1 files changed, 431 insertions, 0 deletions

diff --git a/zebra/rule_netlink.c b/zebra/rule_netlink.c
new file mode 100644
index 0000000..f00aef5
--- /dev/null
+++ b/zebra/rule_netlink.c
@@ -0,0 +1,431 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Zebra Policy Based Routing (PBR) interaction with the kernel using
+ * netlink.
+ * Copyright (C) 2018  Cumulus Networks, Inc.
+ */
+
+#include <zebra.h>
+
+#ifdef HAVE_NETLINK
+
+#include "if.h"
+#include "prefix.h"
+#include "vrf.h"
+
+#include <linux/fib_rules.h>
+#include "zebra/zserv.h"
+#include "zebra/zebra_ns.h"
+#include "zebra/zebra_vrf.h"
+#include "zebra/rt.h"
+#include "zebra/interface.h"
+#include "zebra/debug.h"
+#include "zebra/rtadv.h"
+#include "zebra/kernel_netlink.h"
+#include "zebra/rule_netlink.h"
+#include "zebra/zebra_pbr.h"
+#include "zebra/zebra_errors.h"
+#include "zebra/zebra_dplane.h"
+#include "zebra/zebra_trace.h"
+
+/* definitions */
+
+/* static function declarations */
+
+/* Private functions */
+
+
+/*
+ * netlink_rule_msg_encode
+ *
+ * Encodes netlink RTM_ADDRULE/RTM_DELRULE message to buffer buf of size buflen.
+ *
+ * Returns -1 on failure, 0 when the msg doesn't fit entirely in the buffer
+ * or the number of bytes written to buf.
+ */
+static ssize_t netlink_rule_msg_encode(
+	int cmd, const struct zebra_dplane_ctx *ctx, uint32_t filter_bm,
+	uint32_t priority, uint32_t table, const struct prefix *src_ip,
+	const struct prefix *dst_ip, uint32_t fwmark, uint8_t dsfield,
+	uint8_t ip_protocol, void *buf, size_t buflen)
+{
+	uint8_t protocol = RTPROT_ZEBRA;
+	int family;
+	int bytelen;
+	struct {
+		struct nlmsghdr n;
+		struct fib_rule_hdr frh;
+		char buf[];
+	} *req = buf;
+
+	const char *ifname = dplane_ctx_rule_get_ifname(ctx);
+
+	if (buflen < sizeof(*req))
+		return 0;
+	memset(req, 0, sizeof(*req));
+
+	/* Assume ipv4 if no src/dst set, we only support ipv4/ipv6 */
+	if (PREFIX_FAMILY(src_ip))
+		family = PREFIX_FAMILY(src_ip);
+	else if (PREFIX_FAMILY(dst_ip))
+		family = PREFIX_FAMILY(dst_ip);
+	else
+		family = AF_INET;
+
+	bytelen = (family == AF_INET ? 4 : 16);
+
+	req->n.nlmsg_type = cmd;
+	req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg));
+	req->n.nlmsg_flags = NLM_F_REQUEST;
+
+	req->frh.family = family;
+	req->frh.action = FR_ACT_TO_TBL;
+
+	if (!nl_attr_put(&req->n, buflen, FRA_PROTOCOL, &protocol,
+			 sizeof(protocol)))
+		return 0;
+
+	/* rule's pref # */
+	if (!nl_attr_put32(&req->n, buflen, FRA_PRIORITY, priority))
+		return 0;
+
+	/* interface on which applied */
+	if (!nl_attr_put(&req->n, buflen, FRA_IFNAME, ifname,
+			 strlen(ifname) + 1))
+		return 0;
+
+	/* source IP, if specified */
+	if (filter_bm & PBR_FILTER_SRC_IP) {
+		req->frh.src_len = src_ip->prefixlen;
+		if (!nl_attr_put(&req->n, buflen, FRA_SRC, &src_ip->u.prefix,
+				 bytelen))
+			return 0;
+	}
+
+	/* destination IP, if specified */
+	if (filter_bm & PBR_FILTER_DST_IP) {
+		req->frh.dst_len = dst_ip->prefixlen;
+		if (!nl_attr_put(&req->n, buflen, FRA_DST, &dst_ip->u.prefix,
+				 bytelen))
+			return 0;
+	}
+
+	/* fwmark, if specified */
+	if (filter_bm & PBR_FILTER_FWMARK) {
+		if (!nl_attr_put32(&req->n, buflen, FRA_FWMARK, fwmark))
+			return 0;
+	}
+
+	/* dsfield, if specified; mask off the ECN bits */
+	if (filter_bm & PBR_FILTER_DSCP)
+		req->frh.tos = dsfield & PBR_DSFIELD_DSCP;
+
+	/* protocol to match on */
+	if (filter_bm & PBR_FILTER_IP_PROTOCOL)
+		nl_attr_put8(&req->n, buflen, FRA_IP_PROTO, ip_protocol);
+
+	/* Route table to use to forward, if filter criteria matches. */
+	if (table < 256)
+		req->frh.table = table;
+	else {
+		req->frh.table = RT_TABLE_UNSPEC;
+		if (!nl_attr_put32(&req->n, buflen, FRA_TABLE, table))
+			return 0;
+	}
+
+	if (IS_ZEBRA_DEBUG_KERNEL)
+		zlog_debug(
+			"Tx %s family %s IF %s Pref %u Fwmark %u Src %pFX Dst %pFX Table %u",
+			nl_msg_type_to_str(cmd), nl_family_to_str(family),
+			ifname, priority, fwmark, src_ip, dst_ip, table);
+
+	return NLMSG_ALIGN(req->n.nlmsg_len);
+}
+
+static ssize_t netlink_rule_msg_encoder(struct zebra_dplane_ctx *ctx, void *buf,
+					size_t buflen)
+{
+	int cmd = RTM_NEWRULE;
+
+	if (dplane_ctx_get_op(ctx) == DPLANE_OP_RULE_DELETE)
+		cmd = RTM_DELRULE;
+
+	return netlink_rule_msg_encode(
+		cmd, ctx, dplane_ctx_rule_get_filter_bm(ctx),
+		dplane_ctx_rule_get_priority(ctx),
+		dplane_ctx_rule_get_table(ctx), dplane_ctx_rule_get_src_ip(ctx),
+		dplane_ctx_rule_get_dst_ip(ctx),
+		dplane_ctx_rule_get_fwmark(ctx),
+		dplane_ctx_rule_get_dsfield(ctx),
+		dplane_ctx_rule_get_ipproto(ctx), buf, buflen);
+}
+
+static ssize_t netlink_oldrule_msg_encoder(struct zebra_dplane_ctx *ctx,
+					   void *buf, size_t buflen)
+{
+	return netlink_rule_msg_encode(
+		RTM_DELRULE, ctx, dplane_ctx_rule_get_old_filter_bm(ctx),
+		dplane_ctx_rule_get_old_priority(ctx),
+		dplane_ctx_rule_get_old_table(ctx),
+		dplane_ctx_rule_get_old_src_ip(ctx),
+		dplane_ctx_rule_get_old_dst_ip(ctx),
+		dplane_ctx_rule_get_old_fwmark(ctx),
+		dplane_ctx_rule_get_old_dsfield(ctx),
+		dplane_ctx_rule_get_old_ipproto(ctx), buf, buflen);
+}
+
+/*
+ * Identify valid rule actions for netlink - other actions can't be installed
+ */
+static bool nl_rule_valid_action(uint32_t action)
+{
+	if (action == PBR_ACTION_TABLE)
+		return true;
+	else
+		return false;
+}
+
+/* Public functions */
+
+enum netlink_msg_status
+netlink_put_rule_update_msg(struct nl_batch *bth, struct zebra_dplane_ctx *ctx)
+{
+	enum dplane_op_e op;
+	enum netlink_msg_status ret;
+	struct pbr_rule rule = {};
+
+	op = dplane_ctx_get_op(ctx);
+	if (!(op == DPLANE_OP_RULE_ADD || op == DPLANE_OP_RULE_UPDATE
+	      || op == DPLANE_OP_RULE_DELETE)) {
+		flog_err(
+			EC_ZEBRA_PBR_RULE_UPDATE,
+			"Context received for kernel rule update with incorrect OP code (%u)",
+			op);
+		return FRR_NETLINK_ERROR;
+	}
+
+	/* TODO -- special handling for rules that include actions that
+	 * netlink cannot install. Some of the rule attributes are not
+	 * available in netlink: only try to install valid actions.
+	 */
+	dplane_ctx_rule_get(ctx, &rule, NULL);
+	if (!nl_rule_valid_action(rule.action.flags)) {
+		if (IS_ZEBRA_DEBUG_KERNEL)
+			zlog_debug("%s: skip invalid action %#x", __func__,
+				   rule.action.flags);
+		return 0;
+	}
+
+	ret = netlink_batch_add_msg(bth, ctx, netlink_rule_msg_encoder, false);
+
+	/**
+	 * Delete the old one.
+	 *
+	 * Don't care about this result right?
+	 */
+	if (op == DPLANE_OP_RULE_UPDATE)
+		netlink_batch_add_msg(bth, ctx, netlink_oldrule_msg_encoder,
+				      true);
+
+	return ret;
+}
+
+/*
+ * Handle netlink notification informing a rule add or delete.
+ * Handling of an ADD is TBD.
+ * DELs are notified up, if other attributes indicate it may be a
+ * notification of interest. The expectation is that if this corresponds
+ * to a PBR rule added by FRR, it will be readded.
+ *
+ * If startup and we see a rule we created, delete it as its leftover
+ * from a previous instance and should have been removed on shutdown.
+ *
+ */
+int netlink_rule_change(struct nlmsghdr *h, ns_id_t ns_id, int startup)
+{
+	struct zebra_ns *zns;
+	struct fib_rule_hdr *frh;
+	struct rtattr *tb[FRA_MAX + 1];
+	int len;
+	char *ifname;
+	struct zebra_pbr_rule rule = {};
+	uint8_t proto = 0;
+	uint8_t ip_proto = 0;
+
+	frrtrace(3, frr_zebra, netlink_rule_change, h, ns_id, startup);
+
+	/* Basic validation followed by extracting attributes. */
+	if (h->nlmsg_type != RTM_NEWRULE && h->nlmsg_type != RTM_DELRULE)
+		return 0;
+
+	len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct fib_rule_hdr));
+	if (len < 0) {
+		zlog_err(
+			"%s: Message received from netlink is of a broken size: %d %zu",
+			__func__, h->nlmsg_len,
+			(size_t)NLMSG_LENGTH(sizeof(struct fib_rule_hdr)));
+		return -1;
+	}
+
+	frh = NLMSG_DATA(h);
+
+	if (frh->family != AF_INET && frh->family != AF_INET6) {
+		if (frh->family == RTNL_FAMILY_IPMR
+		    || frh->family == RTNL_FAMILY_IP6MR) {
+			if (IS_ZEBRA_DEBUG_KERNEL)
+				zlog_debug(
+					"Received rule netlink that we are ignoring for family %u, rule change: %u",
+					frh->family, h->nlmsg_type);
+			return 0;
+		}
+		flog_warn(
+			EC_ZEBRA_NETLINK_INVALID_AF,
+			"Invalid address family: %u received from kernel rule change: %u",
+			frh->family, h->nlmsg_type);
+		return 0;
+	}
+	if (frh->action != FR_ACT_TO_TBL)
+		return 0;
+
+	memset(tb, 0, sizeof(tb));
+	netlink_parse_rtattr(tb, FRA_MAX, RTM_RTA(frh), len);
+
+	if (tb[FRA_PRIORITY])
+		rule.rule.priority = *(uint32_t *)RTA_DATA(tb[FRA_PRIORITY]);
+
+	if (tb[FRA_SRC]) {
+		if (frh->family == AF_INET)
+			memcpy(&rule.rule.filter.src_ip.u.prefix4,
+			       RTA_DATA(tb[FRA_SRC]), 4);
+		else
+			memcpy(&rule.rule.filter.src_ip.u.prefix6,
+			       RTA_DATA(tb[FRA_SRC]), 16);
+		rule.rule.filter.src_ip.prefixlen = frh->src_len;
+		rule.rule.filter.src_ip.family = frh->family;
+		rule.rule.filter.filter_bm |= PBR_FILTER_SRC_IP;
+	}
+
+	if (tb[FRA_DST]) {
+		if (frh->family == AF_INET)
+			memcpy(&rule.rule.filter.dst_ip.u.prefix4,
+			       RTA_DATA(tb[FRA_DST]), 4);
+		else
+			memcpy(&rule.rule.filter.dst_ip.u.prefix6,
+			       RTA_DATA(tb[FRA_DST]), 16);
+		rule.rule.filter.dst_ip.prefixlen = frh->dst_len;
+		rule.rule.filter.dst_ip.family = frh->family;
+		rule.rule.filter.filter_bm |= PBR_FILTER_DST_IP;
+	}
+
+	if (tb[FRA_TABLE])
+		rule.rule.action.table = *(uint32_t *)RTA_DATA(tb[FRA_TABLE]);
+	else
+		rule.rule.action.table = frh->table;
+
+	/* TBD: We don't care about rules not specifying an IIF. */
+	if (tb[FRA_IFNAME] == NULL)
+		return 0;
+
+	if (tb[FRA_PROTOCOL])
+		proto = *(uint8_t *)RTA_DATA(tb[FRA_PROTOCOL]);
+
+	if (tb[FRA_IP_PROTO])
+		ip_proto = *(uint8_t *)RTA_DATA(tb[FRA_IP_PROTO]);
+
+	ifname = (char *)RTA_DATA(tb[FRA_IFNAME]);
+	strlcpy(rule.ifname, ifname, sizeof(rule.ifname));
+
+	if (h->nlmsg_type == RTM_NEWRULE) {
+		/*
+		 * If we see a rule at startup we created, delete it now.
+		 * It should have been flushed on a previous shutdown.
+		 */
+		if (startup && proto == RTPROT_ZEBRA) {
+			enum zebra_dplane_result ret;
+
+			ret = dplane_pbr_rule_delete(&rule);
+
+			zlog_debug(
+				"%s: %s leftover rule: family %s IF %s Pref %u Src %pFX Dst %pFX Table %u ip-proto: %u",
+				__func__,
+				((ret == ZEBRA_DPLANE_REQUEST_FAILURE)
+					 ? "Failed to remove"
+					 : "Removed"),
+				nl_family_to_str(frh->family), rule.ifname,
+				rule.rule.priority, &rule.rule.filter.src_ip,
+				&rule.rule.filter.dst_ip,
+				rule.rule.action.table, ip_proto);
+		}
+
+		/* TBD */
+		return 0;
+	}
+
+	zns = zebra_ns_lookup(ns_id);
+
+	/* If we don't know the interface, we don't care. */
+	if (!if_lookup_by_name_per_ns(zns, ifname))
+		return 0;
+
+	if (IS_ZEBRA_DEBUG_KERNEL)
+		zlog_debug(
+			"Rx %s family %s IF %s Pref %u Src %pFX Dst %pFX Table %u ip-proto: %u",
+			nl_msg_type_to_str(h->nlmsg_type),
+			nl_family_to_str(frh->family), rule.ifname,
+			rule.rule.priority, &rule.rule.filter.src_ip,
+			&rule.rule.filter.dst_ip, rule.rule.action.table,
+			ip_proto);
+
+	return kernel_pbr_rule_del(&rule);
+}
+
+/*
+ * Request rules from the kernel
+ */
+static int netlink_request_rules(struct zebra_ns *zns, int family, int type)
+{
+	struct {
+		struct nlmsghdr n;
+		struct fib_rule_hdr frh;
+		char buf[NL_PKT_BUF_SIZE];
+	} req;
+
+	memset(&req, 0, sizeof(req));
+	req.n.nlmsg_type = type;
+	req.n.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST;
+	req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct fib_rule_hdr));
+	req.frh.family = family;
+
+	return netlink_request(&zns->netlink_cmd, &req);
+}
+
+/*
+ * Get to know existing PBR rules in the kernel - typically called at startup.
+ */
+int netlink_rules_read(struct zebra_ns *zns)
+{
+	int ret;
+	struct zebra_dplane_info dp_info;
+
+	zebra_dplane_info_from_zns(&dp_info, zns, true);
+
+	ret = netlink_request_rules(zns, AF_INET, RTM_GETRULE);
+	if (ret < 0)
+		return ret;
+
+	ret = netlink_parse_info(netlink_rule_change, &zns->netlink_cmd,
+				 &dp_info, 0, true);
+	if (ret < 0)
+		return ret;
+
+	ret = netlink_request_rules(zns, AF_INET6, RTM_GETRULE);
+	if (ret < 0)
+		return ret;
+
+	ret = netlink_parse_info(netlink_rule_change, &zns->netlink_cmd,
+				 &dp_info, 0, true);
+
+	return ret;
+}
+
+#endif /* HAVE_NETLINK */