }, [BA_ONLY_TO_CUSTOMER] = { .name = "otc", - .type = EAF_TYPE_INT, + .type = T_INT, .flags = BAF_OPTIONAL | BAF_TRANSITIVE, .encode = bgp_encode_u32, .decode = bgp_decode_otc, }, [BA_MPLS_LABEL_STACK] = { - .name = "mpls_label_stack", - .type = EAF_TYPE_INT_SET, + .name = "bgp_mpls_label_stack", + .type = T_CLIST, + .readonly = 1, .export = bgp_export_mpls_label_stack, .encode = bgp_encode_mpls_label_stack, .decode = bgp_decode_mpls_label_stack, @@ -1145,12 +1205,32 @@ static const struct bgp_attr_desc bgp_attr_table[] = { }, }; -static inline int -bgp_attr_known(uint code) +eattr * +bgp_find_attr(ea_list *attrs, uint code) { - return (code < ARRAY_SIZE(bgp_attr_table)) && bgp_attr_table[code].name; + return ea_find(attrs, BGP_EA_ID(code)); } +void +bgp_register_attrs(void) +{ + for (uint i=0; i<ARRAY_SIZE(bgp_attr_table); i++) + { + if (!bgp_attr_table[i].name) + bgp_attr_table[i] = (union bgp_attr_desc) { + .name = mb_sprintf(&root_pool, "bgp_unknown_0x%02x", i), + .type = T_OPAQUE, + .flags = BAF_OPTIONAL, + .readonly = 1, + .export = bgp_export_unknown, + .encode = bgp_encode_raw, + .decode = bgp_decode_unknown, + .format = bgp_format_unknown, + }; + + ea_register_init(&bgp_attr_table[i].class); + } +} /* * Attribute export @@ -1159,38 +1239,24 @@ bgp_attr_known(uint code) static inline void bgp_export_attr(struct bgp_export_state *s, eattr *a, ea_list *to) { - if (EA_PROTO(a->id) != PROTOCOL_BGP) + const union bgp_attr_desc *desc = bgp_find_attr_desc(a); + if (!desc) return; - uint code = EA_ID(a->id); - - if (bgp_attr_known(code)) - { - const struct bgp_attr_desc *desc = &bgp_attr_table[code]; - - /* The flags might have been zero if the attr was added by filters */ - a->flags = (a->flags & BAF_PARTIAL) | desc->flags; - - /* Set partial bit if new opt-trans attribute is attached to non-local route */ - if ((s->src != NULL) && (a->originated) && - (a->flags & BAF_OPTIONAL) && (a->flags & BAF_TRANSITIVE)) - a->flags |= BAF_PARTIAL; + /* The flags might have been zero if the attr was added locally */ + a->flags = (a->flags & BAF_PARTIAL) | desc->flags; - /* Call specific hook */ - CALL(desc->export, s, a); + /* Set partial bit if new opt-trans attribute is attached to non-local route */ + if ((s->src != NULL) && (a->originated) && + (a->flags & BAF_OPTIONAL) && (a->flags & BAF_TRANSITIVE)) + a->flags |= BAF_PARTIAL; - /* Attribute might become undefined in hook */ - if (a->undef) - return; - } - else - { - /* Don't re-export unknown non-transitive attributes */ - if (!(a->flags & BAF_TRANSITIVE)) - return; + /* Call specific hook */ + CALL(desc->export, s, a); - a->flags |= BAF_PARTIAL; - } + /* Attribute might become undefined in hook */ + if (a->undef) + return; /* Append updated attribute */ to->attrs[to->count++] = *a; @@ -1210,12 +1276,11 @@ bgp_export_attr(struct bgp_export_state *s, eattr *a, ea_list *to) * Result: one sorted attribute list segment, or NULL if attributes are unsuitable. */ static inline ea_list * -bgp_export_attrs(struct bgp_export_state *s, ea_list *attrs) +bgp_export_attrs(struct bgp_export_state *s, ea_list *a) { /* Merge the attribute list */ - ea_list *new = lp_alloc(s->pool, ea_scan(attrs)); - ea_merge(attrs, new); - ea_sort(new); + ea_list *new = ea_normalize(a, 0); + ASSERT_DIE(new); uint i, count; count = new->count; @@ -1239,14 +1304,9 @@ bgp_export_attrs(struct bgp_export_state *s, ea_list *attrs) static inline int bgp_encode_attr(struct bgp_write_state *s, eattr *a, byte *buf, uint size) { - ASSERT(EA_PROTO(a->id) == PROTOCOL_BGP); - - uint code = EA_ID(a->id); - - if (bgp_attr_known(code)) - return bgp_attr_table[code].encode(s, a, buf, size); - else - return bgp_encode_raw(s, a, buf, size); + const union bgp_attr_desc *desc = bgp_find_attr_desc(a); + ASSERT_DIE(desc); + return desc->encode(s, a, buf, size); } /** @@ -1311,7 +1371,7 @@ bgp_cluster_list_loopy(struct bgp_proto *p, ea_list *attrs) } static inline void -bgp_decode_attr(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to) +bgp_decode_attr(struct bgp_parse_state *s, byte code, byte flags, byte *data, uint len, ea_list **to) { /* Handle duplicate attributes; RFC 7606 3 (g) */ if (BIT32_TEST(s->attrs_seen, code)) @@ -1323,24 +1383,15 @@ bgp_decode_attr(struct bgp_parse_state *s, uint code, uint flags, byte *data, ui } BIT32_SET(s->attrs_seen, code); - if (bgp_attr_known(code)) - { - const struct bgp_attr_desc *desc = &bgp_attr_table[code]; - - /* Handle conflicting flags; RFC 7606 3 (c) */ - if (((flags ^ desc->flags) & (BAF_OPTIONAL | BAF_TRANSITIVE)) && - !(desc->flags & BAF_DECODE_FLAGS)) - WITHDRAW("Malformed %s attribute - conflicting flags (%02x)", desc->name, flags); + ASSERT_DIE(bgp_attr_table[code].id); + const union bgp_attr_desc *desc = &bgp_attr_table[code]; - desc->decode(s, code, flags, data, len, to); - } - else /* Unknown attribute */ - { - if (!(flags & BAF_OPTIONAL)) - WITHDRAW("Unknown attribute (code %u) - conflicting flags (%02x)", code, flags); + /* Handle conflicting flags; RFC 7606 3 (c) */ + if (((flags ^ desc->flags) & (BAF_OPTIONAL | BAF_TRANSITIVE)) && + !(desc->flags & BAF_DECODE_FLAGS)) + WITHDRAW("Malformed %s attribute - conflicting flags (%02x, expected %02x)", desc->name, flags, desc->flags); - bgp_decode_unknown(s, code, flags, data, len, to); - } + desc->decode(s, code, flags, data, len, to); } /** @@ -1358,7 +1409,8 @@ bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len) { struct bgp_proto *p = s->proto; ea_list *attrs = NULL; - uint code, flags, alen; + uint alen; + byte code, flags; byte *pos = data; /* Parse the attributes */ @@ -1439,7 +1491,7 @@ bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len) /* If there is no local preference, define one */ if (!BIT32_TEST(s->attrs_seen, BA_LOCAL_PREF)) - bgp_set_attr_u32(&attrs, s->pool, BA_LOCAL_PREF, 0, p->cf->default_local_pref); + bgp_set_attr_u32(&attrs, BA_LOCAL_PREF, 0, p->cf->default_local_pref); return attrs; @@ -1463,20 +1515,20 @@ loop: } void -bgp_finish_attrs(struct bgp_parse_state *s, rta *a) +bgp_finish_attrs(struct bgp_parse_state *s, ea_list **to) { /* AIGP test here instead of in bgp_decode_aigp() - we need to know channel */ if (BIT32_TEST(s->attrs_seen, BA_AIGP) && !s->channel->cf->aigp) { REPORT("Discarding AIGP attribute received on non-AIGP session"); - bgp_unset_attr(&a->eattrs, s->pool, BA_AIGP); + bgp_unset_attr(to, BA_AIGP); } /* Handle OTC ingress procedure, RFC 9234 */ if (bgp_channel_is_role_applicable(s->channel)) { struct bgp_proto *p = s->proto; - eattr *e = bgp_find_attr(a->eattrs, BA_ONLY_TO_CUSTOMER); + eattr *e = bgp_find_attr(*to, BA_ONLY_TO_CUSTOMER); /* Reject routes from downstream if they are leaked */ if (e && (p->cf->local_role == BGP_ROLE_PROVIDER || @@ -1492,7 +1544,7 @@ bgp_finish_attrs(struct bgp_parse_state *s, rta *a) if (!e && (p->cf->local_role == BGP_ROLE_CUSTOMER || p->cf->local_role == BGP_ROLE_PEER || p->cf->local_role == BGP_ROLE_RS_CLIENT)) - bgp_set_attr_u32(&a->eattrs, s->pool, BA_ONLY_TO_CUSTOMER, 0, p->cf->remote_as); + bgp_set_attr_u32(to, BA_ONLY_TO_CUSTOMER, 0, p->cf->remote_as); } } @@ -1512,8 +1564,8 @@ bgp_finish_attrs(struct bgp_parse_state *s, rta *a) HASH_DEFINE_REHASH_FN(RBH, struct bgp_bucket) -void -bgp_init_bucket_table(struct bgp_channel *c) +static void +bgp_init_bucket_table(struct bgp_pending_tx *c) { HASH_INIT(c->bucket_hash, c->pool, 8); @@ -1521,24 +1573,8 @@ bgp_init_bucket_table(struct bgp_channel *c) c->withdraw_bucket = NULL; } -void -bgp_free_bucket_table(struct bgp_channel *c) -{ - HASH_FREE(c->bucket_hash); - - struct bgp_bucket *b; - WALK_LIST_FIRST(b, c->bucket_queue) - { - rem_node(&b->send_node); - mb_free(b); - } - - mb_free(c->withdraw_bucket); - c->withdraw_bucket = NULL; -} - static struct bgp_bucket * -bgp_get_bucket(struct bgp_channel *c, ea_list *new) +bgp_get_bucket(struct bgp_pending_tx *c, ea_list *new) { /* Hash and lookup */ u32 hash = ea_hash(new); @@ -1547,55 +1583,27 @@ bgp_get_bucket(struct bgp_channel *c, ea_list *new) if (b) return b; - uint ea_size = sizeof(ea_list) + new->count * sizeof(eattr); - uint ea_size_aligned = BIRD_ALIGN(ea_size, CPU_STRUCT_ALIGN); - uint size = sizeof(struct bgp_bucket) + ea_size_aligned; - uint i; - byte *dest; - - /* Gather total size of non-inline attributes */ - for (i = 0; i < new->count; i++) - { - eattr *a = &new->attrs[i]; - - if (!(a->type & EAF_EMBEDDED)) - size += BIRD_ALIGN(sizeof(struct adata) + a->u.ptr->length, CPU_STRUCT_ALIGN); - } + /* Scan the list for total size */ + uint ea_size = BIRD_CPU_ALIGN(ea_list_size(new)); + uint size = sizeof(struct bgp_bucket) + ea_size; - /* Create the bucket */ + /* Allocate the bucket */ b = mb_alloc(c->pool, size); *b = (struct bgp_bucket) { }; init_list(&b->prefixes); b->hash = hash; - /* Copy list of extended attributes */ - memcpy(b->eattrs, new, ea_size); - dest = ((byte *) b->eattrs) + ea_size_aligned; - - /* Copy values of non-inline attributes */ - for (i = 0; i < new->count; i++) - { - eattr *a = &b->eattrs->attrs[i]; - - if (!(a->type & EAF_EMBEDDED)) - { - const struct adata *oa = a->u.ptr; - struct adata *na = (struct adata *) dest; - memcpy(na, oa, sizeof(struct adata) + oa->length); - a->u.ptr = na; - dest += BIRD_ALIGN(sizeof(struct adata) + na->length, CPU_STRUCT_ALIGN); - } - } + /* Copy the ea_list */ + ea_list_copy(b->eattrs, new, ea_size); - /* Insert the bucket to send queue and bucket hash */ - add_tail(&c->bucket_queue, &b->send_node); + /* Insert the bucket to bucket hash */ HASH_INSERT2(c->bucket_hash, RBH, c->pool, b); return b; } static struct bgp_bucket * -bgp_get_withdraw_bucket(struct bgp_channel *c) +bgp_get_withdraw_bucket(struct bgp_pending_tx *c) { if (!c->withdraw_bucket) { @@ -1606,25 +1614,45 @@ bgp_get_withdraw_bucket(struct bgp_channel *c) return c->withdraw_bucket; } -void -bgp_free_bucket(struct bgp_channel *c, struct bgp_bucket *b) +static void +bgp_free_bucket(struct bgp_pending_tx *c, struct bgp_bucket *b) { - rem_node(&b->send_node); HASH_REMOVE2(c->bucket_hash, RBH, c->pool, b); mb_free(b); } +int +bgp_done_bucket(struct bgp_channel *bc, struct bgp_bucket *b) +{ + struct bgp_pending_tx *c = bc->ptx; + + /* Won't free the withdraw bucket */ + if (b == c->withdraw_bucket) + return 0; + + if (EMPTY_LIST(b->prefixes)) + rem_node(&b->send_node); + + if (b->px_uc || !EMPTY_LIST(b->prefixes)) + return 0; + + bgp_free_bucket(c, b); + return 1; +} + void -bgp_defer_bucket(struct bgp_channel *c, struct bgp_bucket *b) +bgp_defer_bucket(struct bgp_channel *bc, struct bgp_bucket *b) { + struct bgp_pending_tx *c = bc->ptx; rem_node(&b->send_node); add_tail(&c->bucket_queue, &b->send_node); } void -bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b) +bgp_withdraw_bucket(struct bgp_channel *bc, struct bgp_bucket *b) { - struct bgp_proto *p = (void *) c->c.proto; + struct bgp_proto *p = (void *) bc->c.proto; + struct bgp_pending_tx *c = bc->ptx; struct bgp_bucket *wb = bgp_get_withdraw_bucket(c); log(L_ERR "%s: Attribute list too long", p->p.name); @@ -1633,8 +1661,8 @@ bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b) struct bgp_prefix *px = HEAD(b->prefixes); log(L_ERR "%s: - withdrawing %N", p->p.name, &px->net); - rem_node(&px->buck_node); - add_tail(&wb->prefixes, &px->buck_node); + rem_node(&px->buck_node_xx); + add_tail(&wb->prefixes, &px->buck_node_xx); } } @@ -1645,7 +1673,7 @@ bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b) #define PXH_KEY(px) px->net, px->path_id, px->hash #define PXH_NEXT(px) px->next -#define PXH_EQ(n1,i1,h1,n2,i2,h2) h1 == h2 && i1 == i2 && net_equal(n1, n2) +#define PXH_EQ(n1,i1,h1,n2,i2,h2) h1 == h2 && (add_path_tx ? (i1 == i2) : 1) && net_equal(n1, n2) #define PXH_FN(n,i,h) h #define PXH_REHASH bgp_pxh_rehash @@ -1654,34 +1682,34 @@ bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b) HASH_DEFINE_REHASH_FN(PXH, struct bgp_prefix) -void -bgp_init_prefix_table(struct bgp_channel *c) +static void +bgp_init_prefix_table(struct bgp_channel *bc) { + struct bgp_pending_tx *c = bc->ptx; HASH_INIT(c->prefix_hash, c->pool, 8); - uint alen = net_addr_length[c->c.net_type]; + uint alen = net_addr_length[bc->c.net_type]; c->prefix_slab = alen ? sl_new(c->pool, sizeof(struct bgp_prefix) + alen) : NULL; } -void -bgp_free_prefix_table(struct bgp_channel *c) -{ - HASH_FREE(c->prefix_hash); - - rfree(c->prefix_slab); - c->prefix_slab = NULL; -} - static struct bgp_prefix * -bgp_get_prefix(struct bgp_channel *c, net_addr *net, u32 path_id) +bgp_get_prefix(struct bgp_pending_tx *c, const net_addr *net, struct rte_src *src, int add_path_tx) { + u32 path_id = src->global_id; + u32 path_id_hash = add_path_tx ? path_id : 0; /* We must use a different hash function than the rtable */ - u32 hash = u32_hash(net_hash(net) ^ u32_hash(path_id)); - struct bgp_prefix *px = HASH_FIND(c->prefix_hash, PXH, net, path_id, hash); + u32 hash = u32_hash(net_hash(net) ^ u32_hash(path_id_hash)); + struct bgp_prefix *px = HASH_FIND(c->prefix_hash, PXH, net, path_id_hash, hash); if (px) { - rem_node(&px->buck_node); + if (!add_path_tx && (path_id != px->path_id)) + { + rt_unlock_source(rt_find_source_global(px->path_id)); + rt_lock_source(src); + px->path_id = path_id; + } + return px; } @@ -1694,24 +1722,306 @@ bgp_get_prefix(struct bgp_channel *c, net_addr *net, u32 path_id) px->hash = hash; px->path_id = path_id; net_copy(px->net, net); + rt_lock_source(src); HASH_INSERT2(c->prefix_hash, PXH, c->pool, px); return px; } -void -bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *px) +static void bgp_free_prefix(struct bgp_pending_tx *c, struct bgp_prefix *px); + +static inline int +bgp_update_prefix(struct bgp_channel *c, struct bgp_prefix *px, struct bgp_bucket *b) +{ +#define IS_WITHDRAW_BUCKET(b) ((b) == c->ptx->withdraw_bucket) +#define BPX_TRACE(what) do { \ + if (c->c.debug & D_ROUTES) log(L_TRACE "%s.%s < %s %N %uG %s", \ + c->c.proto->name, c->c.name, what, \ + px->net, px->path_id, IS_WITHDRAW_BUCKET(b) ? "withdraw" : "update"); } while (0) + px->lastmod = current_time(); + + /* Already queued for the same bucket */ + if (px->cur == b) + { + BPX_TRACE("already queued"); + return 0; + } + + /* Unqueue from the old bucket */ + if (px->cur) + { + rem_node(&px->buck_node_xx); + bgp_done_bucket(c, px->cur); + } + + /* The new bucket is the same as we sent before */ + if ((px->last == b) || c->c.out_table && !px->last && IS_WITHDRAW_BUCKET(b)) + { + if (px->cur) + BPX_TRACE("reverted"); + else + BPX_TRACE("already sent"); + + /* Well, we haven't sent anything yet */ + if (!px->last) + bgp_free_prefix(c->ptx, px); + + px->cur = NULL; + return 0; + } + + /* Enqueue the bucket if it has been empty */ + if (!IS_WITHDRAW_BUCKET(b) && EMPTY_LIST(b->prefixes)) + add_tail(&c->ptx->bucket_queue, &b->send_node); + + /* Enqueue to the new bucket and indicate the change */ + add_tail(&b->prefixes, &px->buck_node_xx); + px->cur = b; + + BPX_TRACE("queued"); + return 1; + +#undef BPX_TRACE +} + +static void +bgp_free_prefix(struct bgp_pending_tx *c, struct bgp_prefix *px) { - rem_node(&px->buck_node); HASH_REMOVE2(c->prefix_hash, PXH, c->pool, px); + rt_unlock_source(rt_find_source_global(px->path_id)); + if (c->prefix_slab) sl_free(px); else mb_free(px); } +void +bgp_done_prefix(struct bgp_channel *c, struct bgp_prefix *px, struct bgp_bucket *buck) +{ + /* Cleanup: We're called from bucket senders. */ + ASSERT_DIE(px->cur == buck); + rem_node(&px->buck_node_xx); + + /* We may want to store the updates */ + if (c->c.out_table) + { + /* Nothing to be sent right now */ + px->cur = NULL; + + /* Unref the previous sent version */ + if (px->last) + px->last->px_uc--; + + /* Ref the current sent version */ + if (!IS_WITHDRAW_BUCKET(buck)) + { + px->last = buck; + px->last->px_uc++; + return; + } + + /* Prefixes belonging to the withdraw bucket are freed always */ + } + + bgp_free_prefix(c->ptx, px); +} + +static void +bgp_pending_tx_rfree(resource *r) +{ + struct bgp_pending_tx *ptx = SKIP_BACK(struct bgp_pending_tx, r, r); + + HASH_WALK(ptx->prefix_hash, next, n) + rt_unlock_source(rt_find_source_global(n->path_id)); + HASH_WALK_END; +} + +static void bgp_pending_tx_dump(resource *r UNUSED, unsigned indent UNUSED) { debug("\n"); } + +static struct resclass bgp_pending_tx_class = { + .name = "BGP Pending TX", + .size = sizeof(struct bgp_pending_tx), + .free = bgp_pending_tx_rfree, + .dump = bgp_pending_tx_dump, +}; + +void +bgp_init_pending_tx(struct bgp_channel *c) +{ + ASSERT_DIE(!c->ptx); + + pool *p = rp_new(c->pool, "BGP Pending TX"); + c->ptx = ralloc(p, &bgp_pending_tx_class); + c->ptx->pool = p; + + bgp_init_bucket_table(c->ptx); + bgp_init_prefix_table(c); +} + +void +bgp_free_pending_tx(struct bgp_channel *c) +{ + ASSERT_DIE(c->ptx); + ASSERT_DIE(c->ptx->pool); + + rfree(c->ptx->pool); + c->ptx = NULL; +} + + +/* + * Prefix hash table exporter + */ + +struct bgp_out_export_hook { + struct rt_export_hook h; + u32 hash_iter; /* Iterator over hash */ +}; + +static void +bgp_out_table_feed(void *data) +{ + struct bgp_out_export_hook *hook = data; + struct bgp_channel *bc = SKIP_BACK(struct bgp_channel, prefix_exporter, hook->h.table); + struct bgp_pending_tx *c = bc->ptx; + + int max = 512; + + const net_addr *neq = (hook->h.req->addr_mode == TE_ADDR_EQUAL) ? hook->h.req->addr : NULL; + const net_addr *cand = NULL; + + do { + HASH_WALK_ITER(c->prefix_hash, PXH, n, hook->hash_iter) + { + switch (hook->h.req->addr_mode) + { + case TE_ADDR_IN: + if (!net_in_netX(n->net, hook->h.req->addr)) + continue; + /* fall through */ + case TE_ADDR_NONE: + /* Splitting only for multi-net exports */ + if (--max <= 0) + HASH_WALK_ITER_PUT; + break; + + case TE_ADDR_FOR: + if (!neq) + { + if (net_in_netX(hook->h.req->addr, n->net) && (!cand || (n->net->length > cand->length))) + cand = n->net; + continue; + } + /* fall through */ + case TE_ADDR_EQUAL: + if (!net_equal(n->net, neq)) + continue; + break; + } + + struct bgp_bucket *buck = n->cur ?: n->last; + ea_list *ea = NULL; + if (buck == c->withdraw_bucket) + ea_set_dest(&ea, 0, RTD_UNREACHABLE); + else + { + ea = buck->eattrs; + eattr *eanh = bgp_find_attr(ea, BA_NEXT_HOP); + ASSERT_DIE(eanh); + const ip_addr *nh = (const void *) eanh->u.ptr->data; + + struct nexthop_adata nhad = { + .ad = { .length = sizeof (struct nexthop_adata) - sizeof (struct adata), }, + .nh = { .gw = nh[0], }, + }; + + ea_set_attr(&ea, EA_LITERAL_DIRECT_ADATA(&ea_gen_nexthop, 0, tmp_copy_adata(&nhad.ad))); + } + + struct rte_storage es = { + .rte = { + .attrs = ea, + .net = n->net, + .src = rt_find_source_global(n->path_id), + .sender = NULL, + .lastmod = n->lastmod, + .flags = n->cur ? REF_PENDING : 0, + }, + }; + + struct rt_pending_export rpe = { + .new = &es, .new_best = &es, + }; + + if (hook->h.req->export_bulk) + { + const rte *feed = &es.rte; + hook->h.req->export_bulk(hook->h.req, n->net, &rpe, &rpe, &feed, 1); + } + else if (hook->h.req->export_one) + hook->h.req->export_one(hook->h.req, n->net, &rpe); + else + bug("No export method in export request"); + } + HASH_WALK_ITER_END; + + neq = cand; + cand = NULL; + } while (neq); + + if (hook->hash_iter) + ev_schedule_work(&hook->h.event); + else + rt_set_export_state(&hook->h, TES_READY); +} + +static void +bgp_out_table_export_start(struct rt_exporter *re, struct rt_export_request *req) +{ + req->hook = rt_alloc_export(re, sizeof(struct bgp_out_export_hook)); + req->hook->req = req; + + struct bgp_out_export_hook *hook = SKIP_BACK(struct bgp_out_export_hook, h, req->hook); + + hook->h.event.hook = bgp_out_table_feed; + rt_init_export(re, req->hook); +} + +static void +bgp_out_table_export_done(void *data) +{ + struct bgp_out_export_hook *hook = data; + struct rt_export_request *req = hook->h.req; + void (*stopped)(struct rt_export_request *) = hook->h.stopped; + + rt_export_stopped(&hook->h); + CALL(stopped, req); +} + +static const struct rt_exporter_class bgp_out_table_export_class = { + .start = bgp_out_table_export_start, + .done = bgp_out_table_export_done, +}; + +void +bgp_setup_out_table(struct bgp_channel *c) +{ + ASSERT_DIE(c->c.out_table == NULL); + + c->prefix_exporter = (struct rt_exporter) { + .class = &bgp_out_table_export_class, + .addr_type = c->c.table->addr_type, + .rp = c->c.proto->pool, + }; + + rt_exporter_init(&c->prefix_exporter); + + c->c.out_table = &c->prefix_exporter; +} + /* * BGP protocol glue @@ -1720,9 +2030,8 @@ bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *px) int bgp_preexport(struct channel *C, rte *e) { - struct proto *SRC = e->src->proto; struct bgp_proto *p = (struct bgp_proto *) C->proto; - struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (struct bgp_proto *) SRC : NULL; + struct bgp_proto *src = bgp_rte_proto(e); struct bgp_channel *c = (struct bgp_channel *) C; /* Ignore non-BGP channels */ @@ -1738,8 +2047,20 @@ bgp_preexport(struct channel *C, rte *e) return 0; /* Reject flowspec that failed validation */ - if ((e->attrs->dest == RTD_UNREACHABLE) && net_is_flow(e->net->n.addr)) - return -1; + if (net_is_flow(e->net)) + switch (rt_get_flowspec_valid(e)) + { + case FLOWSPEC_VALID: + break; + case FLOWSPEC_INVALID: + return -1; + case FLOWSPEC_UNKNOWN: + ASSUME((rt_get_source_attr(e) != RTS_BGP) || + !((struct bgp_channel *) SKIP_BACK(struct channel, in_req, e->sender->req))->base_table); + break; + case FLOWSPEC__MAX: + bug("This never happens."); + } /* IBGP route reflection, RFC 4456 */ if (p->is_internal && src->is_internal && (p->local_as == src->local_as)) @@ -1750,14 +2071,14 @@ bgp_preexport(struct channel *C, rte *e) /* Generally, this should be handled when path is received, but we check it also here as rr_cluster_id may be undefined or different in src. */ - if (p->rr_cluster_id && bgp_cluster_list_loopy(p, e->attrs->eattrs)) + if (p->rr_cluster_id && bgp_cluster_list_loopy(p, e->attrs)) return -1; } /* Handle well-known communities, RFC 1997 */ struct eattr *a; if (p->cf->interpret_communities && - (a = bgp_find_attr(e->attrs->eattrs, BA_COMMUNITY))) + (a = bgp_find_attr(e->attrs, BA_COMMUNITY))) { const struct adata *d = a->u.ptr; @@ -1781,7 +2102,7 @@ bgp_preexport(struct channel *C, rte *e) /* Do not export routes marked with OTC to upstream, RFC 9234 */ if (bgp_channel_is_role_applicable(c)) { - a = bgp_find_attr(e->attrs->eattrs, BA_ONLY_TO_CUSTOMER); + a = bgp_find_attr(e->attrs, BA_ONLY_TO_CUSTOMER); if (a && (p->cf->local_role==BGP_ROLE_CUSTOMER || p->cf->local_role==BGP_ROLE_PEER || p->cf->local_role==BGP_ROLE_RS_CLIENT)) @@ -1794,8 +2115,7 @@ bgp_preexport(struct channel *C, rte *e) static ea_list * bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *attrs0, struct linpool *pool) { - struct proto *SRC = e->src->proto; - struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (void *) SRC : NULL; + struct bgp_proto *src = bgp_rte_proto(e); struct bgp_export_state s = { .proto = p, .channel = c, .pool = pool, .src = src, .route = e, .mpls = c->desc->mpls }; ea_list *attrs = attrs0; eattr *a; @@ -1803,7 +2123,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at /* ORIGIN attribute - mandatory, attach if missing */ if (! bgp_find_attr(attrs0, BA_ORIGIN)) - bgp_set_attr_u32(&attrs, pool, BA_ORIGIN, 0, src ? ORIGIN_INCOMPLETE : ORIGIN_IGP); + bgp_set_attr_u32(&attrs, BA_ORIGIN, 0, src ? ORIGIN_INCOMPLETE : ORIGIN_IGP); /* AS_PATH attribute - mandatory */ a = bgp_find_attr(attrs0, BA_AS_PATH); @@ -1818,24 +2138,24 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at { /* IBGP or route server -> just ensure there is one */ if (!a) - bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, &null_adata); + bgp_set_attr_ptr(&attrs, BA_AS_PATH, 0, &null_adata); } else if (p->is_interior) { /* Confederation -> prepend ASN as AS_CONFED_SEQUENCE */ ad = as_path_prepend2(pool, ad, AS_PATH_CONFED_SEQUENCE, p->public_as); - bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad); + bgp_set_attr_ptr(&attrs, BA_AS_PATH, 0, ad); } else /* Regular EBGP (no RS, no confederation) */ { /* Regular EBGP -> prepend ASN as regular sequence */ ad = as_path_prepend2(pool, ad, AS_PATH_SEQUENCE, p->public_as); - bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad); + bgp_set_attr_ptr(&attrs, BA_AS_PATH, 0, ad); /* MULTI_EXIT_DESC attribute - accept only if set in export filter */ a = bgp_find_attr(attrs0, BA_MULTI_EXIT_DISC); if (a && !(a->fresh)) - bgp_unset_attr(&attrs, pool, BA_MULTI_EXIT_DISC); + bgp_unset_attr(&attrs, BA_MULTI_EXIT_DISC); } /* NEXT_HOP attribute - delegated to AF-specific hook */ @@ -1844,7 +2164,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at /* LOCAL_PREF attribute - required for IBGP, attach if missing */ if (p->is_interior && ! bgp_find_attr(attrs0, BA_LOCAL_PREF)) - bgp_set_attr_u32(&attrs, pool, BA_LOCAL_PREF, 0, p->cf->default_local_pref); + bgp_set_attr_u32(&attrs, BA_LOCAL_PREF, 0, p->cf->default_local_pref); /* AIGP attribute - accumulate local metric or originate new one */ u64 metric; @@ -1853,7 +2173,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at (c->cf->aigp_originate && bgp_init_aigp_metric(e, &metric, &ad)))) { ad = bgp_aigp_set_metric(pool, ad, metric); - bgp_set_attr_ptr(&attrs, pool, BA_AIGP, 0, ad); + bgp_set_attr_ptr(&attrs, BA_AIGP, 0, ad); } /* IBGP route reflection, RFC 4456 */ @@ -1861,7 +2181,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at { /* ORIGINATOR_ID attribute - attach if not already set */ if (! bgp_find_attr(attrs0, BA_ORIGINATOR_ID)) - bgp_set_attr_u32(&attrs, pool, BA_ORIGINATOR_ID, 0, src->remote_id); + bgp_set_attr_u32(&attrs, BA_ORIGINATOR_ID, 0, src->remote_id); /* CLUSTER_LIST attribute - prepend cluster ID */ a = bgp_find_attr(attrs0, BA_CLUSTER_LIST); @@ -1876,7 +2196,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at ad = int_set_prepend(pool, ad, p->rr_cluster_id); /* Should be at least one prepended cluster ID */ - bgp_set_attr_ptr(&attrs, pool, BA_CLUSTER_LIST, 0, ad); + bgp_set_attr_ptr(&attrs, BA_CLUSTER_LIST, 0, ad); } /* AS4_* transition attributes, RFC 6793 4.2.2 */ @@ -1885,15 +2205,15 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at a = bgp_find_attr(attrs, BA_AS_PATH); if (a && as_path_contains_as4(a->u.ptr)) { - bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, as_path_to_old(pool, a->u.ptr)); - bgp_set_attr_ptr(&attrs, pool, BA_AS4_PATH, 0, as_path_strip_confed(pool, a->u.ptr)); + bgp_set_attr_ptr(&attrs, BA_AS_PATH, 0, as_path_to_old(pool, a->u.ptr)); + bgp_set_attr_ptr(&attrs, BA_AS4_PATH, 0, as_path_strip_confed(pool, a->u.ptr)); } a = bgp_find_attr(attrs, BA_AGGREGATOR); if (a && aggregator_contains_as4(a->u.ptr)) { - bgp_set_attr_ptr(&attrs, pool, BA_AGGREGATOR, 0, aggregator_to_old(pool, a->u.ptr)); - bgp_set_attr_ptr(&attrs, pool, BA_AS4_AGGREGATOR, 0, a->u.ptr); + bgp_set_attr_ptr(&attrs, BA_AGGREGATOR, 0, aggregator_to_old(pool, a->u.ptr)); + bgp_set_attr_ptr(&attrs, BA_AS4_AGGREGATOR, 0, a->u.ptr); } } @@ -1904,7 +2224,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at if (!a && (p->cf->local_role == BGP_ROLE_PROVIDER || p->cf->local_role == BGP_ROLE_PEER || p->cf->local_role == BGP_ROLE_RS_SERVER)) - bgp_set_attr_u32(&attrs, pool, BA_ONLY_TO_CUSTOMER, 0, p->public_as); + bgp_set_attr_u32(&attrs, BA_ONLY_TO_CUSTOMER, 0, p->public_as); } /* @@ -1918,13 +2238,12 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at } void -bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old) +bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old) { struct bgp_proto *p = (void *) P; struct bgp_channel *c = (void *) C; struct bgp_bucket *buck; - struct bgp_prefix *px; - u32 path; + struct rte_src *path; /* Ignore non-BGP channels */ if (C->channel != &channel_bgp) @@ -1932,71 +2251,53 @@ bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old) if (new) { - struct ea_list *attrs = bgp_update_attrs(p, c, new, new->attrs->eattrs, tmp_linpool); + struct ea_list *attrs = bgp_update_attrs(p, c, new, new->attrs, tmp_linpool); /* Error during attribute processing */ if (!attrs) - log(L_ERR "%s: Invalid route %N withdrawn", p->p.name, n->n.addr); + log(L_ERR "%s: Invalid route %N withdrawn", p->p.name, n); /* If attributes are invalid, we fail back to withdraw */ - buck = attrs ? bgp_get_bucket(c, attrs) : bgp_get_withdraw_bucket(c); - path = new->src->global_id; + buck = attrs ? bgp_get_bucket(c->ptx, attrs) : bgp_get_withdraw_bucket(c->ptx); + path = new->src; } else { - buck = bgp_get_withdraw_bucket(c); - path = old->src->global_id; + buck = bgp_get_withdraw_bucket(c->ptx); + path = old->src; } - px = bgp_get_prefix(c, n->n.addr, c->add_path_tx ? path : 0); - add_tail(&buck->prefixes, &px->buck_node); - - bgp_schedule_packet(p->conn, c, PKT_UPDATE); + if (bgp_update_prefix(c, bgp_get_prefix(c->ptx, n, path, c->add_path_tx), buck)) + bgp_schedule_packet(p->conn, c, PKT_UPDATE); } static inline u32 -bgp_get_neighbor(rte *r) +bgp_get_neighbor(const rte *r) { - eattr *e = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); + eattr *e = ea_find(r->attrs, BGP_EA_ID(BA_AS_PATH)); u32 as; if (e && as_path_get_first_regular(e->u.ptr, &as)) return as; /* If AS_PATH is not defined, we treat rte as locally originated */ - struct bgp_proto *p = (void *) r->src->proto; + struct bgp_proto *p = bgp_rte_proto(r); return p->cf->confederation ?: p->local_as; } static inline int -rte_stale(rte *r) +rte_stale(const rte *r) { - if (r->pflags & BGP_REF_STALE) - return 1; - - if (r->pflags & BGP_REF_NOT_STALE) - return 0; - - /* If staleness is unknown, compute and cache it */ - eattr *a = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); - if (a && int_set_contains(a->u.ptr, BGP_COMM_LLGR_STALE)) - { - r->pflags |= BGP_REF_STALE; - return 1; - } - else - { - r->pflags |= BGP_REF_NOT_STALE; - return 0; - } + eattr *a = ea_find(r->attrs, BGP_EA_ID(BA_COMMUNITY)); + return a && int_set_contains(a->u.ptr, BGP_COMM_LLGR_STALE); } int -bgp_rte_better(rte *new, rte *old) +bgp_rte_better(const rte *new, const rte *old) { - struct bgp_proto *new_bgp = (struct bgp_proto *) new->src->proto; - struct bgp_proto *old_bgp = (struct bgp_proto *) old->src->proto; + struct bgp_proto *new_bgp = bgp_rte_proto(new); + struct bgp_proto *old_bgp = bgp_rte_proto(old); eattr *x, *y; u32 n, o; @@ -2025,8 +2326,8 @@ bgp_rte_better(rte *new, rte *old) return 1; /* Start with local preferences */ - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_LOCAL_PREF)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_LOCAL_PREF)); + x = ea_find(new->attrs, BGP_EA_ID(BA_LOCAL_PREF)); + y = ea_find(old->attrs, BGP_EA_ID(BA_LOCAL_PREF)); n = x ? x->u.data : new_bgp->cf->default_local_pref; o = y ? y->u.data : old_bgp->cf->default_local_pref; if (n > o) @@ -2045,8 +2346,8 @@ bgp_rte_better(rte *new, rte *old) /* RFC 4271 9.1.2.2. a) Use AS path lengths */ if (new_bgp->cf->compare_path_lengths || old_bgp->cf->compare_path_lengths) { - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); + x = ea_find(new->attrs, BGP_EA_ID(BA_AS_PATH)); + y = ea_find(old->attrs, BGP_EA_ID(BA_AS_PATH)); n = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN; o = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN; if (n < o) @@ -2056,8 +2357,8 @@ bgp_rte_better(rte *new, rte *old) } /* RFC 4271 9.1.2.2. b) Use origins */ - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); + x = ea_find(new->attrs, BGP_EA_ID(BA_ORIGIN)); + y = ea_find(old->attrs, BGP_EA_ID(BA_ORIGIN)); n = x ? x->u.data : ORIGIN_INCOMPLETE; o = y ? y->u.data : ORIGIN_INCOMPLETE; if (n < o) @@ -2079,8 +2380,8 @@ bgp_rte_better(rte *new, rte *old) if (new_bgp->cf->med_metric || old_bgp->cf->med_metric || (bgp_get_neighbor(new) == bgp_get_neighbor(old))) { - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_MULTI_EXIT_DISC)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_MULTI_EXIT_DISC)); + x = ea_find(new->attrs, BGP_EA_ID(BA_MULTI_EXIT_DISC)); + y = ea_find(old->attrs, BGP_EA_ID(BA_MULTI_EXIT_DISC)); n = x ? x->u.data : new_bgp->cf->default_med; o = y ? y->u.data : old_bgp->cf->default_med; if (n < o) @@ -2096,8 +2397,8 @@ bgp_rte_better(rte *new, rte *old) return 1; /* RFC 4271 9.1.2.2. e) Compare IGP metrics */ - n = new_bgp->cf->igp_metric ? new->attrs->igp_metric : 0; - o = old_bgp->cf->igp_metric ? old->attrs->igp_metric : 0; + n = new_bgp->cf->igp_metric ? rt_get_igp_metric(new) : 0; + o = old_bgp->cf->igp_metric ? rt_get_igp_metric(old) : 0; if (n < o) return 1; if (n > o) @@ -2105,8 +2406,8 @@ bgp_rte_better(rte *new, rte *old) /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */ /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighbor ID */ - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGINATOR_ID)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGINATOR_ID)); + x = ea_find(new->attrs, BGP_EA_ID(BA_ORIGINATOR_ID)); + y = ea_find(old->attrs, BGP_EA_ID(BA_ORIGINATOR_ID)); n = x ? x->u.data : new_bgp->remote_id; o = y ? y->u.data : old_bgp->remote_id; @@ -2123,8 +2424,8 @@ bgp_rte_better(rte *new, rte *old) return 0; /* RFC 4456 9. b) Compare cluster list lengths */ - x = ea_find(new->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_CLUSTER_LIST)); - y = ea_find(old->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_CLUSTER_LIST)); + x = ea_find(new->attrs, BGP_EA_ID(BA_CLUSTER_LIST)); + y = ea_find(old->attrs, BGP_EA_ID(BA_CLUSTER_LIST)); n = x ? int_set_get_size(x->u.ptr) : 0; o = y ? int_set_get_size(y->u.ptr) : 0; if (n < o) @@ -2138,10 +2439,10 @@ bgp_rte_better(rte *new, rte *old) int -bgp_rte_mergable(rte *pri, rte *sec) +bgp_rte_mergable(const rte *pri, const rte *sec) { - struct bgp_proto *pri_bgp = (struct bgp_proto *) pri->src->proto; - struct bgp_proto *sec_bgp = (struct bgp_proto *) sec->src->proto; + struct bgp_proto *pri_bgp = bgp_rte_proto(pri); + struct bgp_proto *sec_bgp = bgp_rte_proto(sec); eattr *x, *y; u32 p, s; @@ -2158,8 +2459,8 @@ bgp_rte_mergable(rte *pri, rte *sec) return 0; /* Start with local preferences */ - x = ea_find(pri->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_LOCAL_PREF)); - y = ea_find(sec->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_LOCAL_PREF)); + x = ea_find(pri->attrs, BGP_EA_ID(BA_LOCAL_PREF)); + y = ea_find(sec->attrs, BGP_EA_ID(BA_LOCAL_PREF)); p = x ? x->u.data : pri_bgp->cf->default_local_pref; s = y ? y->u.data : sec_bgp->cf->default_local_pref; if (p != s) @@ -2168,8 +2469,8 @@ bgp_rte_mergable(rte *pri, rte *sec) /* RFC 4271 9.1.2.2. a) Use AS path lengths */ if (pri_bgp->cf->compare_path_lengths || sec_bgp->cf->compare_path_lengths) { - x = ea_find(pri->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); - y = ea_find(sec->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); + x = ea_find(pri->attrs, BGP_EA_ID(BA_AS_PATH)); + y = ea_find(sec->attrs, BGP_EA_ID(BA_AS_PATH)); p = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN; s = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN; @@ -2181,8 +2482,8 @@ bgp_rte_mergable(rte *pri, rte *sec) } /* RFC 4271 9.1.2.2. b) Use origins */ - x = ea_find(pri->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); - y = ea_find(sec->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); + x = ea_find(pri->attrs, BGP_EA_ID(BA_ORIGIN)); + y = ea_find(sec->attrs, BGP_EA_ID(BA_ORIGIN)); p = x ? x->u.data : ORIGIN_INCOMPLETE; s = y ? y->u.data : ORIGIN_INCOMPLETE; if (p != s) @@ -2192,8 +2493,8 @@ bgp_rte_mergable(rte *pri, rte *sec) if (pri_bgp->cf->med_metric || sec_bgp->cf->med_metric || (bgp_get_neighbor(pri) == bgp_get_neighbor(sec))) { - x = ea_find(pri->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_MULTI_EXIT_DISC)); - y = ea_find(sec->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_MULTI_EXIT_DISC)); + x = ea_find(pri->attrs, BGP_EA_ID(BA_MULTI_EXIT_DISC)); + y = ea_find(sec->attrs, BGP_EA_ID(BA_MULTI_EXIT_DISC)); p = x ? x->u.data : pri_bgp->cf->default_med; s = y ? y->u.data : sec_bgp->cf->default_med; if (p != s) @@ -2205,8 +2506,8 @@ bgp_rte_mergable(rte *pri, rte *sec) return 0; /* RFC 4271 9.1.2.2. e) Compare IGP metrics */ - p = pri_bgp->cf->igp_metric ? pri->attrs->igp_metric : 0; - s = sec_bgp->cf->igp_metric ? sec->attrs->igp_metric : 0; + p = pri_bgp->cf->igp_metric ? rt_get_igp_metric(pri) : 0; + s = sec_bgp->cf->igp_metric ? rt_get_igp_metric(sec) : 0; if (p != s) return 0; @@ -2219,22 +2520,21 @@ bgp_rte_mergable(rte *pri, rte *sec) static inline int same_group(rte *r, u32 lpref, u32 lasn) { - return (r->attrs->pref == lpref) && (bgp_get_neighbor(r) == lasn); + return (rt_get_preference(r) == lpref) && (bgp_get_neighbor(r) == lasn); } static inline int -use_deterministic_med(rte *r) +use_deterministic_med(struct rte_storage *r) { - struct proto *P = r->src->proto; - return (P->proto == &proto_bgp) && ((struct bgp_proto *) P)->cf->deterministic_med; + struct bgp_proto *p = bgp_rte_proto(&r->rte); + return p && p->cf->deterministic_med; } int -bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) +bgp_rte_recalculate(struct rtable_private *table, net *net, rte *new, rte *old, rte *old_best) { - rte *r, *s; rte *key = new ? new : old; - u32 lpref = key->attrs->pref; + u32 lpref = rt_get_preference(key); u32 lasn = bgp_get_neighbor(key); int old_suppressed = old ? !!(old->pflags & BGP_REF_SUPPRESSED) : 0; @@ -2300,13 +2600,13 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) } /* The default case - find a new best-in-group route */ - r = new; /* new may not be in the list */ - for (s=net->routes; rte_is_valid(s); s=s->next) - if (use_deterministic_med(s) && same_group(s, lpref, lasn)) + rte *r = new; /* new may not be in the list */ + for (struct rte_storage *s = net->routes; rte_is_valid(RTE_OR_NULL(s)); s = s->next) + if (use_deterministic_med(s) && same_group(&s->rte, lpref, lasn)) { - s->pflags |= BGP_REF_SUPPRESSED; - if (!r || bgp_rte_better(s, r)) - r = s; + s->rte.pflags |= BGP_REF_SUPPRESSED; + if (!r || bgp_rte_better(&s->rte, r)) + r = &s->rte; } /* Simple case - the last route in group disappears */ @@ -2318,10 +2618,10 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) new->pflags &= ~BGP_REF_SUPPRESSED; /* Found all existing routes mergable with best-in-group */ - for (s=net->routes; rte_is_valid(s); s=s->next) - if (use_deterministic_med(s) && same_group(s, lpref, lasn)) - if ((s != r) && bgp_rte_mergable(r, s)) - s->pflags &= ~BGP_REF_SUPPRESSED; + for (struct rte_storage *s = net->routes; rte_is_valid(RTE_OR_NULL(s)); s = s->next) + if (use_deterministic_med(s) && same_group(&s->rte, lpref, lasn)) + if ((&s->rte != r) && bgp_rte_mergable(r, &s->rte)) + s->rte.pflags &= ~BGP_REF_SUPPRESSED; /* Found best-in-group */ r->pflags &= ~BGP_REF_SUPPRESSED; @@ -2356,25 +2656,59 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) return !old_suppressed; } -struct rte * -bgp_rte_modify_stale(struct rte *r, struct linpool *pool) +void +bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, + struct rt_pending_export *first, struct rt_pending_export *last, + const rte **feed, uint count) { - eattr *a = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); - const struct adata *ad = a ? a->u.ptr : NULL; - uint flags = a ? a->flags : BAF_PARTIAL; + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); + struct rt_import_hook *irh = c->c.in_req.hook; - if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR)) - return NULL; + /* Find our routes among others */ + for (uint i=0; i<count; i++) + { + const rte *r = feed[i]; + + if ( + !rte_is_valid(r) || /* Not a valid route */ + (r->sender != irh) || /* Not our route */ + (r->stale_cycle == irh->stale_set)) /* A new route, do not mark as stale */ + continue; + + eattr *ea = ea_find(r->attrs, BGP_EA_ID(BA_COMMUNITY)); + const struct adata *ad = ea ? ea->u.ptr : NULL; + uint flags = ea ? ea->flags : BAF_PARTIAL; + + /* LLGR not allowed, withdraw the route */ + if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR)) + { + rte_import(&c->c.in_req, n, NULL, r->src); + continue; + } - if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE)) - return r; + /* Route already marked as LLGR, do nothing */ + if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE)) + continue; - r = rte_cow_rta(r, pool); - bgp_set_attr_ptr(&(r->attrs->eattrs), pool, BA_COMMUNITY, flags, - int_set_add(pool, ad, BGP_COMM_LLGR_STALE)); - r->pflags |= BGP_REF_STALE; + /* Store the tmp_linpool state to aggresively save memory */ + struct lp_state tmpp; + lp_save(tmp_linpool, &tmpp); - return r; + /* Mark the route as LLGR */ + rte e0 = *r; + bgp_set_attr_ptr(&e0.attrs, BA_COMMUNITY, flags, int_set_add(tmp_linpool, ad, BGP_COMM_LLGR_STALE)); + + /* We need to update the route but keep it stale. */ + ASSERT_DIE(irh->stale_set == irh->stale_valid + 1); + irh->stale_set--; + rte_import(&c->c.in_req, n, &e0, r->src); + irh->stale_set++; + + /* Restore the memory state */ + lp_restore(tmp_linpool, &tmpp); + } + + rpe_mark_seen_all(req->hook, first, last, NULL); } @@ -2390,8 +2724,8 @@ bgp_process_as4_attrs(ea_list **attrs, struct linpool *pool) eattr *a4 = bgp_find_attr(*attrs, BA_AS4_AGGREGATOR); /* First, unset AS4_* attributes */ - if (p4) bgp_unset_attr(attrs, pool, BA_AS4_PATH); - if (a4) bgp_unset_attr(attrs, pool, BA_AS4_AGGREGATOR); + if (p4) bgp_unset_attr(attrs, BA_AS4_PATH); + if (a4) bgp_unset_attr(attrs, BA_AS4_AGGREGATOR); /* Handle AGGREGATOR attribute */ if (a2 && a4) @@ -2424,60 +2758,37 @@ bgp_process_as4_attrs(ea_list **attrs, struct linpool *pool) } } -int -bgp_get_attr(const eattr *a, byte *buf, int buflen) -{ - uint i = EA_ID(a->id); - const struct bgp_attr_desc *d; - int len; - - if (bgp_attr_known(i)) - { - d = &bgp_attr_table[i]; - len = bsprintf(buf, "%s", d->name); - buf += len; - if (d->format) - { - *buf++ = ':'; - *buf++ = ' '; - d->format(a, buf, buflen - len - 2); - return GA_FULL; - } - return GA_NAME; - } - - bsprintf(buf, "%02x%s", i, (a->flags & BAF_TRANSITIVE) ? " [t]" : ""); - return GA_NAME; -} - void -bgp_get_route_info(rte *e, byte *buf) +bgp_get_route_info(const rte *e, byte *buf) { - eattr *p = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); - eattr *o = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); + eattr *p = ea_find(e->attrs, BGP_EA_ID(BA_AS_PATH)); + eattr *o = ea_find(e->attrs, BGP_EA_ID(BA_ORIGIN)); u32 origas; - buf += bsprintf(buf, " (%d", e->attrs->pref); + buf += bsprintf(buf, " (%d", rt_get_preference(e)); - if (e->pflags & BGP_REF_SUPPRESSED) - buf += bsprintf(buf, "-"); + if (!net_is_flow(e->net)) + { + if (e->pflags & BGP_REF_SUPPRESSED) + buf += bsprintf(buf, "-"); - if (rte_stale(e)) - buf += bsprintf(buf, "s"); + if (rte_stale(e)) + buf += bsprintf(buf, "s"); - u64 metric = bgp_total_aigp_metric(e); - if (metric < BGP_AIGP_MAX) - { - buf += bsprintf(buf, "/%lu", metric); - } - else if (e->attrs->igp_metric) - { - if (!rte_resolvable(e)) - buf += bsprintf(buf, "/-"); - else if (e->attrs->igp_metric >= IGP_METRIC_UNKNOWN) - buf += bsprintf(buf, "/?"); - else - buf += bsprintf(buf, "/%d", e->attrs->igp_metric); + u64 metric = bgp_total_aigp_metric(e); + if (metric < BGP_AIGP_MAX) + { + buf += bsprintf(buf, "/%lu", metric); + } + else if (metric = rt_get_igp_metric(e)) + { + if (!rte_resolvable(e)) + buf += bsprintf(buf, "/-"); + else if (metric >= IGP_METRIC_UNKNOWN) + buf += bsprintf(buf, "/?"); + else + buf += bsprintf(buf, "/%d", metric); + } } buf += bsprintf(buf, ") ["); diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c index 2e442e16..fffa8cec 100644 --- a/proto/bgp/bgp.c +++ b/proto/bgp/bgp.c @@ -115,7 +115,7 @@ #include "nest/bird.h" #include "nest/iface.h" #include "nest/protocol.h" -#include "nest/route.h" +#include "nest/rt.h" #include "nest/cli.h" #include "nest/locks.h" #include "conf/conf.h" @@ -126,9 +126,13 @@ #include "bgp.h" +static void bgp_listen_create(void *); static list STATIC_LIST_INIT(bgp_sockets); /* Global list of listening sockets */ +static list STATIC_LIST_INIT(bgp_listen_pending); /* Global list of listening socket open requests */ +static event bgp_listen_event = { .hook = bgp_listen_create }; +DOMAIN(rtable) bgp_listen_domain; static void bgp_connect(struct bgp_proto *p); static void bgp_active(struct bgp_proto *p); @@ -139,73 +143,43 @@ static void bgp_update_bfd(struct bgp_proto *p, const struct bfd_options *bfd); static int bgp_incoming_connection(sock *sk, uint dummy UNUSED); static void bgp_listen_sock_err(sock *sk UNUSED, int err); +static void bgp_initiate_disable(struct bgp_proto *p, int err_val); -/** - * bgp_open - open a BGP instance - * @p: BGP instance - * - * This function allocates and configures shared BGP resources, mainly listening - * sockets. Should be called as the last step during initialization (when lock - * is acquired and neighbor is ready). When error, caller should change state to - * PS_DOWN and return immediately. - */ -static int -bgp_open(struct bgp_proto *p) +static void bgp_graceful_restart_feed(struct bgp_channel *c); + + +static inline int +bgp_setup_auth(struct bgp_proto *p, int enable) { - struct bgp_socket *bs = NULL; - struct iface *ifa = p->cf->strict_bind ? p->cf->iface : NULL; - ip_addr addr = p->cf->strict_bind ? p->cf->local_ip : - (p->ipv4 ? IPA_NONE4 : IPA_NONE6); - uint port = p->cf->local_port; - uint flags = p->cf->free_bind ? SKF_FREEBIND : 0; - uint flag_mask = SKF_FREEBIND; + /* Beware. This is done from main_birdloop and protocol birdloop is NOT ENTERED. + * Anyway, we are only accessing: + * - protocol config which can be changed only from main_birdloop (reconfig) + * - protocol listen socket which is always driven by main_birdloop + * - protocol name which is set on reconfig + */ - /* We assume that cf->iface is defined iff cf->local_ip is link-local */ + if (p->cf->password && p->listen.sock) + { + ip_addr prefix = p->cf->remote_ip; + int pxlen = -1; - WALK_LIST(bs, bgp_sockets) - if (ipa_equal(bs->sk->saddr, addr) && - (bs->sk->sport == port) && - (bs->sk->iface == ifa) && - (bs->sk->vrf == p->p.vrf) && - ((bs->sk->flags & flag_mask) == flags)) + if (p->cf->remote_range) { - bs->uc++; - p->sock = bs; - return 0; + prefix = net_prefix(p->cf->remote_range); + pxlen = net_pxlen(p->cf->remote_range); } - sock *sk = sk_new(proto_pool); - sk->type = SK_TCP_PASSIVE; - sk->ttl = 255; - sk->saddr = addr; - sk->sport = port; - sk->iface = ifa; - sk->vrf = p->p.vrf; - sk->flags = flags; - sk->tos = IP_PREC_INTERNET_CONTROL; - sk->rbsize = BGP_RX_BUFFER_SIZE; - sk->tbsize = BGP_TX_BUFFER_SIZE; - sk->rx_hook = bgp_incoming_connection; - sk->err_hook = bgp_listen_sock_err; - - if (sk_open(sk) < 0) - goto err; - - bs = mb_allocz(proto_pool, sizeof(struct bgp_socket)); - bs->sk = sk; - bs->uc = 1; - p->sock = bs; - sk->data = bs; - - add_tail(&bgp_sockets, &bs->n); + int rv = sk_set_md5_auth(p->listen.sock->sk, + p->cf->local_ip, prefix, pxlen, p->cf->iface, + enable ? p->cf->password : NULL, p->cf->setkey); - return 0; + if (rv < 0) + sk_log_error(p->listen.sock->sk, p->p.name); -err: - sk_log_error(sk, p->p.name); - log(L_ERR "%s: Cannot open listening socket", p->p.name); - rfree(sk); - return -1; + return rv; + } + else + return 0; } /** @@ -217,43 +191,155 @@ err: static void bgp_close(struct bgp_proto *p) { - struct bgp_socket *bs = p->sock; + LOCK_DOMAIN(rtable, bgp_listen_domain); - ASSERT(bs && bs->uc); + struct bgp_listen_request *req = &p->listen; + struct bgp_socket *bs = req->sock; - if (--bs->uc) - return; + if (bs) + { + req->sock = NULL; + rem_node(&req->n); - rfree(bs->sk); - rem_node(&bs->n); - mb_free(bs); + if (bs && EMPTY_LIST(bs->requests)) + ev_send(&global_event_list, &bgp_listen_event); + } + + UNLOCK_DOMAIN(rtable, bgp_listen_domain); } -static inline int -bgp_setup_auth(struct bgp_proto *p, int enable) +/** + * bgp_open - open a BGP instance + * @p: BGP instance + * + * This function allocates and configures shared BGP resources, mainly listening + * sockets. Should be called as the last step during initialization (when lock + * is acquired and neighbor is ready). When error, caller should change state to + * PS_DOWN and return immediately. + */ +static void +bgp_open(struct bgp_proto *p) { - if (p->cf->password) - { - ip_addr prefix = p->cf->remote_ip; - int pxlen = -1; + LOCK_DOMAIN(rtable, bgp_listen_domain); - if (p->cf->remote_range) + struct bgp_listen_request *req = &p->listen; + /* We assume that cf->iface is defined iff cf->local_ip is link-local */ + req->iface = p->cf->strict_bind ? p->cf->iface : NULL; + req->vrf = p->p.vrf; + req->addr = p->cf->strict_bind ? p->cf->local_ip : + (p->ipv4 ? IPA_NONE4 : IPA_NONE6); + req->port = p->cf->local_port; + req->flags = p->cf->free_bind ? SKF_FREEBIND : 0; + + BGP_TRACE(D_EVENTS, "Requesting listen socket at %I%J port %u", req->addr, req->iface, req->port); + + add_tail(&bgp_listen_pending, &req->n); + ev_send(&global_event_list, &bgp_listen_event); + + UNLOCK_DOMAIN(rtable, bgp_listen_domain); +} + +static void +bgp_listen_create(void *_ UNUSED) +{ + ASSERT_DIE(birdloop_inside(&main_birdloop)); + uint flag_mask = SKF_FREEBIND; + + while (1) { + LOCK_DOMAIN(rtable, bgp_listen_domain); + + if (EMPTY_LIST(bgp_listen_pending)) { - prefix = net_prefix(p->cf->remote_range); - pxlen = net_pxlen(p->cf->remote_range); + UNLOCK_DOMAIN(rtable, bgp_listen_domain); + break; } - int rv = sk_set_md5_auth(p->sock->sk, - p->cf->local_ip, prefix, pxlen, p->cf->iface, - enable ? p->cf->password : NULL, p->cf->setkey); + /* Get the first request to match */ + struct bgp_listen_request *req = HEAD(bgp_listen_pending); + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, listen, req); + rem_node(&req->n); + + /* First try to find existing socket */ + struct bgp_socket *bs; + WALK_LIST(bs, bgp_sockets) + if (ipa_equal(bs->sk->saddr, req->addr) && + (bs->sk->sport == req->port) && + (bs->sk->iface == req->iface) && + (bs->sk->vrf == req->vrf) && + ((bs->sk->flags & flag_mask) == req->flags)) + break; - if (rv < 0) - sk_log_error(p->sock->sk, p->p.name); + /* Not found any */ + if (NODE_VALID(bs)) + BGP_TRACE(D_EVENTS, "Found a listening socket: %p", bs); + else + { + /* Allocating new socket from global protocol pool. + * We can do this in main_birdloop. */ + sock *sk = sk_new(proto_pool); + sk->type = SK_TCP_PASSIVE; + sk->ttl = 255; + sk->saddr = req->addr; + sk->sport = req->port; + sk->iface = req->iface; + sk->vrf = req->vrf; + sk->flags = req->flags; + sk->tos = IP_PREC_INTERNET_CONTROL; + sk->rbsize = BGP_RX_BUFFER_SIZE; + sk->tbsize = BGP_TX_BUFFER_SIZE; + sk->rx_hook = bgp_incoming_connection; + sk->err_hook = bgp_listen_sock_err; + + if (sk_open(sk, &main_birdloop) < 0) + { + sk_log_error(sk, p->p.name); + log(L_ERR "%s: Cannot open listening socket", p->p.name); + rfree(sk); + UNLOCK_DOMAIN(rtable, bgp_listen_domain); - return rv; + bgp_initiate_disable(p, BEM_NO_SOCKET); + continue; + } + + bs = mb_allocz(proto_pool, sizeof(struct bgp_socket)); + bs->sk = sk; + sk->data = bs; + + init_list(&bs->requests); + add_tail(&bgp_sockets, &bs->n); + + BGP_TRACE(D_EVENTS, "Created new listening socket: %p", bs); + } + + req->sock = bs; + add_tail(&bs->requests, &req->n); + + if (bgp_setup_auth(p, 1) < 0) + { + rem_node(&req->n); + req->sock = NULL; + + UNLOCK_DOMAIN(rtable, bgp_listen_domain); + + bgp_initiate_disable(p, BEM_INVALID_MD5); + continue; + } + + UNLOCK_DOMAIN(rtable, bgp_listen_domain); } - else - return 0; + + /* Cleanup leftover listening sockets */ + LOCK_DOMAIN(rtable, bgp_listen_domain); + struct bgp_socket *bs; + node *nxt; + WALK_LIST_DELSAFE(bs, nxt, bgp_sockets) + if (EMPTY_LIST(bs->requests)) + { + rfree(bs->sk); + rem_node(&bs->n); + mb_free(bs); + } + UNLOCK_DOMAIN(rtable, bgp_listen_domain); } static inline struct bgp_channel * @@ -279,6 +365,8 @@ bgp_startup(struct bgp_proto *p) if (p->postponed_sk) { /* Apply postponed incoming connection */ + sk_reloop(p->postponed_sk, p->p.loop); + bgp_setup_conn(p, &p->incoming_conn); bgp_setup_sk(&p->incoming_conn, p->postponed_sk); bgp_send_open(&p->incoming_conn); @@ -296,13 +384,7 @@ bgp_startup_timeout(timer *t) static void bgp_initiate(struct bgp_proto *p) { - int err_val; - - if (bgp_open(p) < 0) - { err_val = BEM_NO_SOCKET; goto err1; } - - if (bgp_setup_auth(p, 1) < 0) - { err_val = BEM_INVALID_MD5; goto err2; } + bgp_open(p); if (p->cf->bfd) bgp_update_bfd(p, p->cf->bfd); @@ -311,23 +393,28 @@ bgp_initiate(struct bgp_proto *p) { p->start_state = BSS_DELAY; BGP_TRACE(D_EVENTS, "Startup delayed by %d seconds due to errors", p->startup_delay); - bgp_start_timer(p->startup_timer, p->startup_delay); + bgp_start_timer(p, p->startup_timer, p->startup_delay); } else bgp_startup(p); +} - return; - -err2: - bgp_close(p); -err1: - p->p.disabled = 1; - bgp_store_error(p, NULL, BE_MISC, err_val); - - p->neigh = NULL; - proto_notify_state(&p->p, PS_DOWN); - - return; +static void +bgp_initiate_disable(struct bgp_proto *p, int err_val) +{ + PROTO_LOCKED_FROM_MAIN(&p->p) + { + /* The protocol may be already down for another reason. + * Shutdown the protocol only if it isn't already shutting down. */ + switch (p->p.proto_state) + { + case PS_START: + case PS_UP: + p->p.disabled = 1; + bgp_store_error(p, NULL, BE_MISC, err_val); + bgp_stop(p, err_val, NULL, 0); + } + } } /** @@ -340,14 +427,14 @@ err1: * timers. */ void -bgp_start_timer(timer *t, uint value) +bgp_start_timer(struct bgp_proto *p, timer *t, uint value) { if (value) { /* The randomization procedure is specified in RFC 4271 section 10 */ btime time = value S; btime randomize = random() % ((time / 4) + 1); - tm_start(t, time - randomize); + tm_start_in(t, time - randomize, p->p.loop); } else tm_stop(t); @@ -374,8 +461,10 @@ bgp_close_conn(struct bgp_conn *conn) conn->keepalive_timer = NULL; rfree(conn->hold_timer); conn->hold_timer = NULL; + rfree(conn->tx_ev); conn->tx_ev = NULL; + rfree(conn->sk); conn->sk = NULL; @@ -512,9 +601,16 @@ void bgp_stop(struct bgp_proto *p, int subcode, byte *data, uint len) { proto_notify_state(&p->p, PS_STOP); + p->uncork_ev->data = NULL; bgp_graceful_close_conn(&p->outgoing_conn, subcode, data, len); bgp_graceful_close_conn(&p->incoming_conn, subcode, data, len); - ev_schedule(p->event); + + struct bgp_channel *c; + WALK_LIST(c, p->p.channels) + if (c->ptx) + bgp_free_pending_tx(c); + + proto_send_event(&p->p, p->event); } static inline void @@ -607,7 +703,7 @@ bgp_conn_enter_established_state(struct bgp_conn *conn) int active = loc->ready && rem->ready; c->c.disabled = !active; - c->c.reloadable = p->route_refresh || c->cf->import_table; + c->c.reloadable = p->route_refresh || ((c->c.in_keep & RIK_PREFILTER) == RIK_PREFILTER); c->index = active ? num++ : 0; @@ -705,7 +801,7 @@ bgp_conn_enter_close_state(struct bgp_conn *conn) conn->sk->rx_hook = NULL; /* Timeout for CLOSE state, if we cannot send notification soon then we just hangup */ - bgp_start_timer(conn->hold_timer, 10); + bgp_start_timer(p, conn->hold_timer, 10); if (os == BS_ESTABLISHED) bgp_conn_leave_established_state(p); @@ -719,7 +815,7 @@ bgp_conn_enter_idle_state(struct bgp_conn *conn) bgp_close_conn(conn); bgp_conn_set_state(conn, BS_IDLE); - ev_schedule(p->event); + proto_send_event(&p->p, p->event); if (os == BS_ESTABLISHED) bgp_conn_leave_established_state(p); @@ -761,32 +857,28 @@ bgp_handle_graceful_restart(struct bgp_proto *p) { case BGP_GRS_NONE: c->gr_active = BGP_GRS_ACTIVE; - rt_refresh_begin(c->c.table, &c->c); - break; + /* fall through */ case BGP_GRS_ACTIVE: - rt_refresh_end(c->c.table, &c->c); - rt_refresh_begin(c->c.table, &c->c); + rt_refresh_begin(&c->c.in_req); break; case BGP_GRS_LLGR: - rt_refresh_begin(c->c.table, &c->c); - rt_modify_stale(c->c.table, &c->c); + rt_refresh_begin(&c->c.in_req); + bgp_graceful_restart_feed(c); break; } } else { /* Just flush the routes */ - rt_refresh_begin(c->c.table, &c->c); - rt_refresh_end(c->c.table, &c->c); + rt_refresh_begin(&c->c.in_req); + rt_refresh_end(&c->c.in_req); } /* Reset bucket and prefix tables */ - bgp_free_bucket_table(c); - bgp_free_prefix_table(c); - bgp_init_bucket_table(c); - bgp_init_prefix_table(c); + bgp_free_pending_tx(c); + bgp_init_pending_tx(c); c->packets_to_send = 0; } @@ -794,9 +886,56 @@ bgp_handle_graceful_restart(struct bgp_proto *p) ASSERT(p->gr_active_num > 0); proto_notify_state(&p->p, PS_START); - tm_start(p->gr_timer, p->conn->remote_caps->gr_time S); + tm_start_in(p->gr_timer, p->conn->remote_caps->gr_time S, p->p.loop); +} + +static void +bgp_graceful_restart_feed_done(struct rt_export_request *req) +{ + req->hook = NULL; +} + +static void +bgp_graceful_restart_feed_dump_req(struct rt_export_request *req) +{ + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); + debug(" BGP-GR %s.%s export request %p\n", c->c.proto->name, c->c.name, req); +} + +static void +bgp_graceful_restart_feed_log_state_change(struct rt_export_request *req, u8 state) +{ + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); + struct bgp_proto *p = (void *) c->c.proto; + BGP_TRACE(D_EVENTS, "Long-lived graceful restart export state changed to %s", rt_export_state_name(state)); + + if (state == TES_READY) + rt_stop_export(req, bgp_graceful_restart_feed_done); +} + +static void +bgp_graceful_restart_drop_export(struct rt_export_request *req UNUSED, const net_addr *n UNUSED, struct rt_pending_export *rpe UNUSED) +{ /* Nothing to do */ } + +static void +bgp_graceful_restart_feed(struct bgp_channel *c) +{ + c->stale_feed = (struct rt_export_request) { + .name = "BGP-GR", + .list = &global_work_list, + .trace_routes = c->c.debug | c->c.proto->debug, + .dump_req = bgp_graceful_restart_feed_dump_req, + .log_state_change = bgp_graceful_restart_feed_log_state_change, + .export_bulk = bgp_rte_modify_stale, + .export_one = bgp_graceful_restart_drop_export, + }; + + rt_request_export(c->c.table, &c->stale_feed); } + + + /** * bgp_graceful_restart_done - finish active BGP graceful restart * @c: BGP channel @@ -820,7 +959,7 @@ bgp_graceful_restart_done(struct bgp_channel *c) BGP_TRACE(D_EVENTS, "Neighbor graceful restart done"); tm_stop(c->stale_timer); - rt_refresh_end(c->c.table, &c->c); + rt_refresh_end(&c->c.in_req); } /** @@ -861,8 +1000,8 @@ bgp_graceful_restart_timeout(timer *t) /* Channel is in GR, and supports LLGR -> start LLGR */ c->gr_active = BGP_GRS_LLGR; - tm_start(c->stale_timer, c->stale_time S); - rt_modify_stale(c->c.table, &c->c); + tm_start_in(c->stale_timer, c->stale_time S, p->p.loop); + bgp_graceful_restart_feed(c); } } else @@ -900,10 +1039,7 @@ bgp_refresh_begin(struct bgp_channel *c) { log(L_WARN "%s: BEGIN-OF-RR received before END-OF-RIB, ignoring", p->p.name); return; } c->load_state = BFS_REFRESHING; - rt_refresh_begin(c->c.table, &c->c); - - if (c->c.in_table) - rt_refresh_begin(c->c.in_table, &c->c); + rt_refresh_begin(&c->c.in_req); } /** @@ -924,10 +1060,7 @@ bgp_refresh_end(struct bgp_channel *c) { log(L_WARN "%s: END-OF-RR received without prior BEGIN-OF-RR, ignoring", p->p.name); return; } c->load_state = BFS_NONE; - rt_refresh_end(c->c.table, &c->c); - - if (c->c.in_table) - rt_prune_sync(c->c.in_table, 0); + rt_refresh_end(&c->c.in_req); } @@ -941,7 +1074,7 @@ bgp_send_open(struct bgp_conn *conn) bgp_prepare_capabilities(conn); bgp_schedule_packet(conn, NULL, PKT_OPEN); bgp_conn_set_state(conn, BS_OPENSENT); - bgp_start_timer(conn->hold_timer, conn->bgp->cf->initial_hold_time); + bgp_start_timer(conn->bgp, conn->hold_timer, conn->bgp->cf->initial_hold_time); } static void @@ -1018,7 +1151,7 @@ bgp_hold_timeout(timer *t) and perhaps just not processed BGP packets in time. */ if (sk_rx_ready(conn->sk) > 0) - bgp_start_timer(conn->hold_timer, 10); + bgp_start_timer(p, conn->hold_timer, 10); else if ((conn->state == BS_ESTABLISHED) && p->llgr_ready) { BGP_TRACE(D_EVENTS, "Hold timer expired"); @@ -1078,7 +1211,7 @@ bgp_active(struct bgp_proto *p) BGP_TRACE(D_EVENTS, "Connect delayed by %d seconds", delay); bgp_setup_conn(p, conn); bgp_conn_set_state(conn, BS_ACTIVE); - bgp_start_timer(conn->connect_timer, delay); + bgp_start_timer(p, conn->connect_timer, delay); } /** @@ -1109,6 +1242,7 @@ bgp_connect(struct bgp_proto *p) /* Enter Connect state and start establishing c s->tos = IP_PREC_INTERNET_CONTROL; s->password = p->cf->password; s->tx_hook = bgp_connected; + s->flags = p->cf->free_bind ? SKF_FREEBIND : 0; BGP_TRACE(D_EVENTS, "Connecting to %I%J from local address %I%J", s->daddr, ipa_is_link_local(s->daddr) ? p->cf->iface : NULL, s->saddr, ipa_is_link_local(s->saddr) ? s->iface : NULL); @@ -1116,7 +1250,7 @@ bgp_connect(struct bgp_proto *p) /* Enter Connect state and start establishing c bgp_setup_sk(conn, s); bgp_conn_set_state(conn, BS_CONNECT); - if (sk_open(s) < 0) + if (sk_open(s, p->p.loop) < 0) goto err; /* Set minimal receive TTL if needed */ @@ -1125,7 +1259,7 @@ bgp_connect(struct bgp_proto *p) /* Enter Connect state and start establishing c goto err; DBG("BGP: Waiting for connect success\n"); - bgp_start_timer(conn->connect_timer, p->cf->connect_retry_time); + bgp_start_timer(p, conn->connect_timer, p->cf->connect_retry_time); return; err: @@ -1146,12 +1280,17 @@ static struct bgp_proto * bgp_find_proto(sock *sk) { struct bgp_proto *best = NULL; - struct bgp_proto *p; + struct bgp_socket *bs = sk->data; + struct bgp_listen_request *req; /* sk->iface is valid only if src or dst address is link-local */ int link = ipa_is_link_local(sk->saddr) || ipa_is_link_local(sk->daddr); - WALK_LIST(p, proto_list) + LOCK_DOMAIN(rtable, bgp_listen_domain); + + WALK_LIST(req, bs->requests) + { + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, listen, req); if ((p->p.proto == &proto_bgp) && (ipa_equal(p->remote_ip, sk->daddr) || bgp_is_dynamic(p)) && (!p->cf->remote_range || ipa_in_netX(sk->daddr, p->cf->remote_range)) && @@ -1165,7 +1304,9 @@ bgp_find_proto(sock *sk) if (!bgp_is_dynamic(p)) break; } + } + UNLOCK_DOMAIN(rtable, bgp_listen_domain); return best; } @@ -1184,6 +1325,8 @@ bgp_find_proto(sock *sk) static int bgp_incoming_connection(sock *sk, uint dummy UNUSED) { + ASSERT_DIE(birdloop_inside(&main_birdloop)); + struct bgp_proto *p; int acc, hops; @@ -1197,6 +1340,8 @@ bgp_incoming_connection(sock *sk, uint dummy UNUSED) return 0; } + birdloop_enter(p->p.loop); + /* * BIRD should keep multiple incoming connections in OpenSent state (for * details RFC 4271 8.2.1 par 3), but it keeps just one. Duplicate incoming @@ -1226,7 +1371,7 @@ bgp_incoming_connection(sock *sk, uint dummy UNUSED) if (!acc) { rfree(sk); - return 0; + goto leave; } hops = p->cf->multihop ? : 1; @@ -1251,19 +1396,24 @@ bgp_incoming_connection(sock *sk, uint dummy UNUSED) p = bgp_spawn(p, sk->daddr); p->postponed_sk = sk; rmove(sk, p->p.pool); - return 0; + goto leave; } rmove(sk, p->p.pool); + sk_reloop(sk, p->p.loop); + bgp_setup_conn(p, &p->incoming_conn); bgp_setup_sk(&p->incoming_conn, sk); bgp_send_open(&p->incoming_conn); - return 0; + goto leave; err: sk_log_error(sk, p->p.name); log(L_ERR "%s: Incoming connection aborted", p->p.name); rfree(sk); + +leave: + birdloop_leave(p->p.loop); return 0; } @@ -1374,15 +1524,22 @@ static void bgp_update_bfd(struct bgp_proto *p, const struct bfd_options *bfd) { if (bfd && p->bfd_req) + { + BGP_TRACE(D_EVENTS, "Updating existing BFD request"); bfd_update_request(p->bfd_req, bfd); + } if (bfd && !p->bfd_req && !bgp_is_dynamic(p)) + { p->bfd_req = bfd_request_session(p->p.pool, p->remote_ip, p->local_ip, p->cf->multihop ? NULL : p->neigh->iface, - p->p.vrf, bgp_bfd_notify, p, bfd); + p->p.vrf, bgp_bfd_notify, p, p->p.loop, bfd); + BGP_TRACE(D_EVENTS, "Requesting a new BFD session"); + } if (!bfd && p->bfd_req) { + BGP_TRACE(D_EVENTS, "Retracting the BFD request"); rfree(p->bfd_req); p->bfd_req = NULL; } @@ -1398,12 +1555,8 @@ bgp_reload_routes(struct channel *C) if (C->channel != &channel_bgp) return; - ASSERT(p->conn && (p->route_refresh || c->c.in_table)); - - if (c->c.in_table) - channel_schedule_reload(C); - else - bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH); + ASSERT(p->conn && p->route_refresh); + bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH); } static void @@ -1423,6 +1576,12 @@ bgp_feed_begin(struct channel *C, int initial) if (initial && p->cf->gr_mode) c->feed_state = BFS_LOADING; + if (!initial && C->out_table) + { + c->feed_out_table = 1; + return; + } + /* It is refeed and both sides support enhanced route refresh */ if (!initial && p->enhanced_refresh) { @@ -1445,6 +1604,12 @@ bgp_feed_end(struct channel *C) if (C->channel != &channel_bgp) return; + if (c->feed_out_table) + { + c->feed_out_table = 0; + return; + } + /* This should not happen */ if (!p->conn) return; @@ -1467,9 +1632,9 @@ bgp_feed_end(struct channel *C) static void -bgp_start_locked(struct object_lock *lock) +bgp_start_locked(void *_p) { - struct bgp_proto *p = lock->data; + struct bgp_proto *p = _p; const struct bgp_config *cf = p->cf; if (p->p.proto_state != PS_START) @@ -1545,6 +1710,8 @@ bgp_start(struct proto *P) p->last_rx_update = 0; p->event = ev_new_init(p->p.pool, bgp_decision, p); + p->uncork_ev = ev_new_init(p->p.pool, bgp_uncork, p); + p->startup_timer = tm_new_init(p->p.pool, bgp_startup_timeout, p, 0, 0); p->gr_timer = tm_new_init(p->p.pool, bgp_graceful_restart_timeout, p, 0, 0); @@ -1574,8 +1741,11 @@ bgp_start(struct proto *P) lock->iface = p->cf->iface; lock->vrf = p->cf->iface ? NULL : p->p.vrf; lock->type = OBJLOCK_TCP; - lock->hook = bgp_start_locked; - lock->data = p; + lock->event = (event) { + .hook = bgp_start_locked, + .data = p, + }; + lock->target = proto_event_list(P); /* For dynamic BGP, we use inst 1 to avoid collisions with regular BGP */ if (bgp_is_dynamic(p)) @@ -1675,6 +1845,13 @@ done: return p->p.proto_state; } +struct rte_owner_class bgp_rte_owner_class = { + .get_route_info = bgp_get_route_info, + .rte_better = bgp_rte_better, + .rte_mergable = bgp_rte_mergable, + .rte_igp_metric = bgp_rte_igp_metric, +}; + static struct proto * bgp_init(struct proto_config *CF) { @@ -1684,15 +1861,13 @@ bgp_init(struct proto_config *CF) P->rt_notify = bgp_rt_notify; P->preexport = bgp_preexport; - P->neigh_notify = bgp_neigh_notify; + P->iface_sub.neigh_notify = bgp_neigh_notify; P->reload_routes = bgp_reload_routes; P->feed_begin = bgp_feed_begin; P->feed_end = bgp_feed_end; - P->rte_better = bgp_rte_better; - P->rte_mergable = bgp_rte_mergable; - P->rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL; - P->rte_modify = bgp_rte_modify_stale; - P->rte_igp_metric = bgp_rte_igp_metric; + + P->sources.class = &bgp_rte_owner_class; + P->sources.rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL; p->cf = cf; p->is_internal = (cf->local_as == cf->remote_as); @@ -1759,14 +1934,11 @@ bgp_channel_start(struct channel *C) } c->pool = p->p.pool; // XXXX - bgp_init_bucket_table(c); - bgp_init_prefix_table(c); - - if (c->cf->import_table) - channel_setup_in_table(C); if (c->cf->export_table) - channel_setup_out_table(C); + bgp_setup_out_table(c); + + bgp_init_pending_tx(c); c->stale_timer = tm_new_init(c->pool, bgp_long_lived_stale_timeout, c, 0, 0); @@ -1889,7 +2061,7 @@ bgp_default_igp_table(struct bgp_config *cf, struct bgp_channel_config *cc, u32 return cc2->c.table; /* Last, try default table of given type */ - if (tab = cf->c.global->def_tables[type]) + if (tab = rt_get_default_table(cf->c.global, type)) return tab; cf_error("Undefined IGP table"); @@ -1908,7 +2080,7 @@ bgp_default_base_table(struct bgp_config *cf, struct bgp_channel_config *cc) return cc2->c.table; /* Last, try default table of given type */ - struct rtable_config *tab = cf->c.global->def_tables[type]; + struct rtable_config *tab = rt_get_default_table(cf->c.global, type); if (tab) return tab; @@ -2175,7 +2347,6 @@ bgp_channel_reconfigure(struct channel *C, struct channel_config *CC, int *impor (new->llgr_time != old->llgr_time) || (new->ext_next_hop != old->ext_next_hop) || (new->add_path != old->add_path) || - (new->import_table != old->import_table) || (new->export_table != old->export_table) || (TABLE(new, igp_table_ip4) != TABLE(old, igp_table_ip4)) || (TABLE(new, igp_table_ip6) != TABLE(old, igp_table_ip6)) || @@ -2190,11 +2361,13 @@ bgp_channel_reconfigure(struct channel *C, struct channel_config *CC, int *impor (new->aigp != old->aigp) || (new->cost != old->cost)) { - /* import_changed itself does not force ROUTE_REFRESH when import_table is active */ - if (c->c.in_table && (c->c.channel_state == CS_UP)) + /* If import table is active and route refresh is possible, we just ask for route refresh */ + if ((c->c.in_keep & RIK_PREFILTER) && (c->c.channel_state == CS_UP) && p->route_refresh) bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH); - *import_changed = 1; + /* Otherwise we do complete reload */ + else + *import_changed = 1; } if (!ipa_equal(new->next_hop_addr, old->next_hop_addr) || @@ -2562,6 +2735,9 @@ bgp_show_proto_info(struct proto *P) tm_remains(p->conn->hold_timer), p->conn->hold_time); cli_msg(-1006, " Keepalive timer: %t/%u", tm_remains(p->conn->keepalive_timer), p->conn->keepalive_time); + cli_msg(-1006, " TX pending: %d bytes%s", + p->conn->sk->tpos - p->conn->sk->ttx, + ev_active(p->conn->tx_ev) ? " (refill scheduled)" : ""); } #if 0 @@ -2615,6 +2791,22 @@ bgp_show_proto_info(struct proto *P) if (c->base_table) cli_msg(-1006, " Base table: %s", c->base_table->name); + + uint bucket_cnt = 0; + uint prefix_cnt = 0; + struct bgp_bucket *buck; + struct bgp_prefix *px; + if (c->ptx) + WALK_LIST(buck, c->ptx->bucket_queue) + { + bucket_cnt++; + WALK_LIST(px, buck->prefixes) + if (px->cur) + prefix_cnt++; + } + + cli_msg(-1006, " Pending %u attribute sets with total %u prefixes to send", + bucket_cnt, prefix_cnt); } } } @@ -2632,7 +2824,6 @@ struct channel_class channel_bgp = { struct protocol proto_bgp = { .name = "BGP", .template = "bgp%d", - .class = PROTOCOL_BGP, .preference = DEF_PREF_BGP, .channel_mask = NB_IP | NB_VPN | NB_FLOW, .proto_size = sizeof(struct bgp_proto), @@ -2644,12 +2835,12 @@ struct protocol proto_bgp = { .reconfigure = bgp_reconfigure, .copy_config = bgp_copy_config, .get_status = bgp_get_status, - .get_attr = bgp_get_attr, - .get_route_info = bgp_get_route_info, .show_proto_info = bgp_show_proto_info }; void bgp_build(void) { proto_build(&proto_bgp); + bgp_register_attrs(); + bgp_listen_domain = DOMAIN_NEW(rtable, "BGP Listen Sockets"); } diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h index 2808d479..621e1bc5 100644 --- a/proto/bgp/bgp.h +++ b/proto/bgp/bgp.h @@ -14,7 +14,7 @@ #include <stdint.h> #include <setjmp.h> #include "nest/bird.h" -#include "nest/route.h" +#include "nest/rt.h" #include "nest/bfd.h" //#include "lib/lists.h" #include "lib/hash.h" @@ -67,10 +67,10 @@ struct bgp_af_desc { u8 no_igp; const char *name; uint (*encode_nlri)(struct bgp_write_state *s, struct bgp_bucket *buck, byte *buf, uint size); - void (*decode_nlri)(struct bgp_parse_state *s, byte *pos, uint len, rta *a); + void (*decode_nlri)(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a); void (*update_next_hop)(struct bgp_export_state *s, eattr *nh, ea_list **to); uint (*encode_next_hop)(struct bgp_write_state *s, eattr *nh, byte *buf, uint size); - void (*decode_next_hop)(struct bgp_parse_state *s, byte *pos, uint len, rta *a); + void (*decode_next_hop)(struct bgp_parse_state *s, byte *pos, uint len, ea_list **to); }; @@ -162,7 +162,7 @@ struct bgp_channel_config { u8 aigp_originate; /* AIGP is originated automatically */ u32 cost; /* IGP cost for direct next hops */ u8 import_table; /* Use c.in_table as Adj-RIB-In */ - u8 export_table; /* Use c.out_table as Adj-RIB-Out */ + u8 export_table; /* Keep Adj-RIB-Out and export it */ struct rtable_config *igp_table_ip4; /* Table for recursive IPv4 next hop lookups */ struct rtable_config *igp_table_ip6; /* Table for recursive IPv6 next hop lookups */ @@ -220,8 +220,6 @@ struct bgp_channel_config { /* rte->pflags */ #define BGP_REF_SUPPRESSED 0x1 /* Used for deterministic MED comparison */ -#define BGP_REF_STALE 0x2 /* Route is LLGR_STATE */ -#define BGP_REF_NOT_STALE 0x4 /* Route is NOT LLGR_STATE */ struct bgp_af_caps { u32 afi; @@ -266,8 +264,8 @@ struct bgp_caps { struct bgp_socket { node n; /* Node in global bgp_sockets */ + list requests; /* Listen requests */ sock *sk; /* Real listening socket */ - u32 uc; /* Use count */ }; struct bgp_stats { @@ -302,6 +300,16 @@ struct bgp_conn { uint hold_time, keepalive_time; /* Times calculated from my and neighbor's requirements */ }; +struct bgp_listen_request { + node n; /* Node in bgp_socket / pending list */ + struct bgp_socket *sock; /* Assigned socket */ + ip_addr addr; + struct iface *iface; + struct iface *vrf; + uint port; + uint flags; +}; + struct bgp_proto { struct proto p; const struct bgp_config *cf; /* Shortcut to BGP configuration */ @@ -333,9 +341,10 @@ struct bgp_proto { struct bgp_conn incoming_conn; /* Incoming connection we have neither accepted nor rejected yet */ struct object_lock *lock; /* Lock for neighbor connection */ struct neighbor *neigh; /* Neighbor entry corresponding to remote ip, NULL if multihop */ - struct bgp_socket *sock; /* Shared listening socket */ + struct bgp_listen_request listen; /* Shared listening socket */ struct bfd_request *bfd_req; /* BFD request, if BFD is used */ struct birdsock *postponed_sk; /* Postponed incoming socket for dynamic BGP */ + event *uncork_ev; /* Uncork event in case of congestion */ struct bgp_stats stats; /* BGP statistics */ btime last_established; /* Last time of enter/leave of established state */ btime last_rx_update; /* Last time of RX update */ @@ -367,12 +376,8 @@ struct bgp_channel { /* Rest are zeroed when down */ pool *pool; - HASH(struct bgp_bucket) bucket_hash; /* Hash table of route buckets */ - struct bgp_bucket *withdraw_bucket; /* Withdrawn routes */ - list bucket_queue; /* Queue of buckets to send (struct bgp_bucket) */ - - HASH(struct bgp_prefix) prefix_hash; /* Prefixes to be sent */ - slab *prefix_slab; /* Slab holding prefix nodes */ + struct bgp_pending_tx *ptx; /* Routes waiting to be sent */ + struct rt_exporter prefix_exporter; /* Table-like exporter for ptx */ ip_addr next_hop_addr; /* Local address for NEXT_HOP attribute */ ip_addr link_addr; /* Link-local version of next_hop_addr */ @@ -386,17 +391,23 @@ struct bgp_channel { timer *stale_timer; /* Long-lived stale timer for LLGR */ u32 stale_time; /* Stored LLGR stale time from last session */ + struct rt_export_request stale_feed; /* Feeder request for stale route modification */ u8 add_path_rx; /* Session expects receive of ADD-PATH extended NLRI */ u8 add_path_tx; /* Session expects transmit of ADD-PATH extended NLRI */ u8 feed_state; /* Feed state (TX) for EoR, RR packets, see BFS_* */ u8 load_state; /* Load state (RX) for EoR, RR packets, see BFS_* */ + + u8 feed_out_table; /* Refeed into out_table */ }; struct bgp_prefix { - node buck_node; /* Node in per-bucket list */ + node buck_node_xx; /* Node in per-bucket list */ struct bgp_prefix *next; /* Node in prefix hash table */ + struct bgp_bucket *last; /* Last bucket sent with this prefix */ + struct bgp_bucket *cur; /* Current bucket (cur == last) if no update is required */ + btime lastmod; /* Last modification of this prefix */ u32 hash; u32 path_id; net_addr net[0]; @@ -405,11 +416,24 @@ struct bgp_prefix { struct bgp_bucket { node send_node; /* Node in send queue */ struct bgp_bucket *next; /* Node in bucket hash table */ - list prefixes; /* Prefixes in this bucket (struct bgp_prefix) */ + list prefixes; /* Prefixes to send in this bucket (struct bgp_prefix) */ u32 hash; /* Hash over extended attributes */ + u32 px_uc; /* How many prefixes are linking this bucket */ ea_list eattrs[0]; /* Per-bucket extended attributes */ }; +struct bgp_pending_tx { + resource r; + pool *pool; + + HASH(struct bgp_bucket) bucket_hash; /* Hash table of route buckets */ + struct bgp_bucket *withdraw_bucket; /* Withdrawn routes */ + list bucket_queue; /* Queue of buckets to send (struct bgp_bucket) */ + + HASH(struct bgp_prefix) prefix_hash; /* Prefixes to be sent */ + slab *prefix_slab; /* Slab holding prefix nodes */ +}; + struct bgp_export_state { struct bgp_proto *proto; struct bgp_channel *channel; @@ -473,13 +497,12 @@ struct bgp_parse_state { uint err_subcode; jmp_buf err_jmpbuf; - struct hostentry *hostentry; adata *mpls_labels; /* Cached state for bgp_rte_update() */ u32 last_id; struct rte_src *last_src; - rta *cached_rta; + ea_list *cached_ea; }; #define BGP_PORT 179 @@ -524,7 +547,7 @@ bgp_parse_error(struct bgp_parse_state *s, uint subcode) } -void bgp_start_timer(timer *t, uint value); +void bgp_start_timer(struct bgp_proto *p, timer *t, uint value); void bgp_check_config(struct bgp_config *c); void bgp_error(struct bgp_conn *c, unsigned code, unsigned subcode, byte *data, int len); void bgp_close_conn(struct bgp_conn *c); @@ -542,11 +565,17 @@ void bgp_stop(struct bgp_proto *p, int subcode, byte *data, uint len); const char *bgp_format_role_name(u8 role); static inline int -rte_resolvable(rte *rt) +rte_resolvable(const rte *rt) { - return rt->attrs->dest != RTD_UNREACHABLE; + eattr *nhea = ea_find(rt->attrs, &ea_gen_nexthop); + if (!nhea) + return 0; + + struct nexthop_adata *nhad = (void *) nhea->u.ptr; + return NEXTHOP_IS_REACHABLE(nhad) || (nhad->dest != RTD_UNREACHABLE); } +extern struct rte_owner_class bgp_rte_owner_class; #ifdef LOCAL_DEBUG #define BGP_FORCE_DEBUG 1 @@ -562,73 +591,61 @@ rte_resolvable(rte *rt) /* attrs.c */ -static inline eattr * -bgp_find_attr(ea_list *attrs, uint code) -{ - return ea_find(attrs, EA_CODE(PROTOCOL_BGP, code)); -} - eattr * -bgp_set_attr(ea_list **attrs, struct linpool *pool, uint code, uint flags, uintptr_t val); - -static inline void -bgp_set_attr_u32(ea_list **to, struct linpool *pool, uint code, uint flags, u32 val) -{ bgp_set_attr(to, pool, code, flags, (uintptr_t) val); } +bgp_find_attr(ea_list *attrs, uint code); -static inline void -bgp_set_attr_ptr(ea_list **to, struct linpool *pool, uint code, uint flags, const struct adata *val) -{ bgp_set_attr(to, pool, code, flags, (uintptr_t) val); } - -static inline void -bgp_set_attr_data(ea_list **to, struct linpool *pool, uint code, uint flags, void *data, uint len) -{ - struct adata *a = lp_alloc_adata(pool, len); - bmemcpy(a->data, data, len); - bgp_set_attr(to, pool, code, flags, (uintptr_t) a); -} - -#define bgp_unset_attr(to, pool, code) ea_unset_attr(to, pool, 0, code) +void bgp_set_attr_u32(ea_list **to, uint code, uint flags, u32 val); +void bgp_set_attr_ptr(ea_list **to, uint code, uint flags, const struct adata *ad); +void bgp_set_attr_data(ea_list **to, uint code, uint flags, void *data, uint len); +void bgp_unset_attr(ea_list **to, uint code); int bgp_encode_mp_reach_mrt(struct bgp_write_state *s, eattr *a, byte *buf, uint size); int bgp_encode_attrs(struct bgp_write_state *s, ea_list *attrs, byte *buf, byte *end); ea_list * bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len); -void bgp_finish_attrs(struct bgp_parse_state *s, rta *a); +void bgp_finish_attrs(struct bgp_parse_state *s, ea_list **to); + +void bgp_setup_out_table(struct bgp_channel *c); + +void bgp_init_pending_tx(struct bgp_channel *c); +void bgp_free_pending_tx(struct bgp_channel *c); -void bgp_init_bucket_table(struct bgp_channel *c); -void bgp_free_bucket_table(struct bgp_channel *c); -void bgp_free_bucket(struct bgp_channel *c, struct bgp_bucket *b); -void bgp_defer_bucket(struct bgp_channel *c, struct bgp_bucket *b); void bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b); +int bgp_done_bucket(struct bgp_channel *c, struct bgp_bucket *b); -void bgp_init_prefix_table(struct bgp_channel *c); -void bgp_free_prefix_table(struct bgp_channel *c); -void bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *bp); +void bgp_done_prefix(struct bgp_channel *c, struct bgp_prefix *px, struct bgp_bucket *buck); -int bgp_rte_better(struct rte *, struct rte *); -int bgp_rte_mergable(rte *pri, rte *sec); -int bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best); -struct rte *bgp_rte_modify_stale(struct rte *r, struct linpool *pool); -u32 bgp_rte_igp_metric(struct rte *); -void bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old); +int bgp_rte_better(const rte *, const rte *); +int bgp_rte_mergable(const rte *pri, const rte *sec); +int bgp_rte_recalculate(struct rtable_private *table, net *net, rte *new, rte *old, rte *old_best); +void bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *first, struct rt_pending_export *last, const rte **feed, uint count); +u32 bgp_rte_igp_metric(const rte *); +void bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old); int bgp_preexport(struct channel *, struct rte *); -int bgp_get_attr(const struct eattr *e, byte *buf, int buflen); -void bgp_get_route_info(struct rte *, byte *buf); -int bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad); +void bgp_get_route_info(const rte *, byte *); +int bgp_total_aigp_metric_(const rte *e, u64 *metric, const struct adata **ad); + +static inline struct bgp_proto *bgp_rte_proto(const rte *rte) +{ + return (rte->src->owner->class == &bgp_rte_owner_class) ? + SKIP_BACK(struct bgp_proto, p.sources, rte->src->owner) : NULL; +} #define BGP_AIGP_METRIC 1 #define BGP_AIGP_MAX U64(0xffffffffffffffff) static inline u64 -bgp_total_aigp_metric(rte *r) +bgp_total_aigp_metric(const rte *e) { u64 metric = BGP_AIGP_MAX; const struct adata *ad; - bgp_total_aigp_metric_(r, &metric, &ad); + bgp_total_aigp_metric_(e, &metric, &ad); return metric; } +void bgp_register_attrs(void); + /* packets.c */ @@ -640,6 +657,7 @@ void bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type) void bgp_kick_tx(void *vconn); void bgp_tx(struct birdsock *sk); int bgp_rx(struct birdsock *sk, uint size); +void bgp_uncork(void *vp); const char * bgp_error_dsc(unsigned code, unsigned subcode); void bgp_log_error(struct bgp_proto *p, u8 class, char *msg, unsigned code, unsigned subcode, byte *data, unsigned len); @@ -665,27 +683,32 @@ void bgp_update_next_hop(struct bgp_export_state *s, eattr *a, ea_list **to); #define BAF_DECODE_FLAGS 0x0100 /* Private flag - attribute flags are handled by the decode hook */ -#define BA_ORIGIN 0x01 /* RFC 4271 */ /* WM */ -#define BA_AS_PATH 0x02 /* WM */ -#define BA_NEXT_HOP 0x03 /* WM */ -#define BA_MULTI_EXIT_DISC 0x04 /* ON */ -#define BA_LOCAL_PREF 0x05 /* WD */ -#define BA_ATOMIC_AGGR 0x06 /* WD */ -#define BA_AGGREGATOR 0x07 /* OT */ -#define BA_COMMUNITY 0x08 /* RFC 1997 */ /* OT */ -#define BA_ORIGINATOR_ID 0x09 /* RFC 4456 */ /* ON */ -#define BA_CLUSTER_LIST 0x0a /* RFC 4456 */ /* ON */ -#define BA_MP_REACH_NLRI 0x0e /* RFC 4760 */ -#define BA_MP_UNREACH_NLRI 0x0f /* RFC 4760 */ -#define BA_EXT_COMMUNITY 0x10 /* RFC 4360 */ -#define BA_AS4_PATH 0x11 /* RFC 6793 */ -#define BA_AS4_AGGREGATOR 0x12 /* RFC 6793 */ -#define BA_AIGP 0x1a /* RFC 7311 */ -#define BA_LARGE_COMMUNITY 0x20 /* RFC 8092 */ +enum bgp_attr_id { + BA_ORIGIN = 0x01, /* RFC 4271 */ /* WM */ + BA_AS_PATH = 0x02, /* WM */ + BA_NEXT_HOP = 0x03, /* WM */ + BA_MULTI_EXIT_DISC = 0x04, /* ON */ + BA_LOCAL_PREF = 0x05, /* WD */ + BA_ATOMIC_AGGR = 0x06, /* WD */ + BA_AGGREGATOR = 0x07, /* OT */ + BA_COMMUNITY = 0x08, /* RFC 1997 */ /* OT */ + BA_ORIGINATOR_ID = 0x09, /* RFC 4456 */ /* ON */ + BA_CLUSTER_LIST = 0x0a, /* RFC 4456 */ /* ON */ + BA_MP_REACH_NLRI = 0x0e, /* RFC 4760 */ + BA_MP_UNREACH_NLRI = 0x0f, /* RFC 4760 */ + BA_EXT_COMMUNITY = 0x10, /* RFC 4360 */ + BA_AS4_PATH = 0x11, /* RFC 6793 */ + BA_AS4_AGGREGATOR = 0x12, /* RFC 6793 */ + BA_AIGP = 0x1a, /* RFC 7311 */ + BA_LARGE_COMMUNITY = 0x20, /* RFC 8092 */ #define BA_ONLY_TO_CUSTOMER 0x23 /* RFC 9234 */ /* Bird's private internal BGP attributes */ -#define BA_MPLS_LABEL_STACK 0xfe /* MPLS label stack transfer attribute */ + BA_MPLS_LABEL_STACK = 0x100, /* MPLS label stack transfer attribute */ + +/* Maximum */ + BGP_ATTR_MAX, +}; /* BGP connection states */ diff --git a/proto/bgp/config.Y b/proto/bgp/config.Y index a2dfa747..8b0d4356 100644 --- a/proto/bgp/config.Y +++ b/proto/bgp/config.Y @@ -19,18 +19,17 @@ CF_DECLS CF_KEYWORDS(BGP, LOCAL, NEIGHBOR, AS, HOLD, TIME, CONNECT, RETRY, KEEPALIVE, MULTIHOP, STARTUP, VIA, NEXT, HOP, SELF, DEFAULT, PATH, METRIC, ERROR, - START, DELAY, FORGET, WAIT, ENABLE, DISABLE, AFTER, BGP_PATH, - BGP_LOCAL_PREF, BGP_MED, BGP_ORIGIN, BGP_NEXT_HOP, BGP_ATOMIC_AGGR, - BGP_AGGREGATOR, BGP_COMMUNITY, BGP_EXT_COMMUNITY, BGP_LARGE_COMMUNITY, + START, DELAY, FORGET, WAIT, ENABLE, DISABLE, AFTER, + BGP_LOCAL_PREF, BGP_MED, SOURCE, ADDRESS, PASSWORD, RR, RS, CLIENT, CLUSTER, ID, AS4, ADVERTISE, IPV4, CAPABILITIES, LIMIT, PASSIVE, PREFER, OLDER, MISSING, LLADDR, - DROP, IGNORE, ROUTE, REFRESH, INTERPRET, COMMUNITIES, BGP_ORIGINATOR_ID, - BGP_CLUSTER_LIST, IGP, TABLE, GATEWAY, DIRECT, RECURSIVE, MED, TTL, + DROP, IGNORE, ROUTE, REFRESH, INTERPRET, COMMUNITIES, + IGP, TABLE, GATEWAY, DIRECT, RECURSIVE, MED, TTL, SECURITY, DETERMINISTIC, SECONDARY, ALLOW, BFD, ADD, PATHS, RX, TX, GRACEFUL, RESTART, AWARE, CHECK, LINK, PORT, EXTENDED, MESSAGES, SETKEY, STRICT, BIND, CONFEDERATION, MEMBER, MULTICAST, FLOW4, FLOW6, LONG, LIVED, STALE, IMPORT, IBGP, EBGP, MANDATORY, INTERNAL, EXTERNAL, SETS, - DYNAMIC, RANGE, NAME, DIGITS, BGP_AIGP, AIGP, ORIGINATE, COST, ENFORCE, + DYNAMIC, RANGE, NAME, DIGITS, AIGP, ORIGINATE, COST, ENFORCE, FIRST, FREE, VALIDATE, BASE, ROLE, ROLES, PEER, PROVIDER, CUSTOMER, RS_SERVER, RS_CLIENT, REQUIRE, BGP_OTC, GLOBAL) @@ -46,12 +45,14 @@ CF_KEYWORDS(CEASE, PREFIX, LIMIT, HIT, ADMINISTRATIVE, SHUTDOWN, RESET, PEER, CF_GRAMMAR /* Workaround for collisions between keywords and symbols */ -symbol: ROLE | PEER | PROVIDER | CUSTOMER | RS_SERVER | RS_CLIENT ; +toksym: ROLE | PEER | PROVIDER | CUSTOMER | RS_SERVER | RS_CLIENT ; +toksym: BGP_MED | BGP_LOCAL_PREF | SOURCE ; proto: bgp_proto '}' ; bgp_proto_start: proto_start BGP { this_proto = proto_config_new(&proto_bgp, $1); + this_proto->loop_order = DOMAIN_ORDER(proto); BGP_CFG->local_port = BGP_PORT; BGP_CFG->remote_port = BGP_PORT; BGP_CFG->multihop = -1; /* undefined */ @@ -329,43 +330,14 @@ bgp_channel_end: if (!this_channel->table) cf_error("Routing table not specified"); + if (BGP_CC->import_table) + this_channel->in_keep |= RIK_PREFILTER; + this_channel = NULL; }; bgp_proto_channel: bgp_channel_start bgp_channel_opt_list bgp_channel_end; - -dynamic_attr: BGP_ORIGIN - { $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_ENUM_BGP_ORIGIN, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); } ; -dynamic_attr: BGP_PATH - { $$ = f_new_dynamic_attr(EAF_TYPE_AS_PATH, T_PATH, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); } ; -dynamic_attr: BGP_NEXT_HOP - { $$ = f_new_dynamic_attr(EAF_TYPE_IP_ADDRESS, T_IP, EA_CODE(PROTOCOL_BGP, BA_NEXT_HOP)); } ; -dynamic_attr: BGP_MED - { $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(PROTOCOL_BGP, BA_MULTI_EXIT_DISC)); } ; -dynamic_attr: BGP_LOCAL_PREF - { $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(PROTOCOL_BGP, BA_LOCAL_PREF)); } ; -dynamic_attr: BGP_ATOMIC_AGGR - { $$ = f_new_dynamic_attr(EAF_TYPE_OPAQUE, T_ENUM_EMPTY, EA_CODE(PROTOCOL_BGP, BA_ATOMIC_AGGR)); } ; -dynamic_attr: BGP_AGGREGATOR - { $$ = f_new_dynamic_attr(EAF_TYPE_OPAQUE, T_ENUM_EMPTY, EA_CODE(PROTOCOL_BGP, BA_AGGREGATOR)); } ; -dynamic_attr: BGP_COMMUNITY - { $$ = f_new_dynamic_attr(EAF_TYPE_INT_SET, T_CLIST, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); } ; -dynamic_attr: BGP_ORIGINATOR_ID - { $$ = f_new_dynamic_attr(EAF_TYPE_ROUTER_ID, T_QUAD, EA_CODE(PROTOCOL_BGP, BA_ORIGINATOR_ID)); } ; -dynamic_attr: BGP_CLUSTER_LIST - { $$ = f_new_dynamic_attr(EAF_TYPE_INT_SET, T_CLIST, EA_CODE(PROTOCOL_BGP, BA_CLUSTER_LIST)); } ; -dynamic_attr: BGP_EXT_COMMUNITY - { $$ = f_new_dynamic_attr(EAF_TYPE_EC_SET, T_ECLIST, EA_CODE(PROTOCOL_BGP, BA_EXT_COMMUNITY)); } ; -dynamic_attr: BGP_AIGP - { $$ = f_new_dynamic_attr(EAF_TYPE_OPAQUE, T_ENUM_EMPTY, EA_CODE(PROTOCOL_BGP, BA_AIGP)); } ; -dynamic_attr: BGP_LARGE_COMMUNITY - { $$ = f_new_dynamic_attr(EAF_TYPE_LC_SET, T_LCLIST, EA_CODE(PROTOCOL_BGP, BA_LARGE_COMMUNITY)); } ; -dynamic_attr: BGP_OTC - { $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(PROTOCOL_BGP, BA_ONLY_TO_CUSTOMER)); } ; - - - CF_ENUM(T_ENUM_BGP_ORIGIN, ORIGIN_, IGP, EGP, INCOMPLETE) CF_CODE diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c index 16818cf3..546f0809 100644 --- a/proto/bgp/packets.c +++ b/proto/bgp/packets.c @@ -15,8 +15,8 @@ #include "nest/bird.h" #include "nest/iface.h" #include "nest/protocol.h" -#include "nest/route.h" -#include "nest/attrs.h" +#include "nest/rt.h" +#include "lib/attrs.h" #include "proto/mrt/mrt.h" #include "conf/conf.h" #include "lib/unaligned.h" @@ -973,7 +973,7 @@ bgp_rx_open(struct bgp_conn *conn, byte *pkt, uint len) conn->hold_time, conn->keepalive_time, p->remote_as, p->remote_id, conn->as4_session); bgp_schedule_packet(conn, NULL, PKT_KEEPALIVE); - bgp_start_timer(conn->hold_timer, conn->hold_time); + bgp_start_timer(p, conn->hold_timer, conn->hold_time); bgp_conn_enter_openconfirm_state(conn); } @@ -1002,7 +1002,7 @@ bgp_rx_open(struct bgp_conn *conn, byte *pkt, uint len) #define MISMATCHED_AF " - mismatched address family (%I for %s)" static void -bgp_apply_next_hop(struct bgp_parse_state *s, rta *a, ip_addr gw, ip_addr ll) +bgp_apply_next_hop(struct bgp_parse_state *s, ea_list **to, ip_addr gw, ip_addr ll) { struct bgp_proto *p = s->proto; struct bgp_channel *c = s->channel; @@ -1025,10 +1025,18 @@ bgp_apply_next_hop(struct bgp_parse_state *s, rta *a, ip_addr gw, ip_addr ll) if (nbr->scope == SCOPE_HOST) WITHDRAW(BAD_NEXT_HOP " - address %I is local", nbr->addr); - a->dest = RTD_UNICAST; - a->nh.gw = nbr->addr; - a->nh.iface = nbr->iface; - a->igp_metric = c->cf->cost; + ea_set_attr_u32(to, &ea_gen_igp_metric, 0, c->cf->cost); + + struct nexthop_adata nhad = { + .nh = { + .gw = nbr->addr, + .iface = nbr->iface, + }, + .ad = { + .length = sizeof nhad - sizeof nhad.ad, + }, + }; + ea_set_attr_data(to, &ea_gen_nexthop, 0, nhad.ad.data, nhad.ad.length); } else /* GW_RECURSIVE */ { @@ -1037,59 +1045,52 @@ bgp_apply_next_hop(struct bgp_parse_state *s, rta *a, ip_addr gw, ip_addr ll) rtable *tab = ipa_is_ip4(gw) ? c->igp_table_ip4 : c->igp_table_ip6; ip_addr lla = (c->cf->next_hop_prefer == NHP_LOCAL) ? ll : IPA_NONE; - s->hostentry = rt_get_hostentry(tab, gw, lla, c->c.table); - - if (!s->mpls) - rta_apply_hostentry(a, s->hostentry, NULL); - /* With MPLS, hostentry is applied later in bgp_apply_mpls_labels() */ + if (s->mpls) + { + u32 labels[BGP_MPLS_MAX]; + ea_set_hostentry(to, c->c.table, tab, gw, lla, BGP_MPLS_MAX, labels); + } + else + ea_set_hostentry(to, c->c.table, tab, gw, lla, 0, NULL); } } static void -bgp_apply_mpls_labels(struct bgp_parse_state *s, rta *a, u32 *labels, uint lnum) +bgp_apply_mpls_labels(struct bgp_parse_state *s, ea_list **to, u32 lnum, u32 labels[lnum]) { if (lnum > MPLS_MAX_LABEL_STACK) { REPORT("Too many MPLS labels ($u)", lnum); - a->dest = RTD_UNREACHABLE; - a->hostentry = NULL; - a->nh = (struct nexthop) { }; + ea_set_dest(to, 0, RTD_UNREACHABLE); return; } /* Handle implicit NULL as empty MPLS stack */ if ((lnum == 1) && (labels[0] == BGP_MPLS_NULL)) - lnum = 0; + lnum = s->mpls_labels->length = 0; if (s->channel->cf->gw_mode == GW_DIRECT) { - a->nh.labels = lnum; - memcpy(a->nh.label, labels, 4*lnum); + eattr *e = ea_find(*to, &ea_gen_nexthop); + struct { + struct nexthop_adata nhad; + u32 labels[MPLS_MAX_LABEL_STACK]; + } nh; + + memcpy(&nh.nhad, e->u.ptr, sizeof(struct adata) + e->u.ptr->length); + nh.nhad.nh.labels = lnum; + memcpy(nh.labels, labels, lnum * sizeof(u32)); + nh.nhad.ad.length = sizeof nh.nhad + lnum * sizeof(u32); } else /* GW_RECURSIVE */ { - mpls_label_stack ms; - - ms.len = lnum; - memcpy(ms.stack, labels, 4*lnum); - rta_apply_hostentry(a, s->hostentry, &ms); - } -} - -static void -bgp_apply_flow_validation(struct bgp_parse_state *s, const net_addr *n, rta *a) -{ - struct bgp_channel *c = s->channel; - int valid = rt_flowspec_check(c->base_table, c->c.table, n, a, s->proto->is_interior); - a->dest = valid ? RTD_NONE : RTD_UNREACHABLE; - - /* Invalidate cached rta if dest changes */ - if (s->cached_rta && (s->cached_rta->dest != a->dest)) - { - rta_free(s->cached_rta); - s->cached_rta = NULL; + eattr *e = ea_find(*to, &ea_gen_hostentry); + ASSERT_DIE(e); + struct hostentry_adata *head = (void *) e->u.ptr; + memcpy(&head->labels, labels, lnum * sizeof(u32)); + head->ad.length = (void *)(&head->labels[lnum]) - (void *) head->ad.data; } } @@ -1130,7 +1131,7 @@ bgp_use_next_hop(struct bgp_export_state *s, eattr *a) return 0; /* Do not pass NEXT_HOP between different VRFs */ - if (p->p.vrf_set && s->src && s->src->p.vrf_set && (p->p.vrf != s->src->p.vrf)) + if (p->p.vrf && s->src && s->src->p.vrf && (p->p.vrf != s->src->p.vrf)) return 0; /* Keep it when exported to internal peers */ @@ -1143,35 +1144,45 @@ bgp_use_next_hop(struct bgp_export_state *s, eattr *a) return p->neigh && (p->neigh->iface == ifa); } -static inline int +static inline struct nexthop * bgp_use_gateway(struct bgp_export_state *s) { struct bgp_proto *p = s->proto; struct bgp_channel *c = s->channel; - rta *ra = s->route->attrs; + ea_list *ra = s->route->attrs; /* Handle next hop self option - also applies to gateway */ if (c->cf->next_hop_self && bgp_match_src(s, c->cf->next_hop_self)) - return 0; + return NULL; + + eattr *nhea = ea_find(ra, &ea_gen_nexthop); + if (!nhea) + return NULL; /* We need one valid global gateway */ - if ((ra->dest != RTD_UNICAST) || ra->nh.next || ipa_zero(ra->nh.gw) || ipa_is_link_local(ra->nh.gw)) - return 0; + struct nexthop_adata *nhad = (struct nexthop_adata *) nhea->u.ptr; + if (!NEXTHOP_IS_REACHABLE(nhad) || + !NEXTHOP_ONE(nhad) || ipa_zero(nhad->nh.gw) || + ipa_is_link_local(nhad->nh.gw)) + return NULL; /* Check for non-matching AF */ - if ((ipa_is_ip4(ra->nh.gw) != bgp_channel_is_ipv4(c)) && !c->ext_next_hop) - return 0; + if ((ipa_is_ip4(nhad->nh.gw) != bgp_channel_is_ipv4(c)) && !c->ext_next_hop) + return NULL; /* Do not use gateway from different VRF */ - if (p->p.vrf_set && ra->nh.iface && (p->p.vrf != ra->nh.iface->master)) + if (p->p.vrf && nhad->nh.iface && (p->p.vrf != nhad->nh.iface->master)) return 0; /* Use it when exported to internal peers */ if (p->is_interior) - return 1; + return &nhad->nh; /* Use it when forwarded to single-hop BGP peer on on the same iface */ - return p->neigh && (p->neigh->iface == ra->nh.iface); + if (p->neigh && (p->neigh->iface == nhad->nh.iface)) + return &nhad->nh; + + return NULL; } static void @@ -1179,36 +1190,36 @@ bgp_update_next_hop_ip(struct bgp_export_state *s, eattr *a, ea_list **to) { if (!a || !bgp_use_next_hop(s, a)) { - if (bgp_use_gateway(s)) + struct nexthop *nhloc; + if (nhloc = bgp_use_gateway(s)) { - rta *ra = s->route->attrs; - ip_addr nh[1] = { ra->nh.gw }; - bgp_set_attr_data(to, s->pool, BA_NEXT_HOP, 0, nh, 16); + ip_addr nh[1] = { nhloc->gw }; + bgp_set_attr_data(to, BA_NEXT_HOP, 0, nh, 16); if (s->mpls) { u32 implicit_null = BGP_MPLS_NULL; - u32 *labels = ra->nh.labels ? ra->nh.label : &implicit_null; - uint lnum = ra->nh.labels ? ra->nh.labels : 1; - bgp_set_attr_data(to, s->pool, BA_MPLS_LABEL_STACK, 0, labels, lnum * 4); + u32 *labels = nhloc->labels ? nhloc->label : &implicit_null; + uint lnum = nhloc->labels ? nhloc->labels : 1; + bgp_set_attr_data(to, BA_MPLS_LABEL_STACK, 0, labels, lnum * 4); } else - bgp_unset_attr(to, s->pool, BA_MPLS_LABEL_STACK); + bgp_unset_attr(to, BA_MPLS_LABEL_STACK); } else { ip_addr nh[2] = { s->channel->next_hop_addr, s->channel->link_addr }; - bgp_set_attr_data(to, s->pool, BA_NEXT_HOP, 0, nh, ipa_nonzero(nh[1]) ? 32 : 16); + bgp_set_attr_data(to, BA_NEXT_HOP, 0, nh, ipa_nonzero(nh[1]) ? 32 : 16); s->local_next_hop = 1; /* TODO: Use local MPLS assigned label */ if (s->mpls) { u32 implicit_null = BGP_MPLS_NULL; - bgp_set_attr_data(to, s->pool, BA_MPLS_LABEL_STACK, 0, &implicit_null, 4); + bgp_set_attr_data(to, BA_MPLS_LABEL_STACK, 0, &implicit_null, 4); } else - bgp_unset_attr(to, s->pool, BA_MPLS_LABEL_STACK); + bgp_unset_attr(to, BA_MPLS_LABEL_STACK); } } @@ -1270,7 +1281,7 @@ bgp_encode_next_hop_ip(struct bgp_write_state *s, eattr *a, byte *buf, uint size } static void -bgp_decode_next_hop_ip(struct bgp_parse_state *s, byte *data, uint len, rta *a) +bgp_decode_next_hop_ip(struct bgp_parse_state *s, byte *data, uint len, ea_list **to) { struct bgp_channel *c = s->channel; struct adata *ad = lp_alloc_adata(s->pool, 32); @@ -1311,8 +1322,8 @@ bgp_decode_next_hop_ip(struct bgp_parse_state *s, byte *data, uint len, rta *a) // XXXX validate next hop - bgp_set_attr_ptr(&(a->eattrs), s->pool, BA_NEXT_HOP, 0, ad); - bgp_apply_next_hop(s, a, nh[0], nh[1]); + bgp_set_attr_ptr(to, BA_NEXT_HOP, 0, ad); + bgp_apply_next_hop(s, to, nh[0], nh[1]); } static uint @@ -1350,7 +1361,7 @@ bgp_encode_next_hop_vpn(struct bgp_write_state *s, eattr *a, byte *buf, uint siz } static void -bgp_decode_next_hop_vpn(struct bgp_parse_state *s, byte *data, uint len, rta *a) +bgp_decode_next_hop_vpn(struct bgp_parse_state *s, byte *data, uint len, ea_list **to) { struct bgp_channel *c = s->channel; struct adata *ad = lp_alloc_adata(s->pool, 32); @@ -1392,8 +1403,8 @@ bgp_decode_next_hop_vpn(struct bgp_parse_state *s, byte *data, uint len, rta *a) // XXXX validate next hop - bgp_set_attr_ptr(&(a->eattrs), s->pool, BA_NEXT_HOP, 0, ad); - bgp_apply_next_hop(s, a, nh[0], nh[1]); + bgp_set_attr_ptr(to, BA_NEXT_HOP, 0, ad); + bgp_apply_next_hop(s, to, nh[0], nh[1]); } @@ -1405,7 +1416,7 @@ bgp_encode_next_hop_none(struct bgp_write_state *s UNUSED, eattr *a UNUSED, byte } static void -bgp_decode_next_hop_none(struct bgp_parse_state *s UNUSED, byte *data UNUSED, uint len UNUSED, rta *a UNUSED) +bgp_decode_next_hop_none(struct bgp_parse_state *s UNUSED, byte *data UNUSED, uint len UNUSED, ea_list **to UNUSED) { /* * Although we expect no next hop and RFC 7606 7.11 states that attribute @@ -1417,11 +1428,11 @@ bgp_decode_next_hop_none(struct bgp_parse_state *s UNUSED, byte *data UNUSED, ui } static void -bgp_update_next_hop_none(struct bgp_export_state *s, eattr *a, ea_list **to) +bgp_update_next_hop_none(struct bgp_export_state *s UNUSED, eattr *a, ea_list **to) { /* NEXT_HOP shall not pass */ if (a) - bgp_unset_attr(to, s->pool, BA_NEXT_HOP); + bgp_unset_attr(to, BA_NEXT_HOP); } @@ -1430,15 +1441,17 @@ bgp_update_next_hop_none(struct bgp_export_state *s, eattr *a, ea_list **to) */ static void -bgp_rte_update(struct bgp_parse_state *s, const net_addr *n, u32 path_id, rta *a0) +bgp_rte_update(struct bgp_parse_state *s, const net_addr *n, u32 path_id, ea_list *a0) { if (path_id != s->last_id) { + rt_unlock_source(s->last_src); + s->last_src = rt_get_source(&s->proto->p, path_id); s->last_id = path_id; - rta_free(s->cached_rta); - s->cached_rta = NULL; + ea_free(s->cached_ea); + s->cached_ea = NULL; } if (!a0) @@ -1448,23 +1461,20 @@ bgp_rte_update(struct bgp_parse_state *s, const net_addr *n, u32 path_id, rta *a REPORT("Invalid route %N withdrawn", n); /* Route withdraw */ - rte_update3(&s->channel->c, n, NULL, s->last_src); + rte_update(&s->channel->c, n, NULL, s->last_src); return; } /* Prepare cached route attributes */ - if (s->cached_rta == NULL) - { - /* Workaround for rta_lookup() breaking eattrs */ - ea_list *ea = a0->eattrs; - s->cached_rta = rta_lookup(a0); - a0->eattrs = ea; - } + if (s->cached_ea == NULL) + s->cached_ea = ea_lookup(a0, 0); - rta *a = rta_clone(s->cached_rta); - rte *e = rte_get_temp(a, s->last_src); + rte e0 = { + .attrs = s->cached_ea, + .src = s->last_src, + }; - rte_update3(&s->channel->c, n, e, s->last_src); + rte_update(&s->channel->c, n, &e0, s->last_src); } static void @@ -1487,9 +1497,10 @@ bgp_encode_mpls_labels(struct bgp_write_state *s UNUSED, const adata *mpls, byte } static void -bgp_decode_mpls_labels(struct bgp_parse_state *s, byte **pos, uint *len, uint *pxlen, rta *a) +bgp_decode_mpls_labels(struct bgp_parse_state *s, byte **pos, uint *len, uint *pxlen, ea_list **to) { - u32 labels[BGP_MPLS_MAX], label; + u32 labels[BGP_MPLS_MAX]; + u32 label; uint lnum = 0; do { @@ -1508,26 +1519,15 @@ bgp_decode_mpls_labels(struct bgp_parse_state *s, byte **pos, uint *len, uint *p } while (!(label & BGP_MPLS_BOS)); - if (!a) + if (!*to) return; - /* Attach MPLS attribute unless we already have one */ - if (!s->mpls_labels) - { - s->mpls_labels = lp_alloc_adata(s->pool, 4*BGP_MPLS_MAX); - bgp_set_attr_ptr(&(a->eattrs), s->pool, BA_MPLS_LABEL_STACK, 0, s->mpls_labels); - } - - /* Overwrite data in the attribute */ - s->mpls_labels->length = 4*lnum; - memcpy(s->mpls_labels->data, labels, 4*lnum); - /* Update next hop entry in rta */ - bgp_apply_mpls_labels(s, a, labels, lnum); + bgp_apply_mpls_labels(s, to, lnum, labels); /* Attributes were changed, invalidate cached entry */ - rta_free(s->cached_rta); - s->cached_rta = NULL; + rta_free(s->cached_ea); + s->cached_ea = NULL; return; } @@ -1563,14 +1563,14 @@ bgp_encode_nlri_ip4(struct bgp_write_state *s, struct bgp_bucket *buck, byte *bu memcpy(pos, &a, b); ADVANCE(pos, size, b); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_ip4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_ip4(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -1596,7 +1596,7 @@ bgp_decode_nlri_ip4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) /* Decode MPLS labels */ if (s->mpls) - bgp_decode_mpls_labels(s, &pos, &len, &l, a); + bgp_decode_mpls_labels(s, &pos, &len, &l, &a); if (l > IP4_MAX_PREFIX_LENGTH) bgp_parse_error(s, 10); @@ -1648,14 +1648,14 @@ bgp_encode_nlri_ip6(struct bgp_write_state *s, struct bgp_bucket *buck, byte *bu memcpy(pos, &a, b); ADVANCE(pos, size, b); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_ip6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_ip6(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -1681,7 +1681,7 @@ bgp_decode_nlri_ip6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) /* Decode MPLS labels */ if (s->mpls) - bgp_decode_mpls_labels(s, &pos, &len, &l, a); + bgp_decode_mpls_labels(s, &pos, &len, &l, &a); if (l > IP6_MAX_PREFIX_LENGTH) bgp_parse_error(s, 10); @@ -1736,14 +1736,14 @@ bgp_encode_nlri_vpn4(struct bgp_write_state *s, struct bgp_bucket *buck, byte *b memcpy(pos, &a, b); ADVANCE(pos, size, b); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_vpn4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_vpn4(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -1769,7 +1769,7 @@ bgp_decode_nlri_vpn4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) /* Decode MPLS labels */ if (s->mpls) - bgp_decode_mpls_labels(s, &pos, &len, &l, a); + bgp_decode_mpls_labels(s, &pos, &len, &l, &a); /* Decode route distinguisher */ if (l < 64) @@ -1833,14 +1833,14 @@ bgp_encode_nlri_vpn6(struct bgp_write_state *s, struct bgp_bucket *buck, byte *b memcpy(pos, &a, b); ADVANCE(pos, size, b); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_vpn6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_vpn6(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -1866,7 +1866,7 @@ bgp_decode_nlri_vpn6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) /* Decode MPLS labels */ if (s->mpls) - bgp_decode_mpls_labels(s, &pos, &len, &l, a); + bgp_decode_mpls_labels(s, &pos, &len, &l, &a); /* Decode route distinguisher */ if (l < 64) @@ -1920,14 +1920,14 @@ bgp_encode_nlri_flow4(struct bgp_write_state *s, struct bgp_bucket *buck, byte * memcpy(pos, net->data, flen); ADVANCE(pos, size, flen); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_flow4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_flow4(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -1978,10 +1978,6 @@ bgp_decode_nlri_flow4(struct bgp_parse_state *s, byte *pos, uint len, rta *a) net_fill_flow4(n, px, pxlen, pos, flen); ADVANCE(pos, len, flen); - /* Apply validation procedure per RFC 8955 (6) */ - if (a && s->channel->cf->validate) - bgp_apply_flow_validation(s, n, a); - bgp_rte_update(s, n, path_id, a); } } @@ -2012,14 +2008,14 @@ bgp_encode_nlri_flow6(struct bgp_write_state *s, struct bgp_bucket *buck, byte * memcpy(pos, net->data, flen); ADVANCE(pos, size, flen); - bgp_free_prefix(s->channel, px); + bgp_done_prefix(s->channel, px, buck); } return pos - buf; } static void -bgp_decode_nlri_flow6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) +bgp_decode_nlri_flow6(struct bgp_parse_state *s, byte *pos, uint len, ea_list *a) { while (len) { @@ -2070,10 +2066,6 @@ bgp_decode_nlri_flow6(struct bgp_parse_state *s, byte *pos, uint len, rta *a) net_fill_flow6(n, px, pxlen, pos, flen); ADVANCE(pos, len, flen); - /* Apply validation procedure per RFC 8955 (6) */ - if (a && s->channel->cf->validate) - bgp_apply_flow_validation(s, n, a); - bgp_rte_update(s, n, path_id, a); } } @@ -2250,6 +2242,8 @@ bgp_create_ip_reach(struct bgp_write_state *s, struct bgp_bucket *buck, byte *bu * var IPv4 Network Layer Reachability Information */ + ASSERT_DIE(s->channel->ptx->withdraw_bucket != buck); + int lr, la; la = bgp_encode_attrs(s, buck->eattrs, buf+4, buf + MAX_ATTRS_LENGTH); @@ -2271,6 +2265,8 @@ bgp_create_ip_reach(struct bgp_write_state *s, struct bgp_bucket *buck, byte *bu static byte * bgp_create_mp_reach(struct bgp_write_state *s, struct bgp_bucket *buck, byte *buf, byte *end) { + ASSERT_DIE(s->channel->ptx->withdraw_bucket != buck); + /* * 2 B IPv4 Withdrawn Routes Length (zero) * --- IPv4 Withdrawn Routes NLRI (unused) @@ -2409,7 +2405,7 @@ again: ; }; /* Try unreachable bucket */ - if ((buck = c->withdraw_bucket) && !EMPTY_LIST(buck->prefixes)) + if ((buck = c->ptx->withdraw_bucket) && !EMPTY_LIST(buck->prefixes)) { res = (c->afi == BGP_AF_IPV4) && !c->ext_next_hop ? bgp_create_ip_unreach(&s, buck, buf, end): @@ -2419,14 +2415,13 @@ again: ; } /* Try reachable buckets */ - if (!EMPTY_LIST(c->bucket_queue)) + if (!EMPTY_LIST(c->ptx->bucket_queue)) { - buck = HEAD(c->bucket_queue); + buck = HEAD(c->ptx->bucket_queue); /* Cleanup empty buckets */ - if (EMPTY_LIST(buck->prefixes)) + if (bgp_done_bucket(c, buck)) { - bgp_free_bucket(c, buck); lp_restore(tmp_linpool, &tmpp); goto again; } @@ -2435,10 +2430,7 @@ again: ; bgp_create_ip_reach(&s, buck, buf, end): bgp_create_mp_reach(&s, buck, buf, end); - if (EMPTY_LIST(buck->prefixes)) - bgp_free_bucket(c, buck); - else - bgp_defer_bucket(c, buck); + bgp_done_bucket(c, buck); if (!res) { @@ -2522,7 +2514,6 @@ static inline void bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_list *ea, byte *nh, uint nh_len) { struct bgp_channel *c = bgp_get_channel(s->proto, afi); - rta *a = NULL; if (!c) DISCARD(BAD_AFI, BGP_AFI(afi), BGP_SAFI(afi)); @@ -2533,6 +2524,7 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis s->last_id = 0; s->last_src = s->proto->p.main_source; + rt_lock_source(s->last_src); /* * IPv4 BGP and MP-BGP may be used together in one update, therefore we do not @@ -2543,26 +2535,24 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis if (ea) { - a = allocz(RTA_MAX_SIZE); + ea_set_attr_data(&ea, &ea_gen_from, 0, &s->proto->remote_ip, sizeof(ip_addr)); + ea_set_attr_u32(&ea, &ea_gen_preference, 0, c->c.preference); + ea_set_attr_u32(&ea, &ea_gen_source, 0, RTS_BGP); - a->source = RTS_BGP; - a->scope = SCOPE_UNIVERSE; - a->from = s->proto->remote_ip; - a->eattrs = ea; - a->pref = c->c.preference; - - c->desc->decode_next_hop(s, nh, nh_len, a); - bgp_finish_attrs(s, a); + c->desc->decode_next_hop(s, nh, nh_len, &ea); + bgp_finish_attrs(s, &ea); /* Handle withdraw during next hop decoding */ if (s->err_withdraw) - a = NULL; + ea = NULL; } - c->desc->decode_nlri(s, nlri, len, a); + c->desc->decode_nlri(s, nlri, len, ea); + + rta_free(s->cached_ea); + s->cached_ea = NULL; - rta_free(s->cached_rta); - s->cached_rta = NULL; + rt_unlock_source(s->last_src); } static void @@ -2582,7 +2572,7 @@ bgp_rx_update(struct bgp_conn *conn, byte *pkt, uint len) if (conn->state != BS_ESTABLISHED) { bgp_error(conn, 5, fsm_err_subcode[conn->state], NULL, 0); return; } - bgp_start_timer(conn->hold_timer, conn->hold_time); + bgp_start_timer(p, conn->hold_timer, conn->hold_time); struct lp_state tmpp; lp_save(tmp_linpool, &tmpp); @@ -2667,7 +2657,7 @@ bgp_rx_update(struct bgp_conn *conn, byte *pkt, uint len) ea, s.mp_next_hop_data, s.mp_next_hop_len); done: - rta_free(s.cached_rta); + rta_free(s.cached_ea); lp_restore(tmp_linpool, &tmpp); return; } @@ -2923,7 +2913,7 @@ bgp_fire_tx(struct bgp_conn *conn) { conn->packets_to_send &= ~(1 << PKT_KEEPALIVE); BGP_TRACE(D_PACKETS, "Sending KEEPALIVE"); - bgp_start_timer(conn->keepalive_timer, conn->keepalive_time); + bgp_start_timer(p, conn->keepalive_timer, conn->keepalive_time); return bgp_send(conn, PKT_KEEPALIVE, BGP_HEADER_LENGTH); } else while (conn->channels_to_send) @@ -2991,7 +2981,11 @@ bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type) { ASSERT(conn->sk); - DBG("BGP: Scheduling packet type %d\n", type); + struct bgp_proto *p = conn->bgp; + if (c) + BGP_TRACE(D_PACKETS, "Scheduling packet type %d for channel %s", type, c->c.name); + else + BGP_TRACE(D_PACKETS, "Scheduling packet type %d", type); if (c) { @@ -3008,7 +3002,7 @@ bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type) conn->packets_to_send |= 1 << type; if ((conn->sk->tpos == conn->sk->tbuf) && !ev_active(conn->tx_ev)) - ev_schedule(conn->tx_ev); + proto_send_event(&p->p, conn->tx_ev); } void bgp_kick_tx(void *vconn) @@ -3021,7 +3015,7 @@ bgp_kick_tx(void *vconn) ; if (!max && !ev_active(conn->tx_ev)) - ev_schedule(conn->tx_ev); + proto_send_event(&conn->bgp->p, conn->tx_ev); } void @@ -3029,13 +3023,14 @@ bgp_tx(sock *sk) { struct bgp_conn *conn = sk->data; + ASSERT_DIE(birdloop_inside(conn->bgp->p.loop)); DBG("BGP: TX hook\n"); uint max = 1024; while (--max && (bgp_fire_tx(conn) > 0)) ; if (!max && !ev_active(conn->tx_ev)) - ev_schedule(conn->tx_ev); + proto_send_event(&conn->bgp->p, conn->tx_ev); } @@ -3224,7 +3219,7 @@ bgp_rx_keepalive(struct bgp_conn *conn) struct bgp_proto *p = conn->bgp; BGP_TRACE(D_PACKETS, "Got KEEPALIVE"); - bgp_start_timer(conn->hold_timer, conn->hold_time); + bgp_start_timer(p, conn->hold_timer, conn->hold_time); if (conn->state == BS_OPENCONFIRM) { bgp_conn_enter_established_state(conn); return; } @@ -3266,6 +3261,30 @@ bgp_rx_packet(struct bgp_conn *conn, byte *pkt, uint len) } } +void +bgp_uncork(void *vp) +{ + /* The uncork event is run from &main_birdloop and there is no useful way how + * to assign the target loop to it, thus we have to lock it ourselves. */ + + struct bgp_proto *p = vp; + if (!p) + return; + + birdloop_enter(p->p.loop); + + if (p && p->conn && (p->conn->state == BS_ESTABLISHED) && !p->conn->sk->rx_hook) + { + struct birdsock *sk = p->conn->sk; + ASSERT_DIE(sk->rpos > sk->rbuf); + sk_resume_rx(p->p.loop, sk, bgp_rx); + bgp_rx(sk, sk->rpos - sk->rbuf); + BGP_TRACE(D_PACKETS, "Uncorked"); + } + + birdloop_leave(p->p.loop); +} + /** * bgp_rx - handle received data * @sk: socket @@ -3280,6 +3299,7 @@ int bgp_rx(sock *sk, uint size) { struct bgp_conn *conn = sk->data; + struct bgp_proto *p = conn->bgp; byte *pkt_start = sk->rbuf; byte *end = pkt_start + size; uint i, len; @@ -3289,6 +3309,12 @@ bgp_rx(sock *sk, uint size) { if ((conn->state == BS_CLOSE) || (conn->sk != sk)) return 0; + if ((conn->state == BS_ESTABLISHED) && rt_cork_check(conn->bgp->uncork_ev)) + { + sk_pause_rx(p->p.loop, sk); + BGP_TRACE(D_PACKETS, "Corked"); + return 0; + } for(i=0; i<16; i++) if (pkt_start[i] != 0xff) { |