summaryrefslogtreecommitdiffstats
path: root/pimd/pim_mlag.c
diff options
context:
space:
mode:
Diffstat (limited to 'pimd/pim_mlag.c')
-rw-r--r--pimd/pim_mlag.c1098
1 files changed, 1098 insertions, 0 deletions
diff --git a/pimd/pim_mlag.c b/pimd/pim_mlag.c
new file mode 100644
index 0000000..9763a79
--- /dev/null
+++ b/pimd/pim_mlag.c
@@ -0,0 +1,1098 @@
+/*
+ * This is an implementation of PIM MLAG Functionality
+ *
+ * Module name: PIM MLAG
+ *
+ * Author: sathesh Kumar karra <sathk@cumulusnetworks.com>
+ *
+ * Copyright (C) 2019 Cumulus Networks http://www.cumulusnetworks.com
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; see the file COPYING; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+#include <zebra.h>
+
+#include "pimd.h"
+#include "pim_mlag.h"
+#include "pim_upstream.h"
+#include "pim_vxlan.h"
+
+extern struct zclient *zclient;
+
+#define PIM_MLAG_METADATA_LEN 4
+
+/*********************ACtual Data processing *****************************/
+/* TBD: There can be duplicate updates to FIB***/
+#define PIM_MLAG_ADD_OIF_TO_OIL(ch, ch_oil) \
+ do { \
+ if (PIM_DEBUG_MLAG) \
+ zlog_debug( \
+ "%s: add Dual-active Interface to %s " \
+ "to oil:%s", \
+ __func__, ch->interface->name, ch->sg_str); \
+ pim_channel_update_oif_mute(ch_oil, ch->interface->info); \
+ } while (0)
+
+#define PIM_MLAG_DEL_OIF_TO_OIL(ch, ch_oil) \
+ do { \
+ if (PIM_DEBUG_MLAG) \
+ zlog_debug( \
+ "%s: del Dual-active Interface to %s " \
+ "to oil:%s", \
+ __func__, ch->interface->name, ch->sg_str); \
+ pim_channel_update_oif_mute(ch_oil, ch->interface->info); \
+ } while (0)
+
+
+static void pim_mlag_calculate_df_for_ifchannels(struct pim_upstream *up,
+ bool is_df)
+{
+ struct listnode *chnode;
+ struct listnode *chnextnode;
+ struct pim_ifchannel *ch;
+ struct pim_interface *pim_ifp = NULL;
+ struct channel_oil *ch_oil = NULL;
+
+ ch_oil = (up) ? up->channel_oil : NULL;
+
+ if (!ch_oil)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Calculating DF for Dual active if-channel%s",
+ __func__, up->sg_str);
+
+ for (ALL_LIST_ELEMENTS(up->ifchannels, chnode, chnextnode, ch)) {
+ pim_ifp = (ch->interface) ? ch->interface->info : NULL;
+ if (!pim_ifp || !PIM_I_am_DualActive(pim_ifp))
+ continue;
+
+ if (is_df)
+ PIM_MLAG_ADD_OIF_TO_OIL(ch, ch_oil);
+ else
+ PIM_MLAG_DEL_OIF_TO_OIL(ch, ch_oil);
+ }
+}
+
+static void pim_mlag_inherit_mlag_flags(struct pim_upstream *up, bool is_df)
+{
+ struct listnode *listnode;
+ struct pim_upstream *child;
+ struct listnode *chnode;
+ struct listnode *chnextnode;
+ struct pim_ifchannel *ch;
+ struct pim_interface *pim_ifp = NULL;
+ struct channel_oil *ch_oil = NULL;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Updating DF for uptream:%s children", __func__,
+ up->sg_str);
+
+
+ for (ALL_LIST_ELEMENTS(up->ifchannels, chnode, chnextnode, ch)) {
+ pim_ifp = (ch->interface) ? ch->interface->info : NULL;
+ if (!pim_ifp || !PIM_I_am_DualActive(pim_ifp))
+ continue;
+
+ for (ALL_LIST_ELEMENTS_RO(up->sources, listnode, child)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Updating DF for child:%s",
+ __func__, child->sg_str);
+ ch_oil = (child) ? child->channel_oil : NULL;
+
+ if (!ch_oil)
+ continue;
+
+ if (is_df)
+ PIM_MLAG_ADD_OIF_TO_OIL(ch, ch_oil);
+ else
+ PIM_MLAG_DEL_OIF_TO_OIL(ch, ch_oil);
+ }
+ }
+}
+
+/******************************* pim upstream sync **************************/
+/* Update DF role for the upstream entry and return true on role change */
+bool pim_mlag_up_df_role_update(struct pim_instance *pim,
+ struct pim_upstream *up, bool is_df, const char *reason)
+{
+ struct channel_oil *c_oil = up->channel_oil;
+ bool old_is_df = !PIM_UPSTREAM_FLAG_TEST_MLAG_NON_DF(up->flags);
+ struct pim_interface *vxlan_ifp;
+
+ if (is_df == old_is_df) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "%s: Ignoring Role update for %s, since no change",
+ __func__, up->sg_str);
+ return false;
+ }
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("local MLAG mroute %s role changed to %s based on %s",
+ up->sg_str, is_df ? "df" : "non-df", reason);
+
+ if (is_df)
+ PIM_UPSTREAM_FLAG_UNSET_MLAG_NON_DF(up->flags);
+ else
+ PIM_UPSTREAM_FLAG_SET_MLAG_NON_DF(up->flags);
+
+
+ /*
+ * This Upstream entry synced to peer Because of Dual-active
+ * Interface configuration
+ */
+ if (PIM_UPSTREAM_FLAG_TEST_MLAG_INTERFACE(up->flags)) {
+ pim_mlag_inherit_mlag_flags(up, is_df);
+ pim_mlag_calculate_df_for_ifchannels(up, is_df);
+ }
+
+ /* If the DF role has changed check if ipmr-lo needs to be
+ * muted/un-muted. Active-Active devices and vxlan termination
+ * devices (ipmr-lo) are suppressed on the non-DF.
+ * This may leave the mroute with the empty OIL in which case the
+ * the forwarding entry's sole purpose is to just blackhole the flow
+ * headed to the switch.
+ */
+ if (c_oil) {
+ vxlan_ifp = pim_vxlan_get_term_ifp(pim);
+ if (vxlan_ifp)
+ pim_channel_update_oif_mute(c_oil, vxlan_ifp);
+ }
+
+ /* If DF role changed on a (*,G) termination mroute update the
+ * associated DF role on the inherited (S,G) entries
+ */
+ if (pim_addr_is_any(up->sg.src) &&
+ PIM_UPSTREAM_FLAG_TEST_MLAG_VXLAN(up->flags))
+ pim_vxlan_inherit_mlag_flags(pim, up, true /* inherit */);
+
+ return true;
+}
+
+/* Run per-upstream entry DF election and return true on role change */
+static bool pim_mlag_up_df_role_elect(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ bool is_df;
+ uint32_t peer_cost;
+ uint32_t local_cost;
+ bool rv;
+
+ if (!pim_up_mlag_is_local(up))
+ return false;
+
+ /* We are yet to rx a status update from the local MLAG daemon so
+ * we will assume DF status.
+ */
+ if (!(router->mlag_flags & PIM_MLAGF_STATUS_RXED))
+ return pim_mlag_up_df_role_update(pim, up,
+ true /*is_df*/, "mlagd-down");
+
+ /* If not connected to peer assume DF role on the MLAG primary
+ * switch (and non-DF on the secondary switch.
+ */
+ if (!(router->mlag_flags & PIM_MLAGF_PEER_CONN_UP)) {
+ is_df = (router->mlag_role == MLAG_ROLE_PRIMARY) ? true : false;
+ return pim_mlag_up_df_role_update(pim, up,
+ is_df, "peer-down");
+ }
+
+ /* If MLAG peer session is up but zebra is down on the peer
+ * assume DF role.
+ */
+ if (!(router->mlag_flags & PIM_MLAGF_PEER_ZEBRA_UP))
+ return pim_mlag_up_df_role_update(pim, up,
+ true /*is_df*/, "zebra-down");
+
+ /* If we are connected to peer switch but don't have a mroute
+ * from it we have to assume non-DF role to avoid duplicates.
+ * Note: When the peer connection comes up we wait for initial
+ * replay to complete before moving "strays" i.e. local-mlag-mroutes
+ * without a peer reference to non-df role.
+ */
+ if (!PIM_UPSTREAM_FLAG_TEST_MLAG_PEER(up->flags))
+ return pim_mlag_up_df_role_update(pim, up,
+ false /*is_df*/, "no-peer-mroute");
+
+ /* switch with the lowest RPF cost wins. if both switches have the same
+ * cost MLAG role is used as a tie breaker (MLAG primary wins).
+ */
+ peer_cost = up->mlag.peer_mrib_metric;
+ local_cost = pim_up_mlag_local_cost(up);
+ if (local_cost == peer_cost) {
+ is_df = (router->mlag_role == MLAG_ROLE_PRIMARY) ? true : false;
+ rv = pim_mlag_up_df_role_update(pim, up, is_df, "equal-cost");
+ } else {
+ is_df = (local_cost < peer_cost) ? true : false;
+ rv = pim_mlag_up_df_role_update(pim, up, is_df, "cost");
+ }
+
+ return rv;
+}
+
+/* Handle upstream entry add from the peer MLAG switch -
+ * - if a local entry doesn't exist one is created with reference
+ * _MLAG_PEER
+ * - if a local entry exists and has a MLAG OIF DF election is run.
+ * the non-DF switch stop forwarding traffic to MLAG devices.
+ */
+static void pim_mlag_up_peer_add(struct mlag_mroute_add *msg)
+{
+ struct pim_upstream *up;
+ struct pim_instance *pim;
+ int flags = 0;
+ pim_sgaddr sg;
+ struct vrf *vrf;
+
+ memset(&sg, 0, sizeof(sg));
+ sg.src.s_addr = htonl(msg->source_ip);
+ sg.grp.s_addr = htonl(msg->group_ip);
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("peer MLAG mroute add %s:%pSG cost %d",
+ msg->vrf_name, &sg, msg->cost_to_rp);
+
+ /* XXX - this is not correct. we MUST cache updates to avoid losing
+ * an entry because of race conditions with the peer switch.
+ */
+ vrf = vrf_lookup_by_name(msg->vrf_name);
+ if (!vrf) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "peer MLAG mroute add failed %s:%pSG; no vrf",
+ msg->vrf_name, &sg);
+ return;
+ }
+ pim = vrf->info;
+
+ up = pim_upstream_find(pim, &sg);
+ if (up) {
+ /* upstream already exists; create peer reference if it
+ * doesn't already exist.
+ */
+ if (!PIM_UPSTREAM_FLAG_TEST_MLAG_PEER(up->flags))
+ pim_upstream_ref(up, PIM_UPSTREAM_FLAG_MASK_MLAG_PEER,
+ __func__);
+ } else {
+ PIM_UPSTREAM_FLAG_SET_MLAG_PEER(flags);
+ up = pim_upstream_add(pim, &sg, NULL /*iif*/, flags, __func__,
+ NULL /*if_ch*/);
+
+ if (!up) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "peer MLAG mroute add failed %s:%pSG",
+ vrf->name, &sg);
+ return;
+ }
+ }
+ up->mlag.peer_mrib_metric = msg->cost_to_rp;
+ pim_mlag_up_df_role_elect(pim, up);
+}
+
+/* Handle upstream entry del from the peer MLAG switch -
+ * - peer reference is removed. this can result in the upstream
+ * being deleted altogether.
+ * - if a local entry continues to exisy and has a MLAG OIF DF election
+ * is re-run (at the end of which the local entry will be the DF).
+ */
+static struct pim_upstream *pim_mlag_up_peer_deref(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ if (!PIM_UPSTREAM_FLAG_TEST_MLAG_PEER(up->flags))
+ return up;
+
+ PIM_UPSTREAM_FLAG_UNSET_MLAG_PEER(up->flags);
+ up = pim_upstream_del(pim, up, __func__);
+ if (up)
+ pim_mlag_up_df_role_elect(pim, up);
+
+ return up;
+}
+
+static void pim_mlag_up_peer_del(struct mlag_mroute_del *msg)
+{
+ struct pim_upstream *up;
+ struct pim_instance *pim;
+ pim_sgaddr sg;
+ struct vrf *vrf;
+
+ memset(&sg, 0, sizeof(sg));
+ sg.src.s_addr = htonl(msg->source_ip);
+ sg.grp.s_addr = htonl(msg->group_ip);
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("peer MLAG mroute del %s:%pSG", msg->vrf_name, &sg);
+
+ vrf = vrf_lookup_by_name(msg->vrf_name);
+ if (!vrf) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "peer MLAG mroute del skipped %s:%pSG; no vrf",
+ msg->vrf_name, &sg);
+ return;
+ }
+ pim = vrf->info;
+
+ up = pim_upstream_find(pim, &sg);
+ if (!up) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "peer MLAG mroute del skipped %s:%pSG; no up",
+ vrf->name, &sg);
+ return;
+ }
+
+ (void)pim_mlag_up_peer_deref(pim, up);
+}
+
+/* When we lose connection to the local MLAG daemon we can drop all peer
+ * references.
+ */
+static void pim_mlag_up_peer_del_all(void)
+{
+ struct list *temp = list_new();
+ struct pim_upstream *up;
+ struct vrf *vrf;
+ struct pim_instance *pim;
+
+ /*
+ * So why these gyrations?
+ * pim->upstream_head has the list of *,G and S,G
+ * that are in the system. The problem of course
+ * is that it is an ordered list:
+ * (*,G1) -> (S1,G1) -> (S2,G2) -> (S3, G2) -> (*,G2) -> (S1,G2)
+ * And the *,G1 has pointers to S1,G1 and S2,G1
+ * if we delete *,G1 then we have a situation where
+ * S1,G1 and S2,G2 can be deleted as well. Then a
+ * simple ALL_LIST_ELEMENTS will have the next listnode
+ * pointer become invalid and we crash.
+ * So let's grab the list of MLAG_PEER upstreams
+ * add a refcount put on another list and delete safely
+ */
+ RB_FOREACH(vrf, vrf_name_head, &vrfs_by_name) {
+ pim = vrf->info;
+ frr_each (rb_pim_upstream, &pim->upstream_head, up) {
+ if (!PIM_UPSTREAM_FLAG_TEST_MLAG_PEER(up->flags))
+ continue;
+ listnode_add(temp, up);
+ /*
+ * Add a reference since we are adding to this
+ * list for deletion
+ */
+ up->ref_count++;
+ }
+
+ while (temp->count) {
+ up = listnode_head(temp);
+ listnode_delete(temp, up);
+
+ up = pim_mlag_up_peer_deref(pim, up);
+ /*
+ * This is the deletion of the reference added
+ * above
+ */
+ if (up)
+ pim_upstream_del(pim, up, __func__);
+ }
+ }
+
+ list_delete(&temp);
+}
+
+/* Send upstream entry to the local MLAG daemon (which will subsequently
+ * send it to the peer MLAG switch).
+ */
+static void pim_mlag_up_local_add_send(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ struct stream *s = NULL;
+ struct vrf *vrf = pim->vrf;
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP))
+ return;
+
+ s = stream_new(sizeof(struct mlag_mroute_add) + PIM_MLAG_METADATA_LEN);
+ if (!s)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("local MLAG mroute add %s:%s",
+ vrf->name, up->sg_str);
+
+ ++router->mlag_stats.msg.mroute_add_tx;
+
+ stream_putl(s, MLAG_MROUTE_ADD);
+ stream_put(s, vrf->name, VRF_NAMSIZ);
+ stream_putl(s, ntohl(up->sg.src.s_addr));
+ stream_putl(s, ntohl(up->sg.grp.s_addr));
+
+ stream_putl(s, pim_up_mlag_local_cost(up));
+ /* XXX - who is addding*/
+ stream_putl(s, MLAG_OWNER_VXLAN);
+ /* XXX - am_i_DR field should be removed */
+ stream_putc(s, false);
+ stream_putc(s, !(PIM_UPSTREAM_FLAG_TEST_MLAG_NON_DF(up->flags)));
+ stream_putl(s, vrf->vrf_id);
+ /* XXX - this field is a No-op for VXLAN*/
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+
+ stream_fifo_push_safe(router->mlag_fifo, s);
+ pim_mlag_signal_zpthread();
+}
+
+static void pim_mlag_up_local_del_send(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ struct stream *s = NULL;
+ struct vrf *vrf = pim->vrf;
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP))
+ return;
+
+ s = stream_new(sizeof(struct mlag_mroute_del) + PIM_MLAG_METADATA_LEN);
+ if (!s)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("local MLAG mroute del %s:%s",
+ vrf->name, up->sg_str);
+
+ ++router->mlag_stats.msg.mroute_del_tx;
+
+ stream_putl(s, MLAG_MROUTE_DEL);
+ stream_put(s, vrf->name, VRF_NAMSIZ);
+ stream_putl(s, ntohl(up->sg.src.s_addr));
+ stream_putl(s, ntohl(up->sg.grp.s_addr));
+ /* XXX - who is adding */
+ stream_putl(s, MLAG_OWNER_VXLAN);
+ stream_putl(s, vrf->vrf_id);
+ /* XXX - this field is a No-op for VXLAN */
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+
+ /* XXX - is this the the most optimal way to do things */
+ stream_fifo_push_safe(router->mlag_fifo, s);
+ pim_mlag_signal_zpthread();
+}
+
+
+/* Called when a local upstream entry is created or if it's cost changes */
+void pim_mlag_up_local_add(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ pim_mlag_up_df_role_elect(pim, up);
+ /* XXX - need to add some dup checks here */
+ pim_mlag_up_local_add_send(pim, up);
+}
+
+/* Called when local MLAG reference is removed from an upstream entry */
+void pim_mlag_up_local_del(struct pim_instance *pim,
+ struct pim_upstream *up)
+{
+ pim_mlag_up_df_role_elect(pim, up);
+ pim_mlag_up_local_del_send(pim, up);
+}
+
+/* When connection to local MLAG daemon is established all the local
+ * MLAG upstream entries are replayed to it.
+ */
+static void pim_mlag_up_local_replay(void)
+{
+ struct pim_upstream *up;
+ struct vrf *vrf;
+ struct pim_instance *pim;
+
+ RB_FOREACH(vrf, vrf_name_head, &vrfs_by_name) {
+ pim = vrf->info;
+ frr_each (rb_pim_upstream, &pim->upstream_head, up) {
+ if (pim_up_mlag_is_local(up))
+ pim_mlag_up_local_add_send(pim, up);
+ }
+ }
+}
+
+/* on local/peer mlag connection and role changes the DF status needs
+ * to be re-evaluated
+ */
+static void pim_mlag_up_local_reeval(bool mlagd_send, const char *reason_code)
+{
+ struct pim_upstream *up;
+ struct vrf *vrf;
+ struct pim_instance *pim;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s re-run DF election because of %s",
+ __func__, reason_code);
+ RB_FOREACH(vrf, vrf_name_head, &vrfs_by_name) {
+ pim = vrf->info;
+ frr_each (rb_pim_upstream, &pim->upstream_head, up) {
+ if (!pim_up_mlag_is_local(up))
+ continue;
+ /* if role changes re-send to peer */
+ if (pim_mlag_up_df_role_elect(pim, up) &&
+ mlagd_send)
+ pim_mlag_up_local_add_send(pim, up);
+ }
+ }
+}
+
+/*****************PIM Actions for MLAG state changes**********************/
+
+/* notify the anycast VTEP component about state changes */
+static inline void pim_mlag_vxlan_state_update(void)
+{
+ bool enable = !!(router->mlag_flags & PIM_MLAGF_STATUS_RXED);
+ bool peer_state = !!(router->mlag_flags & PIM_MLAGF_PEER_CONN_UP);
+
+ pim_vxlan_mlag_update(enable, peer_state, router->mlag_role,
+ router->peerlink_rif_p, &router->local_vtep_ip);
+
+}
+
+/**************End of PIM Actions for MLAG State changes******************/
+
+
+/********************API to process PIM MLAG Data ************************/
+
+static void pim_mlag_process_mlagd_state_change(struct mlag_status msg)
+{
+ bool role_chg = false;
+ bool state_chg = false;
+ bool notify_vxlan = false;
+ struct interface *peerlink_rif_p;
+ char buf[MLAG_ROLE_STRSIZE];
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg dump: my_role: %s, peer_state: %s",
+ __func__,
+ mlag_role2str(msg.my_role, buf, sizeof(buf)),
+ (msg.peer_state == MLAG_STATE_RUNNING ? "RUNNING"
+ : "DOWN"));
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg ignored mlagd process state down",
+ __func__);
+ return;
+ }
+ ++router->mlag_stats.msg.mlag_status_updates;
+
+ /* evaluate the changes first */
+ if (router->mlag_role != msg.my_role) {
+ role_chg = true;
+ notify_vxlan = true;
+ router->mlag_role = msg.my_role;
+ }
+
+ strlcpy(router->peerlink_rif, msg.peerlink_rif,
+ sizeof(router->peerlink_rif));
+
+ /* XXX - handle the case where we may rx the interface name from the
+ * MLAG daemon before we get the interface from zebra.
+ */
+ peerlink_rif_p = if_lookup_by_name(router->peerlink_rif, VRF_DEFAULT);
+ if (router->peerlink_rif_p != peerlink_rif_p) {
+ router->peerlink_rif_p = peerlink_rif_p;
+ notify_vxlan = true;
+ }
+
+ if (msg.peer_state == MLAG_STATE_RUNNING) {
+ if (!(router->mlag_flags & PIM_MLAGF_PEER_CONN_UP)) {
+ state_chg = true;
+ notify_vxlan = true;
+ router->mlag_flags |= PIM_MLAGF_PEER_CONN_UP;
+ }
+ router->connected_to_mlag = true;
+ } else {
+ if (router->mlag_flags & PIM_MLAGF_PEER_CONN_UP) {
+ ++router->mlag_stats.peer_session_downs;
+ state_chg = true;
+ notify_vxlan = true;
+ router->mlag_flags &= ~PIM_MLAGF_PEER_CONN_UP;
+ }
+ router->connected_to_mlag = false;
+ }
+
+ /* apply the changes */
+ /* when connection to mlagd comes up we hold send mroutes till we have
+ * rxed the status and had a chance to re-valuate DF state
+ */
+ if (!(router->mlag_flags & PIM_MLAGF_STATUS_RXED)) {
+ router->mlag_flags |= PIM_MLAGF_STATUS_RXED;
+ pim_mlag_vxlan_state_update();
+ /* on session up re-eval DF status */
+ pim_mlag_up_local_reeval(false /*mlagd_send*/, "mlagd_up");
+ /* replay all the upstream entries to the local MLAG daemon */
+ pim_mlag_up_local_replay();
+ return;
+ }
+
+ if (notify_vxlan)
+ pim_mlag_vxlan_state_update();
+
+ if (state_chg) {
+ if (!(router->mlag_flags & PIM_MLAGF_PEER_CONN_UP))
+ /* when a connection goes down the primary takes over
+ * DF role for all entries
+ */
+ pim_mlag_up_local_reeval(true /*mlagd_send*/,
+ "peer_down");
+ else
+ /* XXX - when session comes up we need to wait for
+ * PEER_REPLAY_DONE before running re-election on
+ * local-mlag entries that are missing peer reference
+ */
+ pim_mlag_up_local_reeval(true /*mlagd_send*/,
+ "peer_up");
+ } else if (role_chg) {
+ /* MLAG role changed without a state change */
+ pim_mlag_up_local_reeval(true /*mlagd_send*/, "role_chg");
+ }
+}
+
+static void pim_mlag_process_peer_frr_state_change(struct mlag_frr_status msg)
+{
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "%s: msg dump: peer_frr_state: %s", __func__,
+ (msg.frr_state == MLAG_FRR_STATE_UP ? "UP" : "DOWN"));
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg ignored mlagd process state down",
+ __func__);
+ return;
+ }
+ ++router->mlag_stats.msg.peer_zebra_status_updates;
+
+ /* evaluate the changes first */
+ if (msg.frr_state == MLAG_FRR_STATE_UP) {
+ if (!(router->mlag_flags & PIM_MLAGF_PEER_ZEBRA_UP)) {
+ router->mlag_flags |= PIM_MLAGF_PEER_ZEBRA_UP;
+ /* XXX - when peer zebra comes up we need to wait for
+ * for some time to let the peer setup MDTs before
+ * before relinquishing DF status
+ */
+ pim_mlag_up_local_reeval(true /*mlagd_send*/,
+ "zebra_up");
+ }
+ } else {
+ if (router->mlag_flags & PIM_MLAGF_PEER_ZEBRA_UP) {
+ ++router->mlag_stats.peer_zebra_downs;
+ router->mlag_flags &= ~PIM_MLAGF_PEER_ZEBRA_UP;
+ /* when a peer zebra goes down we assume DF role */
+ pim_mlag_up_local_reeval(true /*mlagd_send*/,
+ "zebra_down");
+ }
+ }
+}
+
+static void pim_mlag_process_vxlan_update(struct mlag_vxlan *msg)
+{
+ char addr_buf1[INET_ADDRSTRLEN];
+ char addr_buf2[INET_ADDRSTRLEN];
+ uint32_t local_ip;
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg ignored mlagd process state down",
+ __func__);
+ return;
+ }
+
+ ++router->mlag_stats.msg.vxlan_updates;
+ router->anycast_vtep_ip.s_addr = htonl(msg->anycast_ip);
+ local_ip = htonl(msg->local_ip);
+ if (router->local_vtep_ip.s_addr != local_ip) {
+ router->local_vtep_ip.s_addr = local_ip;
+ pim_mlag_vxlan_state_update();
+ }
+
+ if (PIM_DEBUG_MLAG) {
+ inet_ntop(AF_INET, &router->local_vtep_ip,
+ addr_buf1, INET_ADDRSTRLEN);
+ inet_ntop(AF_INET, &router->anycast_vtep_ip,
+ addr_buf2, INET_ADDRSTRLEN);
+
+ zlog_debug("%s: msg dump: local-ip:%s, anycast-ip:%s",
+ __func__, addr_buf1, addr_buf2);
+ }
+}
+
+static void pim_mlag_process_mroute_add(struct mlag_mroute_add msg)
+{
+ if (PIM_DEBUG_MLAG) {
+ pim_sgaddr sg;
+
+ sg.grp.s_addr = ntohl(msg.group_ip);
+ sg.src.s_addr = ntohl(msg.source_ip);
+
+ zlog_debug(
+ "%s: msg dump: vrf_name: %s, s.ip: 0x%x, g.ip: 0x%x (%pSG) cost: %u",
+ __func__, msg.vrf_name, msg.source_ip, msg.group_ip,
+ &sg, msg.cost_to_rp);
+ zlog_debug(
+ "(%pSG)owner_id: %d, DR: %d, Dual active: %d, vrf_id: 0x%x intf_name: %s",
+ &sg, msg.owner_id, msg.am_i_dr, msg.am_i_dual_active,
+ msg.vrf_id, msg.intf_name);
+ }
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg ignored mlagd process state down",
+ __func__);
+ return;
+ }
+
+ ++router->mlag_stats.msg.mroute_add_rx;
+
+ pim_mlag_up_peer_add(&msg);
+}
+
+static void pim_mlag_process_mroute_del(struct mlag_mroute_del msg)
+{
+ if (PIM_DEBUG_MLAG) {
+ pim_sgaddr sg;
+
+ sg.grp.s_addr = ntohl(msg.group_ip);
+ sg.src.s_addr = ntohl(msg.source_ip);
+ zlog_debug(
+ "%s: msg dump: vrf_name: %s, s.ip: 0x%x, g.ip: 0x%x(%pSG)",
+ __func__, msg.vrf_name, msg.source_ip, msg.group_ip,
+ &sg);
+ zlog_debug("(%pSG)owner_id: %d, vrf_id: 0x%x intf_name: %s",
+ &sg, msg.owner_id, msg.vrf_id, msg.intf_name);
+ }
+
+ if (!(router->mlag_flags & PIM_MLAGF_LOCAL_CONN_UP)) {
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: msg ignored mlagd process state down",
+ __func__);
+ return;
+ }
+
+ ++router->mlag_stats.msg.mroute_del_rx;
+
+ pim_mlag_up_peer_del(&msg);
+}
+
+int pim_zebra_mlag_handle_msg(int cmd, struct zclient *zclient,
+ uint16_t zapi_length, vrf_id_t vrf_id)
+{
+ struct stream *s = zclient->ibuf;
+ struct mlag_msg mlag_msg;
+ char buf[80];
+ int rc = 0;
+ size_t length;
+
+ rc = mlag_lib_decode_mlag_hdr(s, &mlag_msg, &length);
+ if (rc)
+ return (rc);
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Received msg type: %s length: %d, bulk_cnt: %d",
+ __func__,
+ mlag_lib_msgid_to_str(mlag_msg.msg_type, buf,
+ sizeof(buf)),
+ mlag_msg.data_len, mlag_msg.msg_cnt);
+
+ switch (mlag_msg.msg_type) {
+ case MLAG_STATUS_UPDATE: {
+ struct mlag_status msg;
+
+ rc = mlag_lib_decode_mlag_status(s, &msg);
+ if (rc)
+ return (rc);
+ pim_mlag_process_mlagd_state_change(msg);
+ } break;
+ case MLAG_PEER_FRR_STATUS: {
+ struct mlag_frr_status msg;
+
+ rc = mlag_lib_decode_frr_status(s, &msg);
+ if (rc)
+ return (rc);
+ pim_mlag_process_peer_frr_state_change(msg);
+ } break;
+ case MLAG_VXLAN_UPDATE: {
+ struct mlag_vxlan msg;
+
+ rc = mlag_lib_decode_vxlan_update(s, &msg);
+ if (rc)
+ return rc;
+ pim_mlag_process_vxlan_update(&msg);
+ } break;
+ case MLAG_MROUTE_ADD: {
+ struct mlag_mroute_add msg;
+
+ rc = mlag_lib_decode_mroute_add(s, &msg, &length);
+ if (rc)
+ return (rc);
+ pim_mlag_process_mroute_add(msg);
+ } break;
+ case MLAG_MROUTE_DEL: {
+ struct mlag_mroute_del msg;
+
+ rc = mlag_lib_decode_mroute_del(s, &msg, &length);
+ if (rc)
+ return (rc);
+ pim_mlag_process_mroute_del(msg);
+ } break;
+ case MLAG_MROUTE_ADD_BULK: {
+ struct mlag_mroute_add msg;
+ int i;
+
+ for (i = 0; i < mlag_msg.msg_cnt; i++) {
+ rc = mlag_lib_decode_mroute_add(s, &msg, &length);
+ if (rc)
+ return (rc);
+ pim_mlag_process_mroute_add(msg);
+ }
+ } break;
+ case MLAG_MROUTE_DEL_BULK: {
+ struct mlag_mroute_del msg;
+ int i;
+
+ for (i = 0; i < mlag_msg.msg_cnt; i++) {
+ rc = mlag_lib_decode_mroute_del(s, &msg, &length);
+ if (rc)
+ return (rc);
+ pim_mlag_process_mroute_del(msg);
+ }
+ } break;
+ default:
+ break;
+ }
+ return 0;
+}
+
+/****************End of PIM Mesasge processing handler********************/
+
+int pim_zebra_mlag_process_up(ZAPI_CALLBACK_ARGS)
+{
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Received Process-Up from Mlag", __func__);
+
+ /*
+ * Incase of local MLAG restart, PIM needs to replay all the data
+ * since MLAG is empty.
+ */
+ router->connected_to_mlag = true;
+ router->mlag_flags |= PIM_MLAGF_LOCAL_CONN_UP;
+ return 0;
+}
+
+static void pim_mlag_param_reset(void)
+{
+ /* reset the cached params and stats */
+ router->mlag_flags &= ~(PIM_MLAGF_STATUS_RXED |
+ PIM_MLAGF_LOCAL_CONN_UP |
+ PIM_MLAGF_PEER_CONN_UP |
+ PIM_MLAGF_PEER_ZEBRA_UP);
+ router->local_vtep_ip.s_addr = INADDR_ANY;
+ router->anycast_vtep_ip.s_addr = INADDR_ANY;
+ router->mlag_role = MLAG_ROLE_NONE;
+ memset(&router->mlag_stats.msg, 0, sizeof(router->mlag_stats.msg));
+ router->peerlink_rif[0] = '\0';
+}
+
+int pim_zebra_mlag_process_down(ZAPI_CALLBACK_ARGS)
+{
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Received Process-Down from Mlag", __func__);
+
+ /* Local CLAG is down, reset peer data and forward the traffic if
+ * we are DR
+ */
+ if (router->mlag_flags & PIM_MLAGF_PEER_CONN_UP)
+ ++router->mlag_stats.peer_session_downs;
+ if (router->mlag_flags & PIM_MLAGF_PEER_ZEBRA_UP)
+ ++router->mlag_stats.peer_zebra_downs;
+ router->connected_to_mlag = false;
+ pim_mlag_param_reset();
+ /* on mlagd session down re-eval DF status */
+ pim_mlag_up_local_reeval(false /*mlagd_send*/, "mlagd_down");
+ /* flush all peer references */
+ pim_mlag_up_peer_del_all();
+ /* notify the vxlan component */
+ pim_mlag_vxlan_state_update();
+ return 0;
+}
+
+static void pim_mlag_register_handler(struct thread *thread)
+{
+ uint32_t bit_mask = 0;
+
+ if (!zclient)
+ return;
+
+ SET_FLAG(bit_mask, (1 << MLAG_STATUS_UPDATE));
+ SET_FLAG(bit_mask, (1 << MLAG_MROUTE_ADD));
+ SET_FLAG(bit_mask, (1 << MLAG_MROUTE_DEL));
+ SET_FLAG(bit_mask, (1 << MLAG_DUMP));
+ SET_FLAG(bit_mask, (1 << MLAG_MROUTE_ADD_BULK));
+ SET_FLAG(bit_mask, (1 << MLAG_MROUTE_DEL_BULK));
+ SET_FLAG(bit_mask, (1 << MLAG_PIM_CFG_DUMP));
+ SET_FLAG(bit_mask, (1 << MLAG_VXLAN_UPDATE));
+ SET_FLAG(bit_mask, (1 << MLAG_PEER_FRR_STATUS));
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Posting Client Register to MLAG mask: 0x%x",
+ __func__, bit_mask);
+
+ zclient_send_mlag_register(zclient, bit_mask);
+}
+
+void pim_mlag_register(void)
+{
+ if (router->mlag_process_register)
+ return;
+
+ router->mlag_process_register = true;
+
+ thread_add_event(router->master, pim_mlag_register_handler, NULL, 0,
+ NULL);
+}
+
+static void pim_mlag_deregister_handler(struct thread *thread)
+{
+ if (!zclient)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Posting Client De-Register to MLAG from PIM",
+ __func__);
+ router->connected_to_mlag = false;
+ zclient_send_mlag_deregister(zclient);
+}
+
+void pim_mlag_deregister(void)
+{
+ /* if somebody still interested in the MLAG channel skip de-reg */
+ if (router->pim_mlag_intf_cnt || pim_vxlan_do_mlag_reg())
+ return;
+
+ /* not registered; nothing do */
+ if (!router->mlag_process_register)
+ return;
+
+ router->mlag_process_register = false;
+
+ thread_add_event(router->master, pim_mlag_deregister_handler, NULL, 0,
+ NULL);
+}
+
+void pim_if_configure_mlag_dualactive(struct pim_interface *pim_ifp)
+{
+ if (!pim_ifp || !pim_ifp->pim || pim_ifp->activeactive == true)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: Configuring active-active on Interface: %s",
+ __func__, "NULL");
+
+ pim_ifp->activeactive = true;
+ if (pim_ifp->pim)
+ pim_ifp->pim->inst_mlag_intf_cnt++;
+
+ router->pim_mlag_intf_cnt++;
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "%s: Total MLAG configured Interfaces on router: %d, Inst: %d",
+ __func__, router->pim_mlag_intf_cnt,
+ pim_ifp->pim->inst_mlag_intf_cnt);
+
+ if (router->pim_mlag_intf_cnt == 1) {
+ /*
+ * at least one Interface is configured for MLAG, send register
+ * to Zebra for receiving MLAG Updates
+ */
+ pim_mlag_register();
+ }
+}
+
+void pim_if_unconfigure_mlag_dualactive(struct pim_interface *pim_ifp)
+{
+ if (!pim_ifp || !pim_ifp->pim || pim_ifp->activeactive == false)
+ return;
+
+ if (PIM_DEBUG_MLAG)
+ zlog_debug("%s: UnConfiguring active-active on Interface: %s",
+ __func__, "NULL");
+
+ pim_ifp->activeactive = false;
+ pim_ifp->pim->inst_mlag_intf_cnt--;
+
+ router->pim_mlag_intf_cnt--;
+ if (PIM_DEBUG_MLAG)
+ zlog_debug(
+ "%s: Total MLAG configured Interfaces on router: %d, Inst: %d",
+ __func__, router->pim_mlag_intf_cnt,
+ pim_ifp->pim->inst_mlag_intf_cnt);
+
+ if (router->pim_mlag_intf_cnt == 0) {
+ /*
+ * all the Interfaces are MLAG un-configured, post MLAG
+ * De-register to Zebra
+ */
+ pim_mlag_deregister();
+ pim_mlag_param_reset();
+ }
+}
+
+
+void pim_instance_mlag_init(struct pim_instance *pim)
+{
+ if (!pim)
+ return;
+
+ pim->inst_mlag_intf_cnt = 0;
+}
+
+
+void pim_instance_mlag_terminate(struct pim_instance *pim)
+{
+ struct interface *ifp;
+
+ if (!pim)
+ return;
+
+ FOR_ALL_INTERFACES (pim->vrf, ifp) {
+ struct pim_interface *pim_ifp = ifp->info;
+
+ if (!pim_ifp || pim_ifp->activeactive == false)
+ continue;
+
+ pim_if_unconfigure_mlag_dualactive(pim_ifp);
+ }
+ pim->inst_mlag_intf_cnt = 0;
+}
+
+void pim_mlag_terminate(void)
+{
+ stream_free(router->mlag_stream);
+ router->mlag_stream = NULL;
+ stream_fifo_free(router->mlag_fifo);
+ router->mlag_fifo = NULL;
+}
+
+void pim_mlag_init(void)
+{
+ pim_mlag_param_reset();
+ router->pim_mlag_intf_cnt = 0;
+ router->connected_to_mlag = false;
+ router->mlag_fifo = stream_fifo_new();
+ router->zpthread_mlag_write = NULL;
+ router->mlag_stream = stream_new(MLAG_BUF_LIMIT);
+}