rxrpc: Improve skb tracing
authorDavid Howells <dhowells@redhat.com>
Sat, 17 Sep 2016 09:49:14 +0000 (10:49 +0100)
committerDavid Howells <dhowells@redhat.com>
Sat, 17 Sep 2016 10:24:04 +0000 (11:24 +0100)
Improve sk_buff tracing within AF_RXRPC by the following means:

 (1) Use an enum to note the event type rather than plain integers and use
     an array of event names rather than a big multi ?: list.

 (2) Distinguish Rx from Tx packets and account them separately.  This
     requires the call phase to be tracked so that we know what we might
     find in rxtx_buffer[].

 (3) Add a parameter to rxrpc_{new,see,get,free}_skb() to indicate the
     event type.

 (4) A pair of 'rotate' events are added to indicate packets that are about
     to be rotated out of the Rx and Tx windows.

 (5) A pair of 'lost' events are added, along with rxrpc_lose_skb() for
     packet loss injection recording.

Signed-off-by: David Howells <dhowells@redhat.com>
14 files changed:
include/trace/events/rxrpc.h
net/rxrpc/af_rxrpc.c
net/rxrpc/ar-internal.h
net/rxrpc/call_event.c
net/rxrpc/call_object.c
net/rxrpc/conn_event.c
net/rxrpc/input.c
net/rxrpc/local_event.c
net/rxrpc/misc.c
net/rxrpc/output.c
net/rxrpc/peer_event.c
net/rxrpc/recvmsg.c
net/rxrpc/sendmsg.c
net/rxrpc/skbuff.c

index 58732202e9f00d857117a45fb9dcfa33566af6d8..75a5d8bf50e16c3914772d852d03d3c8ba0e931c 100644 (file)
@@ -107,14 +107,14 @@ TRACE_EVENT(rxrpc_call,
            );
 
 TRACE_EVENT(rxrpc_skb,
-           TP_PROTO(struct sk_buff *skb, int op, int usage, int mod_count,
-                    const void *where),
+           TP_PROTO(struct sk_buff *skb, enum rxrpc_skb_trace op,
+                    int usage, int mod_count, const void *where),
 
            TP_ARGS(skb, op, usage, mod_count, where),
 
            TP_STRUCT__entry(
                    __field(struct sk_buff *,           skb             )
-                   __field(int,                        op              )
+                   __field(enum rxrpc_skb_trace,       op              )
                    __field(int,                        usage           )
                    __field(int,                        mod_count       )
                    __field(const void *,               where           )
@@ -130,11 +130,7 @@ TRACE_EVENT(rxrpc_skb,
 
            TP_printk("s=%p %s u=%d m=%d p=%pSR",
                      __entry->skb,
-                     (__entry->op == 0 ? "NEW" :
-                      __entry->op == 1 ? "SEE" :
-                      __entry->op == 2 ? "GET" :
-                      __entry->op == 3 ? "FRE" :
-                      "PUR"),
+                     rxrpc_skb_traces[__entry->op],
                      __entry->usage,
                      __entry->mod_count,
                      __entry->where)
index 09f81befc705661c22dae63ff036b310cd21589b..8dbf7bed2cc453f2b00cf058af081af3966d4bcb 100644 (file)
@@ -45,7 +45,7 @@ u32 rxrpc_epoch;
 atomic_t rxrpc_debug_id;
 
 /* count of skbs currently in use */
-atomic_t rxrpc_n_skbs;
+atomic_t rxrpc_n_tx_skbs, rxrpc_n_rx_skbs;
 
 struct workqueue_struct *rxrpc_workqueue;
 
@@ -867,7 +867,8 @@ static void __exit af_rxrpc_exit(void)
        proto_unregister(&rxrpc_proto);
        rxrpc_destroy_all_calls();
        rxrpc_destroy_all_connections();
-       ASSERTCMP(atomic_read(&rxrpc_n_skbs), ==, 0);
+       ASSERTCMP(atomic_read(&rxrpc_n_tx_skbs), ==, 0);
+       ASSERTCMP(atomic_read(&rxrpc_n_rx_skbs), ==, 0);
        rxrpc_destroy_all_locals();
 
        remove_proc_entry("rxrpc_conns", init_net.proc_net);
index a17341d2df3da0477fb07acad3c9449c4643fcfd..034f525f22358deb8d2c020fa3463a0cba28fe65 100644 (file)
@@ -520,6 +520,7 @@ struct rxrpc_call {
        rxrpc_seq_t             rx_expect_next; /* Expected next packet sequence number */
        u8                      rx_winsize;     /* Size of Rx window */
        u8                      tx_winsize;     /* Maximum size of Tx window */
+       bool                    tx_phase;       /* T if transmission phase, F if receive phase */
        u8                      nr_jumbo_bad;   /* Number of jumbo dups/exceeds-windows */
 
        /* receive-phase ACK management */
@@ -534,6 +535,27 @@ struct rxrpc_call {
        rxrpc_serial_t          acks_latest;    /* serial number of latest ACK received */
 };
 
+enum rxrpc_skb_trace {
+       rxrpc_skb_rx_cleaned,
+       rxrpc_skb_rx_freed,
+       rxrpc_skb_rx_got,
+       rxrpc_skb_rx_lost,
+       rxrpc_skb_rx_received,
+       rxrpc_skb_rx_rotated,
+       rxrpc_skb_rx_purged,
+       rxrpc_skb_rx_seen,
+       rxrpc_skb_tx_cleaned,
+       rxrpc_skb_tx_freed,
+       rxrpc_skb_tx_got,
+       rxrpc_skb_tx_lost,
+       rxrpc_skb_tx_new,
+       rxrpc_skb_tx_rotated,
+       rxrpc_skb_tx_seen,
+       rxrpc_skb__nr_trace
+};
+
+extern const char rxrpc_skb_traces[rxrpc_skb__nr_trace][7];
+
 enum rxrpc_conn_trace {
        rxrpc_conn_new_client,
        rxrpc_conn_new_service,
@@ -642,7 +664,7 @@ extern const char *rxrpc_acks(u8 reason);
 /*
  * af_rxrpc.c
  */
-extern atomic_t rxrpc_n_skbs;
+extern atomic_t rxrpc_n_tx_skbs, rxrpc_n_rx_skbs;
 extern u32 rxrpc_epoch;
 extern atomic_t rxrpc_debug_id;
 extern struct workqueue_struct *rxrpc_workqueue;
@@ -1000,10 +1022,11 @@ int rxrpc_do_sendmsg(struct rxrpc_sock *, struct msghdr *, size_t);
  */
 void rxrpc_kernel_data_consumed(struct rxrpc_call *, struct sk_buff *);
 void rxrpc_packet_destructor(struct sk_buff *);
-void rxrpc_new_skb(struct sk_buff *);
-void rxrpc_see_skb(struct sk_buff *);
-void rxrpc_get_skb(struct sk_buff *);
-void rxrpc_free_skb(struct sk_buff *);
+void rxrpc_new_skb(struct sk_buff *, enum rxrpc_skb_trace);
+void rxrpc_see_skb(struct sk_buff *, enum rxrpc_skb_trace);
+void rxrpc_get_skb(struct sk_buff *, enum rxrpc_skb_trace);
+void rxrpc_free_skb(struct sk_buff *, enum rxrpc_skb_trace);
+void rxrpc_lose_skb(struct sk_buff *, enum rxrpc_skb_trace);
 void rxrpc_purge_queue(struct sk_buff_head *);
 
 /*
index f0cabc48a1b7ae11fd99d90dd4789b4811d32c21..7d1b99824ed9823b09a510dce6256b997b30e645 100644 (file)
@@ -170,7 +170,7 @@ static void rxrpc_resend(struct rxrpc_call *call)
                        continue;
 
                skb = call->rxtx_buffer[ix];
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
                sp = rxrpc_skb(skb);
 
                if (annotation == RXRPC_TX_ANNO_UNACK) {
@@ -199,7 +199,7 @@ static void rxrpc_resend(struct rxrpc_call *call)
                        continue;
 
                skb = call->rxtx_buffer[ix];
-               rxrpc_get_skb(skb);
+               rxrpc_get_skb(skb, rxrpc_skb_tx_got);
                spin_unlock_bh(&call->lock);
                sp = rxrpc_skb(skb);
 
@@ -211,7 +211,7 @@ static void rxrpc_resend(struct rxrpc_call *call)
 
                if (rxrpc_send_data_packet(call->conn, skb) < 0) {
                        call->resend_at = now + 2;
-                       rxrpc_free_skb(skb);
+                       rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
                        return;
                }
 
@@ -219,7 +219,7 @@ static void rxrpc_resend(struct rxrpc_call *call)
                        rxrpc_expose_client_call(call);
                sp->resend_at = now + rxrpc_resend_timeout;
 
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
                spin_lock_bh(&call->lock);
 
                /* We need to clear the retransmit state, but there are two
index 54f30482a7fdcf2ec40c4c9206c785a685a0468c..f50a6094e198a8f252da062d117bdb6dd51ff0a8 100644 (file)
@@ -182,6 +182,7 @@ static struct rxrpc_call *rxrpc_alloc_client_call(struct sockaddr_rxrpc *srx,
                return ERR_PTR(-ENOMEM);
        call->state = RXRPC_CALL_CLIENT_AWAIT_CONN;
        call->service_id = srx->srx_service;
+       call->tx_phase = true;
 
        _leave(" = %p", call);
        return call;
@@ -458,7 +459,9 @@ void rxrpc_release_call(struct rxrpc_sock *rx, struct rxrpc_call *call)
                rxrpc_disconnect_call(call);
 
        for (i = 0; i < RXRPC_RXTX_BUFF_SIZE; i++) {
-               rxrpc_free_skb(call->rxtx_buffer[i]);
+               rxrpc_free_skb(call->rxtx_buffer[i],
+                              (call->tx_phase ? rxrpc_skb_tx_cleaned :
+                               rxrpc_skb_rx_cleaned));
                call->rxtx_buffer[i] = NULL;
        }
 
@@ -552,9 +555,11 @@ void rxrpc_cleanup_call(struct rxrpc_call *call)
 
        /* Clean up the Rx/Tx buffer */
        for (i = 0; i < RXRPC_RXTX_BUFF_SIZE; i++)
-               rxrpc_free_skb(call->rxtx_buffer[i]);
+               rxrpc_free_skb(call->rxtx_buffer[i],
+                              (call->tx_phase ? rxrpc_skb_tx_cleaned :
+                               rxrpc_skb_rx_cleaned));
 
-       rxrpc_free_skb(call->tx_pending);
+       rxrpc_free_skb(call->tx_pending, rxrpc_skb_tx_cleaned);
 
        call_rcu(&call->rcu, rxrpc_rcu_destroy_call);
 }
index 9b19c51831aa1054b59dd8109bd53cad470a72b0..75a15a4c74c313c39901ba2c95257dcc7e9f6748 100644 (file)
@@ -388,7 +388,7 @@ void rxrpc_process_connection(struct work_struct *work)
        /* go through the conn-level event packets, releasing the ref on this
         * connection that each one has when we've finished with it */
        while ((skb = skb_dequeue(&conn->rx_queue))) {
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
                ret = rxrpc_process_event(conn, skb, &abort_code);
                switch (ret) {
                case -EPROTO:
@@ -399,7 +399,7 @@ void rxrpc_process_connection(struct work_struct *work)
                        goto requeue_and_leave;
                case -ECONNABORTED:
                default:
-                       rxrpc_free_skb(skb);
+                       rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
                        break;
                }
        }
@@ -416,7 +416,7 @@ requeue_and_leave:
 protocol_error:
        if (rxrpc_abort_connection(conn, -ret, abort_code) < 0)
                goto requeue_and_leave;
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
        _leave(" [EPROTO]");
        goto out;
 }
index b690220533c60ef2c1e540536c6e4b804a8cdf1f..84bb16d47b85f40594ca15e5acebc2bb66e74a7b 100644 (file)
@@ -50,7 +50,7 @@ static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to)
                call->tx_hard_ack++;
                ix = call->tx_hard_ack & RXRPC_RXTX_BUFF_MASK;
                skb = call->rxtx_buffer[ix];
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_tx_rotated);
                call->rxtx_buffer[ix] = NULL;
                call->rxtx_annotations[ix] = 0;
                skb->next = list;
@@ -66,7 +66,7 @@ static void rxrpc_rotate_tx_window(struct rxrpc_call *call, rxrpc_seq_t to)
                skb = list;
                list = skb->next;
                skb->next = NULL;
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
        }
 }
 
@@ -99,6 +99,7 @@ static bool rxrpc_end_tx_phase(struct rxrpc_call *call, const char *abort_why)
        default:
                break;
        case RXRPC_CALL_CLIENT_AWAIT_REPLY:
+               call->tx_phase = false;
                call->state = RXRPC_CALL_CLIENT_RECV_REPLY;
                break;
        case RXRPC_CALL_SERVER_AWAIT_ACK:
@@ -278,7 +279,7 @@ next_subpacket:
         * Barriers against rxrpc_recvmsg_data() and rxrpc_rotate_rx_window()
         * and also rxrpc_fill_out_ack().
         */
-       rxrpc_get_skb(skb);
+       rxrpc_get_skb(skb, rxrpc_skb_rx_got);
        call->rxtx_annotations[ix] = annotation;
        smp_wmb();
        call->rxtx_buffer[ix] = skb;
@@ -691,13 +692,13 @@ void rxrpc_data_ready(struct sock *udp_sk)
                return;
        }
 
-       rxrpc_new_skb(skb);
+       rxrpc_new_skb(skb, rxrpc_skb_rx_received);
 
        _net("recv skb %p", skb);
 
        /* we'll probably need to checksum it (didn't call sock_recvmsg) */
        if (skb_checksum_complete(skb)) {
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
                __UDP_INC_STATS(&init_net, UDP_MIB_INERRORS, 0);
                _leave(" [CSUM failed]");
                return;
@@ -821,7 +822,7 @@ void rxrpc_data_ready(struct sock *udp_sk)
 discard_unlock:
        rcu_read_unlock();
 discard:
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 out:
        trace_rxrpc_rx_done(0, 0);
        return;
index f073e932500eda4bd7ee4253e4c975fca960982d..190f68bd9e270f2d64969f46d2368480bf3d3ddb 100644 (file)
@@ -90,7 +90,7 @@ void rxrpc_process_local_events(struct rxrpc_local *local)
        if (skb) {
                struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
 
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
                _debug("{%d},{%u}", local->debug_id, sp->hdr.type);
 
                switch (sp->hdr.type) {
@@ -107,7 +107,7 @@ void rxrpc_process_local_events(struct rxrpc_local *local)
                        break;
                }
 
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
        }
 
        _leave("");
index c7065d893d1e5decaa1753962c3a3b79e552d828..026e1f2e83ffc51ab8303161d52152dd208f568c 100644 (file)
@@ -102,6 +102,24 @@ const char *rxrpc_acks(u8 reason)
        return str[reason];
 }
 
+const char rxrpc_skb_traces[rxrpc_skb__nr_trace][7] = {
+       [rxrpc_skb_rx_cleaned]          = "Rx CLN",
+       [rxrpc_skb_rx_freed]            = "Rx FRE",
+       [rxrpc_skb_rx_got]              = "Rx GOT",
+       [rxrpc_skb_rx_lost]             = "Rx *L*",
+       [rxrpc_skb_rx_received]         = "Rx RCV",
+       [rxrpc_skb_rx_purged]           = "Rx PUR",
+       [rxrpc_skb_rx_rotated]          = "Rx ROT",
+       [rxrpc_skb_rx_seen]             = "Rx SEE",
+       [rxrpc_skb_tx_cleaned]          = "Tx CLN",
+       [rxrpc_skb_tx_freed]            = "Tx FRE",
+       [rxrpc_skb_tx_got]              = "Tx GOT",
+       [rxrpc_skb_tx_lost]             = "Tx *L*",
+       [rxrpc_skb_tx_new]              = "Tx NEW",
+       [rxrpc_skb_tx_rotated]          = "Tx ROT",
+       [rxrpc_skb_tx_seen]             = "Tx SEE",
+};
+
 const char rxrpc_conn_traces[rxrpc_conn__nr_trace][4] = {
        [rxrpc_conn_new_client]         = "NWc",
        [rxrpc_conn_new_service]        = "NWs",
index 2c9daeadce87ad2ff60fe8cf0e9ab130e6b5abe3..a2cad5ce741693e937ae103295903830dba25701 100644 (file)
@@ -324,7 +324,7 @@ void rxrpc_reject_packets(struct rxrpc_local *local)
        whdr.type = RXRPC_PACKET_TYPE_ABORT;
 
        while ((skb = skb_dequeue(&local->reject_queue))) {
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
                sp = rxrpc_skb(skb);
 
                if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) {
@@ -343,7 +343,7 @@ void rxrpc_reject_packets(struct rxrpc_local *local)
                        kernel_sendmsg(local->socket, &msg, iov, 2, size);
                }
 
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
        }
 
        _leave("");
index 9e0725f5652bd2faa0ab907933c04c530713e1dd..18276e7cb9e050ae2f176fe0c388378c7a997076 100644 (file)
@@ -155,11 +155,11 @@ void rxrpc_error_report(struct sock *sk)
                _leave("UDP socket errqueue empty");
                return;
        }
-       rxrpc_new_skb(skb);
+       rxrpc_new_skb(skb, rxrpc_skb_rx_received);
        serr = SKB_EXT_ERR(skb);
        if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
                _leave("UDP empty message");
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
                return;
        }
 
@@ -169,7 +169,7 @@ void rxrpc_error_report(struct sock *sk)
                peer = NULL;
        if (!peer) {
                rcu_read_unlock();
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
                _leave(" [no peer]");
                return;
        }
@@ -179,7 +179,7 @@ void rxrpc_error_report(struct sock *sk)
             serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
                rxrpc_adjust_mtu(peer, serr);
                rcu_read_unlock();
-               rxrpc_free_skb(skb);
+               rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
                rxrpc_put_peer(peer);
                _leave(" [MTU update]");
                return;
@@ -187,7 +187,7 @@ void rxrpc_error_report(struct sock *sk)
 
        rxrpc_store_error(peer, serr);
        rcu_read_unlock();
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 
        /* The ref we obtained is passed off to the work item */
        rxrpc_queue_work(&peer->error_distributor);
index 79e65668bc58d9ef8f57c4ffe65f8b723c757c62..6ba4af5a8d95cc6a7d558df5dd76140d076897b4 100644 (file)
@@ -155,6 +155,7 @@ static void rxrpc_end_rx_phase(struct rxrpc_call *call)
                break;
 
        case RXRPC_CALL_SERVER_RECV_REQUEST:
+               call->tx_phase = true;
                call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
                break;
        default:
@@ -185,7 +186,7 @@ static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
        hard_ack++;
        ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
        skb = call->rxtx_buffer[ix];
-       rxrpc_see_skb(skb);
+       rxrpc_see_skb(skb, rxrpc_skb_rx_rotated);
        sp = rxrpc_skb(skb);
        flags = sp->hdr.flags;
        serial = sp->hdr.serial;
@@ -197,7 +198,7 @@ static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
        /* Barrier against rxrpc_input_data(). */
        smp_store_release(&call->rx_hard_ack, hard_ack);
 
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 
        _debug("%u,%u,%02x", hard_ack, top, flags);
        trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
@@ -317,7 +318,7 @@ static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
                        break;
                }
                smp_rmb();
-               rxrpc_see_skb(skb);
+               rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
                sp = rxrpc_skb(skb);
 
                if (!(flags & MSG_PEEK))
index 28d8f73cf11d3500a4662e18c91e5385047a5a72..6a39ee97a0b7d1ec563c94409f0aef66e57fa85f 100644 (file)
@@ -100,7 +100,7 @@ static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
        ASSERTCMP(seq, ==, call->tx_top + 1);
 
        ix = seq & RXRPC_RXTX_BUFF_MASK;
-       rxrpc_get_skb(skb);
+       rxrpc_get_skb(skb, rxrpc_skb_tx_got);
        call->rxtx_annotations[ix] = RXRPC_TX_ANNO_UNACK;
        smp_wmb();
        call->rxtx_buffer[ix] = skb;
@@ -146,7 +146,7 @@ static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
                rxrpc_instant_resend(call, ix);
        }
 
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
        _leave("");
 }
 
@@ -201,7 +201,7 @@ static int rxrpc_send_data(struct rxrpc_sock *rx,
 
        skb = call->tx_pending;
        call->tx_pending = NULL;
-       rxrpc_see_skb(skb);
+       rxrpc_see_skb(skb, rxrpc_skb_tx_seen);
 
        copied = 0;
        do {
@@ -242,7 +242,7 @@ static int rxrpc_send_data(struct rxrpc_sock *rx,
                        if (!skb)
                                goto maybe_error;
 
-                       rxrpc_new_skb(skb);
+                       rxrpc_new_skb(skb, rxrpc_skb_tx_new);
 
                        _debug("ALLOC SEND %p", skb);
 
@@ -352,7 +352,7 @@ out:
        return ret;
 
 call_terminated:
-       rxrpc_free_skb(skb);
+       rxrpc_free_skb(skb, rxrpc_skb_tx_freed);
        _leave(" = %d", -call->error);
        return -call->error;
 
index 620d9ccaf3c1c504394b2c871edfac3a576e3389..5154cbf7e540c55efdd8a8dc94abd294005e4e39 100644 (file)
 #include <net/af_rxrpc.h>
 #include "ar-internal.h"
 
+#define select_skb_count(op) (op >= rxrpc_skb_tx_cleaned ? &rxrpc_n_tx_skbs : &rxrpc_n_rx_skbs)
+
 /*
- * Note the existence of a new-to-us socket buffer (allocated or dequeued).
+ * Note the allocation or reception of a socket buffer.
  */
-void rxrpc_new_skb(struct sk_buff *skb)
+void rxrpc_new_skb(struct sk_buff *skb, enum rxrpc_skb_trace op)
 {
        const void *here = __builtin_return_address(0);
-       int n = atomic_inc_return(&rxrpc_n_skbs);
-       trace_rxrpc_skb(skb, 0, atomic_read(&skb->users), n, here);
+       int n = atomic_inc_return(select_skb_count(op));
+       trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
 }
 
 /*
  * Note the re-emergence of a socket buffer from a queue or buffer.
  */
-void rxrpc_see_skb(struct sk_buff *skb)
+void rxrpc_see_skb(struct sk_buff *skb, enum rxrpc_skb_trace op)
 {
        const void *here = __builtin_return_address(0);
        if (skb) {
-               int n = atomic_read(&rxrpc_n_skbs);
-               trace_rxrpc_skb(skb, 1, atomic_read(&skb->users), n, here);
+               int n = atomic_read(select_skb_count(op));
+               trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
        }
 }
 
 /*
  * Note the addition of a ref on a socket buffer.
  */
-void rxrpc_get_skb(struct sk_buff *skb)
+void rxrpc_get_skb(struct sk_buff *skb, enum rxrpc_skb_trace op)
 {
        const void *here = __builtin_return_address(0);
-       int n = atomic_inc_return(&rxrpc_n_skbs);
-       trace_rxrpc_skb(skb, 2, atomic_read(&skb->users), n, here);
+       int n = atomic_inc_return(select_skb_count(op));
+       trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
        skb_get(skb);
 }
 
 /*
  * Note the destruction of a socket buffer.
  */
-void rxrpc_free_skb(struct sk_buff *skb)
+void rxrpc_free_skb(struct sk_buff *skb, enum rxrpc_skb_trace op)
 {
        const void *here = __builtin_return_address(0);
        if (skb) {
                int n;
                CHECK_SLAB_OKAY(&skb->users);
-               n = atomic_dec_return(&rxrpc_n_skbs);
-               trace_rxrpc_skb(skb, 3, atomic_read(&skb->users), n, here);
+               n = atomic_dec_return(select_skb_count(op));
+               trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
                kfree_skb(skb);
        }
 }
 
+/*
+ * Note the injected loss of a socket buffer.
+ */
+void rxrpc_lose_skb(struct sk_buff *skb, enum rxrpc_skb_trace op)
+{
+       const void *here = __builtin_return_address(0);
+       if (skb) {
+               int n;
+               CHECK_SLAB_OKAY(&skb->users);
+               if (op == rxrpc_skb_tx_lost) {
+                       n = atomic_read(select_skb_count(op));
+                       trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
+               } else {
+                       n = atomic_dec_return(select_skb_count(op));
+                       trace_rxrpc_skb(skb, op, atomic_read(&skb->users), n, here);
+                       kfree_skb(skb);
+               }
+       }
+}
+
 /*
  * Clear a queue of socket buffers.
  */
@@ -74,8 +96,9 @@ void rxrpc_purge_queue(struct sk_buff_head *list)
        const void *here = __builtin_return_address(0);
        struct sk_buff *skb;
        while ((skb = skb_dequeue((list))) != NULL) {
-               int n = atomic_dec_return(&rxrpc_n_skbs);
-               trace_rxrpc_skb(skb, 4, atomic_read(&skb->users), n, here);
+               int n = atomic_dec_return(select_skb_count(rxrpc_skb_rx_purged));
+               trace_rxrpc_skb(skb, rxrpc_skb_rx_purged,
+                               atomic_read(&skb->users), n, here);
                kfree_skb(skb);
        }
 }