fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / net / dccp / output.c
index efd7ffb..3435542 100644 (file)
@@ -1,6 +1,6 @@
 /*
  *  net/dccp/output.c
- * 
+ *
  *  An implementation of the DCCP protocol
  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
  *
@@ -10,7 +10,6 @@
  *     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>
@@ -27,7 +26,7 @@ static inline void dccp_event_ack_sent(struct sock *sk)
        inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
 }
 
-static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
+static void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
 {
        skb_set_owner_w(skb, sk);
        WARN_ON(sk->sk_send_head);
@@ -49,7 +48,7 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                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;
@@ -64,6 +63,10 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                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;
@@ -79,18 +82,21 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                }
 
                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);
 
                /* Build DCCP header and checksum it. */
-               memset(dh, 0, dccp_header_size);
+               dh = dccp_zeroed_hdr(skb, dccp_header_size);
                dh->dccph_type  = dcb->dccpd_type;
                dh->dccph_sport = inet->sport;
                dh->dccph_dport = inet->dport;
                dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
                dh->dccph_ccval = dcb->dccpd_ccval;
+               dh->dccph_cscov = dp->dccps_pcslen;
                /* XXX For now we're using only 48 bits sequence numbers */
                dh->dccph_x     = 1;
 
@@ -110,7 +116,7 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
                        break;
                }
 
-               icsk->icsk_af_ops->send_check(sk, skb->len, skb);
+               icsk->icsk_af_ops->send_check(sk, 0, skb);
 
                if (set_ack)
                        dccp_event_ack_sent(sk);
@@ -119,16 +125,7 @@ static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 
                memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
                err = icsk->icsk_af_ops->queue_xmit(skb, 0);
-               if (err <= 0)
-                       return err;
-
-               /* NET_XMIT_CN is special. It does not guarantee,
-                * that this packet is lost. It tells that device
-                * is about to start to drop packets or already
-                * drops some packets of the same priority and
-                * invokes us to send less aggressively.
-                */
-               return err == NET_XMIT_CN ? 0 : err;
+               return net_xmit_eval(err);
        }
        return -ENOBUFS;
 }
@@ -178,37 +175,29 @@ void dccp_write_space(struct sock *sk)
 /**
  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
  * @sk: socket to wait for
- * @timeo: for how long
  */
-static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
-                             long *timeo)
+static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb)
 {
        struct dccp_sock *dp = dccp_sk(sk);
        DEFINE_WAIT(wait);
-       long delay;
+       unsigned long delay;
        int rc;
 
        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;
                if (signal_pending(current))
                        goto do_interrupted;
 
-               rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
-                                           skb->len);
+               rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
                if (rc <= 0)
                        break;
                delay = msecs_to_jiffies(rc);
-               if (delay > *timeo || delay < 0)
-                       goto do_nonblock;
-
                sk->sk_write_pending++;
                release_sock(sk);
-               *timeo -= schedule_timeout(delay);
+               schedule_timeout(delay);
                lock_sock(sk);
                sk->sk_write_pending--;
        }
@@ -219,45 +208,70 @@ out:
 do_error:
        rc = -EPIPE;
        goto out;
-do_nonblock:
-       rc = -EAGAIN;
-       goto out;
 do_interrupted:
-       rc = sock_intr_errno(*timeo);
+       rc = -EINTR;
        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,
-                                        skb->len);
+       struct dccp_sock *dp = dccp_sk(sk);
+       struct sk_buff *skb;
 
-       if (err > 0)
-               err = dccp_wait_for_ccid(sk, skb, timeo);
+       while ((skb = skb_peek(&sk->sk_write_queue))) {
+               int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb);
+
+               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);
+                       if (err && err != -EINTR)
+                               DCCP_BUG("err=%d after dccp_wait_for_ccid", err);
+               }
 
-       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)
+                               DCCP_BUG("err=%d after ccid_hc_tx_packet_sent",
+                                        err);
+               } else {
+                       dccp_pr_debug("packet discarded\n");
+                       kfree(skb);
+               }
+       }
 }
 
 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
@@ -275,30 +289,32 @@ 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;
 
        dreq = dccp_rsk(req);
+       if (inet_rsk(req)->acked)       /* increase ISS upon retransmission */
+               dccp_inc_seqno(&dreq->dreq_iss);
        DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
        DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
-       dccp_insert_options(sk, skb);
 
-       skb->h.raw = skb_push(skb, dccp_header_size);
+       if (dccp_insert_options(sk, skb)) {
+               kfree_skb(skb);
+               return NULL;
+       }
 
-       dh = dccp_hdr(skb);
-       memset(dh, 0, dccp_header_size);
+       /* Build and checksum header */
+       dh = dccp_zeroed_hdr(skb, dccp_header_size);
 
        dh->dccph_sport = inet_sk(sk)->sport;
        dh->dccph_dport = inet_rsk(req)->rmt_port;
@@ -310,47 +326,46 @@ 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_csum_outgoing(skb);
 
+       /* We use `acked' to remember that a Response was already sent. */
+       inet_rsk(req)->acked = 1;
        DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
        return skb;
 }
 
 EXPORT_SYMBOL_GPL(dccp_make_response);
 
-struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
-                               const enum dccp_reset_codes code)
-                                  
+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;
 
        dccp_inc_seqno(&dp->dccps_gss);
 
        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);
 
-       skb->h.raw = skb_push(skb, dccp_header_size);
+       if (dccp_insert_options(sk, skb)) {
+               kfree_skb(skb);
+               return NULL;
+       }
 
-       dh = dccp_hdr(skb);
-       memset(dh, 0, dccp_header_size);
+       dh = dccp_zeroed_hdr(skb, dccp_header_size);
 
        dh->dccph_sport = inet_sk(sk)->sport;
        dh->dccph_dport = inet_sk(sk)->dport;
@@ -362,14 +377,33 @@ 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, 0, 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);
+                       return net_xmit_eval(err);
+               }
+       }
+
+       return err;
+}
+
 /*
  * Do all connect socket setups that can be done AF independent.
  */
@@ -384,18 +418,25 @@ static inline void dccp_connect_init(struct sock *sk)
        
        dccp_sync_mss(sk, dst_mtu(dst));
 
-       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_update_gss(sk, dp->dccps_iss);
        dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 
+       /* S.GAR - greatest valid acknowledgement number received on a non-Sync;
+        *         initialized to S.ISS (sec. 8.5)                            */
+       dp->dccps_gar = 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)
@@ -405,15 +446,14 @@ 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;
 
        dccp_skb_entail(sk, skb);
        dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
@@ -431,7 +471,8 @@ 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);
@@ -443,8 +484,7 @@ void dccp_send_ack(struct sock *sk)
                }
 
                /* Reserve space for headers */
-               skb_reserve(skb, MAX_DCCP_HEADER);
-               skb->csum = 0;
+               skb_reserve(skb, sk->sk_prot->max_header);
                DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
                dccp_transmit_skb(sk, skb);
        }
@@ -490,21 +530,22 @@ 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->csum = 0;
+       skb_reserve(skb, sk->sk_prot->max_header);
        DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
        DCCP_SKB_CB(skb)->dccpd_seq = seq;
 
        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
@@ -522,13 +563,14 @@ void dccp_send_close(struct sock *sk, const int active)
 
        /* Reserve space for headers and prepare control bits. */
        skb_reserve(skb, sk->sk_prot->max_header);
-       skb->csum = 0;
        DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
                                        DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 
        if (active) {
+               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);
 }