diff options
Diffstat (limited to 'src/plugins/gbp')
| -rw-r--r-- | src/plugins/gbp/gbp.api | 38 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_api.c | 223 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_bridge_domain.c | 67 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_bridge_domain.h | 1 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_classify.c | 2 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_contract.c | 572 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_contract.h | 120 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_endpoint.c | 1137 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_endpoint.h | 166 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_endpoint_group.c | 32 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_endpoint_group.h | 7 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_learn.c | 26 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_policy.c | 103 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_policy_dpo.c | 52 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_policy_dpo.h | 4 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_recirc.c | 15 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_route_domain.c | 6 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_route_domain.h | 1 | ||||
| -rw-r--r-- | src/plugins/gbp/gbp_vxlan.c | 2 |
19 files changed, 2013 insertions, 561 deletions
diff --git a/src/plugins/gbp/gbp.api b/src/plugins/gbp/gbp.api index 1fb9073d190..9af8b35b870 100644 --- a/src/plugins/gbp/gbp.api +++ b/src/plugins/gbp/gbp.api @@ -251,11 +251,47 @@ define gbp_subnet_details vl_api_gbp_subnet_t subnet; }; -typeonly define gbp_contract +typedef gbp_next_hop +{ + vl_api_address_t ip; + vl_api_mac_address_t mac; + u32 bd_id; + u32 rd_id; +}; + +enum gbp_hash_mode +{ + GBP_API_HASH_MODE_SRC_IP, + GBP_API_HASH_MODE_DST_IP, +}; + +typedef gbp_next_hop_set +{ + vl_api_gbp_hash_mode_t hash_mode; + u8 n_nhs; + vl_api_gbp_next_hop_t nhs[8]; +}; + +enum gbp_rule_action +{ + GBP_API_RULE_PERMIT, + GBP_API_RULE_DENY, + GBP_API_RULE_REDIRECT, +}; + +typedef gbp_rule +{ + vl_api_gbp_rule_action_t action; + vl_api_gbp_next_hop_set_t nh_set; +}; + +typedef gbp_contract { u16 src_epg; u16 dst_epg; u32 acl_index; + u8 n_rules; + vl_api_gbp_rule_t rules[n_rules]; }; autoreply define gbp_contract_add_del diff --git a/src/plugins/gbp/gbp_api.c b/src/plugins/gbp/gbp_api.c index 47823d5cd9b..6a000721a01 100644 --- a/src/plugins/gbp/gbp_api.c +++ b/src/plugins/gbp/gbp_api.c @@ -156,16 +156,21 @@ vl_api_gbp_endpoint_add_t_handler (vl_api_gbp_endpoint_add_t * mp) ip_address_decode (&mp->endpoint.tun.src, &tun_src); ip_address_decode (&mp->endpoint.tun.dst, &tun_dst); - rv = gbp_endpoint_update (sw_if_index, ips, &mac, - ntohs (mp->endpoint.epg_id), - gef, &tun_src, &tun_dst, &handle); + rv = gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_CP, + sw_if_index, ips, &mac, + INDEX_INVALID, INDEX_INVALID, + ntohs (mp->endpoint.epg_id), + gef, &tun_src, &tun_dst, &handle); } else { - rv = gbp_endpoint_update (sw_if_index, ips, &mac, - ntohs (mp->endpoint.epg_id), - gef, NULL, NULL, &handle); + rv = gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_CP, + sw_if_index, ips, &mac, + INDEX_INVALID, INDEX_INVALID, + ntohs (mp->endpoint.epg_id), + gef, NULL, NULL, &handle); } + vec_free (ips); BAD_SW_IF_INDEX_LABEL; /* *INDENT-OFF* */ @@ -182,7 +187,7 @@ vl_api_gbp_endpoint_del_t_handler (vl_api_gbp_endpoint_del_t * mp) vl_api_gbp_endpoint_del_reply_t *rmp; int rv = 0; - gbp_endpoint_delete (ntohl (mp->handle)); + gbp_endpoint_unlock (GBP_ENDPOINT_SRC_CP, ntohl (mp->handle)); REPLY_MACRO (VL_API_GBP_ENDPOINT_DEL_REPLY + GBP_MSG_BASE); } @@ -210,6 +215,8 @@ static walk_rc_t gbp_endpoint_send_details (index_t gei, void *args) { vl_api_gbp_endpoint_details_t *mp; + gbp_endpoint_loc_t *gel; + gbp_endpoint_fwd_t *gef; gbp_endpoint_t *ge; gbp_walk_ctx_t *ctx; u8 n_ips, ii; @@ -217,7 +224,7 @@ gbp_endpoint_send_details (index_t gei, void *args) ctx = args; ge = gbp_endpoint_get (gei); - n_ips = vec_len (ge->ge_ips); + n_ips = vec_len (ge->ge_key.gek_ips); mp = vl_msg_api_alloc (sizeof (*mp) + (sizeof (*mp->endpoint.ips) * n_ips)); if (!mp) return 1; @@ -226,28 +233,32 @@ gbp_endpoint_send_details (index_t gei, void *args) mp->_vl_msg_id = ntohs (VL_API_GBP_ENDPOINT_DETAILS + GBP_MSG_BASE); mp->context = ctx->context; + gel = &ge->ge_locs[0]; + gef = &ge->ge_fwd; + if (gbp_endpoint_is_remote (ge)) { - mp->endpoint.sw_if_index = ntohl (ge->tun.ge_parent_sw_if_index); - ip_address_encode (&ge->tun.ge_src, IP46_TYPE_ANY, + mp->endpoint.sw_if_index = ntohl (gel->tun.gel_parent_sw_if_index); + ip_address_encode (&gel->tun.gel_src, IP46_TYPE_ANY, &mp->endpoint.tun.src); - ip_address_encode (&ge->tun.ge_dst, IP46_TYPE_ANY, + ip_address_encode (&gel->tun.gel_dst, IP46_TYPE_ANY, &mp->endpoint.tun.dst); } else { - mp->endpoint.sw_if_index = ntohl (ge->ge_sw_if_index); + mp->endpoint.sw_if_index = ntohl (gef->gef_itf); } - mp->endpoint.epg_id = ntohs (ge->ge_epg_id); + mp->endpoint.epg_id = ntohs (ge->ge_fwd.gef_epg_id); mp->endpoint.n_ips = n_ips; - mp->endpoint.flags = gbp_endpoint_flags_encode (ge->ge_flags); + mp->endpoint.flags = gbp_endpoint_flags_encode (gef->gef_flags); mp->handle = htonl (gei); mp->age = vlib_time_now (vlib_get_main ()) - ge->ge_last_time; - mac_address_encode (&ge->ge_mac, &mp->endpoint.mac); + mac_address_encode (&ge->ge_key.gek_mac, &mp->endpoint.mac); - vec_foreach_index (ii, ge->ge_ips) + vec_foreach_index (ii, ge->ge_key.gek_ips) { - ip_address_encode (&ge->ge_ips[ii], IP46_TYPE_ANY, &mp->endpoint.ips[ii]); + ip_address_encode (&ge->ge_key.gek_ips[ii].fp_addr, + IP46_TYPE_ANY, &mp->endpoint.ips[ii]); } vl_api_send_msg (ctx->reg, (u8 *) mp); @@ -672,20 +683,186 @@ vl_api_gbp_recirc_dump_t_handler (vl_api_gbp_recirc_dump_t * mp) gbp_recirc_walk (gbp_recirc_send_details, &ctx); } +static int +gbp_contract_rule_action_deocde (vl_api_gbp_rule_action_t in, + gbp_rule_action_t * out) +{ + in = clib_net_to_host_u32 (in); + + switch (in) + { + case GBP_API_RULE_PERMIT: + *out = GBP_RULE_PERMIT; + return (0); + case GBP_API_RULE_DENY: + *out = GBP_RULE_DENY; + return (0); + case GBP_API_RULE_REDIRECT: + *out = GBP_RULE_REDIRECT; + return (0); + } + + return (-1); +} + +static int +gbp_hash_mode_decode (vl_api_gbp_hash_mode_t in, gbp_hash_mode_t * out) +{ + in = clib_net_to_host_u32 (in); + + switch (in) + { + case GBP_API_HASH_MODE_SRC_IP: + *out = GBP_HASH_MODE_SRC_IP; + return (0); + case GBP_API_HASH_MODE_DST_IP: + *out = GBP_HASH_MODE_DST_IP; + return (0); + } + + return (-2); +} + +static int +gbp_next_hop_decode (const vl_api_gbp_next_hop_t * in, index_t * gnhi) +{ + ip46_address_t ip; + mac_address_t mac; + index_t grd, gbd; + + gbd = gbp_bridge_domain_find_and_lock (ntohl (in->bd_id)); + + if (INDEX_INVALID == gbd) + return (VNET_API_ERROR_BD_NOT_MODIFIABLE); + + grd = gbp_route_domain_find_and_lock (ntohl (in->rd_id)); + + if (INDEX_INVALID == grd) + return (VNET_API_ERROR_NO_SUCH_FIB); + + ip_address_decode (&in->ip, &ip); + mac_address_decode (&in->mac, &mac); + + *gnhi = gbp_next_hop_alloc (&ip, grd, &mac, gbd); + + return (0); +} + +static int +gbp_next_hop_set_decode (const vl_api_gbp_next_hop_set_t * in, + gbp_hash_mode_t * hash_mode, index_t ** out) +{ + + index_t *gnhis = NULL; + int rv; + u8 ii; + + rv = gbp_hash_mode_decode (in->hash_mode, hash_mode); + + if (0 != rv) + return rv; + + vec_validate (gnhis, in->n_nhs - 1); + + for (ii = 0; ii < in->n_nhs; ii++) + { + rv = gbp_next_hop_decode (&in->nhs[ii], &gnhis[ii]); + + if (0 != rv) + { + vec_free (gnhis); + break; + } + } + + *out = gnhis; + return (rv); +} + +static int +gbp_contract_rule_decode (const vl_api_gbp_rule_t * in, index_t * gui) +{ + gbp_hash_mode_t hash_mode; + gbp_rule_action_t action; + index_t *nhs = NULL; + int rv; + + rv = gbp_contract_rule_action_deocde (in->action, &action); + + if (0 != rv) + return rv; + + if (GBP_RULE_REDIRECT == action) + { + rv = gbp_next_hop_set_decode (&in->nh_set, &hash_mode, &nhs); + + if (0 != rv) + return (rv); + } + else + { + hash_mode = GBP_HASH_MODE_SRC_IP; + } + + *gui = gbp_rule_alloc (action, hash_mode, nhs); + + return (rv); +} + +static int +gbp_contract_rules_decode (u8 n_rules, + const vl_api_gbp_rule_t * rules, index_t ** out) +{ + index_t *guis = NULL; + int rv; + u8 ii; + + if (0 == n_rules) + { + *out = NULL; + return (0); + } + + vec_validate (guis, n_rules - 1); + + for (ii = 0; ii < n_rules; ii++) + { + rv = gbp_contract_rule_decode (&rules[ii], &guis[ii]); + + if (0 != rv) + { + vec_free (guis); + return (rv); + } + } + + *out = guis; + return (rv); +} + static void vl_api_gbp_contract_add_del_t_handler (vl_api_gbp_contract_add_del_t * mp) { vl_api_gbp_contract_add_del_reply_t *rmp; + index_t *rules; int rv = 0; if (mp->is_add) - gbp_contract_update (ntohs (mp->contract.src_epg), - ntohs (mp->contract.dst_epg), - ntohl (mp->contract.acl_index)); + { + rv = gbp_contract_rules_decode (mp->contract.n_rules, + mp->contract.rules, &rules); + if (0 != rv) + goto out; + + rv = gbp_contract_update (ntohs (mp->contract.src_epg), + ntohs (mp->contract.dst_epg), + ntohl (mp->contract.acl_index), rules); + } else - gbp_contract_delete (ntohs (mp->contract.src_epg), - ntohs (mp->contract.dst_epg)); + rv = gbp_contract_delete (ntohs (mp->contract.src_epg), + ntohs (mp->contract.dst_epg)); +out: REPLY_MACRO (VL_API_GBP_CONTRACT_ADD_DEL_REPLY + GBP_MSG_BASE); } @@ -706,7 +883,7 @@ gbp_contract_send_details (gbp_contract_t * gbpc, void *args) mp->contract.src_epg = ntohs (gbpc->gc_key.gck_src); mp->contract.dst_epg = ntohs (gbpc->gc_key.gck_dst); - mp->contract.acl_index = ntohl (gbpc->gc_value.gc_acl_index); + // mp->contract.acl_index = ntohl (gbpc->gc_value.gc_acl_index); vl_api_send_msg (ctx->reg, (u8 *) mp); diff --git a/src/plugins/gbp/gbp_bridge_domain.c b/src/plugins/gbp/gbp_bridge_domain.c index 1a1a7bd3383..298819f87f1 100644 --- a/src/plugins/gbp/gbp_bridge_domain.c +++ b/src/plugins/gbp/gbp_bridge_domain.c @@ -41,6 +41,12 @@ vlib_log_class_t gb_logger; #define GBP_BD_DBG(...) \ vlib_log_debug (gb_logger, __VA_ARGS__); +index_t +gbp_bridge_domain_index (const gbp_bridge_domain_t * gbd) +{ + return (gbd - gbp_bridge_domain_pool); +} + static void gbp_bridge_domain_lock (index_t i) { @@ -96,6 +102,38 @@ gbp_bridge_domain_db_remove (gbp_bridge_domain_t * gb) gbp_bridge_domain_db.gbd_by_bd_index[gb->gb_bd_index] = INDEX_INVALID; } +static u8 * +format_gbp_bridge_domain_ptr (u8 * s, va_list * args) +{ + gbp_bridge_domain_t *gb = va_arg (*args, gbp_bridge_domain_t *); + vnet_main_t *vnm = vnet_get_main (); + + if (NULL != gb) + s = format (s, "[%d] bd:[%d,%d], bvi:%U uu-flood:%U locks:%d", + gb - gbp_bridge_domain_pool, + gb->gb_bd_id, + gb->gb_bd_index, + format_vnet_sw_if_index_name, vnm, gb->gb_bvi_sw_if_index, + format_vnet_sw_if_index_name, vnm, gb->gb_uu_fwd_sw_if_index, + gb->gb_locks); + else + s = format (s, "NULL"); + + return (s); +} + +u8 * +format_gbp_bridge_domain (u8 * s, va_list * args) +{ + index_t gbi = va_arg (*args, index_t); + + s = + format (s, "%U", format_gbp_bridge_domain_ptr, + gbp_bridge_domain_get (gbi)); + + return (s); +} + int gbp_bridge_domain_add_and_lock (u32 bd_id, gbp_bridge_domain_flags_t flags, @@ -158,7 +196,7 @@ gbp_bridge_domain_add_and_lock (u32 bd_id, gb->gb_locks++; } - GBP_BD_DBG ("add: %U", format_gbp_bridge_domain, gb); + GBP_BD_DBG ("add: %U", format_gbp_bridge_domain_ptr, gb); return (0); } @@ -174,7 +212,7 @@ gbp_bridge_domain_unlock (index_t index) if (0 == gb->gb_locks) { - GBP_BD_DBG ("destroy: %U", format_gbp_bridge_domain, gb); + GBP_BD_DBG ("destroy: %U", format_gbp_bridge_domain_ptr, gb); l2fib_del_entry (vnet_sw_interface_get_hw_address (vnet_get_main (), gb->gb_bvi_sw_if_index), @@ -204,8 +242,7 @@ gbp_bridge_domain_delete (u32 bd_id) if (INDEX_INVALID != gbi) { - GBP_BD_DBG ("del: %U", format_gbp_bridge_domain, - gbp_bridge_domain_get (gbi)); + GBP_BD_DBG ("del: %U", format_gbp_bridge_domain, gbi); gbp_bridge_domain_unlock (gbi); return (0); @@ -287,33 +324,13 @@ VLIB_CLI_COMMAND (gbp_bridge_domain_cli_node, static) = { .function = gbp_bridge_domain_cli, }; -u8 * -format_gbp_bridge_domain (u8 * s, va_list * args) -{ - gbp_bridge_domain_t *gb = va_arg (*args, gbp_bridge_domain_t*); - vnet_main_t *vnm = vnet_get_main (); - - if (NULL != gb) - s = format (s, "[%d] bd:[%d,%d], bvi:%U uu-flood:%U locks:%d", - gb - gbp_bridge_domain_pool, - gb->gb_bd_id, - gb->gb_bd_index, - format_vnet_sw_if_index_name, vnm, gb->gb_bvi_sw_if_index, - format_vnet_sw_if_index_name, vnm, gb->gb_uu_fwd_sw_if_index, - gb->gb_locks); - else - s = format (s, "NULL"); - - return (s); -} - static int gbp_bridge_domain_show_one (gbp_bridge_domain_t *gb, void *ctx) { vlib_main_t *vm; vm = ctx; - vlib_cli_output (vm, " %U",format_gbp_bridge_domain, gb); + vlib_cli_output (vm, " %U", format_gbp_bridge_domain_ptr, gb); return (1); } diff --git a/src/plugins/gbp/gbp_bridge_domain.h b/src/plugins/gbp/gbp_bridge_domain.h index 4135c2d8232..cc03320871a 100644 --- a/src/plugins/gbp/gbp_bridge_domain.h +++ b/src/plugins/gbp/gbp_bridge_domain.h @@ -77,6 +77,7 @@ extern int gbp_bridge_domain_add_and_lock (u32 bd_id, extern void gbp_bridge_domain_unlock (index_t gbi); extern index_t gbp_bridge_domain_find_and_lock (u32 bd_id); extern int gbp_bridge_domain_delete (u32 bd_id); +extern index_t gbp_bridge_domain_index (const gbp_bridge_domain_t *); typedef int (*gbp_bridge_domain_cb_t) (gbp_bridge_domain_t * gb, void *ctx); extern void gbp_bridge_domain_walk (gbp_bridge_domain_cb_t bgpe, void *ctx); diff --git a/src/plugins/gbp/gbp_classify.c b/src/plugins/gbp/gbp_classify.c index fb57426e4a7..4846911ed65 100644 --- a/src/plugins/gbp/gbp_classify.c +++ b/src/plugins/gbp/gbp_classify.c @@ -154,7 +154,7 @@ gbp_classify_inline (vlib_main_t * vm, } if (PREDICT_TRUE (NULL != ge0)) - src_epg = ge0->ge_epg_id; + src_epg = ge0->ge_fwd.gef_epg_id; else src_epg = EPG_INVALID; } diff --git a/src/plugins/gbp/gbp_contract.c b/src/plugins/gbp/gbp_contract.c index 32dd94441db..ec545693dd3 100644 --- a/src/plugins/gbp/gbp_contract.c +++ b/src/plugins/gbp/gbp_contract.c @@ -16,21 +16,411 @@ */ #include <plugins/gbp/gbp.h> +#include <plugins/gbp/gbp_bridge_domain.h> +#include <plugins/gbp/gbp_route_domain.h> +#include <plugins/gbp/gbp_policy_dpo.h> + +#include <vnet/dpo/load_balance.h> +#include <vnet/dpo/drop_dpo.h> /** * Single contract DB instance */ gbp_contract_db_t gbp_contract_db; -void -gbp_contract_update (epg_id_t src_epg, epg_id_t dst_epg, u32 acl_index) +gbp_contract_t *gbp_contract_pool; + +vlib_log_class_t gc_logger; + +fib_node_type_t gbp_next_hop_fib_type; + +gbp_rule_t *gbp_rule_pool; +gbp_next_hop_t *gbp_next_hop_pool; + +#define GBP_CONTRACT_DBG(...) \ + vlib_log_notice (gc_logger, __VA_ARGS__); + +index_t +gbp_rule_alloc (gbp_rule_action_t action, + gbp_hash_mode_t hash_mode, index_t * nhs) { - gbp_main_t *gm = &gbp_main; - u32 *acl_vec = 0; - gbp_contract_value_t value = { - .gc_lc_index = ~0, - .gc_acl_index = ~0, + gbp_rule_t *gu; + + pool_get_zero (gbp_rule_pool, gu); + + gu->gu_nhs = nhs; + gu->gu_action = action; + + return (gu - gbp_rule_pool); +} + +index_t +gbp_next_hop_alloc (const ip46_address_t * ip, + index_t grd, const mac_address_t * mac, index_t gbd) +{ + fib_protocol_t fproto; + gbp_next_hop_t *gnh; + + pool_get_zero (gbp_next_hop_pool, gnh); + + fib_node_init (&gnh->gnh_node, gbp_next_hop_fib_type); + + ip46_address_copy (&gnh->gnh_ip, ip); + mac_address_copy (&gnh->gnh_mac, mac); + + gnh->gnh_rd = grd; + gnh->gnh_bd = gbd; + + FOR_EACH_FIB_IP_PROTOCOL (fproto) gnh->gnh_ai[fproto] = INDEX_INVALID; + + return (gnh - gbp_next_hop_pool); +} + +static inline gbp_next_hop_t * +gbp_next_hop_get (index_t gui) +{ + return (pool_elt_at_index (gbp_next_hop_pool, gui)); +} + +static void +gbp_contract_rules_free (index_t * rules) +{ + index_t *gui, *gnhi; + + vec_foreach (gui, rules) + { + gbp_policy_node_t pnode; + fib_protocol_t fproto; + gbp_next_hop_t *gnh; + gbp_rule_t *gu; + + gu = gbp_rule_get (*gui); + + FOR_EACH_GBP_POLICY_NODE (pnode) + { + FOR_EACH_FIB_IP_PROTOCOL (fproto) + { + dpo_reset (&gu->gu_dpo[pnode][fproto]); + dpo_reset (&gu->gu_dpo[pnode][fproto]); + } + } + + vec_foreach (gnhi, gu->gu_nhs) + { + fib_protocol_t fproto; + + gnh = gbp_next_hop_get (*gnhi); + gbp_bridge_domain_unlock (gnh->gnh_bd); + gbp_route_domain_unlock (gnh->gnh_rd); + gbp_endpoint_child_remove (gnh->gnh_ge, gnh->gnh_sibling); + gbp_endpoint_unlock (GBP_ENDPOINT_SRC_RR, gnh->gnh_ge); + + FOR_EACH_FIB_IP_PROTOCOL (fproto) + { + adj_unlock (gnh->gnh_ai[fproto]); + } + } + } + vec_free (rules); +} + +static u8 * +format_gbp_next_hop (u8 * s, va_list * args) +{ + index_t gnhi = va_arg (*args, index_t); + gbp_next_hop_t *gnh; + + gnh = gbp_next_hop_get (gnhi); + + s = format (s, "%U, %U, %U EP:%d", + format_mac_address_t, &gnh->gnh_mac, + format_gbp_bridge_domain, gnh->gnh_bd, + format_ip46_address, &gnh->gnh_ip, IP46_TYPE_ANY, gnh->gnh_ge); + + return (s); +} + +static u8 * +format_gbp_rule_action (u8 * s, va_list * args) +{ + gbp_rule_action_t action = va_arg (*args, gbp_rule_action_t); + + switch (action) + { +#define _(v,a) case GBP_RULE_##v: return (format (s, "%s", a)); + foreach_gbp_rule_action +#undef _ + } + + return (format (s, "unknown")); +} + +static u8 * +format_gbp_hash_mode (u8 * s, va_list * args) +{ + gbp_hash_mode_t action = va_arg (*args, gbp_hash_mode_t); + + switch (action) + { +#define _(v,a) case GBP_HASH_MODE_##v: return (format (s, "%s", a)); + foreach_gbp_hash_mode +#undef _ + } + + return (format (s, "unknown")); +} + +static u8 * +format_gbp_policy_node (u8 * s, va_list * args) +{ + gbp_policy_node_t action = va_arg (*args, gbp_policy_node_t); + + switch (action) + { +#define _(v,a) case GBP_POLICY_NODE_##v: return (format (s, "%s", a)); + foreach_gbp_policy_node +#undef _ + } + + return (format (s, "unknown")); +} + +static u8 * +format_gbp_rule (u8 * s, va_list * args) +{ + index_t gui = va_arg (*args, index_t); + gbp_policy_node_t pnode; + fib_protocol_t fproto; + gbp_rule_t *gu; + index_t *gnhi; + + gu = gbp_rule_get (gui); + s = format (s, "%U", format_gbp_rule_action, gu->gu_action); + + switch (gu->gu_action) + { + case GBP_RULE_PERMIT: + case GBP_RULE_DENY: + break; + case GBP_RULE_REDIRECT: + s = format (s, ", %U", format_gbp_hash_mode, gu->gu_hash_mode); + break; + } + + vec_foreach (gnhi, gu->gu_nhs) + { + s = format (s, "\n [%U]", format_gbp_next_hop, *gnhi); + } + + FOR_EACH_GBP_POLICY_NODE (pnode) + { + s = format (s, "\n policy-%U", format_gbp_policy_node, pnode); + + FOR_EACH_FIB_IP_PROTOCOL (fproto) + { + if (dpo_id_is_valid (&gu->gu_dpo[pnode][fproto])) + { + s = + format (s, "\n %U", format_dpo_id, + &gu->gu_dpo[pnode][fproto], 8); + } + } + } + + return (s); +} + +static void +gbp_contract_mk_adj (gbp_next_hop_t * gnh, fib_protocol_t fproto) +{ + ethernet_header_t *eth; + gbp_endpoint_t *ge; + index_t old_ai; + u8 *rewrite; + + old_ai = gnh->gnh_ai[fproto]; + rewrite = NULL; + vec_validate (rewrite, sizeof (*eth) - 1); + eth = (ethernet_header_t *) rewrite; + + GBP_CONTRACT_DBG ("...mk-adj: %U", format_gbp_next_hop, + gnh - gbp_next_hop_pool); + + ge = gbp_endpoint_get (gnh->gnh_ge); + + eth->type = clib_host_to_net_u16 ((fproto == FIB_PROTOCOL_IP4 ? + ETHERNET_TYPE_IP4 : ETHERNET_TYPE_IP6)); + mac_address_to_bytes (gbp_route_domain_get_local_mac (), eth->src_address); + mac_address_to_bytes (&gnh->gnh_mac, eth->dst_address); + + gnh->gnh_ai[fproto] = + adj_nbr_add_or_lock_w_rewrite (fproto, + fib_proto_to_link (fproto), + &gnh->gnh_ip, ge->ge_fwd.gef_itf, rewrite); + + adj_unlock (old_ai); +} + +static void +gbp_contract_mk_lb (index_t gui, fib_protocol_t fproto) +{ + load_balance_path_t *paths = NULL; + gbp_policy_node_t pnode; + gbp_next_hop_t *gnh; + dpo_proto_t dproto; + gbp_rule_t *gu; + u32 ii; + + u32 policy_nodes[] = { + [GBP_POLICY_NODE_L2] = gbp_policy_port_node.index, + [GBP_POLICY_NODE_IP4] = ip4_gbp_policy_dpo_node.index, + [GBP_POLICY_NODE_IP6] = ip6_gbp_policy_dpo_node.index, }; + + GBP_CONTRACT_DBG ("..mk-lb: %U", format_gbp_rule, gui); + + gu = gbp_rule_get (gui); + dproto = fib_proto_to_dpo (fproto); + + if (GBP_RULE_REDIRECT != gu->gu_action) + return; + + vec_foreach_index (ii, gu->gu_nhs) + { + gnh = gbp_next_hop_get (gu->gu_nhs[ii]); + + gbp_contract_mk_adj (gnh, FIB_PROTOCOL_IP4); + gbp_contract_mk_adj (gnh, FIB_PROTOCOL_IP6); + } + + FOR_EACH_GBP_POLICY_NODE (pnode) + { + vec_validate (paths, vec_len (gu->gu_nhs) - 1); + + vec_foreach_index (ii, gu->gu_nhs) + { + gnh = gbp_next_hop_get (gu->gu_nhs[ii]); + + paths[ii].path_index = FIB_NODE_INDEX_INVALID; + paths[ii].path_weight = 1; + dpo_set (&paths[ii].path_dpo, DPO_ADJACENCY, + dproto, gnh->gnh_ai[fproto]); + } + + // FIXME get algo and sticky bit from contract LB algo + if (!dpo_id_is_valid (&gu->gu_dpo[pnode][fproto])) + { + dpo_id_t dpo = DPO_INVALID; + + dpo_set (&dpo, DPO_LOAD_BALANCE, dproto, + load_balance_create (vec_len (paths), + dproto, IP_FLOW_HASH_DEFAULT)); + dpo_stack_from_node (policy_nodes[pnode], + &gu->gu_dpo[pnode][fproto], &dpo); + dpo_reset (&dpo); + } + + load_balance_multipath_update (&gu->gu_dpo[pnode][fproto], + paths, LOAD_BALANCE_FLAG_NONE); + vec_free (paths); + } +} + +static void +gbp_contract_mk_one_lb (index_t gui) +{ + gbp_contract_mk_lb (gui, FIB_PROTOCOL_IP4); + gbp_contract_mk_lb (gui, FIB_PROTOCOL_IP6); +} + +static int +gbp_contract_next_hop_resolve (index_t gui, index_t gnhi) +{ + gbp_bridge_domain_t *gbd; + gbp_next_hop_t *gnh; + ip46_address_t *ips; + int rv; + + ips = NULL; + gnh = gbp_next_hop_get (gnhi); + gbd = gbp_bridge_domain_get (gnh->gnh_bd); + + gnh->gnh_gu = gui; + vec_add1 (ips, gnh->gnh_ip); + + /* + * source the endpoint this contract needs to forward via. + * give ofrwarding details via the spine proxy. if this EP is known + * to us, then since we source here with a low priority, the learned + * info will take precedenc. + */ + rv = gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_RR, + gbd->gb_uu_fwd_sw_if_index, + ips, + &gnh->gnh_mac, + gnh->gnh_bd, gnh->gnh_rd, EPG_INVALID, + GBP_ENDPOINT_FLAG_NONE, NULL, NULL, + &gnh->gnh_ge); + + if (0 == rv) + { + gnh->gnh_sibling = gbp_endpoint_child_add (gnh->gnh_ge, + gbp_next_hop_fib_type, gnhi); + } + + GBP_CONTRACT_DBG ("..resolve: %d: %d: %U", gui, gnhi, format_gbp_next_hop, + gnhi); + + vec_free (ips); + return (rv); +} + +static void +gbp_contract_rule_resolve (index_t gui) +{ + gbp_rule_t *gu; + index_t *gnhi; + + gu = gbp_rule_get (gui); + + GBP_CONTRACT_DBG ("..resolve: %U", format_gbp_rule, gui); + + vec_foreach (gnhi, gu->gu_nhs) + { + gbp_contract_next_hop_resolve (gui, *gnhi); + } +} + +static void +gbp_contract_resolve (index_t * guis) +{ + index_t *gui; + + vec_foreach (gui, guis) + { + gbp_contract_rule_resolve (*gui); + } +} + +static void +gbp_contract_mk_lbs (index_t * guis) +{ + index_t *gui; + + vec_foreach (gui, guis) + { + gbp_contract_mk_one_lb (*gui); + } +} + +int +gbp_contract_update (epg_id_t src_epg, + epg_id_t dst_epg, u32 acl_index, index_t * rules) +{ + gbp_main_t *gm = &gbp_main; + u32 *acl_vec = NULL; + gbp_contract_t *gc; + index_t gci; uword *p; gbp_contract_key_t key = { @@ -48,59 +438,74 @@ gbp_contract_update (epg_id_t src_epg, epg_id_t dst_epg, u32 acl_index) p = hash_get (gbp_contract_db.gc_hash, key.as_u32); if (p != NULL) { - value.as_u64 = p[0]; + gci = p[0]; + gc = gbp_contract_get (gci); + gbp_contract_rules_free (gc->gc_rules); + gbp_main.acl_plugin.put_lookup_context_index (gc->gc_lc_index); + gc->gc_rules = NULL; } else { - value.gc_lc_index = - gm->acl_plugin.get_lookup_context_index (gm->gbp_acl_user_id, src_epg, - dst_epg); - value.gc_acl_index = acl_index; - hash_set (gbp_contract_db.gc_hash, key.as_u32, value.as_u64); + pool_get_zero (gbp_contract_pool, gc); + gc->gc_key = key; + gci = gc - gbp_contract_pool; + hash_set (gbp_contract_db.gc_hash, key.as_u32, gci); } - if (value.gc_lc_index == ~0) - return; - vec_add1 (acl_vec, acl_index); - gm->acl_plugin.set_acl_vec_for_context (value.gc_lc_index, acl_vec); + GBP_CONTRACT_DBG ("update: %U", format_gbp_contract, gci); + + gc->gc_rules = rules; + gbp_contract_resolve (gc->gc_rules); + gbp_contract_mk_lbs (gc->gc_rules); + + gc->gc_acl_index = acl_index; + gc->gc_lc_index = + gm->acl_plugin.get_lookup_context_index (gm->gbp_acl_user_id, + src_epg, dst_epg); + + vec_add1 (acl_vec, gc->gc_acl_index); + gm->acl_plugin.set_acl_vec_for_context (gc->gc_lc_index, acl_vec); vec_free (acl_vec); + + return (0); } -void +int gbp_contract_delete (epg_id_t src_epg, epg_id_t dst_epg) { - gbp_main_t *gm = &gbp_main; - uword *p; - gbp_contract_value_t value; gbp_contract_key_t key = { .gck_src = src_epg, .gck_dst = dst_epg, }; + gbp_contract_t *gc; + uword *p; p = hash_get (gbp_contract_db.gc_hash, key.as_u32); if (p != NULL) { - value.as_u64 = p[0]; - gm->acl_plugin.put_lookup_context_index (value.gc_lc_index); + gc = gbp_contract_get (p[0]); + + gbp_contract_rules_free (gc->gc_rules); + gbp_main.acl_plugin.put_lookup_context_index (gc->gc_lc_index); + + hash_unset (gbp_contract_db.gc_hash, key.as_u32); + pool_put (gbp_contract_pool, gc); + + return (0); } - hash_unset (gbp_contract_db.gc_hash, key.as_u32); + + return (VNET_API_ERROR_NO_SUCH_ENTRY); } void gbp_contract_walk (gbp_contract_cb_t cb, void *ctx) { - gbp_contract_key_t key; - gbp_contract_value_t value; + gbp_contract_t *gc; /* *INDENT-OFF* */ - hash_foreach(key.as_u32, value.as_u64, gbp_contract_db.gc_hash, + pool_foreach(gc, gbp_contract_pool, ({ - gbp_contract_t gbpc = { - .gc_key = key, - .gc_value = value, - }; - - if (!cb(&gbpc, ctx)) + if (!cb(gc, ctx)) break; })); /* *INDENT-ON* */ @@ -137,7 +542,7 @@ gbp_contract_cli (vlib_main_t * vm, if (add) { - gbp_contract_update (src_epg_id, dst_epg_id, acl_index); + gbp_contract_update (src_epg_id, dst_epg_id, acl_index, NULL); } else { @@ -164,21 +569,49 @@ VLIB_CLI_COMMAND (gbp_contract_cli_node, static) = }; /* *INDENT-ON* */ +static u8 * +format_gbp_contract_key (u8 * s, va_list * args) +{ + gbp_contract_key_t *gck = va_arg (*args, gbp_contract_key_t *); + + s = format (s, "{%d,%d}", gck->gck_src, gck->gck_dst); + + return (s); +} + +u8 * +format_gbp_contract (u8 * s, va_list * args) +{ + index_t gci = va_arg (*args, index_t); + gbp_contract_t *gc; + index_t *gui; + + gc = gbp_contract_get (gci); + + s = format (s, "%U: acl-index:%d", + format_gbp_contract_key, &gc->gc_key, gc->gc_acl_index); + + vec_foreach (gui, gc->gc_rules) + { + s = format (s, "\n %d: %U", *gui, format_gbp_rule, *gui); + } + + return (s); +} + static clib_error_t * gbp_contract_show (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { - gbp_contract_key_t key; - gbp_contract_value_t value; + index_t gci; vlib_cli_output (vm, "Contracts:"); /* *INDENT-OFF* */ - hash_foreach (key.as_u32, value.as_u64, gbp_contract_db.gc_hash, - { - vlib_cli_output (vm, " {%d,%d} -> %d", key.gck_src, - key.gck_dst, value.gc_acl_index); - }); + pool_foreach_index (gci, gbp_contract_pool, + ({ + vlib_cli_output (vm, " [%d] %U", gci, format_gbp_contract, gci); + })); /* *INDENT-ON* */ return (NULL); @@ -199,6 +632,63 @@ VLIB_CLI_COMMAND (gbp_contract_show_node, static) = { }; /* *INDENT-ON* */ +static fib_node_t * +gbp_next_hop_get_node (fib_node_index_t index) +{ + gbp_next_hop_t *gnh; + + gnh = gbp_next_hop_get (index); + + return (&gnh->gnh_node); +} + +static void +gbp_next_hop_last_lock_gone (fib_node_t * node) +{ + ASSERT (0); +} + +static gbp_next_hop_t * +gbp_next_hop_from_fib_node (fib_node_t * node) +{ + ASSERT (gbp_next_hop_fib_type == node->fn_type); + return ((gbp_next_hop_t *) node); +} + +static fib_node_back_walk_rc_t +gbp_next_hop_back_walk_notify (fib_node_t * node, + fib_node_back_walk_ctx_t * ctx) +{ + gbp_next_hop_t *gnh; + + gnh = gbp_next_hop_from_fib_node (node); + + gbp_contract_mk_one_lb (gnh->gnh_gu); + + return (FIB_NODE_BACK_WALK_CONTINUE); +} + +/* + * The FIB path's graph node virtual function table + */ +static const fib_node_vft_t gbp_next_hop_vft = { + .fnv_get = gbp_next_hop_get_node, + .fnv_last_lock = gbp_next_hop_last_lock_gone, + .fnv_back_walk = gbp_next_hop_back_walk_notify, + // .fnv_mem_show = fib_path_memory_show, +}; + +static clib_error_t * +gbp_contract_init (vlib_main_t * vm) +{ + gc_logger = vlib_log_register_class ("gbp", "con"); + gbp_next_hop_fib_type = fib_node_register_new_type (&gbp_next_hop_vft); + + return (NULL); +} + +VLIB_INIT_FUNCTION (gbp_contract_init); + /* * fd.io coding-style-patch-verification: ON * diff --git a/src/plugins/gbp/gbp_contract.h b/src/plugins/gbp/gbp_contract.h index 121192c9c96..c1073510413 100644 --- a/src/plugins/gbp/gbp_contract.h +++ b/src/plugins/gbp/gbp_contract.h @@ -37,24 +37,69 @@ typedef struct gbp_contract_key_t_ }; } gbp_contract_key_t; -/** - * The value for an Contract - */ -typedef struct gbp_contract_value_t_ +typedef struct gbp_next_hop_t_ { - union - { - struct - { - /** - * lookup context and acl index - */ - u32 gc_lc_index; - u32 gc_acl_index; - }; - u64 as_u64; - }; -} gbp_contract_value_t; + fib_node_t gnh_node; + ip46_address_t gnh_ip; + mac_address_t gnh_mac; + index_t gnh_gu; + index_t gnh_bd; + index_t gnh_rd; + u32 gnh_ge; + u32 gnh_sibling; + index_t gnh_ai[FIB_PROTOCOL_IP_MAX]; +} gbp_next_hop_t; + +#define foreach_gbp_hash_mode \ + _(SRC_IP, "src-ip") \ + _(DST_IP, "dst-ip") + +typedef enum gbp_hash_mode_t_ +{ +#define _(v,s) GBP_HASH_MODE_##v, + foreach_gbp_hash_mode +#undef _ +} gbp_hash_mode_t; + +#define foreach_gbp_rule_action \ + _(PERMIT, "permit") \ + _(DENY, "deny") \ + _(REDIRECT, "redirect") + +typedef enum gbp_rule_action_t_ +{ +#define _(v,s) GBP_RULE_##v, + foreach_gbp_rule_action +#undef _ +} gbp_rule_action_t; + +#define foreach_gbp_policy_node \ + _(L2, "L2") \ + _(IP4, "ip4") \ + _(IP6, "ip6") + +typedef enum gbp_policy_node_t_ +{ +#define _(v,s) GBP_POLICY_NODE_##v, + foreach_gbp_policy_node +#undef _ +} gbp_policy_node_t; +#define GBP_POLICY_N_NODES (GBP_POLICY_NODE_IP6+1) + +#define FOR_EACH_GBP_POLICY_NODE(pnode) \ + for (pnode = GBP_POLICY_NODE_L2; pnode < GBP_POLICY_N_NODES; pnode++) + +typedef struct gbp_rule_t_ +{ + gbp_rule_action_t gu_action; + gbp_hash_mode_t gu_hash_mode; + index_t *gu_nhs; + + /** + * DPO of the load-balance object used to redirect + */ + dpo_id_t gu_dpo[GBP_POLICY_N_NODES][FIB_PROTOCOL_IP_MAX]; +} gbp_rule_t; /** * A Group Based Policy Contract. @@ -67,10 +112,13 @@ typedef struct gbp_contract_t_ */ gbp_contract_key_t gc_key; + u32 gc_acl_index; + u32 gc_lc_index; + /** * The ACL to apply for packets from the source to the destination EPG */ - gbp_contract_value_t gc_value; + index_t *gc_rules; } gbp_contract_t; /** @@ -84,21 +132,29 @@ typedef struct gbp_contract_db_t_ uword *gc_hash; } gbp_contract_db_t; -extern void gbp_contract_update (epg_id_t src_epg, - epg_id_t dst_epg, u32 acl_index); -extern void gbp_contract_delete (epg_id_t src_epg, epg_id_t dst_epg); +extern int gbp_contract_update (epg_id_t src_epg, + epg_id_t dst_epg, + u32 acl_index, index_t * rules); +extern int gbp_contract_delete (epg_id_t src_epg, epg_id_t dst_epg); + +extern index_t gbp_rule_alloc (gbp_rule_action_t action, + gbp_hash_mode_t hash_mode, index_t * nhs); +extern index_t gbp_next_hop_alloc (const ip46_address_t * ip, + index_t grd, + const mac_address_t * mac, index_t gbd); typedef int (*gbp_contract_cb_t) (gbp_contract_t * gbpe, void *ctx); extern void gbp_contract_walk (gbp_contract_cb_t bgpe, void *ctx); +extern u8 *format_gbp_contract (u8 * s, va_list * args); /** * DP functions and databases */ extern gbp_contract_db_t gbp_contract_db; -always_inline u64 -gbp_acl_lookup (gbp_contract_key_t * key) +always_inline index_t +gbp_contract_find (gbp_contract_key_t * key) { uword *p; @@ -107,7 +163,23 @@ gbp_acl_lookup (gbp_contract_key_t * key) if (NULL != p) return (p[0]); - return (~0); + return (INDEX_INVALID); +} + +extern gbp_contract_t *gbp_contract_pool; + +always_inline gbp_contract_t * +gbp_contract_get (index_t gci) +{ + return (pool_elt_at_index (gbp_contract_pool, gci)); +} + +extern gbp_rule_t *gbp_rule_pool; + +always_inline gbp_rule_t * +gbp_rule_get (index_t gui) +{ + return (pool_elt_at_index (gbp_rule_pool, gui)); } #endif diff --git a/src/plugins/gbp/gbp_endpoint.c b/src/plugins/gbp/gbp_endpoint.c index 47adbb57371..e9ced3f9e03 100644 --- a/src/plugins/gbp/gbp_endpoint.c +++ b/src/plugins/gbp/gbp_endpoint.c @@ -31,6 +31,7 @@ #include <vnet/l2/l2_fib.h> #include <vnet/fib/fib_table.h> #include <vnet/ip/ip_neighbor.h> +#include <vnet/fib/fib_walk.h> static const char *gbp_endpoint_attr_names[] = GBP_ENDPOINT_ATTR_NAMES; @@ -39,6 +40,8 @@ static const char *gbp_endpoint_attr_names[] = GBP_ENDPOINT_ATTR_NAMES; */ gbp_ep_db_t gbp_ep_db; +fib_node_type_t gbp_endpoint_fib_type; + vlib_log_class_t gbp_ep_logger; #define GBP_ENDPOINT_DBG(...) \ @@ -89,7 +92,7 @@ format_gbp_endpoint_flags (u8 * s, va_list * args) int gbp_endpoint_is_remote (const gbp_endpoint_t * ge) { - return (ge->ge_flags & GBP_ENDPOINT_FLAG_REMOTE); + return (ge->ge_fwd.gef_flags & GBP_ENDPOINT_FLAG_REMOTE); } static void @@ -194,264 +197,522 @@ ip46_address_get_type (const ip46_address_t * a) return (ip46_address_is_ip4 (a) ? IP46_TYPE_IP4 : IP46_TYPE_IP6); } -static ip46_type_t -ip46_address_get_len (const ip46_address_t * a) +static int +gbp_endpoint_ip_is_equal (const fib_prefix_t * fp, const ip46_address_t * ip) { - return (ip46_address_is_ip4 (a) ? 32 : 128); + return (ip46_address_is_equal (ip, &fp->fp_addr)); +} + +static void +gbp_endpoint_ips_update (gbp_endpoint_t * ge, + const ip46_address_t * ips, + const gbp_route_domain_t * grd) +{ + const ip46_address_t *ip; + index_t gei, grdi; + + gei = gbp_endpoint_index (ge); + grdi = gbp_route_domain_index (grd); + + ASSERT ((ge->ge_key.gek_grd == INDEX_INVALID) || + (ge->ge_key.gek_grd == grdi)); + + vec_foreach (ip, ips) + { + if (~0 == vec_search_with_function (ge->ge_key.gek_ips, ip, + gbp_endpoint_ip_is_equal)) + { + fib_prefix_t *pfx; + + vec_add2 (ge->ge_key.gek_ips, pfx, 1); + fib_prefix_from_ip46_addr (ip, pfx); + + gbp_endpoint_add_ip (&pfx->fp_addr, + grd->grd_fib_index[pfx->fp_proto], gei); + } + ge->ge_key.gek_grd = grdi; + } } static gbp_endpoint_t * -gbp_endpoint_alloc (epg_id_t epg_id, - index_t ggi, u32 sw_if_index, gbp_endpoint_flags_t flags, - const ip46_address_t * tun_src, - const ip46_address_t * tun_dst) +gbp_endpoint_alloc (const ip46_address_t * ips, + const gbp_route_domain_t * grd, + const mac_address_t * mac, + const gbp_bridge_domain_t * gbd) { gbp_endpoint_t *ge; + index_t gei; pool_get_zero (gbp_endpoint_pool, ge); - ge->ge_epg = ggi; - ge->ge_epg_id = epg_id; - ge->ge_flags = flags; - ge->ge_sw_if_index = sw_if_index; + fib_node_init (&ge->ge_node, gbp_endpoint_fib_type); + gei = gbp_endpoint_index (ge); + ge->ge_key.gek_gbd = + ge->ge_key.gek_grd = ge->ge_fwd.gef_itf = INDEX_INVALID; ge->ge_last_time = vlib_time_now (vlib_get_main ()); + ge->ge_key.gek_gbd = gbp_bridge_domain_index (gbd); + + if (NULL != mac) + { + mac_address_copy (&ge->ge_key.gek_mac, mac); + gbp_endpoint_add_mac (mac, gbd->gb_bd_index, gei); + } + gbp_endpoint_ips_update (ge, ips, grd); + + return (ge); +} + +static int +gbp_endpoint_loc_is_equal (gbp_endpoint_loc_t * a, gbp_endpoint_loc_t * b) +{ + return (a->gel_src == b->gel_src); +} + +static int +gbp_endpoint_loc_cmp_for_sort (gbp_endpoint_loc_t * a, gbp_endpoint_loc_t * b) +{ + return (a->gel_src - b->gel_src); +} + +static gbp_endpoint_loc_t * +gbp_endpoint_loc_find (gbp_endpoint_t * ge, gbp_endpoint_src_t src) +{ + gbp_endpoint_loc_t gel = { + .gel_src = src, + }; + u32 pos; + + pos = vec_search_with_function (ge->ge_locs, &gel, + gbp_endpoint_loc_is_equal); + + if (~0 != pos) + return (&ge->ge_locs[pos]); + + return NULL; +} + +static int +gbp_endpoint_loc_unlock (gbp_endpoint_t * ge, gbp_endpoint_loc_t * gel) +{ + u32 pos; + + gel->gel_locks--; + + if (0 == gel->gel_locks) + { + pos = gel - ge->ge_locs; + + vec_del1 (ge->ge_locs, pos); + if (vec_len (ge->ge_locs) > 1) + vec_sort_with_function (ge->ge_locs, gbp_endpoint_loc_cmp_for_sort); + + /* This could be the last lock, so don't access the EP from + * this point on */ + fib_node_unlock (&ge->ge_node); + + return (1); + } + return (0); +} + +static void +gbp_endpoint_loc_destroy (gbp_endpoint_loc_t * gel) +{ + gbp_endpoint_group_unlock (gel->gel_epg); + + if (gel->gel_flags & GBP_ENDPOINT_FLAG_REMOTE) + { + vxlan_gbp_tunnel_unlock (gel->gel_sw_if_index); + } +} + +static gbp_endpoint_loc_t * +gbp_endpoint_loc_find_or_add (gbp_endpoint_t * ge, gbp_endpoint_src_t src) +{ + gbp_endpoint_loc_t gel = { + .gel_src = src, + .gel_epg = INDEX_INVALID, + .gel_sw_if_index = INDEX_INVALID, + .gel_locks = 0, + }; + u32 pos; + + pos = vec_search_with_function (ge->ge_locs, &gel, + gbp_endpoint_loc_is_equal); + + if (~0 == pos) + { + vec_add1 (ge->ge_locs, gel); + + if (vec_len (ge->ge_locs) > 1) + { + vec_sort_with_function (ge->ge_locs, gbp_endpoint_loc_cmp_for_sort); + + pos = vec_search_with_function (ge->ge_locs, &gel, + gbp_endpoint_loc_is_equal); + } + else + pos = 0; + + /* + * it's the sources and children that lock the endpoints + */ + fib_node_lock (&ge->ge_node); + } + + return (&ge->ge_locs[pos]); +} + +/** + * Find an EP inthe DBs and check that if we find it in the L2 DB + * it has the same IPs as this update + */ +static int +gbp_endpoint_find_for_update (const ip46_address_t * ips, + const gbp_route_domain_t * grd, + const mac_address_t * mac, + const gbp_bridge_domain_t * gbd, + gbp_endpoint_t ** ge) +{ + gbp_endpoint_t *l2_ge, *l3_ge, *tmp; + + l2_ge = l3_ge = NULL; + + if (NULL != mac && !mac_address_is_zero (mac)) + { + ASSERT (gbd); + l2_ge = gbp_endpoint_find_mac (mac->bytes, gbd->gb_bd_index); + } + if (NULL != ips && !ip46_address_is_zero (ips)) + { + const ip46_address_t *ip; + fib_protocol_t fproto; + + ASSERT (grd); + vec_foreach (ip, ips) + { + fproto = fib_proto_from_ip46 (ip46_address_get_type (ip)); + + tmp = gbp_endpoint_find_ip (ip, grd->grd_fib_index[fproto]); + + if (NULL == tmp) + /* not found */ + continue; + else if (NULL == l3_ge) + /* first match against an IP address */ + l3_ge = tmp; + else if (tmp == l3_ge) + /* another match against IP address that is the same endpoint */ + continue; + else + { + /* + * a match agains a different endpoint. + * this means the KEY of the EP is changing which is not allowed + */ + return (-1); + } + } + } + + if (NULL == l2_ge && NULL == l3_ge) + /* not found */ + *ge = NULL; + else if (NULL == l2_ge) + /* found at L3 */ + *ge = l3_ge; + else if (NULL == l3_ge) + /* found at L2 */ + *ge = l2_ge; + else + { + /* found both L3 and L2 - they must be the same else the KEY + * is changing + */ + if (l2_ge == l3_ge) + *ge = l2_ge; + else + return (-1); + } + + return (0); +} - gbp_endpoint_group_find_and_lock (epg_id); +static gbp_endpoint_src_t +gbp_endpoint_get_best_src (const gbp_endpoint_t * ge) +{ + if (0 == vec_len (ge->ge_locs)) + return (GBP_ENDPOINT_SRC_MAX); + + return (ge->ge_locs[0].gel_src); +} + +static void +gbp_endpoint_n_learned (int n) +{ + gbp_n_learnt_endpoints += n; + + if (n > 0 && 1 == gbp_n_learnt_endpoints) + { + vlib_process_signal_event (vlib_get_main (), + gbp_scanner_node.index, + GBP_ENDPOINT_SCAN_START, 0); + } + if (n < 0 && 0 == gbp_n_learnt_endpoints) + { + vlib_process_signal_event (vlib_get_main (), + gbp_scanner_node.index, + GBP_ENDPOINT_SCAN_STOP, 0); + } +} + +static void +gbp_endpoint_loc_update (gbp_endpoint_loc_t * gel, + u32 sw_if_index, + index_t ggi, + gbp_endpoint_flags_t flags, + const ip46_address_t * tun_src, + const ip46_address_t * tun_dst) +{ + int was_learnt, is_learnt; + + gel->gel_locks++; + was_learnt = ! !(gel->gel_flags & GBP_ENDPOINT_FLAG_REMOTE); + gel->gel_flags = flags; + is_learnt = ! !(gel->gel_flags & GBP_ENDPOINT_FLAG_REMOTE); + + gbp_endpoint_n_learned (is_learnt - was_learnt); - if (gbp_endpoint_is_remote (ge)) + if (INDEX_INVALID == gel->gel_epg) + { + gel->gel_epg = ggi; + if (INDEX_INVALID != gel->gel_epg) + { + gbp_endpoint_group_lock (gel->gel_epg); + } + } + else + { + ASSERT (gel->gel_epg == ggi); + } + + if (gel->gel_flags & GBP_ENDPOINT_FLAG_REMOTE) { if (NULL != tun_src) - ip46_address_copy (&ge->tun.ge_src, tun_src); + ip46_address_copy (&gel->tun.gel_src, tun_src); if (NULL != tun_dst) - ip46_address_copy (&ge->tun.ge_dst, tun_dst); + ip46_address_copy (&gel->tun.gel_dst, tun_dst); /* * the input interface may be the parent GBP-vxlan interface, * create a child vlxan-gbp tunnel and use that as the endpoint's * interface. */ + if (~0 != gel->gel_sw_if_index) + vxlan_gbp_tunnel_unlock (gel->gel_sw_if_index); + switch (gbp_vxlan_tunnel_get_type (sw_if_index)) { case GBP_VXLAN_TEMPLATE_TUNNEL: - ge->tun.ge_parent_sw_if_index = sw_if_index; - ge->ge_sw_if_index = + gel->tun.gel_parent_sw_if_index = sw_if_index; + gel->gel_sw_if_index = gbp_vxlan_tunnel_clone_and_lock (sw_if_index, - &ge->tun.ge_src, - &ge->tun.ge_dst); + &gel->tun.gel_src, + &gel->tun.gel_dst); break; case VXLAN_GBP_TUNNEL: - ge->tun.ge_parent_sw_if_index = + gel->tun.gel_parent_sw_if_index = vxlan_gbp_tunnel_get_parent (sw_if_index); - ge->ge_sw_if_index = sw_if_index; - vxlan_gbp_tunnel_lock (ge->ge_sw_if_index); + gel->gel_sw_if_index = sw_if_index; + vxlan_gbp_tunnel_lock (gel->gel_sw_if_index); break; } } - - return (ge); + else + { + gel->gel_sw_if_index = sw_if_index; + } } -int -gbp_endpoint_update (u32 sw_if_index, - const ip46_address_t * ips, - const mac_address_t * mac, - epg_id_t epg_id, - gbp_endpoint_flags_t flags, - const ip46_address_t * tun_src, - const ip46_address_t * tun_dst, u32 * handle) +static void +gbb_endpoint_fwd_reset (gbp_endpoint_t * ge) { - gbp_endpoint_group_t *gg; - gbp_endpoint_t *ge; - index_t ggi, gei; + const gbp_route_domain_t *grd; + const gbp_bridge_domain_t *gbd; + gbp_endpoint_fwd_t *gef; + const fib_prefix_t *pfx; + index_t *ai; + index_t gei; + + gei = gbp_endpoint_index (ge); + gbd = gbp_bridge_domain_get (ge->ge_key.gek_gbd); + gef = &ge->ge_fwd; + + vec_foreach (pfx, ge->ge_key.gek_ips) + { + u32 fib_index; - if (~0 == sw_if_index) - return (VNET_API_ERROR_INVALID_SW_IF_INDEX); + grd = gbp_route_domain_get (ge->ge_key.gek_grd); + fib_index = grd->grd_fib_index[pfx->fp_proto]; - ge = NULL; - ggi = gbp_endpoint_group_find_and_lock (epg_id); + bd_add_del_ip_mac (gbd->gb_bd_index, fib_proto_to_ip46 (pfx->fp_proto), + &pfx->fp_addr, &ge->ge_key.gek_mac, 0); - if (INDEX_INVALID == ggi) - return (VNET_API_ERROR_NO_SUCH_ENTRY); + /* + * remove a host route + */ + if (gbp_endpoint_is_remote (ge)) + { + fib_table_entry_special_remove (fib_index, pfx, FIB_SOURCE_PLUGIN_HI); + } - gg = gbp_endpoint_group_get (ggi); + fib_table_entry_delete (fib_index, pfx, FIB_SOURCE_PLUGIN_LOW); + } + vec_foreach (ai, gef->gef_adjs) + { + adj_unlock (*ai); + } - /* - * L2 EP - */ - if (NULL != mac && !mac_address_is_zero (mac)) + if (INDEX_INVALID != gef->gef_itf) { - /* - * find an existing endpoint matching one of the key types - */ - ge = gbp_endpoint_find_mac (mac->bytes, gg->gg_bd_index); - if (NULL == ge) - { - /* - * new entry - */ - ge = gbp_endpoint_alloc (epg_id, ggi, sw_if_index, flags, - tun_src, tun_dst); - gei = gbp_endpoint_index (ge); - mac_address_copy (&ge->ge_mac, mac); - - ge->ge_itf = gbp_itf_add_and_lock (ge->ge_sw_if_index, - gg->gg_bd_index); - - gbp_itf_set_l2_input_feature (ge->ge_itf, gei, - L2INPUT_FEAT_GBP_FWD); + l2fib_del_entry (ge->ge_key.gek_mac.bytes, + gbd->gb_bd_index, gef->gef_itf); + gbp_itf_set_l2_input_feature (gef->gef_itf, gei, (L2INPUT_FEAT_NONE)); + gbp_itf_set_l2_output_feature (gef->gef_itf, gei, L2OUTPUT_FEAT_NONE); - if (gbp_endpoint_is_remote (ge)) - { - gbp_itf_set_l2_output_feature (ge->ge_itf, gei, - L2OUTPUT_FEAT_GBP_POLICY_MAC); - } - else - { - gbp_endpoint_add_itf (ge->ge_sw_if_index, gei); - gbp_itf_set_l2_output_feature (ge->ge_itf, gei, - L2OUTPUT_FEAT_GBP_POLICY_PORT); - } + gbp_itf_unlock (gef->gef_itf); + gef->gef_itf = INDEX_INVALID; + } - gbp_endpoint_add_mac (mac, gg->gg_bd_index, gei); + vec_free (gef->gef_adjs); +} - l2fib_add_entry (mac->bytes, gg->gg_bd_index, ge->ge_sw_if_index, - L2FIB_ENTRY_RESULT_FLAG_STATIC); - } - else - { - /* - * update existing entry.. - */ - ge->ge_flags = flags; - gei = gbp_endpoint_index (ge); - goto out; - } - } +static void +gbb_endpoint_fwd_recalc (gbp_endpoint_t * ge) +{ + const gbp_route_domain_t *grd; + const gbp_bridge_domain_t *gbd; + const gbp_endpoint_group_t *gg; + gbp_endpoint_loc_t *gel; + gbp_endpoint_fwd_t *gef; + const fib_prefix_t *pfx; + index_t gei; /* - * L3 EP + * locations are sort in source priority order */ - if (NULL != ips && !ip46_address_is_zero (ips)) + gei = gbp_endpoint_index (ge); + gel = &ge->ge_locs[0]; + gef = &ge->ge_fwd; + gbd = gbp_bridge_domain_get (ge->ge_key.gek_gbd); + + gef->gef_flags = gel->gel_flags; + + if (INDEX_INVALID != gel->gel_epg) { - const ip46_address_t *ip; - fib_protocol_t fproto; - gbp_endpoint_t *l3_ge; - u32 ii; + gg = gbp_endpoint_group_get (gel->gel_epg); + gef->gef_epg_id = gg->gg_id; + } + else + { + gg = NULL; + } - /* - * look for a matching EP by any of the address - * An EP's IP addresses cannot change so we can search based on - * the first - */ - fproto = fib_proto_from_ip46 (ip46_address_get_type (&ips[0])); + gef->gef_itf = gbp_itf_add_and_lock (gel->gel_sw_if_index, + gbd->gb_bd_index); + + if (!mac_address_is_zero (&ge->ge_key.gek_mac)) + { + gbp_itf_set_l2_input_feature (gef->gef_itf, gei, L2INPUT_FEAT_GBP_FWD); - l3_ge = gbp_endpoint_find_ip (&ips[0], - gbp_endpoint_group_get_fib_index (gg, - fproto)); - if (NULL == l3_ge) + if (gbp_endpoint_is_remote (ge)) { - if (NULL == ge) - { - ge = gbp_endpoint_alloc (epg_id, ggi, sw_if_index, flags, - tun_src, tun_dst); - ge->ge_itf = gbp_itf_add_and_lock (sw_if_index, ~0); - } - /* else - * L2 EP with IPs - */ + gbp_itf_set_l2_output_feature (gef->gef_itf, gei, + L2OUTPUT_FEAT_GBP_POLICY_MAC); } else { - /* modify */ - ge = l3_ge; - ge->ge_flags = flags; - gei = gbp_endpoint_index (ge); - goto out; + gbp_endpoint_add_itf (gef->gef_itf, gei); + gbp_itf_set_l2_output_feature (gef->gef_itf, gei, + L2OUTPUT_FEAT_GBP_POLICY_PORT); } + l2fib_add_entry (ge->ge_key.gek_mac.bytes, + gbd->gb_bd_index, + gef->gef_itf, L2FIB_ENTRY_RESULT_FLAG_STATIC); + } - gei = gbp_endpoint_index (ge); - ge->ge_ips = ips; - vec_validate (ge->ge_adjs, vec_len (ips) - 1); - - vec_foreach_index (ii, ge->ge_ips) + vec_foreach (pfx, ge->ge_key.gek_ips) + { + ethernet_header_t *eth; + u32 ip_sw_if_index; + u32 fib_index; + u8 *rewrite; + index_t ai; + + rewrite = NULL; + grd = gbp_route_domain_get (ge->ge_key.gek_grd); + fib_index = grd->grd_fib_index[pfx->fp_proto]; + + bd_add_del_ip_mac (gbd->gb_bd_index, fib_proto_to_ip46 (pfx->fp_proto), + &pfx->fp_addr, &ge->ge_key.gek_mac, 1); + + /* + * add a host route via the EPG's BVI we need this because the + * adj fib does not install, due to cover refinement check, since + * the BVI's prefix is /32 + */ + vec_validate (rewrite, sizeof (*eth) - 1); + eth = (ethernet_header_t *) rewrite; + + eth->type = clib_host_to_net_u16 ((pfx->fp_proto == FIB_PROTOCOL_IP4 ? + ETHERNET_TYPE_IP4 : + ETHERNET_TYPE_IP6)); + + if (gbp_endpoint_is_remote (ge)) { - ethernet_header_t *eth; - ip46_type_t ip_type; - u32 ip_sw_if_index; - u8 *rewrite; - - rewrite = NULL; - ip = &ge->ge_ips[ii]; - ip_type = ip46_address_get_type (ip); - fproto = fib_proto_from_ip46 (ip_type); - - bd_add_del_ip_mac (gg->gg_bd_index, ip_type, ip, &ge->ge_mac, 1); - - // FIXME - check error - gbp_endpoint_add_ip (ip, - gbp_endpoint_group_get_fib_index (gg, fproto), - gei); - /* - * add a host route via the EPG's BVI we need this because the - * adj fib does not install, due to cover refinement check, since - * the BVI's prefix is /32 + * for dynamic EPs we must add the IP adjacency via the learned + * tunnel since the BD will not contain the EP's MAC since it was + * L3 learned. The dst MAC address used is the 'BD's MAC'. */ - fib_prefix_t pfx = { - .fp_proto = fproto, - .fp_len = ip46_address_get_len (ip), - .fp_addr = *ip, - }; - vec_validate (rewrite, sizeof (*eth) - 1); - eth = (ethernet_header_t *) rewrite; - - eth->type = clib_host_to_net_u16 ((fproto == FIB_PROTOCOL_IP4 ? - ETHERNET_TYPE_IP4 : - ETHERNET_TYPE_IP6)); - - if (gbp_endpoint_is_remote (ge)) - { - /* - * for dynamic EPs we msut add the IP adjacency via the learned - * tunnel since the BD will not contain the EP's MAC since it was - * L3 learned. The dst MAC address used is the 'BD's MAC'. - */ - ip_sw_if_index = ge->ge_sw_if_index; + ip_sw_if_index = gef->gef_itf; - mac_address_to_bytes (gbp_route_domain_get_local_mac (), - eth->src_address); - mac_address_to_bytes (gbp_route_domain_get_remote_mac (), - eth->dst_address); - } - else - { - /* - * for the static EPs we add the IP adjacency via the BVI - * knowing that the BD has the MAC address to route to and - * that policy will be applied on egress to the EP's port - */ - ip_sw_if_index = gbp_endpoint_group_get_bvi (gg); + mac_address_to_bytes (gbp_route_domain_get_local_mac (), + eth->src_address); + mac_address_to_bytes (gbp_route_domain_get_remote_mac (), + eth->dst_address); + } + else + { + /* + * for the static EPs we add the IP adjacency via the BVI + * knowing that the BD has the MAC address to route to and + * that policy will be applied on egress to the EP's port + */ + ip_sw_if_index = gbd->gb_bvi_sw_if_index; - clib_memcpy (eth->src_address, - vnet_sw_interface_get_hw_address (vnet_get_main (), - ip_sw_if_index), - sizeof (eth->src_address)); - mac_address_to_bytes (&ge->ge_mac, eth->dst_address); - } + clib_memcpy (eth->src_address, + vnet_sw_interface_get_hw_address (vnet_get_main (), + ip_sw_if_index), + sizeof (eth->src_address)); + mac_address_to_bytes (&ge->ge_key.gek_mac, eth->dst_address); + } - fib_table_entry_path_add - (gbp_endpoint_group_get_fib_index (gg, fproto), - &pfx, FIB_SOURCE_PLUGIN_LOW, - FIB_ENTRY_FLAG_NONE, - fib_proto_to_dpo (fproto), ip, ip_sw_if_index, - ~0, 1, NULL, FIB_ROUTE_PATH_FLAG_NONE); + fib_table_entry_path_add (fib_index, pfx, + FIB_SOURCE_PLUGIN_LOW, + FIB_ENTRY_FLAG_NONE, + fib_proto_to_dpo (pfx->fp_proto), + &pfx->fp_addr, ip_sw_if_index, + ~0, 1, NULL, FIB_ROUTE_PATH_FLAG_NONE); - ge->ge_adjs[ii] = adj_nbr_add_or_lock_w_rewrite (fproto, - fib_proto_to_link - (fproto), ip, - ip_sw_if_index, - rewrite); + ai = adj_nbr_add_or_lock_w_rewrite (pfx->fp_proto, + fib_proto_to_link (pfx->fp_proto), + &pfx->fp_addr, + ip_sw_if_index, rewrite); + vec_add1 (gef->gef_adjs, ai); + if (NULL != gg) + { if (gbp_endpoint_is_remote (ge)) { dpo_id_t policy_dpo = DPO_INVALID; @@ -460,13 +721,13 @@ gbp_endpoint_update (u32 sw_if_index, * interpose a policy DPO from the endpoint so that policy * is applied */ - gbp_policy_dpo_add_or_lock (fib_proto_to_dpo (fproto), + gbp_policy_dpo_add_or_lock (fib_proto_to_dpo (pfx->fp_proto), gg->gg_id, ~0, &policy_dpo); - fib_table_entry_special_dpo_add - (gbp_endpoint_group_get_fib_index (gg, fproto), - &pfx, - FIB_SOURCE_PLUGIN_HI, FIB_ENTRY_FLAG_INTERPOSE, &policy_dpo); + fib_table_entry_special_dpo_add (fib_index, pfx, + FIB_SOURCE_PLUGIN_HI, + FIB_ENTRY_FLAG_INTERPOSE, + &policy_dpo); } /* @@ -474,68 +735,148 @@ gbp_endpoint_update (u32 sw_if_index, * that if this EP has moved from some other place in the * 'fabric', upstream devices are informed */ - if (!(gbp_endpoint_is_remote (ge)) && ~0 != gg->gg_uplink_sw_if_index) + if (!gbp_endpoint_is_remote (ge) && ~0 != gg->gg_uplink_sw_if_index) { - gbp_endpoint_add_itf (sw_if_index, gei); - if (ip46_address_is_ip4 (ip)) + gbp_endpoint_add_itf (gef->gef_itf, gei); + if (FIB_PROTOCOL_IP4 == pfx->fp_proto) send_ip4_garp_w_addr (vlib_get_main (), - &ip->ip4, gg->gg_uplink_sw_if_index); + &pfx->fp_addr.ip4, + gg->gg_uplink_sw_if_index); else send_ip6_na_w_addr (vlib_get_main (), - &ip->ip6, gg->gg_uplink_sw_if_index); + &pfx->fp_addr.ip6, + gg->gg_uplink_sw_if_index); } } + } + + if (!gbp_endpoint_is_remote (ge)) + { + /* + * non-remote endpoints (i.e. those not arriving on iVXLAN + * tunnels) need to be classifed based on the the input interface. + * We enable the GBP-FWD feature only if the group has an uplink + * interface (on which the GBP-FWD feature would send UU traffic). + */ + l2input_feat_masks_t feats = L2INPUT_FEAT_GBP_SRC_CLASSIFY; + + if (NULL != gg && ~0 != gg->gg_uplink_sw_if_index) + feats |= L2INPUT_FEAT_GBP_FWD; + gbp_itf_set_l2_input_feature (gef->gef_itf, gei, feats); } - if (NULL == ge) - return (0); + /* + * update children with the new forwarding info + */ + fib_node_back_walk_ctx_t bw_ctx = { + .fnbw_reason = FIB_NODE_BW_REASON_FLAG_EVALUATE, + .fnbw_flags = FIB_NODE_BW_FLAG_FORCE_SYNC, + }; + + fib_walk_sync (gbp_endpoint_fib_type, gei, &bw_ctx); +} + +int +gbp_endpoint_update_and_lock (gbp_endpoint_src_t src, + u32 sw_if_index, + const ip46_address_t * ips, + const mac_address_t * mac, + index_t gbdi, index_t grdi, epg_id_t epg_id, + gbp_endpoint_flags_t flags, + const ip46_address_t * tun_src, + const ip46_address_t * tun_dst, u32 * handle) +{ + gbp_bridge_domain_t *gbd; + gbp_endpoint_group_t *gg; + gbp_endpoint_src_t best; + gbp_route_domain_t *grd; + gbp_endpoint_loc_t *gel; + gbp_endpoint_t *ge; + index_t ggi, gei; + int rv; + + if (~0 == sw_if_index) + return (VNET_API_ERROR_INVALID_SW_IF_INDEX); + + ge = NULL; + gg = NULL; /* - * count the number of dynamic entries and kick off the scanner - * process is this is our first. + * we need to determine the bridge-domain, either from the EPG or + * the BD passed */ - if (gbp_endpoint_is_remote (ge)) + if (EPG_INVALID != epg_id) { - gbp_n_learnt_endpoints++; + ggi = gbp_endpoint_group_find (epg_id); - if (1 == gbp_n_learnt_endpoints) - { - vlib_process_signal_event (vlib_get_main (), - gbp_scanner_node.index, - GBP_ENDPOINT_SCAN_START, 0); - } + if (INDEX_INVALID == ggi) + return (VNET_API_ERROR_NO_SUCH_ENTRY); + + gg = gbp_endpoint_group_get (ggi); + gbdi = gg->gg_gbd; + grdi = gg->gg_rd; + } + else + { + if (INDEX_INVALID == gbdi) + return (VNET_API_ERROR_NO_SUCH_ENTRY); + if (INDEX_INVALID == grdi) + return (VNET_API_ERROR_NO_SUCH_FIB); + ggi = INDEX_INVALID; + } + + gbd = gbp_bridge_domain_get (gbdi); + grd = gbp_route_domain_get (grdi); + rv = gbp_endpoint_find_for_update (ips, grd, mac, gbd, &ge); + + if (0 != rv) + return (rv); + + if (NULL == ge) + { + ge = gbp_endpoint_alloc (ips, grd, mac, gbd); } else { + gbp_endpoint_ips_update (ge, ips, grd); + } + + best = gbp_endpoint_get_best_src (ge); + gei = gbp_endpoint_index (ge); + gel = gbp_endpoint_loc_find_or_add (ge, src); + + gbp_endpoint_loc_update (gel, sw_if_index, ggi, flags, tun_src, tun_dst); + + if (src <= best) + { /* - * non-remote endpoints (i.e. those not arriving on iVXLAN - * tunnels) need to be classifed based on the the input interface. - * We enable the GBP-FWD feature only is the group has an uplink - * interface (on which the GBP-FWD feature would send UU traffic). + * either the best source has been updated or we have a new best source + */ + gbb_endpoint_fwd_reset (ge); + gbb_endpoint_fwd_recalc (ge); + } + else + { + /* + * an update to a lower priority source, so we need do nothing */ - l2input_feat_masks_t feats = L2INPUT_FEAT_GBP_SRC_CLASSIFY; - - if (~0 != gg->gg_uplink_sw_if_index) - feats |= L2INPUT_FEAT_GBP_FWD; - gbp_itf_set_l2_input_feature (ge->ge_itf, gbp_endpoint_index (ge), - feats); } -out: if (handle) - *handle = (ge - gbp_endpoint_pool); + *handle = gei; - gbp_endpoint_group_unlock (ggi); GBP_ENDPOINT_INFO ("update: %U", format_gbp_endpoint, gei); return (0); } void -gbp_endpoint_delete (index_t gei) +gbp_endpoint_unlock (gbp_endpoint_src_t src, index_t gei) { - gbp_endpoint_group_t *gg; + gbp_endpoint_loc_t *gel, gel_copy; + gbp_endpoint_src_t best; gbp_endpoint_t *ge; + int removed; if (pool_is_free_index (gbp_endpoint_pool, gei)) return; @@ -543,86 +884,90 @@ gbp_endpoint_delete (index_t gei) GBP_ENDPOINT_INFO ("delete: %U", format_gbp_endpoint, gei); ge = gbp_endpoint_get (gei); - gg = gbp_endpoint_group_get (ge->ge_epg); - - gbp_endpoint_del_mac (&ge->ge_mac, gg->gg_bd_index); - l2fib_del_entry (ge->ge_mac.bytes, gg->gg_bd_index, ge->ge_sw_if_index); - gbp_itf_set_l2_input_feature (ge->ge_itf, gei, (L2INPUT_FEAT_NONE)); - gbp_itf_set_l2_output_feature (ge->ge_itf, gei, L2OUTPUT_FEAT_NONE); - - if (NULL != ge->ge_ips) - { - const ip46_address_t *ip; - index_t *ai; - - vec_foreach (ai, ge->ge_adjs) - { - adj_unlock (*ai); - } - vec_foreach (ip, ge->ge_ips) - { - fib_protocol_t fproto; - ip46_type_t ip_type; - ip_type = ip46_address_get_type (ip); - fproto = fib_proto_from_ip46 (ip_type); + gel = gbp_endpoint_loc_find (ge, src); - gbp_endpoint_del_ip (ip, - gbp_endpoint_group_get_fib_index (gg, fproto)); - - bd_add_del_ip_mac (gg->gg_bd_index, ip_type, ip, &ge->ge_mac, 0); + if (NULL == gel) + return; - /* - * remove a host route via the EPG's BVI - */ - fib_prefix_t pfx = { - .fp_proto = fproto, - .fp_len = ip46_address_get_len (ip), - .fp_addr = *ip, - }; + /* + * lock the EP so we can control when it is deleted + */ + fib_node_lock (&ge->ge_node); + best = gbp_endpoint_get_best_src (ge); - if (gbp_endpoint_is_remote (ge)) - { - fib_table_entry_special_remove - (gbp_endpoint_group_get_fib_index (gg, fproto), - &pfx, FIB_SOURCE_PLUGIN_HI); - } + /* + * copy the location info since we'll lose it when it's removed from + * the vector + */ + clib_memcpy (&gel_copy, gel, sizeof (gel_copy)); - fib_table_entry_path_remove - (gbp_endpoint_group_get_fib_index (gg, fproto), - &pfx, FIB_SOURCE_PLUGIN_LOW, - fib_proto_to_dpo (fproto), ip, - (gbp_endpoint_is_remote (ge) ? - ge->ge_sw_if_index : - gbp_endpoint_group_get_bvi (gg)), - ~0, 1, FIB_ROUTE_PATH_FLAG_NONE); - } - } + /* + * remove the source we no longer need + */ + removed = gbp_endpoint_loc_unlock (ge, gel); - if (ge->ge_flags & GBP_ENDPOINT_FLAG_LEARNT) + if (src == best) { - gbp_n_learnt_endpoints--; - - if (0 == gbp_n_learnt_endpoints) + /* + * we have removed the old best source => recalculate fwding + */ + if (0 == vec_len (ge->ge_locs)) + { + /* + * if there are no more sources left, then we need only release + * the fwding resources held and then this EP is gawn. + */ + gbb_endpoint_fwd_reset (ge); + } + else { - vlib_process_signal_event (vlib_get_main (), - gbp_scanner_node.index, - GBP_ENDPOINT_SCAN_STOP, 0); + /* + * else there are more sources. release the old and get new + * fwding objects + */ + gbb_endpoint_fwd_reset (ge); + gbb_endpoint_fwd_recalc (ge); } } + /* + * else + * we removed a lower priority source so we need to do nothing + */ - gbp_itf_unlock (ge->ge_itf); - if (gbp_endpoint_is_remote (ge)) - { - vxlan_gbp_tunnel_unlock (ge->ge_sw_if_index); - } - gbp_endpoint_group_unlock (ge->ge_epg); - pool_put (gbp_endpoint_pool, ge); + /* + * clear up any resources held by the source + */ + if (removed) + gbp_endpoint_loc_destroy (&gel_copy); + + /* + * remove the lock taken above + */ + fib_node_unlock (&ge->ge_node); + /* + * We may have removed the last source and so this EP is now TOAST + * DO NOTHING BELOW HERE + */ +} + +u32 +gbp_endpoint_child_add (index_t gei, + fib_node_type_t type, fib_node_index_t index) +{ + return (fib_node_child_add (gbp_endpoint_fib_type, gei, type, index)); +} + +void +gbp_endpoint_child_remove (index_t gei, u32 sibling) +{ + return (fib_node_child_remove (gbp_endpoint_fib_type, gei, sibling)); } typedef struct gbp_endpoint_flush_ctx_t_ { u32 sw_if_index; + gbp_endpoint_src_t src; index_t *geis; } gbp_endpoint_flush_ctx_t; @@ -630,12 +975,13 @@ static walk_rc_t gbp_endpoint_flush_cb (index_t gei, void *args) { gbp_endpoint_flush_ctx_t *ctx = args; + gbp_endpoint_loc_t *gel; gbp_endpoint_t *ge; ge = gbp_endpoint_get (gei); + gel = gbp_endpoint_loc_find (ge, ctx->src); - if (gbp_endpoint_is_remote (ge) && - ctx->sw_if_index == ge->tun.ge_parent_sw_if_index) + if ((NULL != gel) && ctx->sw_if_index == gel->tun.gel_parent_sw_if_index) { vec_add1 (ctx->geis, gei); } @@ -647,18 +993,23 @@ gbp_endpoint_flush_cb (index_t gei, void *args) * remove all learnt endpoints using the interface */ void -gbp_endpoint_flush (u32 sw_if_index) +gbp_endpoint_flush (gbp_endpoint_src_t src, u32 sw_if_index) { gbp_endpoint_flush_ctx_t ctx = { .sw_if_index = sw_if_index, + .src = src, }; index_t *gei; + GBP_ENDPOINT_INFO ("flush: %U %U", + format_gbp_endpoint_src, src, + format_vnet_sw_if_index_name, vnet_get_main (), + sw_if_index); gbp_endpoint_walk (gbp_endpoint_flush_cb, &ctx); vec_foreach (gei, ctx.geis) { - gbp_endpoint_delete (*gei); + gbp_endpoint_unlock (src, *gei); } vec_free (ctx.geis); @@ -724,8 +1075,12 @@ gbp_endpoint_cli (vlib_main_t * vm, return clib_error_return (0, "EPG-ID must be specified"); rv = - gbp_endpoint_update (sw_if_index, ips, &mac, epg_id, - GBP_ENDPOINT_FLAG_NONE, NULL, NULL, &handle); + gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_CP, + sw_if_index, ips, &mac, + INDEX_INVALID, INDEX_INVALID, + epg_id, + GBP_ENDPOINT_FLAG_NONE, + NULL, NULL, &handle); if (rv) return clib_error_return (0, "GBP Endpoint update returned %d", rv); @@ -737,7 +1092,7 @@ gbp_endpoint_cli (vlib_main_t * vm, if (INDEX_INVALID == handle) return clib_error_return (0, "handle must be specified"); - gbp_endpoint_delete (handle); + gbp_endpoint_unlock (GBP_ENDPOINT_SRC_CP, handle); } vec_free (ips); @@ -745,7 +1100,6 @@ gbp_endpoint_cli (vlib_main_t * vm, return (NULL); } - /*? * Configure a GBP Endpoint * @@ -762,33 +1116,100 @@ VLIB_CLI_COMMAND (gbp_endpoint_cli_node, static) = { /* *INDENT-ON* */ u8 * -format_gbp_endpoint (u8 * s, va_list * args) +format_gbp_endpoint_src (u8 * s, va_list * args) { - index_t gei = va_arg (*args, index_t); - const ip46_address_t *ip; - gbp_endpoint_t *ge; + gbp_endpoint_src_t action = va_arg (*args, gbp_endpoint_src_t); - ge = gbp_endpoint_get (gei); + switch (action) + { +#define _(v,a) case GBP_ENDPOINT_SRC_##v: return (format (s, "%s", a)); + foreach_gbp_endpoint_src +#undef _ + } + + return (format (s, "unknown")); +} + +static u8 * +format_gbp_endpoint_fwd (u8 * s, va_list * args) +{ + gbp_endpoint_fwd_t *gef = va_arg (*args, gbp_endpoint_fwd_t *); + + s = format (s, "fwd:"); + s = format (s, "\n itf:[%U]", format_gbp_itf, gef->gef_itf); + if (GBP_ENDPOINT_FLAG_NONE != gef->gef_flags) + { + s = format (s, " flags:%U", format_gbp_endpoint_flags, gef->gef_flags); + } + + return (s); +} + +static u8 * +format_gbp_endpoint_key (u8 * s, va_list * args) +{ + gbp_endpoint_key_t *gek = va_arg (*args, gbp_endpoint_key_t *); + const fib_prefix_t *pfx; - s = format (s, "[@%d] ", gei); - s = format (s, "IPs:["); + s = format (s, "ips:["); - vec_foreach (ip, ge->ge_ips) + vec_foreach (pfx, gek->gek_ips) { - s = format (s, "%U, ", format_ip46_address, ip, IP46_TYPE_ANY); + s = format (s, "%U, ", format_fib_prefix, pfx); } s = format (s, "]"); - s = format (s, " MAC:%U", format_mac_address_t, &ge->ge_mac); - s = format (s, " EPG-ID:%d", ge->ge_epg_id); - if (GBP_ENDPOINT_FLAG_NONE != ge->ge_flags) + s = format (s, " mac:%U", format_mac_address_t, &gek->gek_mac); + + return (s); +} + +static u8 * +format_gbp_endpoint_loc (u8 * s, va_list * args) +{ + gbp_endpoint_loc_t *gel = va_arg (*args, gbp_endpoint_loc_t *); + + s = format (s, "%U", format_gbp_endpoint_src, gel->gel_src); + s = + format (s, "\n %U", format_vnet_sw_if_index_name, vnet_get_main (), + gel->gel_sw_if_index); + s = format (s, " EPG:%d", gel->gel_epg); + + if (GBP_ENDPOINT_FLAG_NONE != gel->gel_flags) + { + s = format (s, " flags:%U", format_gbp_endpoint_flags, gel->gel_flags); + } + if (GBP_ENDPOINT_FLAG_REMOTE & gel->gel_flags) { - s = format (s, " flags:%U", format_gbp_endpoint_flags, ge->ge_flags); + s = format (s, " tun:["); + s = format (s, "parent:%U", format_vnet_sw_if_index_name, + vnet_get_main (), gel->tun.gel_parent_sw_if_index); + s = format (s, " {%U,%U}]", + format_ip46_address, &gel->tun.gel_src, IP46_TYPE_ANY, + format_ip46_address, &gel->tun.gel_dst, IP46_TYPE_ANY); } - s = format (s, " itf:[%U]", format_gbp_itf, ge->ge_itf); + return (s); +} + +u8 * +format_gbp_endpoint (u8 * s, va_list * args) +{ + index_t gei = va_arg (*args, index_t); + gbp_endpoint_loc_t *gel; + gbp_endpoint_t *ge; + + ge = gbp_endpoint_get (gei); + + s = format (s, "[@%d] %U", gei, format_gbp_endpoint_key, &ge->ge_key); s = format (s, " last-time:[%f]", ge->ge_last_time); + vec_foreach (gel, ge->ge_locs) + { + s = format (s, "\n %U", format_gbp_endpoint_loc, gel); + } + s = format (s, "\n %U", format_gbp_endpoint_fwd, &ge->ge_fwd); + return s; } @@ -895,16 +1316,16 @@ VLIB_CLI_COMMAND (gbp_endpoint_show_node, static) = { static void gbp_endpoint_check (index_t gei, f64 start_time) { + gbp_endpoint_loc_t *gel; gbp_endpoint_t *ge; ge = gbp_endpoint_get (gei); + gel = gbp_endpoint_loc_find (ge, GBP_ENDPOINT_SRC_DP); - GBP_ENDPOINT_DBG ("scan at:%f -> %U", start_time, format_gbp_endpoint, gei); - - if ((ge->ge_flags & GBP_ENDPOINT_FLAG_LEARNT) && + if ((NULL != gel) && ((start_time - ge->ge_last_time) > GBP_ENDPOINT_INACTIVE_TIME)) { - gbp_endpoint_delete (gei); + gbp_endpoint_unlock (GBP_ENDPOINT_SRC_DP, gei); } } @@ -1031,12 +1452,77 @@ gbp_endpoint_scan_threshold (void) return (GBP_ENDPOINT_INACTIVE_TIME); } -#define GBP_EP_HASH_NUM_BUCKETS (2 * 1024) -#define GBP_EP_HASH_MEMORY_SIZE (1 << 20) +static fib_node_t * +gbp_endpoint_get_node (fib_node_index_t index) +{ + gbp_endpoint_t *ge; + + ge = gbp_endpoint_get (index); + + return (&ge->ge_node); +} + +static gbp_endpoint_t * +gbp_endpoint_from_fib_node (fib_node_t * node) +{ + ASSERT (gbp_endpoint_fib_type == node->fn_type); + return ((gbp_endpoint_t *) node); +} + +static void +gbp_endpoint_last_lock_gone (fib_node_t * node) +{ + const gbp_bridge_domain_t *gbd; + const gbp_route_domain_t *grd; + const fib_prefix_t *pfx; + gbp_endpoint_t *ge; + + ge = gbp_endpoint_from_fib_node (node); + + ASSERT (0 == vec_len (ge->ge_locs)); + + gbd = gbp_bridge_domain_get (ge->ge_key.gek_gbd); + + /* + * we have removed the last source. this EP is toast + */ + if (INDEX_INVALID != ge->ge_key.gek_gbd) + { + gbp_endpoint_del_mac (&ge->ge_key.gek_mac, gbd->gb_bd_index); + } + vec_foreach (pfx, ge->ge_key.gek_ips) + { + grd = gbp_route_domain_get (ge->ge_key.gek_grd); + gbp_endpoint_del_ip (&pfx->fp_addr, grd->grd_fib_index[pfx->fp_proto]); + } + pool_put (gbp_endpoint_pool, ge); +} + +static fib_node_back_walk_rc_t +gbp_endpoint_back_walk_notify (fib_node_t * node, + fib_node_back_walk_ctx_t * ctx) +{ + ASSERT (0); + + return (FIB_NODE_BACK_WALK_CONTINUE); +} + +/* + * The FIB path's graph node virtual function table + */ +static const fib_node_vft_t gbp_endpoint_vft = { + .fnv_get = gbp_endpoint_get_node, + .fnv_last_lock = gbp_endpoint_last_lock_gone, + .fnv_back_walk = gbp_endpoint_back_walk_notify, + // .fnv_mem_show = fib_path_memory_show, +}; static clib_error_t * gbp_endpoint_init (vlib_main_t * vm) { +#define GBP_EP_HASH_NUM_BUCKETS (2 * 1024) +#define GBP_EP_HASH_MEMORY_SIZE (1 << 20) + clib_bihash_init_24_8 (&gbp_ep_db.ged_by_ip_rd, "GBP Endpoints - IP/RD", GBP_EP_HASH_NUM_BUCKETS, GBP_EP_HASH_MEMORY_SIZE); @@ -1046,6 +1532,7 @@ gbp_endpoint_init (vlib_main_t * vm) GBP_EP_HASH_NUM_BUCKETS, GBP_EP_HASH_MEMORY_SIZE); gbp_ep_logger = vlib_log_register_class ("gbp", "ep"); + gbp_endpoint_fib_type = fib_node_register_new_type (&gbp_endpoint_vft); return (NULL); } diff --git a/src/plugins/gbp/gbp_endpoint.h b/src/plugins/gbp/gbp_endpoint.h index bd157c99f1b..d56e91daa99 100644 --- a/src/plugins/gbp/gbp_endpoint.h +++ b/src/plugins/gbp/gbp_endpoint.h @@ -54,65 +54,152 @@ typedef enum gbp_endpoint_flags_t_ extern u8 *format_gbp_endpoint_flags (u8 * s, va_list * args); /** - * A Group Based Policy Endpoint. - * This is typically a VM or container. If the endpoint is local (i.e. on - * the same compute node as VPP) then there is one interface per-endpoint. - * If the EP is remote,e.g. reachable over a [vxlan] tunnel, then there - * will be multiple EPs reachable over the tunnel and they can be distinguished - * via either their MAC or IP Address[es]. + * Sources of Endpoints in priority order. The best (lowest value) source + * provides the forwarding information */ -typedef struct gbp_endpoint_t_ +#define foreach_gbp_endpoint_src \ + _(CP, "control-plane") \ + _(DP, "data-plane") \ + _(RR, "recursive-resolution") + +typedef enum gbp_endpoint_src_t_ +{ +#define _(v,s) GBP_ENDPOINT_SRC_##v, + foreach_gbp_endpoint_src +#undef _ +} gbp_endpoint_src_t; + +#define GBP_ENDPOINT_SRC_MAX (GBP_ENDPOINT_SRC_RR+1) + +extern u8 *format_gbp_endpoint_src (u8 * s, va_list * args); + +/** + * This is the identity of an endpoint, as such it is information + * about an endpoint that is idempotent. + * The ID is used to add the EP into the various data-bases for retrieval. + */ +typedef struct gbp_endpoint_key_t_ { /** - * The interface on which the EP is connected + * A vector of ip addresses that belong to the endpoint. + * Together with the route EPG's RD this forms the EP's L3 key */ - index_t ge_itf; - u32 ge_sw_if_index; + fib_prefix_t *gek_ips; /** - * A vector of ip addresses that below to the endpoint + * MAC address of the endpoint. + * Together with the route EPG's BD this forms the EP's L2 key */ - const ip46_address_t *ge_ips; + mac_address_t gek_mac; /** - * MAC address of the endpoint + * Index of the Bridge-Domain */ - mac_address_t ge_mac; + index_t gek_gbd; /** - * Index of the Endpoint's Group + * Index of the Route-Domain */ - index_t ge_epg; + index_t gek_grd; +} gbp_endpoint_key_t; + +/** + * Information about the location of the endpoint provided by a source + * of endpoints + */ +typedef struct gbp_endpoint_loc_t_ +{ + /** + * The source providing this location information + */ + gbp_endpoint_src_t gel_src; /** - * Endpoint Group's ID + * The interface on which the EP is connected */ - index_t ge_epg_id; + u32 gel_sw_if_index; /** * Endpoint flags */ - gbp_endpoint_flags_t ge_flags; + gbp_endpoint_flags_t gel_flags; /** - * The L3 adj, if created + * Endpoint Group. */ - index_t *ge_adjs; + index_t gel_epg; /** - * The last time a packet from seen from this end point + * number of times this source has locked this */ - f64 ge_last_time; + u32 gel_locks; /** * Tunnel info for remote endpoints */ struct { - u32 ge_parent_sw_if_index; - ip46_address_t ge_src; - ip46_address_t ge_dst; + u32 gel_parent_sw_if_index; + ip46_address_t gel_src; + ip46_address_t gel_dst; } tun; +} gbp_endpoint_loc_t; + +/** + * And endpoints current forwarding state + */ +typedef struct gbp_endpoint_fwd_t_ +{ + /** + * The interface on which the EP is connected + */ + index_t gef_itf; + + /** + * The L3 adj, if created + */ + index_t *gef_adjs; + + /** + * Endpoint Group's ID. cached for fast DP access. + */ + epg_id_t gef_epg_id; + + gbp_endpoint_flags_t gef_flags; +} gbp_endpoint_fwd_t; + +/** + * A Group Based Policy Endpoint. + * This is typically a VM or container. If the endpoint is local (i.e. on + * the same compute node as VPP) then there is one interface per-endpoint. + * If the EP is remote,e.g. reachable over a [vxlan] tunnel, then there + * will be multiple EPs reachable over the tunnel and they can be distinguished + * via either their MAC or IP Address[es]. + */ +typedef struct gbp_endpoint_t_ +{ + /** + * A FIB node that allows the tracking of children. + */ + fib_node_t ge_node; + + /** + * The key/ID of this EP + */ + gbp_endpoint_key_t ge_key; + + /** + * Location information provided by the various sources. + * These are sorted based on source priority. + */ + gbp_endpoint_loc_t *ge_locs; + + gbp_endpoint_fwd_t ge_fwd; + + /** + * The last time a packet from seen from this end point + */ + f64 ge_last_time; } gbp_endpoint_t; extern u8 *format_gbp_endpoint (u8 * s, va_list * args); @@ -127,22 +214,31 @@ typedef struct gbp_ep_by_ip_itf_db_t_ clib_bihash_16_8_t ged_by_mac_bd; } gbp_ep_db_t; -extern int gbp_endpoint_update (u32 sw_if_index, - const ip46_address_t * ip, - const mac_address_t * mac, - epg_id_t epg_id, - gbp_endpoint_flags_t flags, - const ip46_address_t * tun_src, - const ip46_address_t * tun_dst, u32 * handle); -extern void gbp_endpoint_delete (index_t gbpei); +extern int gbp_endpoint_update_and_lock (gbp_endpoint_src_t src, + u32 sw_if_index, + const ip46_address_t * ip, + const mac_address_t * mac, + index_t gbd, index_t grd, + epg_id_t epg_id, + gbp_endpoint_flags_t flags, + const ip46_address_t * tun_src, + const ip46_address_t * tun_dst, + u32 * handle); +extern void gbp_endpoint_unlock (gbp_endpoint_src_t src, index_t gbpei); +extern u32 gbp_endpoint_child_add (index_t gei, + fib_node_type_t type, + fib_node_index_t index); +extern void gbp_endpoint_child_remove (index_t gei, u32 sibling); typedef walk_rc_t (*gbp_endpoint_cb_t) (index_t gbpei, void *ctx); extern void gbp_endpoint_walk (gbp_endpoint_cb_t cb, void *ctx); extern void gbp_endpoint_scan (vlib_main_t * vm); extern f64 gbp_endpoint_scan_threshold (void); extern int gbp_endpoint_is_remote (const gbp_endpoint_t * ge); +extern int gbp_endpoint_is_learnt (const gbp_endpoint_t * ge); + -extern void gbp_endpoint_flush (u32 sw_if_index); +extern void gbp_endpoint_flush (gbp_endpoint_src_t src, u32 sw_if_index); /** * DP functions and databases diff --git a/src/plugins/gbp/gbp_endpoint_group.c b/src/plugins/gbp/gbp_endpoint_group.c index ee4af2c0060..834f865bc92 100644 --- a/src/plugins/gbp/gbp_endpoint_group.c +++ b/src/plugins/gbp/gbp_endpoint_group.c @@ -44,7 +44,7 @@ gbp_endpoint_group_get (index_t i) return (pool_elt_at_index (gbp_endpoint_group_pool, i)); } -static void +void gbp_endpoint_group_lock (index_t i) { gbp_endpoint_group_t *gg; @@ -66,21 +66,6 @@ gbp_endpoint_group_find (epg_id_t epg_id) return (INDEX_INVALID); } -index_t -gbp_endpoint_group_find_and_lock (epg_id_t epg_id) -{ - uword *p; - - p = hash_get (gbp_endpoint_group_db.gg_hash, epg_id); - - if (NULL != p) - { - gbp_endpoint_group_lock (p[0]); - return p[0]; - } - return (INDEX_INVALID); -} - int gbp_endpoint_group_add_and_lock (epg_id_t epg_id, u32 bd_id, u32 rd_id, u32 uplink_sw_if_index) @@ -165,6 +150,9 @@ gbp_endpoint_group_unlock (index_t ggi) { gbp_endpoint_group_t *gg; + if (INDEX_INVALID == ggi) + return; + gg = gbp_endpoint_group_get (ggi); gg->gg_locks--; @@ -227,7 +215,7 @@ gbp_endpoint_group_get_bd_id (const gbp_endpoint_group_t * gg) } index_t -gbp_endpoint_group_get_fib_index (gbp_endpoint_group_t * gg, +gbp_endpoint_group_get_fib_index (const gbp_endpoint_group_t * gg, fib_protocol_t fproto) { const gbp_route_domain_t *grd; @@ -237,16 +225,6 @@ gbp_endpoint_group_get_fib_index (gbp_endpoint_group_t * gg, return (grd->grd_fib_index[fproto]); } -u32 -gbp_endpoint_group_get_bvi (gbp_endpoint_group_t * gg) -{ - const gbp_bridge_domain_t *gb; - - gb = gbp_bridge_domain_get (gg->gg_gbd); - - return (gb->gb_bvi_sw_if_index); -} - void gbp_endpoint_group_walk (gbp_endpoint_group_cb_t cb, void *ctx) { diff --git a/src/plugins/gbp/gbp_endpoint_group.h b/src/plugins/gbp/gbp_endpoint_group.h index 7116a058aee..763a80e4d87 100644 --- a/src/plugins/gbp/gbp_endpoint_group.h +++ b/src/plugins/gbp/gbp_endpoint_group.h @@ -74,16 +74,15 @@ extern int gbp_endpoint_group_add_and_lock (epg_id_t epg_id, u32 bd_id, u32 rd_id, u32 uplink_sw_if_index); -extern index_t gbp_endpoint_group_find_and_lock (epg_id_t epg_id); extern index_t gbp_endpoint_group_find (epg_id_t epg_id); extern int gbp_endpoint_group_delete (epg_id_t epg_id); extern void gbp_endpoint_group_unlock (index_t index); +extern void gbp_endpoint_group_lock (index_t index); extern u32 gbp_endpoint_group_get_bd_id (const gbp_endpoint_group_t *); extern gbp_endpoint_group_t *gbp_endpoint_group_get (index_t i); -extern index_t gbp_endpoint_group_get_fib_index (gbp_endpoint_group_t * gg, - fib_protocol_t fproto); -extern u32 gbp_endpoint_group_get_bvi (gbp_endpoint_group_t * gg); +extern index_t gbp_endpoint_group_get_fib_index (const gbp_endpoint_group_t * + gg, fib_protocol_t fproto); typedef int (*gbp_endpoint_group_cb_t) (gbp_endpoint_group_t * gbpe, void *ctx); diff --git a/src/plugins/gbp/gbp_learn.c b/src/plugins/gbp/gbp_learn.c index cd6a15d955d..762b463223e 100644 --- a/src/plugins/gbp/gbp_learn.c +++ b/src/plugins/gbp/gbp_learn.c @@ -110,11 +110,14 @@ gbp_learn_l2_cp (const gbp_learn_l2_t * gl2) * flip the source and dst, since that's how it was received, this API * takes how it's sent */ - gbp_endpoint_update (gl2->sw_if_index, ips, - &gl2->mac, gl2->epg, - (GBP_ENDPOINT_FLAG_LEARNT | - GBP_ENDPOINT_FLAG_REMOTE), - &gl2->outer_dst, &gl2->outer_src, NULL); + gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_DP, + gl2->sw_if_index, ips, + &gl2->mac, INDEX_INVALID, + INDEX_INVALID, gl2->epg, + (GBP_ENDPOINT_FLAG_LEARNT | + GBP_ENDPOINT_FLAG_REMOTE), + &gl2->outer_dst, &gl2->outer_src, NULL); + vec_free (ips); } static void @@ -273,7 +276,7 @@ gbp_learn_l2 (vlib_main_t * vm, /* * check for new EP or a moved EP */ - if (NULL == ge0 || ge0->ge_sw_if_index != sw_if_index0) + if (NULL == ge0 || ge0->ge_fwd.gef_itf != sw_if_index0) { /* @@ -415,10 +418,13 @@ gbp_learn_l3_cp (const gbp_learn_l3_t * gl3) vec_add1 (ips, gl3->ip); - gbp_endpoint_update (gl3->sw_if_index, ips, NULL, gl3->epg, - (GBP_ENDPOINT_FLAG_REMOTE | - GBP_ENDPOINT_FLAG_LEARNT), - &gl3->outer_dst, &gl3->outer_src, NULL); + gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_DP, + gl3->sw_if_index, ips, NULL, + INDEX_INVALID, INDEX_INVALID, gl3->epg, + (GBP_ENDPOINT_FLAG_REMOTE | + GBP_ENDPOINT_FLAG_LEARNT), + &gl3->outer_dst, &gl3->outer_src, NULL); + vec_free (ips); } static void diff --git a/src/plugins/gbp/gbp_policy.c b/src/plugins/gbp/gbp_policy.c index 6d84a994d41..3b204ff4da8 100644 --- a/src/plugins/gbp/gbp_policy.c +++ b/src/plugins/gbp/gbp_policy.c @@ -14,6 +14,7 @@ */ #include <plugins/gbp/gbp.h> +#include <plugins/gbp/gbp_policy_dpo.h> #include <vnet/vxlan-gbp/vxlan_gbp_packet.h> @@ -67,6 +68,44 @@ typedef struct gbp_policy_trace_t_ u32 allowed; } gbp_policy_trace_t; +always_inline dpo_proto_t +ethertype_to_dpo_proto (u16 etype) +{ + etype = clib_net_to_host_u16 (etype); + + switch (etype) + { + case ETHERNET_TYPE_IP4: + return (DPO_PROTO_IP4); + case ETHERNET_TYPE_IP6: + return (DPO_PROTO_IP6); + } + + return (DPO_PROTO_NONE); +} + +always_inline u32 +gbp_rule_l2_redirect (const gbp_rule_t * gu, vlib_buffer_t * b0) +{ + const ethernet_header_t *eth0; + const dpo_id_t *dpo; + dpo_proto_t dproto; + + eth0 = vlib_buffer_get_current (b0); + /* pop the ethernet header to prepare for L3 rewrite */ + vlib_buffer_advance (b0, vnet_buffer (b0)->l2.l2_len); + + dproto = ethertype_to_dpo_proto (eth0->type); + dpo = &gu->gu_dpo[GBP_POLICY_NODE_L2][dproto]; + + /* save the LB index for the next node and reset the IP flow hash + * so it's recalculated */ + vnet_buffer (b0)->ip.adj_index[VLIB_TX] = dpo->dpoi_index; + vnet_buffer (b0)->ip.flow_hash = 0; + + return (dpo->dpoi_next_node); +} + static uword gbp_policy_inline (vlib_main_t * vm, vlib_node_runtime_t * node, @@ -93,12 +132,12 @@ gbp_policy_inline (vlib_main_t * vm, const gbp_endpoint_t *ge0; gbp_policy_next_t next0; gbp_contract_key_t key0; - gbp_contract_value_t value0 = { - .as_u64 = ~0, - }; + gbp_contract_t *gc0; u32 bi0, sw_if_index0; vlib_buffer_t *b0; + index_t gci0; + gc0 = NULL; next0 = GBP_POLICY_NEXT_DENY; bi0 = from[0]; to_next[0] = bi0; @@ -136,7 +175,7 @@ gbp_policy_inline (vlib_main_t * vm, vnet_buffer (b0)->l2.bd_index); if (NULL != ge0) - key0.gck_dst = ge0->ge_epg_id; + key0.gck_dst = ge0->ge_fwd.gef_epg_id; else /* If you cannot determine the destination EP then drop */ goto trace; @@ -161,9 +200,9 @@ gbp_policy_inline (vlib_main_t * vm, } else { - value0.as_u64 = gbp_acl_lookup (&key0); + gci0 = gbp_contract_find (&key0); - if (~0 != value0.gc_lc_index) + if (INDEX_INVALID != gci0) { fa_5tuple_opaque_t pkt_5tuple0; u8 action0 = 0; @@ -173,6 +212,7 @@ gbp_policy_inline (vlib_main_t * vm, u16 ether_type0; u8 is_ip60 = 0; + gc0 = gbp_contract_get (gci0); l2_len0 = vnet_buffer (b0)->l2.l2_len; h0 = vlib_buffer_get_current (b0); @@ -185,14 +225,14 @@ gbp_policy_inline (vlib_main_t * vm, */ acl_plugin_fill_5tuple_inline (gm-> acl_plugin.p_acl_main, - value0.gc_lc_index, b0, + gc0->gc_lc_index, b0, is_ip60, /* is_input */ 0, /* is_l2_path */ 1, &pkt_5tuple0); acl_plugin_match_5tuple_inline (gm-> acl_plugin.p_acl_main, - value0.gc_lc_index, + gc0->gc_lc_index, &pkt_5tuple0, is_ip60, &action0, &acl_pos_p0, &acl_match_p0, @@ -201,17 +241,30 @@ gbp_policy_inline (vlib_main_t * vm, if (action0 > 0) { - vnet_buffer2 (b0)->gbp.flags |= VXLAN_GBP_GPFLAGS_A; + gbp_rule_t *gu; - next0 = - vnet_l2_feature_next (b0, - gpm->l2_output_feat_next - [is_port_based], - (is_port_based ? - L2OUTPUT_FEAT_GBP_POLICY_PORT - : - L2OUTPUT_FEAT_GBP_POLICY_MAC)); - ; + vnet_buffer2 (b0)->gbp.flags |= VXLAN_GBP_GPFLAGS_A; + gu = gbp_rule_get (gc0->gc_rules[rule_match_p0]); + + switch (gu->gu_action) + { + case GBP_RULE_PERMIT: + next0 = vnet_l2_feature_next + (b0, + gpm->l2_output_feat_next + [is_port_based], + (is_port_based ? + L2OUTPUT_FEAT_GBP_POLICY_PORT : + L2OUTPUT_FEAT_GBP_POLICY_MAC)); + break; + case GBP_RULE_DENY: + ASSERT (0); + next0 = 0; + break; + case GBP_RULE_REDIRECT: + next0 = gbp_rule_l2_redirect (gu, b0); + break; + } } } } @@ -237,8 +290,8 @@ gbp_policy_inline (vlib_main_t * vm, vlib_add_trace (vm, node, b0, sizeof (*t)); t->src_epg = key0.gck_src; t->dst_epg = key0.gck_dst; - t->acl_index = value0.gc_acl_index; - t->allowed = (next0 != GBP_POLICY_NEXT_DENY); + t->acl_index = (gc0 ? gc0->gc_acl_index : ~0), + t->allowed = (next0 != GBP_POLICY_NEXT_DENY); } /* verify speculative enqueue, maybe switch current next frame */ @@ -308,15 +361,7 @@ VLIB_REGISTER_NODE (gbp_policy_mac_node) = { .vector_size = sizeof (u32), .format_trace = format_gbp_policy_trace, .type = VLIB_NODE_TYPE_INTERNAL, - - .n_errors = ARRAY_LEN(gbp_policy_error_strings), - .error_strings = gbp_policy_error_strings, - - .n_next_nodes = GBP_POLICY_N_NEXT, - - .next_nodes = { - [GBP_POLICY_NEXT_DENY] = "error-drop", - }, + .sibling_of = "gbp-policy-port", }; VLIB_NODE_FUNCTION_MULTIARCH (gbp_policy_mac_node, gbp_policy_mac); diff --git a/src/plugins/gbp/gbp_policy_dpo.c b/src/plugins/gbp/gbp_policy_dpo.c index fd9dbce8bfa..7c53d1bcedb 100644 --- a/src/plugins/gbp/gbp_policy_dpo.c +++ b/src/plugins/gbp/gbp_policy_dpo.c @@ -236,6 +236,23 @@ typedef enum GBP_POLICY_N_NEXT, } gbp_policy_next_t; +always_inline u32 +gbp_rule_l3_redirect (const gbp_rule_t * gu, vlib_buffer_t * b0, int is_ip6) +{ + gbp_policy_node_t pnode; + const dpo_id_t *dpo; + dpo_proto_t dproto; + + pnode = (is_ip6 ? GBP_POLICY_NODE_IP6 : GBP_POLICY_NODE_IP4); + dproto = (is_ip6 ? DPO_PROTO_IP6 : DPO_PROTO_IP4); + dpo = &gu->gu_dpo[pnode][dproto]; + + /* The flow hash is still valid as this is a IP packet being switched */ + vnet_buffer (b0)->ip.adj_index[VLIB_TX] = dpo->dpoi_index; + + return (dpo->dpoi_next_node); +} + always_inline uword gbp_policy_dpo_inline (vlib_main_t * vm, vlib_node_runtime_t * node, @@ -243,6 +260,7 @@ gbp_policy_dpo_inline (vlib_main_t * vm, { gbp_main_t *gm = &gbp_main; u32 n_left_from, next_index, *from, *to_next; + gbp_rule_t *gu; from = vlib_frame_vector_args (from_frame); n_left_from = from_frame->n_vectors; @@ -260,10 +278,9 @@ gbp_policy_dpo_inline (vlib_main_t * vm, const gbp_policy_dpo_t *gpd0; u32 bi0, next0; gbp_contract_key_t key0; - gbp_contract_value_t value0 = { - .as_u64 = ~0, - }; + gbp_contract_t *gc0; vlib_buffer_t *b0; + index_t gci0; bi0 = from[0]; to_next[0] = bi0; @@ -275,6 +292,7 @@ gbp_policy_dpo_inline (vlib_main_t * vm, b0 = vlib_get_buffer (vm, bi0); + gc0 = NULL; gpd0 = gbp_policy_dpo_get_i (vnet_buffer (b0)->ip.adj_index[VLIB_TX]); vnet_buffer (b0)->ip.adj_index[VLIB_TX] = gpd0->gpd_dpo.dpoi_index; @@ -301,9 +319,9 @@ gbp_policy_dpo_inline (vlib_main_t * vm, } else { - value0.as_u64 = gbp_acl_lookup (&key0); + gci0 = gbp_contract_find (&key0); - if (~0 != value0.gc_lc_index) + if (INDEX_INVALID != gci0) { fa_5tuple_opaque_t pkt_5tuple0; u8 action0 = 0; @@ -312,16 +330,17 @@ gbp_policy_dpo_inline (vlib_main_t * vm, /* * tests against the ACL */ + gc0 = gbp_contract_get (gci0); acl_plugin_fill_5tuple_inline (gm-> acl_plugin.p_acl_main, - value0.gc_lc_index, b0, + gc0->gc_lc_index, b0, is_ip6, /* is_input */ 1, /* is_l2_path */ 0, &pkt_5tuple0); acl_plugin_match_5tuple_inline (gm-> acl_plugin.p_acl_main, - value0.gc_lc_index, + gc0->gc_lc_index, &pkt_5tuple0, is_ip6, &action0, &acl_pos_p0, &acl_match_p0, @@ -330,8 +349,23 @@ gbp_policy_dpo_inline (vlib_main_t * vm, if (action0 > 0) { + vnet_buffer2 (b0)->gbp.flags |= VXLAN_GBP_GPFLAGS_A; - next0 = gpd0->gpd_dpo.dpoi_next_node; + gu = gbp_rule_get (gc0->gc_rules[rule_match_p0]); + + switch (gu->gu_action) + { + case GBP_RULE_PERMIT: + next0 = gpd0->gpd_dpo.dpoi_next_node; + break; + case GBP_RULE_DENY: + ASSERT (0); + next0 = 0; + break; + case GBP_RULE_REDIRECT: + next0 = gbp_rule_l3_redirect (gu, b0, is_ip6); + break; + } } } } @@ -352,7 +386,7 @@ gbp_policy_dpo_inline (vlib_main_t * vm, tr = vlib_add_trace (vm, node, b0, sizeof (*tr)); tr->src_epg = key0.gck_src; tr->dst_epg = key0.gck_dst; - tr->acl_index = value0.gc_acl_index; + tr->acl_index = (gc0 ? gc0->gc_acl_index : ~0); tr->a_bit = vnet_buffer2 (b0)->gbp.flags & VXLAN_GBP_GPFLAGS_A; } diff --git a/src/plugins/gbp/gbp_policy_dpo.h b/src/plugins/gbp/gbp_policy_dpo.h index d6a8986d622..8ee086d1cb0 100644 --- a/src/plugins/gbp/gbp_policy_dpo.h +++ b/src/plugins/gbp/gbp_policy_dpo.h @@ -59,6 +59,10 @@ extern gbp_policy_dpo_t *gbp_policy_dpo_get (index_t index); extern dpo_type_t gbp_policy_dpo_get_type (void); +extern vlib_node_registration_t ip4_gbp_policy_dpo_node; +extern vlib_node_registration_t ip6_gbp_policy_dpo_node; +extern vlib_node_registration_t gbp_policy_port_node; + /* * fd.io coding-style-patch-verification: ON * diff --git a/src/plugins/gbp/gbp_recirc.c b/src/plugins/gbp/gbp_recirc.c index 57ba4083671..59588eafacd 100644 --- a/src/plugins/gbp/gbp_recirc.c +++ b/src/plugins/gbp/gbp_recirc.c @@ -66,11 +66,12 @@ gbp_recirc_add (u32 sw_if_index, epg_id_t epg_id, u8 is_ext) fib_protocol_t fproto; index_t ggi; - ggi = gbp_endpoint_group_find_and_lock (epg_id); + ggi = gbp_endpoint_group_find (epg_id); if (INDEX_INVALID == ggi) return (VNET_API_ERROR_NO_SUCH_ENTRY); + gbp_endpoint_group_lock (ggi); pool_get (gbp_recirc_pool, gr); clib_memset (gr, 0, sizeof (*gr)); gri = gr - gbp_recirc_pool; @@ -119,10 +120,12 @@ gbp_recirc_add (u32 sw_if_index, epg_id_t epg_id, u8 is_ext) mac_address_from_bytes (&mac, vnet_sw_interface_get_hw_address (vnet_get_main (), gr->gr_sw_if_index)); - gbp_endpoint_update (gr->gr_sw_if_index, - NULL, &mac, gr->gr_epg, - GBP_ENDPOINT_FLAG_NONE, - NULL, NULL, &gr->gr_ep); + gbp_endpoint_update_and_lock (GBP_ENDPOINT_SRC_CP, + gr->gr_sw_if_index, + NULL, &mac, INDEX_INVALID, + INDEX_INVALID, gr->gr_epg, + GBP_ENDPOINT_FLAG_NONE, + NULL, NULL, &gr->gr_ep); vnet_feature_enable_disable ("ip4-unicast", "ip4-gbp-src-classify", gr->gr_sw_if_index, 1, 0, 0); @@ -172,7 +175,7 @@ gbp_recirc_delete (u32 sw_if_index) if (gr->gr_is_ext) { - gbp_endpoint_delete (gr->gr_ep); + gbp_endpoint_unlock (GBP_ENDPOINT_SRC_CP, gr->gr_ep); vnet_feature_enable_disable ("ip4-unicast", "ip4-gbp-src-classify", gr->gr_sw_if_index, 0, 0, 0); diff --git a/src/plugins/gbp/gbp_route_domain.c b/src/plugins/gbp/gbp_route_domain.c index c030ee51ada..df3959edca6 100644 --- a/src/plugins/gbp/gbp_route_domain.c +++ b/src/plugins/gbp/gbp_route_domain.c @@ -62,6 +62,12 @@ vlib_log_class_t grd_logger; #define GBP_BD_DBG(...) \ vlib_log_debug (grd_logger, __VA_ARGS__); +index_t +gbp_route_domain_index (const gbp_route_domain_t * grd) +{ + return (grd - gbp_route_domain_pool); +} + gbp_route_domain_t * gbp_route_domain_get (index_t i) { diff --git a/src/plugins/gbp/gbp_route_domain.h b/src/plugins/gbp/gbp_route_domain.h index f7fc4a46d0b..ba5d5e49aca 100644 --- a/src/plugins/gbp/gbp_route_domain.h +++ b/src/plugins/gbp/gbp_route_domain.h @@ -62,6 +62,7 @@ extern int gbp_route_domain_add_and_lock (u32 rd_id, extern void gbp_route_domain_unlock (index_t grdi); extern index_t gbp_route_domain_find_and_lock (u32 rd_id); extern index_t gbp_route_domain_find (u32 rd_id); +extern index_t gbp_route_domain_index (const gbp_route_domain_t *); extern int gbp_route_domain_delete (u32 rd_id); extern gbp_route_domain_t *gbp_route_domain_get (index_t i); diff --git a/src/plugins/gbp/gbp_vxlan.c b/src/plugins/gbp/gbp_vxlan.c index 7464ccf2332..2b264f813ed 100644 --- a/src/plugins/gbp/gbp_vxlan.c +++ b/src/plugins/gbp/gbp_vxlan.c @@ -781,7 +781,7 @@ gbp_vxlan_tunnel_del (u32 vni) GBP_VXLAN_TUN_DBG ("del: %U", format_gbp_vxlan_tunnel, gt - gbp_vxlan_tunnel_pool); - gbp_endpoint_flush (gt->gt_sw_if_index); + gbp_endpoint_flush (GBP_ENDPOINT_SRC_DP, gt->gt_sw_if_index); ASSERT (0 == vec_len (gt->gt_tuns)); vec_free (gt->gt_tuns); |