From f3ab4896ed13733a22e9637395973fc1808823e1 Mon Sep 17 00:00:00 2001 From: Marco Varlese Date: Mon, 19 Feb 2018 15:23:13 +0100 Subject: SCTP: congestion control This patch addresses the requirements depicted by section 7.1.1 and 7.1.2 of the RFC 4960. Specifically, it implements the Slow-start and Congestion-avoidance policies. The patch also took care of correctly implementing some 'formatting' functions required - for instance - in packet(s) tracing. Change-Id: I68eade1b30345de3acb3ac8a653a5ef76eb6d2ac Signed-off-by: Marco Varlese --- src/vnet/sctp/sctp.c | 187 ++++++++++++++++++++------------------------------- 1 file changed, 73 insertions(+), 114 deletions(-) (limited to 'src/vnet/sctp/sctp.c') diff --git a/src/vnet/sctp/sctp.c b/src/vnet/sctp/sctp.c index 224c97d419e..b81d4d0e9ea 100644 --- a/src/vnet/sctp/sctp.c +++ b/src/vnet/sctp/sctp.c @@ -42,6 +42,8 @@ sctp_connection_bind (u32 session_index, transport_endpoint_t * tep) ip_copy (&listener->sub_conn[MAIN_SCTP_SUB_CONN_IDX].connection.lcl_ip, &tep->ip, tep->is_ip4); + listener->sub_conn[MAIN_SCTP_SUB_CONN_IDX].PMTU = + vnet_sw_interface_get_mtu (vnet_get_main (), tep->sw_if_index, VLIB_TX); listener->sub_conn[MAIN_SCTP_SUB_CONN_IDX].connection.is_ip4 = tep->is_ip4; listener->sub_conn[MAIN_SCTP_SUB_CONN_IDX].connection.proto = TRANSPORT_PROTO_SCTP; @@ -178,25 +180,44 @@ format_sctp_state (u8 * s, va_list * args) u8 * format_sctp_connection_id (u8 * s, va_list * args) { - /* - sctp_connection_t *sctp_conn = va_arg (*args, sctp_connection_t *); - if (!sctp_conn) - return s; - if (sctp_conn->c_is_ip4) - { - s = format (s, "[#%d][%s] %U:%d->%U:%d", sctp_conn->c_thread_index, "T", - format_ip4_address, &sctp_conn->c_lcl_ip4, - clib_net_to_host_u16 (sctp_conn->c_lcl_port), format_ip4_address, - &sctp_conn->c_rmt_ip4, clib_net_to_host_u16 (sctp_conn->c_rmt_port)); - } - else - { - s = format (s, "[#%d][%s] %U:%d->%U:%d", sctp_conn->c_thread_index, "T", - format_ip6_address, &sctp_conn->c_lcl_ip6, - clib_net_to_host_u16 (sctp_conn->c_lcl_port), format_ip6_address, - &sctp_conn->c_rmt_ip6, clib_net_to_host_u16 (sctp_conn->c_rmt_port)); - } - */ + sctp_connection_t *sctp_conn = va_arg (*args, sctp_connection_t *); + if (!sctp_conn) + return s; + + u8 i; + for (i = 0; i < MAX_SCTP_CONNECTIONS; i++) + { + if (sctp_conn->sub_conn[i].connection.is_ip4) + { + s = format (s, "%U[#%d][%s] %U:%d->%U:%d", + s, + sctp_conn->sub_conn[i].connection.thread_index, + "T", + format_ip4_address, + &sctp_conn->sub_conn[i].connection.lcl_ip.ip4, + clib_net_to_host_u16 (sctp_conn->sub_conn[i]. + connection.lcl_port), + format_ip4_address, + &sctp_conn->sub_conn[i].connection.rmt_ip.ip4, + clib_net_to_host_u16 (sctp_conn->sub_conn[i]. + connection.rmt_port)); + } + else + { + s = format (s, "%U[#%d][%s] %U:%d->%U:%d", + s, + sctp_conn->sub_conn[i].connection.thread_index, + "T", + format_ip6_address, + &sctp_conn->sub_conn[i].connection.lcl_ip.ip6, + clib_net_to_host_u16 (sctp_conn->sub_conn[i]. + connection.lcl_port), + format_ip6_address, + &sctp_conn->sub_conn[i].connection.rmt_ip.ip6, + clib_net_to_host_u16 (sctp_conn->sub_conn[i]. + connection.rmt_port)); + } + } return s; } @@ -235,48 +256,11 @@ sctp_init_snd_vars (sctp_connection_t * sctp_conn) time_now = sctp_time_now (); sctp_conn->local_initial_tsn = random_u32 (&time_now); - sctp_conn->remote_initial_tsn = 0x0; - sctp_conn->last_rcvd_tsn = sctp_conn->remote_initial_tsn; + sctp_conn->last_unacked_tsn = sctp_conn->local_initial_tsn; sctp_conn->next_tsn = sctp_conn->local_initial_tsn + 1; -} - -/** - * Update max segment size we're able to process. - * - * The value is constrained by our interface's MTU and IP options. It is - * also what we advertise to our peer. - */ -void -sctp_update_rcv_mss (sctp_connection_t * sctp_conn) -{ - sctp_conn->smallest_PMTU = DEFAULT_A_RWND; /* TODO find our iface MTU */ - sctp_conn->a_rwnd = DEFAULT_A_RWND - sizeof (sctp_full_hdr_t); - sctp_conn->rcv_opts.a_rwnd = sctp_conn->a_rwnd; - sctp_conn->rcv_a_rwnd = sctp_conn->a_rwnd; /* This will be updated by our congestion algos */ -} - -void -sctp_init_mss (sctp_connection_t * sctp_conn) -{ - SCTP_DBG ("CONN_INDEX = %u", - sctp_conn->sub_conn[MAIN_SCTP_SUB_CONN_IDX].connection.c_index); - - u16 default_a_rwnd = 536; - sctp_update_rcv_mss (sctp_conn); - - /* TODO cache mss and consider PMTU discovery */ - sctp_conn->snd_a_rwnd = - clib_min (sctp_conn->rcv_opts.a_rwnd, sctp_conn->a_rwnd); - if (sctp_conn->snd_a_rwnd < sizeof (sctp_full_hdr_t)) - { - SCTP_ADV_DBG ("sctp_conn->snd_a_rwnd < sizeof(sctp_full_hdr_t)"); - /* Assume that at least the min default mss works */ - sctp_conn->snd_a_rwnd = default_a_rwnd; - sctp_conn->rcv_opts.a_rwnd = default_a_rwnd; - } - - ASSERT (sctp_conn->snd_a_rwnd > sizeof (sctp_full_hdr_t)); + sctp_conn->remote_initial_tsn = 0x0; + sctp_conn->last_rcvd_tsn = sctp_conn->remote_initial_tsn; } always_inline sctp_connection_t * @@ -384,6 +368,8 @@ sctp_connection_open (transport_endpoint_t * rmt) clib_spinlock_lock_if_init (&tm->half_open_lock); sctp_conn = sctp_half_open_connection_new (thread_id); + sctp_conn->sub_conn[idx].PMTU = + vnet_sw_interface_get_mtu (vnet_get_main (), rmt->sw_if_index, VLIB_TX); transport_connection_t *trans_conn = &sctp_conn->sub_conn[idx].connection; ip_copy (&trans_conn->rmt_ip, &rmt->ip, rmt->is_ip4); @@ -462,7 +448,8 @@ sctp_check_outstanding_data_chunks (sctp_connection_t * sctp_conn) sctp_conn->sub_conn[i].enqueue_state != SCTP_ERROR_ENQUEUED) { SCTP_DBG_OUTPUT - ("Connection %u has still DATA to be enqueued inboud / outboud"); + ("Connection %u has still DATA to be enqueued inboud / outboud", + sctp_conn->sub_conn[i].connection.c_index); return 1; } @@ -485,6 +472,7 @@ void sctp_session_close (u32 conn_index, u32 thread_index) { ASSERT (thread_index == 0); + sctp_connection_t *sctp_conn; sctp_conn = sctp_connection_get (conn_index, thread_index); if (sctp_conn != NULL) @@ -506,80 +494,40 @@ sctp_session_cleanup (u32 conn_index, u32 thread_index) } /** - * Update snd_mss to reflect the effective segment size that we can send + * Compute maximum segment size for session layer. */ -void -sctp_update_snd_mss (sctp_connection_t * sctp_conn) -{ - /* The overhead for the sctp_header_t and sctp_chunks_common_hdr_t - * (the sum equals to sctp_full_hdr_t) is already taken into account - * for the sctp_conn->a_rwnd computation. - * So let's not account it again here. - */ - sctp_conn->snd_hdr_length = - sizeof (sctp_payload_data_chunk_t) - sizeof (sctp_full_hdr_t); - sctp_conn->snd_a_rwnd = - clib_min (sctp_conn->a_rwnd, - sctp_conn->rcv_opts.a_rwnd) - sctp_conn->snd_hdr_length; - - SCTP_DBG ("sctp_conn->snd_a_rwnd = %u, sctp_conn->snd_hdr_length = %u ", - sctp_conn->snd_a_rwnd, sctp_conn->snd_hdr_length); - - ASSERT (sctp_conn->snd_a_rwnd > 0); -} - u16 sctp_session_send_mss (transport_connection_t * trans_conn) { - SCTP_DBG ("CONN_INDEX: %u", trans_conn->c_index); - sctp_connection_t *sctp_conn = sctp_get_connection_from_transport (trans_conn); - if (trans_conn == NULL) - { - SCTP_DBG ("trans_conn == NULL"); - return 0; - } - if (sctp_conn == NULL) { SCTP_DBG ("sctp_conn == NULL"); return 0; } - /* Ensure snd_mss does accurately reflect the amount of data we can push - * in a segment. This also makes sure that options are updated according to - * the current state of the connection. */ - sctp_update_snd_mss (sctp_conn); - return sctp_conn->snd_a_rwnd; + update_cwnd (sctp_conn); + update_smallest_pmtu_idx (sctp_conn); + + return sctp_conn->sub_conn[sctp_conn->smallest_PMTU_idx].cwnd; } u16 sctp_snd_space (sctp_connection_t * sctp_conn) { - /* TODO: This requires a real implementation */ - if (sctp_conn == NULL) - { - SCTP_DBG ("sctp_conn == NULL"); - return 0; - } - - if (sctp_conn->state != SCTP_STATE_ESTABLISHED) - { - SCTP_DBG_STATE_MACHINE - ("Trying to send DATA while not in SCTP_STATE_ESTABLISHED"); - return 0; - } - - return sctp_conn->snd_a_rwnd; + /* Finally, let's subtract the DATA chunk headers overhead */ + return sctp_conn->sub_conn[sctp_conn->smallest_PMTU_idx].cwnd - + sizeof (sctp_payload_data_chunk_t) - sizeof (sctp_full_hdr_t); } +/** + * Compute TX window session is allowed to fill. + */ u32 sctp_session_send_space (transport_connection_t * trans_conn) { - SCTP_DBG ("CONN_INDEX: %u", trans_conn->c_index); - sctp_connection_t *sctp_conn = sctp_get_connection_from_transport (trans_conn); @@ -610,13 +558,25 @@ sctp_session_get_listener (u32 listener_index) u8 * format_sctp_session (u8 * s, va_list * args) { - return NULL; + u32 tci = va_arg (*args, u32); + u32 thread_index = va_arg (*args, u32); + u32 verbose = va_arg (*args, u32); + sctp_connection_t *tc; + + tc = sctp_connection_get (tci, thread_index); + if (tc) + s = format (s, "%U", format_sctp_connection, tc, verbose); + else + s = format (s, "empty\n"); + return s; } u8 * format_sctp_listener_session (u8 * s, va_list * args) { - return NULL; + u32 tci = va_arg (*args, u32); + sctp_connection_t *tc = sctp_listener_get (tci); + return format (s, "%U", format_sctp_connection_id, tc); } void @@ -849,7 +809,6 @@ const static transport_proto_vft_t sctp_proto = { .push_header = sctp_push_header, .send_mss = sctp_session_send_mss, .send_space = sctp_session_send_space, - .tx_fifo_offset = NULL, //sctp_session_tx_fifo_offset, .update_time = sctp_update_time, .get_connection = sctp_session_get_transport, .get_listener = sctp_session_get_listener, -- cgit 1.2.3-korg