]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - net/dccp/output.c
Merge branch 'tj-upstream-fixes' of git://htj.dyndns.org/libata-tj into tmp
[linux-2.6-omap-h63xx.git] / net / dccp / output.c
index 29250749f16f86044d1f5fec4a592271fa12f856..7102e3aed4ca2885928e990cf8b934e87d6fe2e4 100644 (file)
  *     2 of the License, or (at your option) any later version.
  */
 
-#include <linux/config.h>
 #include <linux/dccp.h>
+#include <linux/kernel.h>
 #include <linux/skbuff.h>
 
+#include <net/inet_sock.h>
 #include <net/sock.h>
 
 #include "ackvec.h"
@@ -25,21 +26,29 @@ static inline void dccp_event_ack_sent(struct sock *sk)
        inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
 }
 
+static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
+{
+       skb_set_owner_w(skb, sk);
+       WARN_ON(sk->sk_send_head);
+       sk->sk_send_head = skb;
+}
+
 /*
  * All SKB's seen here are completely headerless. It is our
  * job to build the DCCP header, and pass the packet down to
  * IP so it can do the same plus pass the packet off to the
  * device.
  */
-int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
+static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 {
        if (likely(skb != NULL)) {
                const struct inet_sock *inet = inet_sk(sk);
+               const struct inet_connection_sock *icsk = inet_csk(sk);
                struct dccp_sock *dp = dccp_sk(sk);
                struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
                struct dccp_hdr *dh;
                /* XXX For now we're using only 48 bits sequence numbers */
-               const int dccp_header_size = sizeof(*dh) +
+               const u32 dccp_header_size = sizeof(*dh) +
                                             sizeof(struct dccp_hdr_ext) +
                                          dccp_packet_hdr_len(dcb->dccpd_type);
                int err, set_ack = 1;
@@ -50,22 +59,38 @@ int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                switch (dcb->dccpd_type) {
                case DCCP_PKT_DATA:
                        set_ack = 0;
+                       /* fall through */
+               case DCCP_PKT_DATAACK:
                        break;
+
+               case DCCP_PKT_REQUEST:
+                       set_ack = 0;
+                       /* fall through */
+
                case DCCP_PKT_SYNC:
                case DCCP_PKT_SYNCACK:
                        ackno = dcb->dccpd_seq;
+                       /* fall through */
+               default:
+                       /*
+                        * Only data packets should come through with skb->sk
+                        * set.
+                        */
+                       WARN_ON(skb->sk);
+                       skb_set_owner_w(skb, sk);
                        break;
                }
 
                dcb->dccpd_seq = dp->dccps_gss;
-               dccp_insert_options(sk, skb);
+
+               if (dccp_insert_options(sk, skb)) {
+                       kfree_skb(skb);
+                       return -EPROTO;
+               }
                
                skb->h.raw = skb_push(skb, dccp_header_size);
                dh = dccp_hdr(skb);
 
-               if (!skb->sk)
-                       skb_set_owner_w(skb, sk);
-
                /* Build DCCP header and checksum it. */
                memset(dh, 0, dccp_header_size);
                dh->dccph_type  = dcb->dccpd_type;
@@ -92,8 +117,7 @@ int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                        break;
                }
 
-               dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr,
-                                                     inet->daddr);
+               icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 
                if (set_ack)
                        dccp_event_ack_sent(sk);
@@ -101,7 +125,7 @@ int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 
                memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
-               err = ip_queue_xmit(skb, 0);
+               err = icsk->icsk_af_ops->queue_xmit(skb, 0);
                if (err <= 0)
                        return err;
 
@@ -118,20 +142,13 @@ int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 
 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 {
+       struct inet_connection_sock *icsk = inet_csk(sk);
        struct dccp_sock *dp = dccp_sk(sk);
-       int mss_now;
-
-       /*
-        * FIXME: we really should be using the af_specific thing to support
-        *        IPv6.
-        * mss_now = pmtu - tp->af_specific->net_header_len -
-        *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
-        */
-       mss_now = pmtu - sizeof(struct iphdr) - sizeof(struct dccp_hdr) -
-                 sizeof(struct dccp_hdr_ext);
+       int mss_now = (pmtu - icsk->icsk_af_ops->net_header_len -
+                      sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 
        /* Now subtract optional transport overhead */
-       mss_now -= dp->dccps_ext_header_len;
+       mss_now -= icsk->icsk_ext_hdr_len;
 
        /*
         * FIXME: this should come from the CCID infrastructure, where, say,
@@ -144,12 +161,14 @@ unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
        mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 
        /* And store cached results */
-       dp->dccps_pmtu_cookie = pmtu;
+       icsk->icsk_pmtu_cookie = pmtu;
        dp->dccps_mss_cache = mss_now;
 
        return mss_now;
 }
 
+EXPORT_SYMBOL_GPL(dccp_sync_mss);
+
 void dccp_write_space(struct sock *sk)
 {
        read_lock(&sk->sk_callback_lock);
@@ -179,7 +198,7 @@ static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
        while (1) {
                prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 
-               if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
+               if (sk->sk_err)
                        goto do_error;
                if (!*timeo)
                        goto do_nonblock;
@@ -215,42 +234,77 @@ do_interrupted:
        goto out;
 }
 
-int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
+static void dccp_write_xmit_timer(unsigned long data) {
+       struct sock *sk = (struct sock *)data;
+       struct dccp_sock *dp = dccp_sk(sk);
+
+       bh_lock_sock(sk);
+       if (sock_owned_by_user(sk))
+               sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
+       else
+               dccp_write_xmit(sk, 0);
+       bh_unlock_sock(sk);
+       sock_put(sk);
+}
+
+void dccp_write_xmit(struct sock *sk, int block)
 {
-       const struct dccp_sock *dp = dccp_sk(sk);
-       int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
+       struct dccp_sock *dp = dccp_sk(sk);
+       struct sk_buff *skb;
+       long timeo = 30000;     /* If a packet is taking longer than 2 secs
+                                  we have other issues */
+
+       while ((skb = skb_peek(&sk->sk_write_queue))) {
+               int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
                                         skb->len);
 
-       if (err > 0)
-               err = dccp_wait_for_ccid(sk, skb, timeo);
+               if (err > 0) {
+                       if (!block) {
+                               sk_reset_timer(sk, &dp->dccps_xmit_timer,
+                                               msecs_to_jiffies(err)+jiffies);
+                               break;
+                       } else
+                               err = dccp_wait_for_ccid(sk, skb, &timeo);
+                       if (err) {
+                               printk(KERN_CRIT "%s:err at dccp_wait_for_ccid"
+                                                " %d\n", __FUNCTION__, err);
+                               dump_stack();
+                       }
+               }
 
-       if (err == 0) {
-               struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
-               const int len = skb->len;
+               skb_dequeue(&sk->sk_write_queue);
+               if (err == 0) {
+                       struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
+                       const int len = skb->len;
 
-               if (sk->sk_state == DCCP_PARTOPEN) {
-                       /* See 8.1.5.  Handshake Completion */
-                       inet_csk_schedule_ack(sk);
-                       inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
+                       if (sk->sk_state == DCCP_PARTOPEN) {
+                               /* See 8.1.5.  Handshake Completion */
+                               inet_csk_schedule_ack(sk);
+                               inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
                                                  inet_csk(sk)->icsk_rto,
                                                  DCCP_RTO_MAX);
-                       dcb->dccpd_type = DCCP_PKT_DATAACK;
-               } else if (dccp_ack_pending(sk))
-                       dcb->dccpd_type = DCCP_PKT_DATAACK;
-               else
-                       dcb->dccpd_type = DCCP_PKT_DATA;
-
-               err = dccp_transmit_skb(sk, skb);
-               ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
-       } else
-               kfree_skb(skb);
-
-       return err;
+                               dcb->dccpd_type = DCCP_PKT_DATAACK;
+                       } else if (dccp_ack_pending(sk))
+                               dcb->dccpd_type = DCCP_PKT_DATAACK;
+                       else
+                               dcb->dccpd_type = DCCP_PKT_DATA;
+
+                       err = dccp_transmit_skb(sk, skb);
+                       ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
+                       if (err) {
+                               printk(KERN_CRIT "%s:err from "
+                                                "ccid_hc_tx_packet_sent %d\n",
+                                                __FUNCTION__, err);
+                               dump_stack();
+                       }
+               } else
+                       kfree(skb);
+       }
 }
 
 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 {
-       if (inet_sk_rebuild_header(sk) != 0)
+       if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
                return -EHOSTUNREACH; /* Routing failure or similar. */
 
        return dccp_transmit_skb(sk, (skb_cloned(skb) ?
@@ -263,17 +317,16 @@ struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 {
        struct dccp_hdr *dh;
        struct dccp_request_sock *dreq;
-       const int dccp_header_size = sizeof(struct dccp_hdr) +
+       const u32 dccp_header_size = sizeof(struct dccp_hdr) +
                                     sizeof(struct dccp_hdr_ext) +
                                     sizeof(struct dccp_hdr_response);
-       struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
-                                              dccp_header_size, 1,
+       struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
                                           GFP_ATOMIC);
        if (skb == NULL)
                return NULL;
 
        /* Reserve space for headers. */
-       skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
+       skb_reserve(skb, sk->sk_prot->max_header);
 
        skb->dst = dst_clone(dst);
        skb->csum = 0;
@@ -281,7 +334,11 @@ struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
        dreq = dccp_rsk(req);
        DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
        DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
-       dccp_insert_options(sk, skb);
+
+       if (dccp_insert_options(sk, skb)) {
+               kfree_skb(skb);
+               return NULL;
+       }
 
        skb->h.raw = skb_push(skb, dccp_header_size);
 
@@ -298,30 +355,28 @@ struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
        dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
        dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 
-       dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
-                                             inet_rsk(req)->rmt_addr);
-
        DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
        return skb;
 }
 
-struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
-                               const enum dccp_reset_codes code)
+EXPORT_SYMBOL_GPL(dccp_make_response);
+
+static struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
+                                      const enum dccp_reset_codes code)
                                   
 {
        struct dccp_hdr *dh;
        struct dccp_sock *dp = dccp_sk(sk);
-       const int dccp_header_size = sizeof(struct dccp_hdr) +
+       const u32 dccp_header_size = sizeof(struct dccp_hdr) +
                                     sizeof(struct dccp_hdr_ext) +
                                     sizeof(struct dccp_hdr_reset);
-       struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
-                                              dccp_header_size, 1,
+       struct sk_buff *skb = sock_wmalloc(sk, sk->sk_prot->max_header, 1,
                                           GFP_ATOMIC);
        if (skb == NULL)
                return NULL;
 
        /* Reserve space for headers. */
-       skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
+       skb_reserve(skb, sk->sk_prot->max_header);
 
        skb->dst = dst_clone(dst);
        skb->csum = 0;
@@ -331,7 +386,11 @@ struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
        DCCP_SKB_CB(skb)->dccpd_reset_code = code;
        DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
        DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
-       dccp_insert_options(sk, skb);
+
+       if (dccp_insert_options(sk, skb)) {
+               kfree_skb(skb);
+               return NULL;
+       }
 
        skb->h.raw = skb_push(skb, dccp_header_size);
 
@@ -348,19 +407,40 @@ struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
        dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 
        dccp_hdr_reset(skb)->dccph_reset_code = code;
-
-       dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
-                                             inet_sk(sk)->daddr);
+       inet_csk(sk)->icsk_af_ops->send_check(sk, skb->len, skb);
 
        DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
        return skb;
 }
 
+int dccp_send_reset(struct sock *sk, enum dccp_reset_codes code)
+{
+       /*
+        * FIXME: what if rebuild_header fails?
+        * Should we be doing a rebuild_header here?
+        */
+       int err = inet_sk_rebuild_header(sk);
+
+       if (err == 0) {
+               struct sk_buff *skb = dccp_make_reset(sk, sk->sk_dst_cache,
+                                                     code);
+               if (skb != NULL) {
+                       memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
+                       err = inet_csk(sk)->icsk_af_ops->queue_xmit(skb, 0);
+                       if (err == NET_XMIT_CN)
+                               err = 0;
+               }
+       }
+
+       return err;
+}
+
 /*
  * Do all connect socket setups that can be done AF independent.
  */
 static inline void dccp_connect_init(struct sock *sk)
 {
+       struct dccp_sock *dp = dccp_sk(sk);
        struct dst_entry *dst = __sk_dst_get(sk);
        struct inet_connection_sock *icsk = inet_csk(sk);
 
@@ -369,12 +449,21 @@ static inline void dccp_connect_init(struct sock *sk)
        
        dccp_sync_mss(sk, dst_mtu(dst));
 
-       /*
-        * FIXME: set dp->{dccps_swh,dccps_swl}, with
-        * something like dccp_inc_seq
-        */
+       dccp_update_gss(sk, dp->dccps_iss);
+       /*
+        * SWL and AWL are initially adjusted so that they are not less than
+        * the initial Sequence Numbers received and sent, respectively:
+        *      SWL := max(GSR + 1 - floor(W/4), ISR),
+        *      AWL := max(GSS - W' + 1, ISS).
+        * These adjustments MUST be applied only at the beginning of the
+        * connection.
+        */
+       dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 
        icsk->icsk_retransmits = 0;
+       init_timer(&dp->dccps_xmit_timer);
+       dp->dccps_xmit_timer.data = (unsigned long)sk;
+       dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
 }
 
 int dccp_connect(struct sock *sk)
@@ -384,19 +473,17 @@ int dccp_connect(struct sock *sk)
 
        dccp_connect_init(sk);
 
-       skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
+       skb = alloc_skb(sk->sk_prot->max_header, sk->sk_allocation);
        if (unlikely(skb == NULL))
                return -ENOBUFS;
 
        /* Reserve space for headers. */
-       skb_reserve(skb, MAX_DCCP_HEADER);
+       skb_reserve(skb, sk->sk_prot->max_header);
 
        DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
        skb->csum = 0;
-       skb_set_owner_w(skb, sk);
 
-       BUG_TRAP(sk->sk_send_head == NULL);
-       sk->sk_send_head = skb;
+       dccp_skb_entail(sk, skb);
        dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
        DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 
@@ -406,11 +493,14 @@ int dccp_connect(struct sock *sk)
        return 0;
 }
 
+EXPORT_SYMBOL_GPL(dccp_connect);
+
 void dccp_send_ack(struct sock *sk)
 {
        /* If we have been reset, we may not send again. */
        if (sk->sk_state != DCCP_CLOSED) {
-               struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
+               struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header,
+                                               GFP_ATOMIC);
 
                if (skb == NULL) {
                        inet_csk_schedule_ack(sk);
@@ -422,10 +512,9 @@ void dccp_send_ack(struct sock *sk)
                }
 
                /* Reserve space for headers */
-               skb_reserve(skb, MAX_DCCP_HEADER);
+               skb_reserve(skb, sk->sk_prot->max_header);
                skb->csum = 0;
                DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
-               skb_set_owner_w(skb, sk);
                dccp_transmit_skb(sk, skb);
        }
 }
@@ -470,22 +559,23 @@ void dccp_send_sync(struct sock *sk, const u64 seq,
         * dccp_transmit_skb() will set the ownership to this
         * sock.
         */
-       struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
+       struct sk_buff *skb = alloc_skb(sk->sk_prot->max_header, GFP_ATOMIC);
 
        if (skb == NULL)
                /* FIXME: how to make sure the sync is sent? */
                return;
 
        /* Reserve space for headers and prepare control bits. */
-       skb_reserve(skb, MAX_DCCP_HEADER);
+       skb_reserve(skb, sk->sk_prot->max_header);
        skb->csum = 0;
        DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
        DCCP_SKB_CB(skb)->dccpd_seq = seq;
 
-       skb_set_owner_w(skb, sk);
        dccp_transmit_skb(sk, skb);
 }
 
+EXPORT_SYMBOL_GPL(dccp_send_sync);
+
 /*
  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
@@ -495,7 +585,7 @@ void dccp_send_close(struct sock *sk, const int active)
 {
        struct dccp_sock *dp = dccp_sk(sk);
        struct sk_buff *skb;
-       const unsigned int prio = active ? GFP_KERNEL : GFP_ATOMIC;
+       const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 
        skb = alloc_skb(sk->sk_prot->max_header, prio);
        if (skb == NULL)
@@ -507,11 +597,11 @@ void dccp_send_close(struct sock *sk, const int active)
        DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
                                        DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 
-       skb_set_owner_w(skb, sk);
        if (active) {
-               BUG_TRAP(sk->sk_send_head == NULL);
-               sk->sk_send_head = skb;
+               dccp_write_xmit(sk, 1);
+               dccp_skb_entail(sk, skb);
                dccp_transmit_skb(sk, skb_clone(skb, prio));
+               /* FIXME do we need a retransmit timer here? */
        } else
                dccp_transmit_skb(sk, skb);
 }