tipc: clean up socket layer message reception
authorJon Paul Maloy <jon.maloy@ericsson.com>
Wed, 22 Jul 2015 14:11:20 +0000 (10:11 -0400)
committerDavid S. Miller <davem@davemloft.net>
Sun, 26 Jul 2015 23:31:50 +0000 (16:31 -0700)
When a message is received in a socket, one of the call chains
tipc_sk_rcv()->tipc_sk_enqueue()->filter_rcv()(->tipc_sk_proto_rcv())
or
tipc_sk_backlog_rcv()->filter_rcv()(->tipc_sk_proto_rcv())
are followed. At each of these levels we may encounter situations
where the message may need to be rejected, or a new message
produced for transfer back to the sender. Despite recent
improvements, the current code for doing this is perceived
as awkward and hard to follow.

Leveraging the two previous commits in this series, we now
introduce a more uniform handling of such situations. We
let each of the functions in the chain itself produce/reverse
the message to be returned to the sender, but also perform the
actual forwarding. This simplifies the necessary logics within
each function.

Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/msg.c
net/tipc/msg.h
net/tipc/socket.c
net/tipc/socket.h

index b6cc58e..562c926 100644 (file)
@@ -520,17 +520,15 @@ exit:
 /**
  * tipc_msg_lookup_dest(): try to find new destination for named message
  * @skb: the buffer containing the message.
- * @dnode: return value: next-hop node, if destination found
- * @err: return value: error code to use, if message to be rejected
+ * @err: error code to be used by caller if lookup fails
  * Does not consume buffer
  * Returns true if a destination is found, false otherwise
  */
-bool tipc_msg_lookup_dest(struct net *net, struct sk_buff *skb,
-                         u32 *dnode, int *err)
+bool tipc_msg_lookup_dest(struct net *net, struct sk_buff *skb, int *err)
 {
        struct tipc_msg *msg = buf_msg(skb);
-       u32 dport;
-       u32 own_addr = tipc_own_addr(net);
+       u32 dport, dnode;
+       u32 onode = tipc_own_addr(net);
 
        if (!msg_isdata(msg))
                return false;
@@ -543,15 +541,15 @@ bool tipc_msg_lookup_dest(struct net *net, struct sk_buff *skb,
                return false;
        if (msg_reroute_cnt(msg))
                return false;
-       *dnode = addr_domain(net, msg_lookup_scope(msg));
+       dnode = addr_domain(net, msg_lookup_scope(msg));
        dport = tipc_nametbl_translate(net, msg_nametype(msg),
-                                      msg_nameinst(msg), dnode);
+                                      msg_nameinst(msg), &dnode);
        if (!dport)
                return false;
        msg_incr_reroute_cnt(msg);
-       if (*dnode != own_addr)
-               msg_set_prevnode(msg, own_addr);
-       msg_set_destnode(msg, *dnode);
+       if (dnode != onode)
+               msg_set_prevnode(msg, onode);
+       msg_set_destnode(msg, dnode);
        msg_set_destport(msg, dport);
        *err = TIPC_OK;
        return true;
index d0834bc..234fb05 100644 (file)
@@ -798,8 +798,7 @@ bool tipc_msg_make_bundle(struct sk_buff **skb, struct tipc_msg *msg,
 bool tipc_msg_extract(struct sk_buff *skb, struct sk_buff **iskb, int *pos);
 int tipc_msg_build(struct tipc_msg *mhdr, struct msghdr *m,
                   int offset, int dsz, int mtu, struct sk_buff_head *list);
-bool tipc_msg_lookup_dest(struct net *net, struct sk_buff *skb, u32 *dnode,
-                         int *err);
+bool tipc_msg_lookup_dest(struct net *net, struct sk_buff *skb, int *err);
 struct sk_buff *tipc_msg_reassemble(struct sk_buff_head *list);
 
 static inline u16 buf_seqno(struct sk_buff *skb)
index 71d88ad..1060d52 100644 (file)
@@ -1520,82 +1520,81 @@ static void tipc_data_ready(struct sock *sk)
  * @tsk: TIPC socket
  * @skb: pointer to message buffer. Set to NULL if buffer is consumed
  *
- * Returns 0 (TIPC_OK) if everything ok, -TIPC_ERR_NO_PORT otherwise
+ * Returns true if everything ok, false otherwise
  */
-static int filter_connect(struct tipc_sock *tsk, struct sk_buff **skb)
+static bool filter_connect(struct tipc_sock *tsk, struct sk_buff *skb)
 {
        struct sock *sk = &tsk->sk;
        struct net *net = sock_net(sk);
        struct socket *sock = sk->sk_socket;
-       struct tipc_msg *msg = buf_msg(*skb);
-       int retval = -TIPC_ERR_NO_PORT;
+       struct tipc_msg *hdr = buf_msg(skb);
 
-       if (msg_mcast(msg))
-               return retval;
+       if (unlikely(msg_mcast(hdr)))
+               return false;
 
        switch ((int)sock->state) {
        case SS_CONNECTED:
+
                /* Accept only connection-based messages sent by peer */
-               if (tsk_peer_msg(tsk, msg)) {
-                       if (unlikely(msg_errcode(msg))) {
-                               sock->state = SS_DISCONNECTING;
-                               tsk->connected = 0;
-                               /* let timer expire on it's own */
-                               tipc_node_remove_conn(net, tsk_peer_node(tsk),
-                                                     tsk->portid);
-                       }
-                       retval = TIPC_OK;
+               if (unlikely(!tsk_peer_msg(tsk, hdr)))
+                       return false;
+
+               if (unlikely(msg_errcode(hdr))) {
+                       sock->state = SS_DISCONNECTING;
+                       tsk->connected = 0;
+                       /* Let timer expire on it's own */
+                       tipc_node_remove_conn(net, tsk_peer_node(tsk),
+                                             tsk->portid);
                }
-               break;
+               return true;
+
        case SS_CONNECTING:
-               /* Accept only ACK or NACK message */
 
-               if (unlikely(!msg_connected(msg)))
-                       break;
+               /* Accept only ACK or NACK message */
+               if (unlikely(!msg_connected(hdr)))
+                       return false;
 
-               if (unlikely(msg_errcode(msg))) {
+               if (unlikely(msg_errcode(hdr))) {
                        sock->state = SS_DISCONNECTING;
                        sk->sk_err = ECONNREFUSED;
-                       retval = TIPC_OK;
-                       break;
+                       return true;
                }
 
-               if (unlikely(msg_importance(msg) > TIPC_CRITICAL_IMPORTANCE)) {
+               if (unlikely(!msg_isdata(hdr))) {
                        sock->state = SS_DISCONNECTING;
                        sk->sk_err = EINVAL;
-                       retval = TIPC_OK;
-                       break;
+                       return true;
                }
 
-               tipc_sk_finish_conn(tsk, msg_origport(msg), msg_orignode(msg));
-               msg_set_importance(&tsk->phdr, msg_importance(msg));
+               tipc_sk_finish_conn(tsk, msg_origport(hdr), msg_orignode(hdr));
+               msg_set_importance(&tsk->phdr, msg_importance(hdr));
                sock->state = SS_CONNECTED;
 
-               /* If an incoming message is an 'ACK-', it should be
-                * discarded here because it doesn't contain useful
-                * data. In addition, we should try to wake up
-                * connect() routine if sleeping.
-                */
-               if (msg_data_sz(msg) == 0) {
-                       kfree_skb(*skb);
-                       *skb = NULL;
-                       if (waitqueue_active(sk_sleep(sk)))
-                               wake_up_interruptible(sk_sleep(sk));
-               }
-               retval = TIPC_OK;
-               break;
+               /* If 'ACK+' message, add to socket receive queue */
+               if (msg_data_sz(hdr))
+                       return true;
+
+               /* If empty 'ACK-' message, wake up sleeping connect() */
+               if (waitqueue_active(sk_sleep(sk)))
+                       wake_up_interruptible(sk_sleep(sk));
+
+               /* 'ACK-' message is neither accepted nor rejected: */
+               msg_set_dest_droppable(hdr, 1);
+               return false;
+
        case SS_LISTENING:
        case SS_UNCONNECTED:
+
                /* Accept only SYN message */
-               if (!msg_connected(msg) && !(msg_errcode(msg)))
-                       retval = TIPC_OK;
+               if (!msg_connected(hdr) && !(msg_errcode(hdr)))
+                       return true;
                break;
        case SS_DISCONNECTING:
                break;
        default:
                pr_err("Unknown socket state %u\n", sock->state);
        }
-       return retval;
+       return false;
 }
 
 /**
@@ -1630,61 +1629,70 @@ static unsigned int rcvbuf_limit(struct sock *sk, struct sk_buff *buf)
 /**
  * filter_rcv - validate incoming message
  * @sk: socket
- * @skb: pointer to message. Set to NULL if buffer is consumed.
+ * @skb: pointer to message.
  *
  * Enqueues message on receive queue if acceptable; optionally handles
  * disconnect indication for a connected socket.
  *
  * Called with socket lock already taken
  *
- * Returns 0 (TIPC_OK) if message was ok, -TIPC error code if rejected
+ * Returns true if message was added to socket receive queue, otherwise false
  */
-static int filter_rcv(struct sock *sk, struct sk_buff **skb)
+static bool filter_rcv(struct sock *sk, struct sk_buff *skb)
 {
        struct socket *sock = sk->sk_socket;
        struct tipc_sock *tsk = tipc_sk(sk);
-       struct tipc_msg *msg = buf_msg(*skb);
-       unsigned int limit = rcvbuf_limit(sk, *skb);
-       int rc = TIPC_OK;
+       struct tipc_msg *hdr = buf_msg(skb);
+       unsigned int limit = rcvbuf_limit(sk, skb);
+       int err = TIPC_OK;
+       int usr = msg_user(hdr);
 
-       if (unlikely(msg_user(msg) == CONN_MANAGER)) {
-               tipc_sk_proto_rcv(tsk, *skb);
-               return TIPC_OK;
+       if (unlikely(msg_user(hdr) == CONN_MANAGER)) {
+               tipc_sk_proto_rcv(tsk, skb);
+               return false;
        }
 
-       if (unlikely(msg_user(msg) == SOCK_WAKEUP)) {
-               kfree_skb(*skb);
+       if (unlikely(usr == SOCK_WAKEUP)) {
+               kfree_skb(skb);
                tsk->link_cong = 0;
                sk->sk_write_space(sk);
-               *skb = NULL;
-               return TIPC_OK;
+               return false;
        }
 
-       /* Reject message if it is wrong sort of message for socket */
-       if (msg_type(msg) > TIPC_DIRECT_MSG)
-               return -TIPC_ERR_NO_PORT;
+       /* Drop if illegal message type */
+       if (unlikely(msg_type(hdr) > TIPC_DIRECT_MSG)) {
+               kfree_skb(skb);
+               return false;
+       }
 
-       if (sock->state == SS_READY) {
-               if (msg_connected(msg))
-                       return -TIPC_ERR_NO_PORT;
-       } else {
-               rc = filter_connect(tsk, skb);
-               if (rc != TIPC_OK || !*skb)
-                       return rc;
+       /* Reject if wrong message type for current socket state */
+       if (unlikely(sock->state == SS_READY)) {
+               if (msg_connected(hdr)) {
+                       err = TIPC_ERR_NO_PORT;
+                       goto reject;
+               }
+       } else if (unlikely(!filter_connect(tsk, skb))) {
+               err = TIPC_ERR_NO_PORT;
+               goto reject;
        }
 
        /* Reject message if there isn't room to queue it */
-       if (sk_rmem_alloc_get(sk) + (*skb)->truesize >= limit)
-               return -TIPC_ERR_OVERLOAD;
+       if (unlikely(sk_rmem_alloc_get(sk) + skb->truesize >= limit)) {
+               err = TIPC_ERR_OVERLOAD;
+               goto reject;
+       }
 
        /* Enqueue message */
-       TIPC_SKB_CB(*skb)->handle = NULL;
-       __skb_queue_tail(&sk->sk_receive_queue, *skb);
-       skb_set_owner_r(*skb, sk);
+       TIPC_SKB_CB(skb)->handle = NULL;
+       __skb_queue_tail(&sk->sk_receive_queue, skb);
+       skb_set_owner_r(skb, sk);
 
        sk->sk_data_ready(sk);
-       *skb = NULL;
-       return TIPC_OK;
+       return true;
+
+reject:
+       tipc_sk_respond(sk, skb, err);
+       return false;
 }
 
 /**
@@ -1698,22 +1706,10 @@ static int filter_rcv(struct sock *sk, struct sk_buff **skb)
  */
 static int tipc_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 {
-       int err;
-       atomic_t *dcnt;
-       u32 dnode = msg_prevnode(buf_msg(skb));
-       struct tipc_sock *tsk = tipc_sk(sk);
-       struct net *net = sock_net(sk);
-       uint truesize = skb->truesize;
+       unsigned int truesize = skb->truesize;
 
-       err = filter_rcv(sk, &skb);
-       if (likely(!skb)) {
-               dcnt = &tsk->dupl_rcvcnt;
-               if (atomic_read(dcnt) < TIPC_CONN_OVERLOAD_LIMIT)
-                       atomic_add(truesize, dcnt);
-               return 0;
-       }
-       if (!err || tipc_msg_reverse(tsk_own_node(tsk), &skb, -err))
-               tipc_node_xmit_skb(net, skb, dnode, tsk->portid);
+       if (likely(filter_rcv(sk, skb)))
+               atomic_add(truesize, &tipc_sk(sk)->dupl_rcvcnt);
        return 0;
 }
 
@@ -1723,45 +1719,43 @@ static int tipc_backlog_rcv(struct sock *sk, struct sk_buff *skb)
  * @inputq: list of incoming buffers with potentially different destinations
  * @sk: socket where the buffers should be enqueued
  * @dport: port number for the socket
- * @_skb: returned buffer to be forwarded or rejected, if applicable
  *
  * Caller must hold socket lock
- *
- * Returns TIPC_OK if all buffers enqueued, otherwise -TIPC_ERR_OVERLOAD
- * or -TIPC_ERR_NO_PORT
  */
-static int tipc_sk_enqueue(struct sk_buff_head *inputq, struct sock *sk,
-                          u32 dport, struct sk_buff **_skb)
+static void tipc_sk_enqueue(struct sk_buff_head *inputq, struct sock *sk,
+                           u32 dport)
 {
        unsigned int lim;
        atomic_t *dcnt;
-       int err;
        struct sk_buff *skb;
        unsigned long time_limit = jiffies + 2;
 
        while (skb_queue_len(inputq)) {
                if (unlikely(time_after_eq(jiffies, time_limit)))
-                       return TIPC_OK;
+                       return;
+
                skb = tipc_skb_dequeue(inputq, dport);
                if (unlikely(!skb))
-                       return TIPC_OK;
+                       return;
+
+               /* Add message directly to receive queue if possible */
                if (!sock_owned_by_user(sk)) {
-                       err = filter_rcv(sk, &skb);
-                       if (likely(!skb))
-                               continue;
-                       *_skb = skb;
-                       return err;
+                       filter_rcv(sk, skb);
+                       continue;
                }
+
+               /* Try backlog, compensating for double-counted bytes */
                dcnt = &tipc_sk(sk)->dupl_rcvcnt;
                if (sk->sk_backlog.len)
                        atomic_set(dcnt, 0);
                lim = rcvbuf_limit(sk, skb) + atomic_read(dcnt);
                if (likely(!sk_add_backlog(sk, skb, lim)))
                        continue;
-               *_skb = skb;
-               return -TIPC_ERR_OVERLOAD;
+
+               /* Overload => reject message back to sender */
+               tipc_sk_respond(sk, skb, TIPC_ERR_OVERLOAD);
+               break;
        }
-       return TIPC_OK;
 }
 
 /**
@@ -1769,51 +1763,46 @@ static int tipc_sk_enqueue(struct sk_buff_head *inputq, struct sock *sk,
  * @inputq: buffer list containing the buffers
  * Consumes all buffers in list until inputq is empty
  * Note: may be called in multiple threads referring to the same queue
- * Returns 0 if last buffer was accepted, otherwise -EHOSTUNREACH
- * Only node local calls check the return value, sending single-buffer queues
  */
-int tipc_sk_rcv(struct net *net, struct sk_buff_head *inputq)
+void tipc_sk_rcv(struct net *net, struct sk_buff_head *inputq)
 {
        u32 dnode, dport = 0;
        int err;
-       struct sk_buff *skb;
        struct tipc_sock *tsk;
-       struct tipc_net *tn;
        struct sock *sk;
+       struct sk_buff *skb;
 
        while (skb_queue_len(inputq)) {
-               err = -TIPC_ERR_NO_PORT;
-               skb = NULL;
                dport = tipc_skb_peek_port(inputq, dport);
                tsk = tipc_sk_lookup(net, dport);
+
                if (likely(tsk)) {
                        sk = &tsk->sk;
                        if (likely(spin_trylock_bh(&sk->sk_lock.slock))) {
-                               err = tipc_sk_enqueue(inputq, sk, dport, &skb);
+                               tipc_sk_enqueue(inputq, sk, dport);
                                spin_unlock_bh(&sk->sk_lock.slock);
-                               dport = 0;
                        }
                        sock_put(sk);
-               } else {
-                       skb = tipc_skb_dequeue(inputq, dport);
-               }
-               if (likely(!skb))
                        continue;
-               if (tipc_msg_lookup_dest(net, skb, &dnode, &err))
-                       goto xmit;
-               if (!err) {
-                       dnode = msg_destnode(buf_msg(skb));
-                       goto xmit;
-               } else {
-                       dnode = msg_prevnode(buf_msg(skb));
                }
-               tn = net_generic(net, tipc_net_id);
-               if (!tipc_msg_reverse(tn->own_addr, &skb, -err))
+
+               /* No destination socket => dequeue skb if still there */
+               skb = tipc_skb_dequeue(inputq, dport);
+               if (!skb)
+                       return;
+
+               /* Try secondary lookup if unresolved named message */
+               err = TIPC_ERR_NO_PORT;
+               if (tipc_msg_lookup_dest(net, skb, &err))
+                       goto xmit;
+
+               /* Prepare for message rejection */
+               if (!tipc_msg_reverse(tipc_own_addr(net), &skb, err))
                        continue;
 xmit:
+               dnode = msg_destnode(buf_msg(skb));
                tipc_node_xmit_skb(net, skb, dnode, dport);
        }
-       return err ? -EHOSTUNREACH : 0;
 }
 
 static int tipc_wait_for_connect(struct socket *sock, long *timeo_p)
@@ -2082,7 +2071,10 @@ static int tipc_shutdown(struct socket *sock, int how)
        struct net *net = sock_net(sk);
        struct tipc_sock *tsk = tipc_sk(sk);
        struct sk_buff *skb;
-       u32 dnode;
+       u32 dnode = tsk_peer_node(tsk);
+       u32 dport = tsk_peer_port(tsk);
+       u32 onode = tipc_own_addr(net);
+       u32 oport = tsk->portid;
        int res;
 
        if (how != SHUT_RDWR)
@@ -2108,9 +2100,8 @@ restart:
                } else {
                        skb = tipc_msg_create(TIPC_CRITICAL_IMPORTANCE,
                                              TIPC_CONN_MSG, SHORT_H_SIZE,
-                                             0, dnode, tsk_own_node(tsk),
-                                             tsk_peer_port(tsk),
-                                             tsk->portid, TIPC_CONN_SHUTDOWN);
+                                             0, dnode, onode, dport, oport,
+                                             TIPC_CONN_SHUTDOWN);
                        tipc_node_xmit_skb(net, skb, dnode, tsk->portid);
                }
                tsk->connected = 0;
index bf65513..4241f22 100644 (file)
@@ -44,7 +44,7 @@
                                  SKB_TRUESIZE(TIPC_MAX_USER_MSG_SIZE))
 int tipc_socket_init(void);
 void tipc_socket_stop(void);
-int tipc_sk_rcv(struct net *net, struct sk_buff_head *inputq);
+void tipc_sk_rcv(struct net *net, struct sk_buff_head *inputq);
 void tipc_sk_mcast_rcv(struct net *net, struct sk_buff_head *arrvq,
                       struct sk_buff_head *inputq);
 void tipc_sk_reinit(struct net *net);