Initial implementation of transmit buffering. Tested using netem with delay and loss also.
Signed off by: Ian McDonald <[EMAIL PROTECTED]> ---- diff --git a/include/linux/dccp.h b/include/linux/dccp.h index 088529f..4a2f845 100644 --- a/include/linux/dccp.h +++ b/include/linux/dccp.h @@ -411,6 +411,8 @@ struct dccp_ackvec; * @dccps_role - Role of this sock, one of %dccp_role * @dccps_ndp_count - number of Non Data Packets since last data packet * @dccps_hc_rx_ackvec - rx half connection ack vector + * @dccps_xmit_timer - timer for when CCID is not ready to send + * @dccps_xmit_lock - lock for transmitting */ struct dccp_sock { /* inet_connection_sock has to be the first member of dccp_sock */ @@ -443,6 +445,8 @@ struct dccp_sock { enum dccp_role dccps_role:2; __u8 dccps_hc_rx_insert_options:1; __u8 dccps_hc_tx_insert_options:1; + struct timer_list dccps_xmit_timer; + spinlock_t dccps_xmit_lock; }; static inline struct dccp_sock *dccp_sk(const struct sock *sk) diff --git a/net/dccp/dccp.h b/net/dccp/dccp.h index 93f26dd..82431c0 100644 --- a/net/dccp/dccp.h +++ b/net/dccp/dccp.h @@ -5,7 +5,7 @@ * * An implementation of the DCCP protocol * Copyright (c) 2005 Arnaldo Carvalho de Melo <[EMAIL PROTECTED]> - * Copyright (c) 2005 Ian McDonald <[EMAIL PROTECTED]> + * Copyright (c) 2005-6 Ian McDonald <[EMAIL PROTECTED]> * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 as @@ -126,7 +126,7 @@ extern void dccp_send_delayed_ack(struct extern void dccp_send_sync(struct sock *sk, const u64 seq, const enum dccp_pkt_type pkt_type); -extern int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo); +void dccp_write_xmit(struct sock *sk, int block); extern void dccp_write_space(struct sock *sk); extern void dccp_init_xmit_timers(struct sock *sk); diff --git a/net/dccp/output.c b/net/dccp/output.c index efd7ffb..90b802d 100644 --- a/net/dccp/output.c +++ b/net/dccp/output.c @@ -2,7 +2,8 @@ * net/dccp/output.c * * An implementation of the DCCP protocol - * Arnaldo Carvalho de Melo <[EMAIL PROTECTED]> + * Copyright (c) 2005 Arnaldo Carvalho de Melo <[EMAIL PROTECTED]> + * Copyright (c) 2006 Ian McDonald <[EMAIL PROTECTED]> * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License @@ -191,7 +192,7 @@ static int dccp_wait_for_ccid(struct soc 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; @@ -207,9 +208,11 @@ static int dccp_wait_for_ccid(struct soc goto do_nonblock; sk->sk_write_pending++; + spin_unlock(&dp->dccps_xmit_lock); release_sock(sk); *timeo -= schedule_timeout(delay); lock_sock(sk); + spin_lock(&dp->dccps_xmit_lock); sk->sk_write_pending--; } out: @@ -227,37 +230,56 @@ do_interrupted: goto out; } -int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo) -{ - const struct dccp_sock *dp = dccp_sk(sk); - int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb, - skb->len); +static void dccp_write_xmit_timer(unsigned long sk) { + dccp_write_xmit((struct sock *)sk,0); +} - if (err > 0) - err = dccp_wait_for_ccid(sk, skb, timeo); +void dccp_write_xmit(struct sock *sk, int block) +{ + struct dccp_sock *dp = dccp_sk(sk); + struct sk_buff *skb; + long timeo = 2000; /* FIXME imcdnzl - 2 second default */ - if (err == 0) { - struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb); - const int len = skb->len; + spin_lock(&dp->dccps_xmit_lock); + + 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) { + if (likely(!block)) { + mod_timer(&dp->dccps_xmit_timer, + msecs_to_jiffies(err)+jiffies); + goto xmit_out; + } else { + err = dccp_wait_for_ccid(sk, skb, &timeo); + } + } - 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, + 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, 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); + } else + kfree(skb); + } +xmit_out: + spin_unlock(&dp->dccps_xmit_lock); } int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb) @@ -396,6 +418,10 @@ static inline void dccp_connect_init(str 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; + dp->dccps_xmit_lock = SPIN_LOCK_UNLOCKED; } int dccp_connect(struct sock *sk) @@ -527,8 +553,11 @@ void dccp_send_close(struct sock *sk, co 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)); - } else + /* FIXME do we need a retransmit timer here? */ + } else { dccp_transmit_skb(sk, skb); + } } diff --git a/net/dccp/proto.c b/net/dccp/proto.c index 65b11ea..119ecc5 100644 --- a/net/dccp/proto.c +++ b/net/dccp/proto.c @@ -385,6 +385,7 @@ int dccp_sendmsg(struct kiocb *iocb, str lock_sock(sk); timeo = sock_sndtimeo(sk, noblock); + /* FIXME imcdnzl - we should be using timeo for memory allocation */ /* * We have to use sk_stream_wait_connect here to set sk_write_pending, @@ -407,19 +408,13 @@ int dccp_sendmsg(struct kiocb *iocb, str if (rc != 0) goto out_discard; - rc = dccp_write_xmit(sk, skb, &timeo); - /* - * XXX we don't use sk_write_queue, so just discard the packet. - * Current plan however is to _use_ sk_write_queue with - * an algorith similar to tcp_sendmsg, where the main difference - * is that in DCCP we have to respect packet boundaries, so - * no coalescing of skbs. - * - * This bug was _quickly_ found & fixed by just looking at an OSTRA - * generated callgraph 8) -acme - */ + skb_queue_tail(&sk->sk_write_queue, skb); + release_sock(sk); + dccp_write_xmit(sk,0); + goto out_end; out_release: release_sock(sk); +out_end: return rc ? : len; out_discard: kfree_skb(skb); @@ -591,6 +586,7 @@ static int dccp_close_state(struct sock void dccp_close(struct sock *sk, long timeout) { + struct dccp_sock *dp = dccp_sk(sk); struct sk_buff *skb; lock_sock(sk); @@ -606,6 +602,8 @@ void dccp_close(struct sock *sk, long ti goto adjudge_to_death; } + del_timer_sync(&dp->dccps_xmit_timer); + /* * We need to flush the recv. buffs. We do this only on the * descriptor close, not protocol-sourced closes, because the - To unsubscribe from this list: send the line "unsubscribe dccp" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html