tipc: remove implicit message delivery in node_unlock()
[cascardo/linux.git] / net / tipc / node.c
index 77effb2..9e9b093 100644 (file)
 #include "name_distr.h"
 #include "socket.h"
 #include "bcast.h"
+#include "discover.h"
 
-static void node_lost_contact(struct tipc_node *n_ptr);
+/* Node FSM states and events:
+ */
+enum {
+       SELF_DOWN_PEER_DOWN    = 0xdd,
+       SELF_UP_PEER_UP        = 0xaa,
+       SELF_DOWN_PEER_LEAVING = 0xd1,
+       SELF_UP_PEER_COMING    = 0xac,
+       SELF_COMING_PEER_UP    = 0xca,
+       SELF_LEAVING_PEER_DOWN = 0x1d,
+       NODE_FAILINGOVER       = 0xf0,
+       NODE_SYNCHING          = 0xcc
+};
+
+enum {
+       SELF_ESTABL_CONTACT_EVT = 0xece,
+       SELF_LOST_CONTACT_EVT   = 0x1ce,
+       PEER_ESTABL_CONTACT_EVT = 0x9ece,
+       PEER_LOST_CONTACT_EVT   = 0x91ce,
+       NODE_FAILOVER_BEGIN_EVT = 0xfbe,
+       NODE_FAILOVER_END_EVT   = 0xfee,
+       NODE_SYNCH_BEGIN_EVT    = 0xcbe,
+       NODE_SYNCH_END_EVT      = 0xcee
+};
+
+static void __tipc_node_link_down(struct tipc_node *n, int *bearer_id,
+                                 struct sk_buff_head *xmitq,
+                                 struct tipc_media_addr **maddr);
+static void tipc_node_link_down(struct tipc_node *n, int bearer_id,
+                               bool delete);
+static void node_lost_contact(struct tipc_node *n, struct sk_buff_head *inputq);
 static void node_established_contact(struct tipc_node *n_ptr);
 static void tipc_node_delete(struct tipc_node *node);
 static void tipc_node_timeout(unsigned long data);
+static void tipc_node_fsm_evt(struct tipc_node *n, int evt);
 
 struct tipc_sock_conn {
        u32 port;
@@ -111,7 +142,7 @@ struct tipc_node *tipc_node_find(struct net *net, u32 addr)
        return NULL;
 }
 
-struct tipc_node *tipc_node_create(struct net *net, u32 addr)
+struct tipc_node *tipc_node_create(struct net *net, u32 addr, u16 capabilities)
 {
        struct tipc_net *tn = net_generic(net, tipc_net_id);
        struct tipc_node *n_ptr, *temp_node;
@@ -127,6 +158,7 @@ struct tipc_node *tipc_node_create(struct net *net, u32 addr)
        }
        n_ptr->addr = addr;
        n_ptr->net = net;
+       n_ptr->capabilities = capabilities;
        kref_init(&n_ptr->kref);
        spin_lock_init(&n_ptr->lock);
        INIT_HLIST_NODE(&n_ptr->hash);
@@ -141,7 +173,7 @@ struct tipc_node *tipc_node_create(struct net *net, u32 addr)
                        break;
        }
        list_add_tail_rcu(&n_ptr->list, &temp_node->list);
-       n_ptr->action_flags = TIPC_WAIT_PEER_LINKS_DOWN;
+       n_ptr->state = SELF_DOWN_PEER_LEAVING;
        n_ptr->signature = INVALID_NODE_SIG;
        n_ptr->active_links[0] = INVALID_BEARER_ID;
        n_ptr->active_links[1] = INVALID_BEARER_ID;
@@ -247,9 +279,8 @@ void tipc_node_remove_conn(struct net *net, u32 dnode, u32 port)
 static void tipc_node_timeout(unsigned long data)
 {
        struct tipc_node *n = (struct tipc_node *)data;
+       struct tipc_link_entry *le;
        struct sk_buff_head xmitq;
-       struct tipc_link *l;
-       struct tipc_media_addr *maddr;
        int bearer_id;
        int rc = 0;
 
@@ -257,17 +288,16 @@ static void tipc_node_timeout(unsigned long data)
 
        for (bearer_id = 0; bearer_id < MAX_BEARERS; bearer_id++) {
                tipc_node_lock(n);
-               l = n->links[bearer_id].link;
-               if (l) {
+               le = &n->links[bearer_id];
+               if (le->link) {
                        /* Link tolerance may change asynchronously: */
-                       tipc_node_calculate_timer(n, l);
-                       rc = tipc_link_timeout(l, &xmitq);
-                       if (rc & TIPC_LINK_DOWN_EVT)
-                               tipc_link_reset(l);
+                       tipc_node_calculate_timer(n, le->link);
+                       rc = tipc_link_timeout(le->link, &xmitq);
                }
                tipc_node_unlock(n);
-               maddr = &n->links[bearer_id].maddr;
-               tipc_bearer_xmit(n->net, bearer_id, &xmitq, maddr);
+               tipc_bearer_xmit(n->net, bearer_id, &xmitq, &le->maddr);
+               if (rc & TIPC_LINK_DOWN_EVT)
+                       tipc_node_link_down(n, bearer_id, false);
        }
        if (!mod_timer(&n->timer, jiffies + n->keepalive_intv))
                tipc_node_get(n);
@@ -275,68 +305,96 @@ static void tipc_node_timeout(unsigned long data)
 }
 
 /**
- * tipc_node_link_up - handle addition of link
- *
+ * __tipc_node_link_up - handle addition of link
+ * Node lock must be held by caller
  * Link becomes active (alone or shared) or standby, depending on its priority.
  */
-void tipc_node_link_up(struct tipc_node *n, int bearer_id)
+static void __tipc_node_link_up(struct tipc_node *n, int bearer_id,
+                               struct sk_buff_head *xmitq)
 {
        int *slot0 = &n->active_links[0];
        int *slot1 = &n->active_links[1];
-       struct tipc_link_entry *links = n->links;
-       struct tipc_link *l = n->links[bearer_id].link;
+       struct tipc_link *ol = node_active_link(n, 0);
+       struct tipc_link *nl = n->links[bearer_id].link;
 
-       /* Leave room for tunnel header when returning 'mtu' to users: */
-       links[bearer_id].mtu = l->mtu - INT_H_SIZE;
+       if (!nl || !tipc_link_is_up(nl))
+               return;
 
+       if (n->working_links > 1) {
+               pr_warn("Attempt to establish 3rd link to %x\n", n->addr);
+               return;
+       }
        n->working_links++;
        n->action_flags |= TIPC_NOTIFY_LINK_UP;
-       n->link_id = l->peer_bearer_id << 16 | l->bearer_id;
+       n->link_id = nl->peer_bearer_id << 16 | bearer_id;
+
+       /* Leave room for tunnel header when returning 'mtu' to users: */
+       n->links[bearer_id].mtu = nl->mtu - INT_H_SIZE;
+
+       tipc_bearer_add_dest(n->net, bearer_id, n->addr);
 
        pr_debug("Established link <%s> on network plane %c\n",
-                l->name, l->net_plane);
+                nl->name, nl->net_plane);
 
-       /* No active links ? => take both active slots */
-       if (*slot0 < 0) {
+       /* First link? => give it both slots */
+       if (!ol) {
                *slot0 = bearer_id;
                *slot1 = bearer_id;
+               tipc_link_build_bcast_sync_msg(nl, xmitq);
                node_established_contact(n);
                return;
        }
 
-       /* Lower prio than current active ? => no slot */
-       if (l->priority < links[*slot0].link->priority) {
-               pr_debug("New link <%s> becomes standby\n", l->name);
-               return;
-       }
-       tipc_link_dup_queue_xmit(links[*slot0].link, l);
-
-       /* Same prio as current active ? => take one slot */
-       if (l->priority == links[*slot0].link->priority) {
+       /* Second link => redistribute slots */
+       if (nl->priority > ol->priority) {
+               pr_debug("Old link <%s> becomes standby\n", ol->name);
                *slot0 = bearer_id;
-               return;
+               *slot1 = bearer_id;
+       } else if (nl->priority == ol->priority) {
+               *slot0 = bearer_id;
+       } else {
+               pr_debug("New link <%s> is standby\n", nl->name);
        }
 
-       /* Higher prio than current active => take both active slots */
-       pr_debug("Old link <%s> now standby\n", links[*slot0].link->name);
-       *slot0 = bearer_id;
-       *slot1 = bearer_id;
+       /* Prepare synchronization with first link */
+       tipc_link_tnl_prepare(ol, nl, SYNCH_MSG, xmitq);
+}
+
+/**
+ * tipc_node_link_up - handle addition of link
+ *
+ * Link becomes active (alone or shared) or standby, depending on its priority.
+ */
+static void tipc_node_link_up(struct tipc_node *n, int bearer_id,
+                             struct sk_buff_head *xmitq)
+{
+       tipc_node_lock(n);
+       __tipc_node_link_up(n, bearer_id, xmitq);
+       tipc_node_unlock(n);
 }
 
 /**
- * tipc_node_link_down - handle loss of link
+ * __tipc_node_link_down - handle loss of link
  */
-void tipc_node_link_down(struct tipc_node *n, int bearer_id)
+static void __tipc_node_link_down(struct tipc_node *n, int *bearer_id,
+                                 struct sk_buff_head *xmitq,
+                                 struct tipc_media_addr **maddr)
 {
+       struct tipc_link_entry *le = &n->links[*bearer_id];
        int *slot0 = &n->active_links[0];
        int *slot1 = &n->active_links[1];
        int i, highest = 0;
-       struct tipc_link *l, *_l;
+       struct tipc_link *l, *_l, *tnl;
+
+       l = n->links[*bearer_id].link;
+       if (!l || tipc_link_is_reset(l))
+               return;
 
-       l = n->links[bearer_id].link;
        n->working_links--;
        n->action_flags |= TIPC_NOTIFY_LINK_DOWN;
-       n->link_id = l->peer_bearer_id << 16 | l->bearer_id;
+       n->link_id = l->peer_bearer_id << 16 | *bearer_id;
+
+       tipc_bearer_remove_dest(n->net, *bearer_id, n->addr);
 
        pr_debug("Lost link <%s> on network plane %c\n",
                 l->name, l->net_plane);
@@ -348,6 +406,8 @@ void tipc_node_link_down(struct tipc_node *n, int bearer_id)
                _l = n->links[i].link;
                if (!_l || !tipc_link_is_up(_l))
                        continue;
+               if (_l == l)
+                       continue;
                if (_l->priority < highest)
                        continue;
                if (_l->priority > highest) {
@@ -358,10 +418,43 @@ void tipc_node_link_down(struct tipc_node *n, int bearer_id)
                }
                *slot1 = i;
        }
-       if (tipc_node_is_up(n))
-               tipc_link_failover_send_queue(l);
-       else
-               node_lost_contact(n);
+
+       if (!tipc_node_is_up(n)) {
+               tipc_link_reset(l);
+               node_lost_contact(n, &le->inputq);
+               return;
+       }
+
+       /* There is still a working link => initiate failover */
+       tnl = node_active_link(n, 0);
+       n->sync_point = tnl->rcv_nxt + (U16_MAX / 2 - 1);
+       tipc_link_tnl_prepare(l, tnl, FAILOVER_MSG, xmitq);
+       tipc_link_reset(l);
+       tipc_link_fsm_evt(l, LINK_FAILOVER_BEGIN_EVT);
+       tipc_node_fsm_evt(n, NODE_FAILOVER_BEGIN_EVT);
+       *maddr = &n->links[tnl->bearer_id].maddr;
+       *bearer_id = tnl->bearer_id;
+}
+
+static void tipc_node_link_down(struct tipc_node *n, int bearer_id, bool delete)
+{
+       struct tipc_link_entry *le = &n->links[bearer_id];
+       struct tipc_media_addr *maddr;
+       struct sk_buff_head xmitq;
+
+       __skb_queue_head_init(&xmitq);
+
+       tipc_node_lock(n);
+       __tipc_node_link_down(n, &bearer_id, &xmitq, &maddr);
+       if (delete && le->link) {
+               kfree(le->link);
+               le->link = NULL;
+               n->link_cnt--;
+       }
+       tipc_node_unlock(n);
+
+       tipc_bearer_xmit(n->net, bearer_id, &xmitq, maddr);
+       tipc_sk_rcv(n->net, &le->inputq);
 }
 
 bool tipc_node_is_up(struct tipc_node *n)
@@ -369,38 +462,145 @@ bool tipc_node_is_up(struct tipc_node *n)
        return n->active_links[0] != INVALID_BEARER_ID;
 }
 
-void tipc_node_check_dest(struct tipc_node *n, struct tipc_bearer *b,
-                         bool *link_up, bool *addr_match,
-                         struct tipc_media_addr *maddr)
+void tipc_node_check_dest(struct net *net, u32 onode,
+                         struct tipc_bearer *b,
+                         u16 capabilities, u32 signature,
+                         struct tipc_media_addr *maddr,
+                         bool *respond, bool *dupl_addr)
 {
-       struct tipc_link *l = n->links[b->identity].link;
-       struct tipc_media_addr *curr = &n->links[b->identity].maddr;
+       struct tipc_node *n;
+       struct tipc_link *l;
+       struct tipc_media_addr *curr_maddr;
+       struct sk_buff_head *inputq;
+       bool addr_match = false;
+       bool sign_match = false;
+       bool link_up = false;
+       bool accept_addr = false;
+       bool reset = true;
+       *dupl_addr = false;
+       *respond = false;
+
+       n = tipc_node_create(net, onode, capabilities);
+       if (!n)
+               return;
 
-       *link_up = l && tipc_link_is_up(l);
-       *addr_match = l && !memcmp(curr, maddr, sizeof(*maddr));
-}
+       tipc_node_lock(n);
+
+       curr_maddr = &n->links[b->identity].maddr;
+       inputq = &n->links[b->identity].inputq;
+
+       /* Prepare to validate requesting node's signature and media address */
+       l = n->links[b->identity].link;
+       link_up = l && tipc_link_is_up(l);
+       addr_match = l && !memcmp(curr_maddr, maddr, sizeof(*maddr));
+       sign_match = (signature == n->signature);
+
+       /* These three flags give us eight permutations: */
+
+       if (sign_match && addr_match && link_up) {
+               /* All is fine. Do nothing. */
+               reset = false;
+       } else if (sign_match && addr_match && !link_up) {
+               /* Respond. The link will come up in due time */
+               *respond = true;
+       } else if (sign_match && !addr_match && link_up) {
+               /* Peer has changed i/f address without rebooting.
+                * If so, the link will reset soon, and the next
+                * discovery will be accepted. So we can ignore it.
+                * It may also be an cloned or malicious peer having
+                * chosen the same node address and signature as an
+                * existing one.
+                * Ignore requests until the link goes down, if ever.
+                */
+               *dupl_addr = true;
+       } else if (sign_match && !addr_match && !link_up) {
+               /* Peer link has changed i/f address without rebooting.
+                * It may also be a cloned or malicious peer; we can't
+                * distinguish between the two.
+                * The signature is correct, so we must accept.
+                */
+               accept_addr = true;
+               *respond = true;
+       } else if (!sign_match && addr_match && link_up) {
+               /* Peer node rebooted. Two possibilities:
+                *  - Delayed re-discovery; this link endpoint has already
+                *    reset and re-established contact with the peer, before
+                *    receiving a discovery message from that node.
+                *    (The peer happened to receive one from this node first).
+                *  - The peer came back so fast that our side has not
+                *    discovered it yet. Probing from this side will soon
+                *    reset the link, since there can be no working link
+                *    endpoint at the peer end, and the link will re-establish.
+                *  Accept the signature, since it comes from a known peer.
+                */
+               n->signature = signature;
+       } else if (!sign_match && addr_match && !link_up) {
+               /*  The peer node has rebooted.
+                *  Accept signature, since it is a known peer.
+                */
+               n->signature = signature;
+               *respond = true;
+       } else if (!sign_match && !addr_match && link_up) {
+               /* Peer rebooted with new address, or a new/duplicate peer.
+                * Ignore until the link goes down, if ever.
+                */
+               *dupl_addr = true;
+       } else if (!sign_match && !addr_match && !link_up) {
+               /* Peer rebooted with new address, or it is a new peer.
+                * Accept signature and address.
+                */
+               n->signature = signature;
+               accept_addr = true;
+               *respond = true;
+       }
 
-bool tipc_node_update_dest(struct tipc_node *n,  struct tipc_bearer *b,
-                          struct tipc_media_addr *maddr)
-{
-       struct tipc_link *l = n->links[b->identity].link;
-       struct tipc_media_addr *curr = &n->links[b->identity].maddr;
-       struct sk_buff_head *inputq = &n->links[b->identity].inputq;
+       if (!accept_addr)
+               goto exit;
 
+       /* Now create new link if not already existing */
        if (!l) {
                l = tipc_link_create(n, b, maddr, inputq, &n->bclink.namedq);
-               if (!l)
-                       return false;
+               if (!l) {
+                       *respond = false;
+                       goto exit;
+               }
                tipc_node_calculate_timer(n, l);
-               if (n->link_cnt == 1) {
+               if (n->link_cnt == 1)
                        if (!mod_timer(&n->timer, jiffies + n->keepalive_intv))
                                tipc_node_get(n);
-               }
        }
        memcpy(&l->media_addr, maddr, sizeof(*maddr));
-       memcpy(curr, maddr, sizeof(*maddr));
-       tipc_link_reset(l);
-       return true;
+       memcpy(curr_maddr, maddr, sizeof(*maddr));
+exit:
+       tipc_node_unlock(n);
+       if (reset)
+               tipc_node_link_down(n, b->identity, false);
+       tipc_node_put(n);
+}
+
+void tipc_node_delete_links(struct net *net, int bearer_id)
+{
+       struct tipc_net *tn = net_generic(net, tipc_net_id);
+       struct tipc_node *n;
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(n, &tn->node_list, list) {
+               tipc_node_link_down(n, bearer_id, true);
+       }
+       rcu_read_unlock();
+}
+
+static void tipc_node_reset_links(struct tipc_node *n)
+{
+       char addr_string[16];
+       int i;
+
+       pr_warn("Resetting all links to %s\n",
+               tipc_addr_string_fill(addr_string, n->addr));
+
+       for (i = 0; i < MAX_BEARERS; i++) {
+               tipc_node_link_down(n, i, false);
+       }
 }
 
 void tipc_node_attach_link(struct tipc_node *n_ptr, struct tipc_link *l_ptr)
@@ -421,18 +621,216 @@ void tipc_node_detach_link(struct tipc_node *n_ptr, struct tipc_link *l_ptr)
        }
 }
 
+/* tipc_node_fsm_evt - node finite state machine
+ * Determines when contact is allowed with peer node
+ */
+static void tipc_node_fsm_evt(struct tipc_node *n, int evt)
+{
+       int state = n->state;
+
+       switch (state) {
+       case SELF_DOWN_PEER_DOWN:
+               switch (evt) {
+               case SELF_ESTABL_CONTACT_EVT:
+                       state = SELF_UP_PEER_COMING;
+                       break;
+               case PEER_ESTABL_CONTACT_EVT:
+                       state = SELF_COMING_PEER_UP;
+                       break;
+               case SELF_LOST_CONTACT_EVT:
+               case PEER_LOST_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_END_EVT:
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_FAILOVER_BEGIN_EVT:
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case SELF_UP_PEER_UP:
+               switch (evt) {
+               case SELF_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_LEAVING;
+                       break;
+               case PEER_LOST_CONTACT_EVT:
+                       state = SELF_LEAVING_PEER_DOWN;
+                       break;
+               case NODE_SYNCH_BEGIN_EVT:
+                       state = NODE_SYNCHING;
+                       break;
+               case NODE_FAILOVER_BEGIN_EVT:
+                       state = NODE_FAILINGOVER;
+                       break;
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+               case NODE_SYNCH_END_EVT:
+               case NODE_FAILOVER_END_EVT:
+                       break;
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case SELF_DOWN_PEER_LEAVING:
+               switch (evt) {
+               case PEER_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_DOWN;
+                       break;
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+               case SELF_LOST_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_END_EVT:
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_FAILOVER_BEGIN_EVT:
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case SELF_UP_PEER_COMING:
+               switch (evt) {
+               case PEER_ESTABL_CONTACT_EVT:
+                       state = SELF_UP_PEER_UP;
+                       break;
+               case SELF_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_LEAVING;
+                       break;
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_LOST_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_END_EVT:
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_FAILOVER_BEGIN_EVT:
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case SELF_COMING_PEER_UP:
+               switch (evt) {
+               case SELF_ESTABL_CONTACT_EVT:
+                       state = SELF_UP_PEER_UP;
+                       break;
+               case PEER_LOST_CONTACT_EVT:
+                       state = SELF_LEAVING_PEER_DOWN;
+                       break;
+               case SELF_LOST_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_END_EVT:
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_FAILOVER_BEGIN_EVT:
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case SELF_LEAVING_PEER_DOWN:
+               switch (evt) {
+               case SELF_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_DOWN;
+                       break;
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+               case PEER_LOST_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_END_EVT:
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_FAILOVER_BEGIN_EVT:
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case NODE_FAILINGOVER:
+               switch (evt) {
+               case SELF_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_LEAVING;
+                       break;
+               case PEER_LOST_CONTACT_EVT:
+                       state = SELF_LEAVING_PEER_DOWN;
+                       break;
+               case NODE_FAILOVER_END_EVT:
+                       state = SELF_UP_PEER_UP;
+                       break;
+               case NODE_FAILOVER_BEGIN_EVT:
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+                       break;
+               case NODE_SYNCH_BEGIN_EVT:
+               case NODE_SYNCH_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       case NODE_SYNCHING:
+               switch (evt) {
+               case SELF_LOST_CONTACT_EVT:
+                       state = SELF_DOWN_PEER_LEAVING;
+                       break;
+               case PEER_LOST_CONTACT_EVT:
+                       state = SELF_LEAVING_PEER_DOWN;
+                       break;
+               case NODE_SYNCH_END_EVT:
+                       state = SELF_UP_PEER_UP;
+                       break;
+               case NODE_FAILOVER_BEGIN_EVT:
+                       state = NODE_FAILINGOVER;
+                       break;
+               case NODE_SYNCH_BEGIN_EVT:
+               case SELF_ESTABL_CONTACT_EVT:
+               case PEER_ESTABL_CONTACT_EVT:
+                       break;
+               case NODE_FAILOVER_END_EVT:
+               default:
+                       goto illegal_evt;
+               }
+               break;
+       default:
+               pr_err("Unknown node fsm state %x\n", state);
+               break;
+       }
+       n->state = state;
+       return;
+
+illegal_evt:
+       pr_err("Illegal node fsm evt %x in state %x\n", evt, state);
+}
+
+bool tipc_node_filter_pkt(struct tipc_node *n, struct tipc_msg *hdr)
+{
+       int state = n->state;
+
+       if (likely(state == SELF_UP_PEER_UP))
+               return true;
+
+       if (state == SELF_LEAVING_PEER_DOWN)
+               return false;
+
+       if (state == SELF_DOWN_PEER_LEAVING) {
+               if (msg_peer_node_is_up(hdr))
+                       return false;
+       }
+
+       return true;
+}
+
 static void node_established_contact(struct tipc_node *n_ptr)
 {
+       tipc_node_fsm_evt(n_ptr, SELF_ESTABL_CONTACT_EVT);
        n_ptr->action_flags |= TIPC_NOTIFY_NODE_UP;
        n_ptr->bclink.oos_state = 0;
        n_ptr->bclink.acked = tipc_bclink_get_last_sent(n_ptr->net);
        tipc_bclink_add_node(n_ptr->net, n_ptr->addr);
 }
 
-static void node_lost_contact(struct tipc_node *n_ptr)
+static void node_lost_contact(struct tipc_node *n_ptr,
+                             struct sk_buff_head *inputq)
 {
        char addr_string[16];
        struct tipc_sock_conn *conn, *safe;
+       struct tipc_link *l;
        struct list_head *conns = &n_ptr->conn_sks;
        struct sk_buff *skb;
        struct tipc_net *tn = net_generic(n_ptr->net, tipc_net_id);
@@ -458,21 +856,13 @@ static void node_lost_contact(struct tipc_node *n_ptr)
 
        /* Abort any ongoing link failover */
        for (i = 0; i < MAX_BEARERS; i++) {
-               struct tipc_link *l_ptr = n_ptr->links[i].link;
-               if (!l_ptr)
-                       continue;
-               l_ptr->exec_mode = TIPC_LINK_OPEN;
-               l_ptr->failover_checkpt = 0;
-               l_ptr->failover_pkts = 0;
-               kfree_skb(l_ptr->failover_skb);
-               l_ptr->failover_skb = NULL;
-               tipc_link_reset_fragments(l_ptr);
+               l = n_ptr->links[i].link;
+               if (l)
+                       tipc_link_fsm_evt(l, LINK_FAILOVER_END_EVT);
        }
 
-       n_ptr->action_flags &= ~TIPC_WAIT_OWN_LINKS_DOWN;
-
        /* Prevent re-contact with node until cleanup is done */
-       n_ptr->action_flags |= TIPC_WAIT_PEER_LINKS_DOWN;
+       tipc_node_fsm_evt(n_ptr, SELF_LOST_CONTACT_EVT);
 
        /* Notify publications from this node */
        n_ptr->action_flags |= TIPC_NOTIFY_NODE_DOWN;
@@ -483,10 +873,8 @@ static void node_lost_contact(struct tipc_node *n_ptr)
                                      SHORT_H_SIZE, 0, tn->own_addr,
                                      conn->peer_node, conn->port,
                                      conn->peer_port, TIPC_ERR_NO_NODE);
-               if (likely(skb)) {
-                       skb_queue_tail(n_ptr->inputq, skb);
-                       n_ptr->action_flags |= TIPC_MSG_EVT;
-               }
+               if (likely(skb))
+                       skb_queue_tail(inputq, skb);
                list_del(&conn->list);
                kfree(conn);
        }
@@ -533,27 +921,20 @@ void tipc_node_unlock(struct tipc_node *node)
        u32 flags = node->action_flags;
        u32 link_id = 0;
        struct list_head *publ_list;
-       struct sk_buff_head *inputq = node->inputq;
-       struct sk_buff_head *namedq;
 
-       if (likely(!flags || (flags == TIPC_MSG_EVT))) {
-               node->action_flags = 0;
+       if (likely(!flags)) {
                spin_unlock_bh(&node->lock);
-               if (flags == TIPC_MSG_EVT)
-                       tipc_sk_rcv(net, inputq);
                return;
        }
 
        addr = node->addr;
        link_id = node->link_id;
-       namedq = node->namedq;
        publ_list = &node->publ_list;
 
-       node->action_flags &= ~(TIPC_MSG_EVT |
-                               TIPC_NOTIFY_NODE_DOWN | TIPC_NOTIFY_NODE_UP |
+       node->action_flags &= ~(TIPC_NOTIFY_NODE_DOWN | TIPC_NOTIFY_NODE_UP |
                                TIPC_NOTIFY_LINK_DOWN | TIPC_NOTIFY_LINK_UP |
                                TIPC_WAKEUP_BCAST_USERS | TIPC_BCAST_MSG_EVT |
-                               TIPC_NAMED_MSG_EVT | TIPC_BCAST_RESET);
+                               TIPC_BCAST_RESET);
 
        spin_unlock_bh(&node->lock);
 
@@ -574,17 +955,11 @@ void tipc_node_unlock(struct tipc_node *node)
                tipc_nametbl_withdraw(net, TIPC_LINK_STATE, addr,
                                      link_id, addr);
 
-       if (flags & TIPC_MSG_EVT)
-               tipc_sk_rcv(net, inputq);
-
-       if (flags & TIPC_NAMED_MSG_EVT)
-               tipc_named_rcv(net, namedq);
-
        if (flags & TIPC_BCAST_MSG_EVT)
                tipc_bclink_input(net);
 
        if (flags & TIPC_BCAST_RESET)
-               tipc_link_reset_all(node);
+               tipc_node_reset_links(node);
 }
 
 /* Caller should hold node lock for the passed node */
@@ -661,9 +1036,9 @@ int tipc_node_xmit(struct net *net, struct sk_buff_head *list,
                l = tipc_node_select_link(n, selector, &bearer_id, &maddr);
                if (likely(l))
                        rc = tipc_link_xmit(l, list, &xmitq);
-               if (unlikely(rc == -ENOBUFS))
-                       tipc_link_reset(l);
                tipc_node_unlock(n);
+               if (unlikely(rc == -ENOBUFS))
+                       tipc_node_link_down(n, bearer_id, false);
                tipc_node_put(n);
        }
        if (likely(!rc)) {
@@ -699,6 +1074,201 @@ int tipc_node_xmit_skb(struct net *net, struct sk_buff *skb, u32 dnode,
        return 0;
 }
 
+/**
+ * tipc_node_check_state - check and if necessary update node state
+ * @skb: TIPC packet
+ * @bearer_id: identity of bearer delivering the packet
+ * Returns true if state is ok, otherwise consumes buffer and returns false
+ */
+static bool tipc_node_check_state(struct tipc_node *n, struct sk_buff *skb,
+                                 int bearer_id, struct sk_buff_head *xmitq)
+{
+       struct tipc_msg *hdr = buf_msg(skb);
+       int usr = msg_user(hdr);
+       int mtyp = msg_type(hdr);
+       u16 oseqno = msg_seqno(hdr);
+       u16 iseqno = msg_seqno(msg_get_wrapped(hdr));
+       u16 exp_pkts = msg_msgcnt(hdr);
+       u16 rcv_nxt, syncpt, dlv_nxt;
+       int state = n->state;
+       struct tipc_link *l, *pl = NULL;
+       struct tipc_media_addr *maddr;
+       int i, pb_id;
+
+       l = n->links[bearer_id].link;
+       if (!l)
+               return false;
+       rcv_nxt = l->rcv_nxt;
+
+
+       if (likely((state == SELF_UP_PEER_UP) && (usr != TUNNEL_PROTOCOL)))
+               return true;
+
+       /* Find parallel link, if any */
+       for (i = 0; i < MAX_BEARERS; i++) {
+               if ((i != bearer_id) && n->links[i].link) {
+                       pl = n->links[i].link;
+                       break;
+               }
+       }
+
+       /* Update node accesibility if applicable */
+       if (state == SELF_UP_PEER_COMING) {
+               if (!tipc_link_is_up(l))
+                       return true;
+               if (!msg_peer_link_is_up(hdr))
+                       return true;
+               tipc_node_fsm_evt(n, PEER_ESTABL_CONTACT_EVT);
+       }
+
+       if (state == SELF_DOWN_PEER_LEAVING) {
+               if (msg_peer_node_is_up(hdr))
+                       return false;
+               tipc_node_fsm_evt(n, PEER_LOST_CONTACT_EVT);
+       }
+
+       /* Ignore duplicate packets */
+       if (less(oseqno, rcv_nxt))
+               return true;
+
+       /* Initiate or update failover mode if applicable */
+       if ((usr == TUNNEL_PROTOCOL) && (mtyp == FAILOVER_MSG)) {
+               syncpt = oseqno + exp_pkts - 1;
+               if (pl && tipc_link_is_up(pl)) {
+                       pb_id = pl->bearer_id;
+                       __tipc_node_link_down(n, &pb_id, xmitq, &maddr);
+                       tipc_skb_queue_splice_tail_init(pl->inputq, l->inputq);
+               }
+               /* If pkts arrive out of order, use lowest calculated syncpt */
+               if (less(syncpt, n->sync_point))
+                       n->sync_point = syncpt;
+       }
+
+       /* Open parallel link when tunnel link reaches synch point */
+       if ((n->state == NODE_FAILINGOVER) && !tipc_link_is_failingover(l)) {
+               if (!more(rcv_nxt, n->sync_point))
+                       return true;
+               tipc_node_fsm_evt(n, NODE_FAILOVER_END_EVT);
+               if (pl)
+                       tipc_link_fsm_evt(pl, LINK_FAILOVER_END_EVT);
+               return true;
+       }
+
+       /* Initiate or update synch mode if applicable */
+       if ((usr == TUNNEL_PROTOCOL) && (mtyp == SYNCH_MSG)) {
+               syncpt = iseqno + exp_pkts - 1;
+               if (!tipc_link_is_up(l)) {
+                       tipc_link_fsm_evt(l, LINK_ESTABLISH_EVT);
+                       __tipc_node_link_up(n, bearer_id, xmitq);
+               }
+               if (n->state == SELF_UP_PEER_UP) {
+                       n->sync_point = syncpt;
+                       tipc_link_fsm_evt(l, LINK_SYNCH_BEGIN_EVT);
+                       tipc_node_fsm_evt(n, NODE_SYNCH_BEGIN_EVT);
+               }
+               if (less(syncpt, n->sync_point))
+                       n->sync_point = syncpt;
+       }
+
+       /* Open tunnel link when parallel link reaches synch point */
+       if ((n->state == NODE_SYNCHING) && tipc_link_is_synching(l)) {
+               if (pl)
+                       dlv_nxt = mod(pl->rcv_nxt - skb_queue_len(pl->inputq));
+               if (!pl || more(dlv_nxt, n->sync_point)) {
+                       tipc_link_fsm_evt(l, LINK_SYNCH_END_EVT);
+                       tipc_node_fsm_evt(n, NODE_SYNCH_END_EVT);
+                       return true;
+               }
+               if ((usr == TUNNEL_PROTOCOL) && (mtyp == SYNCH_MSG))
+                       return true;
+               if (usr == LINK_PROTOCOL)
+                       return true;
+               return false;
+       }
+       return true;
+}
+
+/**
+ * tipc_rcv - process TIPC packets/messages arriving from off-node
+ * @net: the applicable net namespace
+ * @skb: TIPC packet
+ * @bearer: pointer to bearer message arrived on
+ *
+ * Invoked with no locks held. Bearer pointer must point to a valid bearer
+ * structure (i.e. cannot be NULL), but bearer can be inactive.
+ */
+void tipc_rcv(struct net *net, struct sk_buff *skb, struct tipc_bearer *b)
+{
+       struct sk_buff_head xmitq;
+       struct tipc_node *n;
+       struct tipc_msg *hdr = buf_msg(skb);
+       int usr = msg_user(hdr);
+       int bearer_id = b->identity;
+       struct tipc_link_entry *le;
+       int rc = 0;
+
+       __skb_queue_head_init(&xmitq);
+
+       /* Ensure message is well-formed */
+       if (unlikely(!tipc_msg_validate(skb)))
+               goto discard;
+
+       /* Handle arrival of a non-unicast link packet */
+       if (unlikely(msg_non_seq(hdr))) {
+               if (usr ==  LINK_CONFIG)
+                       tipc_disc_rcv(net, skb, b);
+               else
+                       tipc_bclink_rcv(net, skb);
+               return;
+       }
+
+       /* Locate neighboring node that sent packet */
+       n = tipc_node_find(net, msg_prevnode(hdr));
+       if (unlikely(!n))
+               goto discard;
+       le = &n->links[bearer_id];
+
+       tipc_node_lock(n);
+
+       /* Is reception permitted at the moment ? */
+       if (!tipc_node_filter_pkt(n, hdr))
+               goto unlock;
+
+       if (unlikely(msg_user(hdr) == LINK_PROTOCOL))
+               tipc_bclink_sync_state(n, hdr);
+
+       /* Release acked broadcast packets */
+       if (unlikely(n->bclink.acked != msg_bcast_ack(hdr)))
+               tipc_bclink_acknowledge(n, msg_bcast_ack(hdr));
+
+       /* Check and if necessary update node state */
+       if (likely(tipc_node_check_state(n, skb, bearer_id, &xmitq))) {
+               rc = tipc_link_rcv(le->link, skb, &xmitq);
+               skb = NULL;
+       }
+unlock:
+       tipc_node_unlock(n);
+
+       if (unlikely(rc & TIPC_LINK_UP_EVT))
+               tipc_node_link_up(n, bearer_id, &xmitq);
+
+       if (unlikely(rc & TIPC_LINK_DOWN_EVT))
+               tipc_node_link_down(n, bearer_id, false);
+
+       if (unlikely(!skb_queue_empty(&n->bclink.namedq)))
+               tipc_named_rcv(net, &n->bclink.namedq);
+
+       if (!skb_queue_empty(&le->inputq))
+               tipc_sk_rcv(net, &le->inputq);
+
+       if (!skb_queue_empty(&xmitq))
+               tipc_bearer_xmit(net, bearer_id, &xmitq, &le->maddr);
+
+       tipc_node_put(n);
+discard:
+       kfree_skb(skb);
+}
+
 int tipc_nl_node_dump(struct sk_buff *skb, struct netlink_callback *cb)
 {
        int err;