[WIP] QPPB | QoS Policy Propagation via BGP (using XDP/BCC/BPF)

bgpd/bgp_attr.c

@@ -1076,6 +1076,7 @@ struct attr *bgp_attr_default_set(struct attr *attr, struct bgp *bgp,
 	attr->flag |= ATTR_FLAG_BIT(BGP_ATTR_AS_PATH);
 	attr->weight = BGP_ATTR_DEFAULT_WEIGHT;
 	attr->tag = 0;
+	attr->dscp = 0;
 	attr->label_index = BGP_INVALID_LABEL_INDEX;
 	attr->label = MPLS_INVALID_LABEL;
 	attr->flag |= ATTR_FLAG_BIT(BGP_ATTR_NEXT_HOP);

bgpd/bgp_attr.h

@@ -309,6 +309,9 @@ struct attr {
 	/* SR-TE Color */
 	uint32_t srte_color;
 
+	/* DSCP tag, used by the QPPB plugin */
+	uint8_t dscp;
+
 	/* Nexthop type */
 	enum nexthop_types_t nh_type;
 

bgpd/bgp_route.c

@@ -16,6 +16,7 @@
 #include "stream.h"
 #include "filter.h"
 #include "log.h"
+#include "dscp.h"
 #include "routemap.h"
 #include "buffer.h"
 #include "sockunion.h"
@@ -10156,7 +10157,7 @@ void route_vty_out_detail(struct vty *vty, struct bgp *bgp, struct bgp_dest *bn,
 	json_object *json_peer = NULL;
 	json_object *json_string = NULL;
 	json_object *json_adv_to = NULL;
-	int first = 0;
+	int first = 0, dscp = attr->dscp >> 2;
 	struct listnode *node, *nnode;
 	struct peer *peer;
 	bool addpath_capable;
@@ -10838,6 +10839,18 @@ void route_vty_out_detail(struct vty *vty, struct bgp *bgp, struct bgp_dest *bn,
 		}
 	}
 
+	/* XXX: on Cisco
+	 * R1# sh ip cef 5.5.0.0 de
+	 * 5.5.0.0/24, epoch 0
+	 *     QOS: Precedence critical (5)
+	 *     recursive via ...
+	 *           nexthop ... iface
+	 */
+	if (dscp) {
+		vty_out(vty, "      QOS: Precedence %s (%d)\n",
+			dscp_enum_str(dscp), dscp);
+	}
+
 	/* Line 5 display Extended-community */
 	if (attr->flag & ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES)) {
 		if (json_paths) {

bgpd/bgp_routemap.c

@@ -107,6 +107,7 @@ o Cisco route-map
       metric-type       :  Not yet
       origin            :  Done
       tag               :  Done
+      dscp              :  Done
       weight            :  Done
       table             :  Done
 
@@ -3567,6 +3568,30 @@ static const struct route_map_rule_cmd route_set_tag_cmd = {
 	route_map_rule_tag_free,
 };
 
+/* Set dscp to object. object must be pointer to struct bgp_path_info */
+static enum route_map_cmd_result_t
+route_set_dscp(void *rule, const struct prefix *prefix, void *object)
+{
+	struct bgp_path_info *path;
+	uint8_t *rawDscp;
+
+	rawDscp = rule;
+	path = object;
+
+	/* Set dscp value */
+	path->attr->dscp = *rawDscp;
+
+	return RMAP_OKAY;
+}
+
+/* Route map commands for dscp set. */
+static const struct route_map_rule_cmd route_set_dscp_cmd = {
+	"dscp",
+	route_set_dscp,
+	route_map_rule_dscp_compile,
+	route_map_rule_dscp_free,
+};
+
 /* Set label-index to object. object must be pointer to struct bgp_path_info */
 static enum route_map_cmd_result_t
 route_set_label_index(void *rule, const struct prefix *prefix, void *object)
@@ -7840,6 +7865,9 @@ void bgp_route_map_init(void)
 	route_map_set_tag_hook(generic_set_add);
 	route_map_no_set_tag_hook(generic_set_delete);
 
+	route_map_set_dscp_hook(generic_set_add);
+	route_map_no_set_dscp_hook(generic_set_delete);
+
 	route_map_install_match(&route_match_peer_cmd);
 	route_map_install_match(&route_match_alias_cmd);
 	route_map_install_match(&route_match_local_pref_cmd);
@@ -7903,6 +7931,7 @@ void bgp_route_map_init(void)
 	route_map_install_set(&route_set_ecommunity_color_cmd);
 	route_map_install_set(&route_set_ecommunity_none_cmd);
 	route_map_install_set(&route_set_tag_cmd);
+	route_map_install_set(&route_set_dscp_cmd);
 	route_map_install_set(&route_set_label_index_cmd);
 	route_map_install_set(&route_set_l3vpn_nexthop_encapsulation_cmd);
 

bgpd/bgp_zebra.c

@@ -1233,7 +1233,7 @@ static void bgp_zebra_announce_parse_nexthop(
 	struct bgp_path_info *info, const struct prefix *p, struct bgp *bgp,
 	struct zapi_route *api, unsigned int *valid_nh_count, afi_t afi,
 	safi_t safi, uint32_t *nhg_id, uint32_t *metric, route_tag_t *tag,
-	bool *allow_recursion)
+	uint8_t *dscp, bool *allow_recursion)
 {
 	struct zapi_nexthop *api_nh;
 	int nh_family;
@@ -1334,13 +1334,21 @@ static void bgp_zebra_announce_parse_nexthop(
 						 p, mpinfo_cp))
 				continue;
 
-			/* metric/tag is only allowed to be
+			/* metric/tag/dscp is only allowed to be
 			 * overridden on 1st nexthop */
 			if (mpinfo == info) {
 				if (metric)
 					*metric = mpinfo_cp->attr->med;
 				if (tag)
 					*tag = mpinfo_cp->attr->tag;
+				if (dscp) {
+					/* used by hook from the QPPB plugin */
+					*dscp = mpinfo_cp->attr->dscp;
+					/* expose value back to CLI `show bgp ipv4 nh`
+					 * displayed by route_vty_out_detail()`
+					 */
+					mpinfo->attr->dscp = *dscp;
+				}
 			}
 		}
 
@@ -1532,7 +1540,7 @@ bgp_zebra_announce_actual(struct bgp_dest *dest, struct bgp_path_info *info,
 	struct zapi_route api = { 0 };
 	unsigned int valid_nh_count = 0;
 	bool allow_recursion = false;
-	uint8_t distance;
+	uint8_t distance, dscp = 0;
 	struct peer *peer;
 	uint32_t metric;
 	route_tag_t tag;
@@ -1589,7 +1597,7 @@ bgp_zebra_announce_actual(struct bgp_dest *dest, struct bgp_path_info *info,
 
 	bgp_zebra_announce_parse_nexthop(info, p, bgp, &api, &valid_nh_count,
 					 table->afi, table->safi, &nhg_id,
-					 &metric, &tag, &allow_recursion);
+					 &metric, &tag, &dscp, &allow_recursion);
 
 	is_add = (valid_nh_count || nhg_id) ? true : false;
 
@@ -1637,6 +1645,11 @@ bgp_zebra_announce_actual(struct bgp_dest *dest, struct bgp_path_info *info,
 	SET_FLAG(api.message, ZAPI_MESSAGE_METRIC);
 	api.metric = metric;
 
+	if (dscp) {
+		SET_FLAG(api.message, ZAPI_MESSAGE_DSCP);
+		api.dscp = dscp;
+	}
+
 	if (tag) {
 		SET_FLAG(api.message, ZAPI_MESSAGE_TAG);
 		api.tag = tag;

configure.ac

@@ -749,6 +749,8 @@ AC_ARG_ENABLE([bgp-bmp],
   AS_HELP_STRING([--disable-bgp-bmp],[turn off BGP BMP support]))
 AC_ARG_ENABLE([snmp],
   AS_HELP_STRING([--enable-snmp], [enable SNMP support for agentx]))
+AC_ARG_ENABLE([qppb],
+  AS_HELP_STRING([--enable-qppb], [enable QPPB support for BGP (requires libbpf)]))
 AC_ARG_ENABLE([config_rollbacks],
   AS_HELP_STRING([--enable-config-rollbacks], [enable configuration rollbacks (requires sqlite3)]))
 AC_ARG_ENABLE([sysrepo],
@@ -2079,6 +2081,18 @@ if test "$enable_config_rollbacks" = "yes"; then
   ])
 fi
 
+dnl ---------------
+dnl configuration QPPB
+dnl ---------------
+QPPB=false
+if test "$enable_qppb" = "yes"; then
+  PKG_CHECK_MODULES([LIBBPF], [libbpf], [
+    QPPB=true
+  ], [
+    AC_MSG_ERROR([--enable-qppb given but libbpf was not found on your system.])
+  ])
+fi
+
 dnl ---------------
 dnl sysrepo
 dnl ---------------
@@ -2748,6 +2762,7 @@ AM_CONDITIONAL([ENABLE_BGP_VNC], [test "$enable_bgp_vnc" != "no"])
 AM_CONDITIONAL([BGP_BMP], [$bgpd_bmp])
 dnl northbound
 AM_CONDITIONAL([SQLITE3], [$SQLITE3])
+AM_CONDITIONAL([QPPB], [$QPPB])
 AM_CONDITIONAL([SYSREPO], [test "$enable_sysrepo" = "yes"])
 AM_CONDITIONAL([GRPC], [test "$enable_grpc" = "yes"])
 AM_CONDITIONAL([ZEROMQ], [test "$ZEROMQ" = "true"])

lib/dscp.c

@@ -0,0 +1,82 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * DSCP manipulation routines
+ * Copyright (C) 2023 VyOS Inc.
+ * Volodymyr Huti
+ */
+
+#include "dscp.h"
+
+enum dscp_val {
+	DSCP_CS0 = 0x00,
+	DSCP_CS1 = 0x08,
+	DSCP_CS2 = 0x10,
+	DSCP_CS3 = 0x18,
+	DSCP_CS4 = 0x20,
+	DSCP_CS5 = 0x28,
+	DSCP_CS6 = 0x30,
+	DSCP_CS7 = 0x38,
+	DSCP_AF11 = 0x0A,
+	DSCP_AF12 = 0x0C,
+	DSCP_AF13 = 0x0E,
+	DSCP_AF21 = 0x12,
+	DSCP_AF22 = 0x14,
+	DSCP_AF23 = 0x16,
+	DSCP_AF31 = 0x1A,
+	DSCP_AF32 = 0x1C,
+	DSCP_AF33 = 0x1E,
+	DSCP_AF41 = 0x22,
+	DSCP_AF42 = 0x34,
+	DSCP_AF43 = 0x26,
+	DSCP_EF = 0x2E,
+	DSCP_VOICE = 0x2C,
+	DSCP_MAX = DSCP_CS7 + 1,
+	DSCP_ERR
+};
+
+static const struct {
+	const char *name;
+	int val;
+} dscp_enum_int_map[] = {
+	{"cs0", DSCP_CS0},	    {"cs1", DSCP_CS1},	 {"cs2", DSCP_CS2},
+	{"cs3", DSCP_CS3},	    {"cs4", DSCP_CS4},	 {"cs5", DSCP_CS5},
+	{"cs6", DSCP_CS6},	    {"cs7", DSCP_CS7},	 {"af11", DSCP_AF11},
+	{"af12", DSCP_AF12},	    {"af13", DSCP_AF13}, {"af21", DSCP_AF21},
+	{"af22", DSCP_AF22},	    {"af23", DSCP_AF23}, {"af31", DSCP_AF31},
+	{"af32", DSCP_AF32},	    {"af33", DSCP_AF33}, {"af41", DSCP_AF41},
+	{"af42", DSCP_AF42},	    {"af43", DSCP_AF43}, {"ef", DSCP_EF},
+	{"voice-admit", DSCP_VOICE}};
+
+static const char *dscp_int_enum_map[DSCP_MAX] = {
+	[DSCP_CS0] = "cs0",	     [DSCP_CS1] = "cs1",   [DSCP_CS2] = "cs2",
+	[DSCP_CS3] = "cs3",	     [DSCP_CS4] = "cs4",   [DSCP_CS5] = "cs5",
+	[DSCP_CS6] = "cs6",	     [DSCP_CS7] = "cs7",   [DSCP_AF11] = "af11",
+	[DSCP_AF12] = "af12",	     [DSCP_AF13] = "af13", [DSCP_AF21] = "af21",
+	[DSCP_AF22] = "af22",	     [DSCP_AF23] = "af23", [DSCP_AF31] = "af31",
+	[DSCP_AF32] = "af32",	     [DSCP_AF33] = "af33", [DSCP_AF41] = "af41",
+	[DSCP_AF42] = "af42",	     [DSCP_AF43] = "af43", [DSCP_EF] = "ef",
+	[DSCP_VOICE] = "voice-admit"};
+
+static const int dscp_msize =
+	(sizeof(dscp_enum_int_map) / sizeof(*dscp_enum_int_map));
+
+/* Decodes a standardized DSCP into its representative value */
+uint8_t dscp_decode_enum(const char *name)
+{
+	int dscp_val = -1;
+
+	for (int i = 0; i < dscp_msize; ++i) {
+		if (!strcmp(dscp_enum_int_map[i].name, name)) {
+			dscp_val = dscp_enum_int_map[i].val;
+			break;
+		}
+	}
+
+	return dscp_val;
+}
+
+const char *dscp_enum_str(int dscp)
+{
+	return (dscp < DSCP_MAX) ? dscp_int_enum_map[dscp] : NULL;
+}
+

lib/dscp.h

@@ -0,0 +1,28 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * DSCP manipulation routines header
+ * Copyright (C) 2023 VyOS Inc.
+ * Volodymyr Huti
+ */
+
+#ifndef FRR_DSCP_H
+#define FRR_DSCP_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <zebra.h>
+#include "vty.h"
+
+#define DSFIELD_DSCP (0xfc) /* Upper 6 bits of DS field: DSCP */
+#define DSFIELD_ECN (0x03)  /* Lower 2 bits of DS field: BCN */
+
+extern uint8_t dscp_decode_enum(const char *dscp);
+extern const char *dscp_enum_str(int dscp);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FRR_DSCP_H */

lib/pbr.h

@@ -43,9 +43,6 @@ struct pbr_filter {
 #define PBR_FILTER_VLAN_FLAGS     (1 << 11)
 #define PBR_FILTER_VLAN_ID        (1 << 12)
 
-#define PBR_DSFIELD_DSCP (0xfc) /* Upper 6 bits of DS field: DSCP */
-#define PBR_DSFIELD_ECN (0x03)	/* Lower 2 bits of DS field: BCN */
-
 #define PBR_PCP (0x07)		/* 3-bit value 0..7 for prioritization*/
 
 #define PBR_VLAN_FLAGS_NO_WILD	  0

lib/routemap.c

@@ -20,6 +20,7 @@
 #include "table.h"
 #include "json.h"
 #include "jhash.h"
+#include "dscp.h"
 
 #include "lib/routemap_clippy.c"
 
@@ -446,6 +447,24 @@ void route_map_no_set_tag_hook(int (*func)(struct route_map_index *index,
 	rmap_match_set_hook.no_set_tag = func;
 }
 
+/* set dscp */
+void route_map_set_dscp_hook(int (*func)(struct route_map_index *index,
+					 const char *command, const char *arg,
+					 char *errmsg, size_t errmsg_len))
+{
+
+	rmap_match_set_hook.set_dscp = func;
+}
+
+/* no set dscp */
+void route_map_no_set_dscp_hook(int (*func)(struct route_map_index *index,
+					    const char *command,
+					    const char *arg, char *errmsg,
+					    size_t errmsg_len))
+{
+	rmap_match_set_hook.no_set_dscp = func;
+}
+
 int generic_match_add(struct route_map_index *index,
 		      const char *command, const char *arg,
 		      route_map_event_t type,
@@ -3228,6 +3247,33 @@ void *route_map_rule_tag_compile(const char *arg)
 	return tag;
 }
 
+void *route_map_rule_dscp_compile(const char *dscp)
+{
+	unsigned long ul_dscp;
+	char *pend = NULL;
+	uint8_t *shifted_dscp;
+
+	assert(dscp);
+	ul_dscp = strtoul(dscp, &pend, 0);
+	if (pend && *pend)
+		ul_dscp = dscp_decode_enum(dscp);
+
+	if (ul_dscp > (DSFIELD_DSCP >> 2)) {
+		zlog_err("Invalid dscp value: %s%s", dscp,
+			((pend && *pend) ? "" : " (numeric value must be in range 0-63)"));
+		return NULL;
+	}
+
+	shifted_dscp = XMALLOC(MTYPE_ROUTE_MAP_COMPILED, sizeof(*shifted_dscp));
+	*shifted_dscp = (ul_dscp << 2) & DSFIELD_DSCP;
+	return shifted_dscp;
+}
+
+void route_map_rule_dscp_free(void *rule)
+{
+	XFREE(MTYPE_ROUTE_MAP_COMPILED, rule);
+}
+
 void route_map_rule_tag_free(void *rule)
 {
 	XFREE(MTYPE_ROUTE_MAP_COMPILED, rule);