d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
198f362e32
freebsd-skq/sys/netinet/sctp_cc_functions.c

2374 lines
69 KiB
C
Raw Normal View History

- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*-
* Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved.
Use consistent text at the begining of the files. MFC after: 3 days
2012-05-23 11:26:28 +00:00
* Copyright (c) 2008-2012, by Randall Stewart. All rights reserved.
* Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved.
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* a) Redistributions of source code must retain the above copyright notice,
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
* this list of conditions and the following disclaimer.
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
*
* b) Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
* the documentation and/or other materials provided with the distribution.
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
*
* c) Neither the name of Cisco Systems, Inc. nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
Use consistent text at the begining of the files. MFC after: 3 days
2012-05-23 11:26:28 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
#include <netinet/sctp_os.h>
#include <netinet/sctp_var.h>
#include <netinet/sctp_sysctl.h>
#include <netinet/sctp_pcb.h>
#include <netinet/sctp_header.h>
#include <netinet/sctputil.h>
#include <netinet/sctp_output.h>
#include <netinet/sctp_input.h>
#include <netinet/sctp_indata.h>
#include <netinet/sctp_uio.h>
#include <netinet/sctp_timer.h>
#include <netinet/sctp_auth.h>
#include <netinet/sctp_asconf.h>
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
#include <netinet/sctp_dtrace_declare.h>
Make the initial congestion window configurable via sysctl. MFC after: 2 weeks.
2010-09-17 18:53:07 +00:00
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
#define SHIFT_MPTCP_MULTI_N 40
#define SHIFT_MPTCP_MULTI_Z 16
#define SHIFT_MPTCP_MULTI 8
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
static void
sctp_enforce_cwnd_limit(struct sctp_association *assoc, struct sctp_nets *net)
{
if ((assoc->max_cwnd > 0) &&
(net->cwnd > assoc->max_cwnd) &&
(net->cwnd > (net->mtu - sizeof(struct sctphdr)))) {
net->cwnd = assoc->max_cwnd;
if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) {
net->cwnd = net->mtu - sizeof(struct sctphdr);
}
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
{
Make the initial congestion window configurable via sysctl. MFC after: 2 weeks.
2010-09-17 18:53:07 +00:00
struct sctp_association *assoc;
uint32_t cwnd_in_mtu;
assoc = &stcb->asoc;
cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd);
Provide a possibility to configure the inital congestion window to the value defined in RFC 4960. MFC after: 3 months.
2010-12-22 19:04:14 +00:00
if (cwnd_in_mtu == 0) {
/* Using 0 means that the value of RFC 4960 is used. */
net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
} else {
/*
* We take the minimum of the burst limit and the initial
* congestion window.
*/
if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst))
cwnd_in_mtu = assoc->max_burst;
net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu;
}
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
(stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
/* In case of resource pooling initialize appropriately */
net->cwnd /= assoc->numnets;
if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) {
net->cwnd = net->mtu - sizeof(struct sctphdr);
}
}
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(assoc, net);
Make the initial congestion window configurable via sysctl. MFC after: 2 weeks.
2010-09-17 18:53:07 +00:00
net->ssthresh = assoc->peers_rwnd;
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
SDT_PROBE(sctp, cwnd, net, init,
stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
0, net->cwnd);
Make the initial congestion window configurable via sysctl. MFC after: 2 weeks.
2010-09-17 18:53:07 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) &
(SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_cwnd_update_after_fr(struct sctp_tcb *stcb,
struct sctp_association *asoc)
{
struct sctp_nets *net;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
uint32_t t_ssthresh, t_cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
uint64_t t_ucwnd_sbw;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
/* MT FIXME: Don't compute this over and over again */
t_ssthresh = 0;
t_cwnd = 0;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
t_ucwnd_sbw = 0;
if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) ||
(asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
t_ssthresh += net->ssthresh;
t_cwnd += net->cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
if (net->lastsa > 0) {
t_ucwnd_sbw += (uint64_t) net->cwnd / (uint64_t) net->lastsa;
}
}
if (t_ucwnd_sbw == 0) {
t_ucwnd_sbw = 1;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
}
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*-
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* (net->fast_retran_loss_recovery == 0)))
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if ((asoc->fast_retran_loss_recovery == 0) ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
(asoc->sctp_cmt_on_off > 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* out of a RFC2582 Fast recovery window? */
if (net->net_ack > 0) {
/*
* per section 7.2.3, are there any
* destinations that had a fast retransmit
* to them. If so what we need to do is
* adjust ssthresh and cwnd.
*/
struct sctp_tmit_chunk *lchk;
int old_cwnd = net->cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) ||
(asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) {
if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) {
net->ssthresh = (uint32_t) (((uint64_t) 4 *
(uint64_t) net->mtu *
(uint64_t) net->ssthresh) /
(uint64_t) t_ssthresh);
}
if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2) {
uint32_t srtt;
srtt = net->lastsa;
/*
* lastsa>>3; we don't need
* to devide ...
*/
if (srtt == 0) {
srtt = 1;
}
/*
* Short Version => Equal to
* Contel Version MBe
*/
net->ssthresh = (uint32_t) (((uint64_t) 4 *
(uint64_t) net->mtu *
(uint64_t) net->cwnd) /
((uint64_t) srtt *
t_ucwnd_sbw));
/* INCREASE FACTOR */ ;
}
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
if ((net->cwnd > t_cwnd / 2) &&
(net->ssthresh < net->cwnd - t_cwnd / 2)) {
net->ssthresh = net->cwnd - t_cwnd / 2;
}
if (net->ssthresh < net->mtu) {
net->ssthresh = net->mtu;
}
} else {
net->ssthresh = net->cwnd / 2;
if (net->ssthresh < (net->mtu * 2)) {
net->ssthresh = 2 * net->mtu;
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
net->cwnd = net->ssthresh;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(asoc, net);
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
SDT_PROBE(sctp, cwnd, net, fr,
stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
old_cwnd, net->cwnd);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
SCTP_CWND_LOG_FROM_FR);
}
lchk = TAILQ_FIRST(&asoc->send_queue);
net->partial_bytes_acked = 0;
/* Turn on fast recovery window */
asoc->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
asoc->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
/*
* CMT fast recovery -- per destination
* recovery variable.
*/
net->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
net->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
Fix and cleanup the debug information. This has no user-visible changes. Thanks to Irene Ruengeler for proving a patch. MFC after: 3 days
2015-05-28 16:00:23 +00:00
stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_1);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_timer_start(SCTP_TIMER_TYPE_SEND,
stcb->sctp_ep, stcb, net);
}
} else if (net->net_ack > 0) {
/*
* Mark a peg that we WOULD have done a cwnd
* reduction but RFC2582 prevented this action.
*/
SCTP_STAT_INCR(sctps_fastretransinrtt);
}
}
}
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
/* Defines for instantaneous bw decisions */
#define SCTP_INST_LOOSING 1 /* Loosing to other flows */
#define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */
#define SCTP_INST_GAINING 3 /* Gaining, step down possible */
static int
cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw,
uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind)
{
uint64_t oth, probepoint;
probepoint = (((uint64_t) net->cwnd) << 32);
if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) {
/*
* rtt increased we don't update bw.. so we don't update the
* rtt either.
*/
/* Probe point 5 */
probepoint |= ((5 << 16) | 1);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) {
if (net->cc_mod.rtcc.last_step_state == 5)
net->cc_mod.rtcc.step_cnt++;
else
net->cc_mod.rtcc.step_cnt = 1;
net->cc_mod.rtcc.last_step_state = 5;
if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) ||
((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) &&
((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) {
/* Try a step down */
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
if (net->cwnd > (4 * net->mtu)) {
net->cwnd -= net->mtu;
net->cc_mod.rtcc.vol_reduce++;
} else {
net->cc_mod.rtcc.step_cnt = 0;
}
}
}
return (1);
}
if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) {
/*
* rtt decreased, there could be more room. we update both
* the bw and the rtt here to lock this in as a good step
* down.
*/
/* Probe point 6 */
probepoint |= ((6 << 16) | 0);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if (net->cc_mod.rtcc.steady_step) {
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
if ((net->cc_mod.rtcc.last_step_state == 5) &&
(net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) {
/* Step down worked */
net->cc_mod.rtcc.step_cnt = 0;
return (1);
} else {
net->cc_mod.rtcc.last_step_state = 6;
net->cc_mod.rtcc.step_cnt = 0;
}
}
net->cc_mod.rtcc.lbw = nbw;
net->cc_mod.rtcc.lbw_rtt = net->rtt;
net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
if (inst_ind == SCTP_INST_GAINING)
return (1);
else if (inst_ind == SCTP_INST_NEUTRAL)
return (1);
else
return (0);
}
/*
* Ok bw and rtt remained the same .. no update to any
*/
/* Probe point 7 */
probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) {
if (net->cc_mod.rtcc.last_step_state == 5)
net->cc_mod.rtcc.step_cnt++;
else
net->cc_mod.rtcc.step_cnt = 1;
net->cc_mod.rtcc.last_step_state = 5;
if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) ||
((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) &&
((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) {
/* Try a step down */
if (net->cwnd > (4 * net->mtu)) {
net->cwnd -= net->mtu;
net->cc_mod.rtcc.vol_reduce++;
return (1);
} else {
net->cc_mod.rtcc.step_cnt = 0;
}
}
}
if (inst_ind == SCTP_INST_GAINING)
return (1);
else if (inst_ind == SCTP_INST_NEUTRAL)
return (1);
else
return ((int)net->cc_mod.rtcc.ret_from_eq);
}
static int
cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset,
uint64_t vtag, uint8_t inst_ind)
{
uint64_t oth, probepoint;
/* Bandwidth decreased. */
probepoint = (((uint64_t) net->cwnd) << 32);
if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) {
/* rtt increased */
/* Did we add more */
if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) &&
(inst_ind != SCTP_INST_LOOSING)) {
/* We caused it maybe.. back off? */
/* PROBE POINT 1 */
probepoint |= ((1 << 16) | 1);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if (net->cc_mod.rtcc.ret_from_eq) {
/*
* Switch over to CA if we are less
* aggressive
*/
net->ssthresh = net->cwnd - 1;
net->partial_bytes_acked = 0;
}
return (1);
}
/* Probe point 2 */
probepoint |= ((2 << 16) | 0);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
/* Someone else - fight for more? */
if (net->cc_mod.rtcc.steady_step) {
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
/*
* Did we voluntarily give up some? if so take one
* back please
*/
if ((net->cc_mod.rtcc.vol_reduce) &&
(inst_ind != SCTP_INST_GAINING)) {
net->cwnd += net->mtu;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.vol_reduce--;
}
net->cc_mod.rtcc.last_step_state = 2;
net->cc_mod.rtcc.step_cnt = 0;
}
goto out_decision;
} else if (net->rtt < net->cc_mod.rtcc.lbw_rtt - rtt_offset) {
/* bw & rtt decreased */
/* Probe point 3 */
probepoint |= ((3 << 16) | 0);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if (net->cc_mod.rtcc.steady_step) {
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
if ((net->cc_mod.rtcc.vol_reduce) &&
(inst_ind != SCTP_INST_GAINING)) {
net->cwnd += net->mtu;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.vol_reduce--;
}
net->cc_mod.rtcc.last_step_state = 3;
net->cc_mod.rtcc.step_cnt = 0;
}
goto out_decision;
}
/* The bw decreased but rtt stayed the same */
/* Probe point 4 */
probepoint |= ((4 << 16) | 0);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if (net->cc_mod.rtcc.steady_step) {
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
if ((net->cc_mod.rtcc.vol_reduce) &&
(inst_ind != SCTP_INST_GAINING)) {
net->cwnd += net->mtu;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.vol_reduce--;
}
net->cc_mod.rtcc.last_step_state = 4;
net->cc_mod.rtcc.step_cnt = 0;
}
out_decision:
net->cc_mod.rtcc.lbw = nbw;
net->cc_mod.rtcc.lbw_rtt = net->rtt;
net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
if (inst_ind == SCTP_INST_GAINING) {
return (1);
} else {
return (0);
}
}
static int
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t vtag)
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
{
uint64_t oth, probepoint;
/*
* BW increased, so update and return 0, since all actions in our
* table say to do the normal CC update. Note that we pay no
* attention to the inst_ind since our overall sum is increasing.
*/
/* PROBE POINT 0 */
probepoint = (((uint64_t) net->cwnd) << 32);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
probepoint);
if (net->cc_mod.rtcc.steady_step) {
oth = net->cc_mod.rtcc.vol_reduce;
oth <<= 16;
oth |= net->cc_mod.rtcc.step_cnt;
oth <<= 16;
oth |= net->cc_mod.rtcc.last_step_state;
SDT_PROBE(sctp, cwnd, net, rttstep,
vtag,
((net->cc_mod.rtcc.lbw << 32) | nbw),
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
oth,
probepoint);
net->cc_mod.rtcc.last_step_state = 0;
net->cc_mod.rtcc.step_cnt = 0;
net->cc_mod.rtcc.vol_reduce = 0;
}
net->cc_mod.rtcc.lbw = nbw;
net->cc_mod.rtcc.lbw_rtt = net->rtt;
net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
return (0);
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
/* RTCC Algoritm to limit growth of cwnd, return
* true if you want to NOT allow cwnd growth
*/
static int
cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw)
{
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
uint64_t bw_offset, rtt_offset;
uint64_t probepoint, rtt, vtag;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
uint64_t bytes_for_this_rtt, inst_bw;
uint64_t div, inst_off;
int bw_shift;
uint8_t inst_ind;
int ret;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
/*-
* Here we need to see if we want
* to limit cwnd growth due to increase
* in overall rtt but no increase in bw.
* We use the following table to figure
* out what we should do. When we return
* 0, cc update goes on as planned. If we
* return 1, then no cc update happens and cwnd
* stays where it is at.
* ----------------------------------
* BW | RTT | Action
* *********************************
* INC | INC | return 0
* ----------------------------------
* INC | SAME | return 0
* ----------------------------------
* INC | DECR | return 0
* ----------------------------------
* SAME | INC | return 1
* ----------------------------------
* SAME | SAME | return 1
* ----------------------------------
* SAME | DECR | return 0
* ----------------------------------
* DECR | INC | return 0 or 1 based on if we caused.
* ----------------------------------
* DECR | SAME | return 0
* ----------------------------------
* DECR | DECR | return 0
* ----------------------------------
*
* We are a bit fuzz on what an increase or
* decrease is. For BW it is the same if
* it did not change within 1/64th. For
* RTT it stayed the same if it did not
* change within 1/32nd
*/
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw);
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
rtt = stcb->asoc.my_vtag;
vtag = (rtt << 32) | (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport);
probepoint = (((uint64_t) net->cwnd) << 32);
rtt = net->rtt;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
if (net->cc_mod.rtcc.rtt_set_this_sack) {
net->cc_mod.rtcc.rtt_set_this_sack = 0;
bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc;
net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes;
if (net->rtt) {
div = net->rtt / 1000;
if (div) {
inst_bw = bytes_for_this_rtt / div;
inst_off = inst_bw >> bw_shift;
if (inst_bw > nbw)
inst_ind = SCTP_INST_GAINING;
else if ((inst_bw + inst_off) < nbw)
inst_ind = SCTP_INST_LOOSING;
else
inst_ind = SCTP_INST_NEUTRAL;
probepoint |= ((0xb << 16) | inst_ind);
} else {
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
inst_ind = net->cc_mod.rtcc.last_inst_ind;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
inst_bw = bytes_for_this_rtt / (uint64_t) (net->rtt);
/* Can't determine do not change */
probepoint |= ((0xc << 16) | inst_ind);
}
} else {
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
inst_ind = net->cc_mod.rtcc.last_inst_ind;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
inst_bw = bytes_for_this_rtt;
/* Can't determine do not change */
probepoint |= ((0xd << 16) | inst_ind);
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
((nbw << 32) | inst_bw),
((net->cc_mod.rtcc.lbw_rtt << 32) | rtt),
net->flight_size,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
probepoint);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
} else {
/* No rtt measurement, use last one */
inst_ind = net->cc_mod.rtcc.last_inst_ind;
}
bw_offset = net->cc_mod.rtcc.lbw >> bw_shift;
if (nbw > net->cc_mod.rtcc.lbw + bw_offset) {
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
ret = cc_bw_increase(stcb, net, nbw, vtag);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
goto out;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt);
if (nbw < net->cc_mod.rtcc.lbw - bw_offset) {
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind);
goto out;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
/*
* If we reach here then we are in a situation where the bw stayed
* the same.
*/
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind);
out:
net->cc_mod.rtcc.last_inst_ind = inst_ind;
return (ret);
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb,
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
struct sctp_association *asoc,
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
int accum_moved, int reneged_all SCTP_UNUSED, int will_exit, int use_rtcc)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
struct sctp_nets *net;
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
int old_cwnd;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
uint32_t t_ssthresh, t_cwnd, incr;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
uint64_t t_ucwnd_sbw;
uint64_t t_path_mptcp;
uint64_t mptcp_like_alpha;
uint32_t srtt;
uint64_t max_path;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
/* MT FIXME: Don't compute this over and over again */
t_ssthresh = 0;
t_cwnd = 0;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
t_ucwnd_sbw = 0;
t_path_mptcp = 0;
mptcp_like_alpha = 1;
if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
(stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) ||
(stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) {
max_path = 0;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
t_ssthresh += net->ssthresh;
t_cwnd += net->cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
/* lastsa>>3; we don't need to devide ... */
srtt = net->lastsa;
if (srtt > 0) {
uint64_t tmp;
t_ucwnd_sbw += (uint64_t) net->cwnd / (uint64_t) srtt;
t_path_mptcp += (((uint64_t) net->cwnd) << SHIFT_MPTCP_MULTI_Z) /
(((uint64_t) net->mtu) * (uint64_t) srtt);
tmp = (((uint64_t) net->cwnd) << SHIFT_MPTCP_MULTI_N) /
((uint64_t) net->mtu * (uint64_t) (srtt * srtt));
if (tmp > max_path) {
max_path = tmp;
}
}
}
if (t_path_mptcp > 0) {
mptcp_like_alpha = max_path / (t_path_mptcp * t_path_mptcp);
} else {
mptcp_like_alpha = 1;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
}
}
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
if (t_ssthresh == 0) {
t_ssthresh = 1;
}
if (t_ucwnd_sbw == 0) {
t_ucwnd_sbw = 1;
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/******************************/
/* update cwnd and Early FR */
/******************************/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code. Need to debug.
*/
if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->will_exit_fast_recovery = 1;
}
}
#endif
/* if nothing was acked on this destination skip it */
if (net->net_ack == 0) {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
}
continue;
}
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code
*/
/*
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery
* && net->will_exit_fast_recovery == 0) { @@@ Do something
* } else if (sctp_cmt_on_off == 0 &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* asoc->fast_retran_loss_recovery && will_exit == 0) {
*/
#endif
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (asoc->fast_retran_loss_recovery &&
(will_exit == 0) &&
(asoc->sctp_cmt_on_off == 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* If we are in loss recovery we skip any cwnd
* update
*/
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
return;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
/*
* Did any measurements go on for this network?
*/
if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) {
uint64_t nbw;
/*
* At this point our bw_bytes has been updated by
* incoming sack information.
*
* But our bw may not yet be set.
*
*/
if ((net->cc_mod.rtcc.new_tot_time / 1000) > 0) {
nbw = net->cc_mod.rtcc.bw_bytes / (net->cc_mod.rtcc.new_tot_time / 1000);
} else {
nbw = net->cc_mod.rtcc.bw_bytes;
}
if (net->cc_mod.rtcc.lbw) {
if (cc_bw_limit(stcb, net, nbw)) {
/* Hold here, no update */
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
continue;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
} else {
uint64_t vtag, probepoint;
probepoint = (((uint64_t) net->cwnd) << 32);
probepoint |= ((0xa << 16) | 0);
vtag = (net->rtt << 32) |
(((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) |
(stcb->rport);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
nbw,
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
probepoint);
net->cc_mod.rtcc.lbw = nbw;
net->cc_mod.rtcc.lbw_rtt = net->rtt;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
if (net->cc_mod.rtcc.rtt_set_this_sack) {
net->cc_mod.rtcc.rtt_set_this_sack = 0;
net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes;
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* CMT: CUC algorithm. Update cwnd if pseudo-cumack has
* moved.
*/
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (accum_moved ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* If the cumulative ack moved we can proceed */
if (net->cwnd <= net->ssthresh) {
/* We are in slow start */
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
if (net->flight_size + net->net_ack >= net->cwnd) {
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
uint32_t limit;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
old_cwnd = net->cwnd;
switch (asoc->sctp_cmt_on_off) {
case SCTP_CMT_RPV1:
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
limit = (uint32_t) (((uint64_t) net->mtu *
(uint64_t) SCTP_BASE_SYSCTL(sctp_L2_abc_variable) *
(uint64_t) net->ssthresh) /
(uint64_t) t_ssthresh);
incr = (uint32_t) (((uint64_t) net->net_ack *
(uint64_t) net->ssthresh) /
(uint64_t) t_ssthresh);
if (incr > limit) {
incr = limit;
}
if (incr == 0) {
incr = 1;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
break;
case SCTP_CMT_RPV2:
/*
* lastsa>>3; we don't need
* to divide ...
*/
srtt = net->lastsa;
if (srtt == 0) {
srtt = 1;
}
limit = (uint32_t) (((uint64_t) net->mtu *
(uint64_t) SCTP_BASE_SYSCTL(sctp_L2_abc_variable) *
(uint64_t) net->cwnd) /
((uint64_t) srtt * t_ucwnd_sbw));
/* INCREASE FACTOR */
incr = (uint32_t) (((uint64_t) net->net_ack *
(uint64_t) net->cwnd) /
((uint64_t) srtt * t_ucwnd_sbw));
/* INCREASE FACTOR */
if (incr > limit) {
incr = limit;
}
if (incr == 0) {
incr = 1;
}
break;
case SCTP_CMT_MPTCP:
limit = (uint32_t) (((uint64_t) net->mtu *
mptcp_like_alpha *
(uint64_t) SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >>
SHIFT_MPTCP_MULTI);
incr = (uint32_t) (((uint64_t) net->net_ack *
mptcp_like_alpha) >>
SHIFT_MPTCP_MULTI);
if (incr > limit) {
incr = limit;
}
if (incr > net->net_ack) {
incr = net->net_ack;
}
if (incr > net->mtu) {
incr = net->mtu;
}
break;
default:
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
incr = net->net_ack;
if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) {
incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
break;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
net->cwnd += incr;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(asoc, net);
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, incr,
SCTP_CWND_LOG_FROM_SS);
}
SDT_PROBE(sctp, cwnd, net, ack,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_SS);
}
}
} else {
/* We are in congestion avoidance */
- Define changes in sctp.h - Bug in CA that does not get us incrementing the PBA properly which made us more conservative. - comment updated in sctp_input.c - memsets added before we log - added arg to hmac id's MFC after: 2 weeks
2008-05-20 09:51:36 +00:00
/*
* Add to pba
*/
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
net->partial_bytes_acked += net->net_ack;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
- Define changes in sctp.h - Bug in CA that does not get us incrementing the PBA properly which made us more conservative. - comment updated in sctp_input.c - memsets added before we log - added arg to hmac id's MFC after: 2 weeks
2008-05-20 09:51:36 +00:00
if ((net->flight_size + net->net_ack >= net->cwnd) &&
(net->partial_bytes_acked >= net->cwnd)) {
net->partial_bytes_acked -= net->cwnd;
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
old_cwnd = net->cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
switch (asoc->sctp_cmt_on_off) {
case SCTP_CMT_RPV1:
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
incr = (uint32_t) (((uint64_t) net->mtu *
(uint64_t) net->ssthresh) /
(uint64_t) t_ssthresh);
if (incr == 0) {
incr = 1;
}
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
break;
case SCTP_CMT_RPV2:
/*
* lastsa>>3; we don't need
* to divide ...
*/
srtt = net->lastsa;
if (srtt == 0) {
srtt = 1;
}
incr = (uint32_t) ((uint64_t) net->mtu *
(uint64_t) net->cwnd /
((uint64_t) srtt *
t_ucwnd_sbw));
/* INCREASE FACTOR */
if (incr == 0) {
incr = 1;
}
break;
case SCTP_CMT_MPTCP:
incr = (uint32_t) ((mptcp_like_alpha *
(uint64_t) net->cwnd) >>
SHIFT_MPTCP_MULTI);
if (incr > net->mtu) {
incr = net->mtu;
}
break;
default:
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
incr = net->mtu;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
break;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
}
net->cwnd += incr;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(asoc, net);
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
SDT_PROBE(sctp, cwnd, net, ack,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Define changes in sctp.h - Bug in CA that does not get us incrementing the PBA properly which made us more conservative. - comment updated in sctp_input.c - memsets added before we log - added arg to hmac id's MFC after: 2 weeks
2008-05-20 09:51:36 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_FROM_CA);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_CA);
}
}
}
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_NO_CUMACK);
}
}
}
}
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
static void
sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net)
{
int old_cwnd;
old_cwnd = net->cwnd;
net->cwnd = net->mtu;
SDT_PROBE(sctp, cwnd, net, ack,
stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
old_cwnd, net->cwnd);
SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
Using %p in a format string requires a void *. MFC after: 10 days
2012-09-05 18:52:01 +00:00
(void *)net, net->cwnd);
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
int old_cwnd = net->cwnd;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
uint32_t t_ssthresh, t_cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
uint64_t t_ucwnd_sbw;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
/* MT FIXME: Don't compute this over and over again */
t_ssthresh = 0;
t_cwnd = 0;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
(stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
struct sctp_nets *lnet;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
uint32_t srtt;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
t_ucwnd_sbw = 0;
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) {
t_ssthresh += lnet->ssthresh;
t_cwnd += lnet->cwnd;
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
srtt = lnet->lastsa;
/* lastsa>>3; we don't need to divide ... */
if (srtt > 0) {
t_ucwnd_sbw += (uint64_t) lnet->cwnd / (uint64_t) srtt;
}
}
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
if (t_ssthresh < 1) {
t_ssthresh = 1;
}
Implement Resource Pooling V2 and an MPTCP like congestion control. Based on a patch received from Martin Becke. MFC after: 2 weeks.
2011-05-04 21:27:05 +00:00
if (t_ucwnd_sbw < 1) {
t_ucwnd_sbw = 1;
}
if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) {
net->ssthresh = (uint32_t) (((uint64_t) 4 *
(uint64_t) net->mtu *
(uint64_t) net->ssthresh) /
(uint64_t) t_ssthresh);
} else {
uint64_t cc_delta;
srtt = net->lastsa;
/* lastsa>>3; we don't need to divide ... */
if (srtt == 0) {
srtt = 1;
}
cc_delta = t_ucwnd_sbw * (uint64_t) srtt / 2;
if (cc_delta < t_cwnd) {
net->ssthresh = (uint32_t) ((uint64_t) t_cwnd - cc_delta);
} else {
net->ssthresh = net->mtu;
}
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
}
if ((net->cwnd > t_cwnd / 2) &&
(net->ssthresh < net->cwnd - t_cwnd / 2)) {
net->ssthresh = net->cwnd - t_cwnd / 2;
}
if (net->ssthresh < net->mtu) {
net->ssthresh = net->mtu;
}
} else {
net->ssthresh = max(net->cwnd / 2, 4 * net->mtu);
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->cwnd = net->mtu;
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
net->partial_bytes_acked = 0;
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
SDT_PROBE(sctp, cwnd, net, to,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
}
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net,
int in_window, int num_pkt_lost, int use_rtcc)
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
{
int old_cwnd = net->cwnd;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) {
/* Data center Congestion Control */
if (in_window == 0) {
/*
* Go to CA with the cwnd at the point we sent the
* TSN that was marked with a CE.
*/
if (net->ecn_prev_cwnd < net->cwnd) {
/* Restore to prev cwnd */
net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost);
} else {
/* Just cut in 1/2 */
net->cwnd /= 2;
}
/* Drop to CA */
net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
}
} else {
/*
* Further tuning down required over the drastic
* orginal cut
*/
net->ssthresh -= (net->mtu * num_pkt_lost);
net->cwnd -= (net->mtu * num_pkt_lost);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
}
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
SCTP_STAT_INCR(sctps_ecnereducedcwnd);
} else {
if (in_window == 0) {
SCTP_STAT_INCR(sctps_ecnereducedcwnd);
net->ssthresh = net->cwnd / 2;
if (net->ssthresh < net->mtu) {
net->ssthresh = net->mtu;
/*
* here back off the timer as well, to slow
* us down
*/
net->RTO <<= 1;
}
net->cwnd = net->ssthresh;
SDT_PROBE(sctp, cwnd, net, ecn,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
}
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
}
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb,
struct sctp_nets *net, struct sctp_pktdrop_chunk *cp,
uint32_t * bottle_bw, uint32_t * on_queue)
{
uint32_t bw_avail;
1) Move per John Baldwin to mp_maxid 2) Some signed/unsigned errors found by Mac OS compiler (from Michael) 3) a couple of copyright updates on the effected files. MFC after: 3 months
2011-02-03 19:22:21 +00:00
unsigned int incr;
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
int old_cwnd = net->cwnd;
/* get bottle neck bw */
*bottle_bw = ntohl(cp->bottle_bw);
/* and whats on queue */
*on_queue = ntohl(cp->current_onq);
/*
* adjust the on-queue if our flight is more it could be that the
* router has not yet gotten data "in-flight" to it
*/
Small cleanup which addresses a warning regaring the truncation of a 64-bit entity to a 32-bit entity. This issue was reported by Peter Kasting from Google. MFC after: 3 days
2014-09-16 13:48:46 +00:00
if (*on_queue < net->flight_size) {
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
*on_queue = net->flight_size;
Small cleanup which addresses a warning regaring the truncation of a 64-bit entity to a 32-bit entity. This issue was reported by Peter Kasting from Google. MFC after: 3 days
2014-09-16 13:48:46 +00:00
}
/* rtt is measured in micro seconds, bottle_bw in bytes per second */
bw_avail = (uint32_t) (((uint64_t) (*bottle_bw) * net->rtt) / (uint64_t) 1000000);
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
if (bw_avail > *bottle_bw) {
/*
* Cap the growth to no more than the bottle neck. This can
* happen as RTT slides up due to queues. It also means if
* you have more than a 1 second RTT with a empty queue you
* will be limited to the bottle_bw per second no matter if
* other points have 1/2 the RTT and you could get more
* out...
*/
bw_avail = *bottle_bw;
}
if (*on_queue > bw_avail) {
/*
* No room for anything else don't allow anything else to be
* "added to the fire".
*/
int seg_inflight, seg_onqueue, my_portion;
net->partial_bytes_acked = 0;
/* how much are we over queue size? */
incr = *on_queue - bw_avail;
if (stcb->asoc.seen_a_sack_this_pkt) {
/*
* undo any cwnd adjustment that the sack might have
* made
*/
net->cwnd = net->prev_cwnd;
}
/* Now how much of that is mine? */
seg_inflight = net->flight_size / net->mtu;
seg_onqueue = *on_queue / net->mtu;
my_portion = (incr * seg_inflight) / seg_onqueue;
/* Have I made an adjustment already */
if (net->cwnd > net->flight_size) {
/*
* for this flight I made an adjustment we need to
* decrease the portion by a share our previous
* adjustment.
*/
int diff_adj;
diff_adj = net->cwnd - net->flight_size;
if (diff_adj > my_portion)
my_portion = 0;
else
my_portion -= diff_adj;
}
/*
* back down to the previous cwnd (assume we have had a sack
* before this packet). minus what ever portion of the
* overage is my fault.
*/
net->cwnd -= my_portion;
/* we will NOT back down more than 1 MTU */
if (net->cwnd <= net->mtu) {
net->cwnd = net->mtu;
}
/* force into CA */
net->ssthresh = net->cwnd - 1;
} else {
/*
* Take 1/4 of the space left or max burst up .. whichever
* is less.
*/
Change infrastructure for SCTP_MAX_BURST to allow compliance with the latest socket API ID. Especially it can be disabled. Full compliance needs changing the structure used in the socket option. Since this breaks the API, it will be a seperate commit which will not be MFCed to stable/8. MFC after: 3 months.
2011-01-26 19:49:03 +00:00
incr = (bw_avail - *on_queue) >> 2;
if ((stcb->asoc.max_burst > 0) &&
(stcb->asoc.max_burst * net->mtu < incr)) {
incr = stcb->asoc.max_burst * net->mtu;
}
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
net->cwnd += incr;
}
if (net->cwnd > bw_avail) {
/* We can't exceed the pipe size */
net->cwnd = bw_avail;
}
if (net->cwnd < net->mtu) {
/* We always have 1 MTU */
net->cwnd = net->mtu;
}
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
if (net->cwnd - old_cwnd != 0) {
/* log only changes */
Adds new dtrace for cwnd functions and lay's groundwork for future dtrace points (rwnd flightsize etc). MFC after: 2 months
2010-11-25 13:39:55 +00:00
SDT_PROBE(sctp, cwnd, net, pd,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
SCTP_CWND_LOG_FROM_SAT);
}
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
sctp_cwnd_update_after_output(struct sctp_tcb *stcb,
struct sctp_nets *net, int burst_limit)
{
int old_cwnd = net->cwnd;
if (net->ssthresh < net->cwnd)
net->ssthresh = net->cwnd;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
if (burst_limit) {
net->cwnd = (net->flight_size + (burst_limit * net->mtu));
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
SDT_PROBE(sctp, cwnd, net, bl,
stcb->asoc.my_vtag,
((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
net,
old_cwnd, net->cwnd);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST);
}
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
}
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
static void
sctp_cwnd_update_after_sack(struct sctp_tcb *stcb,
struct sctp_association *asoc,
int accum_moved, int reneged_all, int will_exit)
{
/* Passing a zero argument in last disables the rtcc algoritm */
sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0);
}
static void
sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net,
int in_window, int num_pkt_lost)
{
/* Passing a zero argument in last disables the rtcc algoritm */
sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0);
}
/* Here starts the RTCCVAR type CC invented by RRS which
* is a slight mod to RFC2581. We reuse a common routine or
* two since these algoritms are so close and need to
* remain the same.
*/
static void
sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net,
int in_window, int num_pkt_lost)
{
sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1);
}
static
void
sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net,
struct sctp_tmit_chunk *tp1)
{
net->cc_mod.rtcc.bw_bytes += tp1->send_size;
}
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
struct sctp_nets *net)
{
if (net->cc_mod.rtcc.tls_needs_set > 0) {
/* We had a bw measurment going on */
struct timeval ltls;
SCTP_GETPTIME_TIMEVAL(&ltls);
timevalsub(&ltls, &net->cc_mod.rtcc.tls);
net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec;
}
}
static void
sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb,
struct sctp_nets *net)
{
uint64_t vtag, probepoint;
if (net->cc_mod.rtcc.lbw) {
/* Clear the old bw.. we went to 0 in-flight */
vtag = (net->rtt << 32) | (((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) |
(stcb->rport);
probepoint = (((uint64_t) net->cwnd) << 32);
/* Probe point 8 */
probepoint |= ((8 << 16) | 0);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
((net->cc_mod.rtcc.lbw << 32) | 0),
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
net->flight_size,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
probepoint);
net->cc_mod.rtcc.lbw_rtt = 0;
net->cc_mod.rtcc.cwnd_at_bw_set = 0;
net->cc_mod.rtcc.lbw = 0;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0;
net->cc_mod.rtcc.vol_reduce = 0;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
net->cc_mod.rtcc.bw_tot_time = 0;
net->cc_mod.rtcc.bw_bytes = 0;
net->cc_mod.rtcc.tls_needs_set = 0;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
if (net->cc_mod.rtcc.steady_step) {
net->cc_mod.rtcc.vol_reduce = 0;
net->cc_mod.rtcc.step_cnt = 0;
net->cc_mod.rtcc.last_step_state = 0;
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
if (net->cc_mod.rtcc.ret_from_eq) {
/* less aggressive one - reset cwnd too */
uint32_t cwnd_in_mtu, cwnd;
cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd);
if (cwnd_in_mtu == 0) {
/*
* Using 0 means that the value of RFC 4960
* is used.
*/
cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
} else {
/*
* We take the minimum of the burst limit
* and the initial congestion window.
*/
if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst))
cwnd_in_mtu = stcb->asoc.max_burst;
cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu;
}
if (net->cwnd > cwnd) {
/*
* Only set if we are not a timeout (i.e.
* down to 1 mtu)
*/
net->cwnd = cwnd;
}
}
}
}
static void
sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb,
struct sctp_nets *net)
{
uint64_t vtag, probepoint;
sctp_set_initial_cc_param(stcb, net);
stcb->asoc.use_precise_time = 1;
probepoint = (((uint64_t) net->cwnd) << 32);
probepoint |= ((9 << 16) | 0);
vtag = (net->rtt << 32) |
(((uint32_t) (stcb->sctp_ep->sctp_lport)) << 16) |
(stcb->rport);
SDT_PROBE(sctp, cwnd, net, rttvar,
vtag,
0,
0,
0,
probepoint);
net->cc_mod.rtcc.lbw_rtt = 0;
net->cc_mod.rtcc.cwnd_at_bw_set = 0;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.vol_reduce = 0;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
net->cc_mod.rtcc.lbw = 0;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.vol_reduce = 0;
net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
net->cc_mod.rtcc.bw_tot_time = 0;
net->cc_mod.rtcc.bw_bytes = 0;
net->cc_mod.rtcc.tls_needs_set = 0;
net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret);
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step);
net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn);
net->cc_mod.rtcc.step_cnt = 0;
net->cc_mod.rtcc.last_step_state = 0;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
}
static int
sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget,
struct sctp_cc_option *cc_opt)
{
struct sctp_nets *net;
if (setorget == 1) {
/* a set */
if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) {
if ((cc_opt->aid_value.assoc_value != 0) &&
(cc_opt->aid_value.assoc_value != 1)) {
return (EINVAL);
}
TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value;
}
} else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) {
if ((cc_opt->aid_value.assoc_value != 0) &&
(cc_opt->aid_value.assoc_value != 1)) {
return (EINVAL);
}
TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value;
}
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
} else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) {
TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value;
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
} else {
return (EINVAL);
}
} else {
/* a get */
if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) {
net = TAILQ_FIRST(&stcb->asoc.nets);
if (net == NULL) {
return (EFAULT);
}
cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq;
} else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) {
net = TAILQ_FIRST(&stcb->asoc.nets);
if (net == NULL) {
return (EFAULT);
}
cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn;
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
} else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) {
net = TAILQ_FIRST(&stcb->asoc.nets);
if (net == NULL) {
return (EFAULT);
}
cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step;
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
} else {
return (EINVAL);
}
}
return (0);
}
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
struct sctp_nets *net)
{
if (net->cc_mod.rtcc.tls_needs_set == 0) {
SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls);
net->cc_mod.rtcc.tls_needs_set = 2;
}
}
static void
sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb,
struct sctp_association *asoc,
int accum_moved, int reneged_all, int will_exit)
{
/* Passing a one argument at the last enables the rtcc algoritm */
sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1);
}
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED,
struct sctp_nets *net,
struct timeval *now SCTP_UNUSED)
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
{
net->cc_mod.rtcc.rtt_set_this_sack = 1;
}
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
/* Here starts Sally Floyds HS-TCP */
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
struct sctp_hs_raise_drop {
int32_t cwnd;
int32_t increase;
int32_t drop_percent;
};
#define SCTP_HS_TABLE_SIZE 73
struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = {
{38, 1, 50}, /* 0 */
{118, 2, 44}, /* 1 */
{221, 3, 41}, /* 2 */
{347, 4, 38}, /* 3 */
{495, 5, 37}, /* 4 */
{663, 6, 35}, /* 5 */
{851, 7, 34}, /* 6 */
{1058, 8, 33}, /* 7 */
{1284, 9, 32}, /* 8 */
{1529, 10, 31}, /* 9 */
{1793, 11, 30}, /* 10 */
{2076, 12, 29}, /* 11 */
{2378, 13, 28}, /* 12 */
{2699, 14, 28}, /* 13 */
{3039, 15, 27}, /* 14 */
{3399, 16, 27}, /* 15 */
{3778, 17, 26}, /* 16 */
{4177, 18, 26}, /* 17 */
{4596, 19, 25}, /* 18 */
{5036, 20, 25}, /* 19 */
{5497, 21, 24}, /* 20 */
{5979, 22, 24}, /* 21 */
{6483, 23, 23}, /* 22 */
{7009, 24, 23}, /* 23 */
{7558, 25, 22}, /* 24 */
{8130, 26, 22}, /* 25 */
{8726, 27, 22}, /* 26 */
{9346, 28, 21}, /* 27 */
{9991, 29, 21}, /* 28 */
{10661, 30, 21}, /* 29 */
{11358, 31, 20}, /* 30 */
{12082, 32, 20}, /* 31 */
{12834, 33, 20}, /* 32 */
{13614, 34, 19}, /* 33 */
{14424, 35, 19}, /* 34 */
{15265, 36, 19}, /* 35 */
{16137, 37, 19}, /* 36 */
{17042, 38, 18}, /* 37 */
{17981, 39, 18}, /* 38 */
{18955, 40, 18}, /* 39 */
{19965, 41, 17}, /* 40 */
{21013, 42, 17}, /* 41 */
{22101, 43, 17}, /* 42 */
{23230, 44, 17}, /* 43 */
{24402, 45, 16}, /* 44 */
{25618, 46, 16}, /* 45 */
{26881, 47, 16}, /* 46 */
{28193, 48, 16}, /* 47 */
{29557, 49, 15}, /* 48 */
{30975, 50, 15}, /* 49 */
{32450, 51, 15}, /* 50 */
{33986, 52, 15}, /* 51 */
{35586, 53, 14}, /* 52 */
{37253, 54, 14}, /* 53 */
{38992, 55, 14}, /* 54 */
{40808, 56, 14}, /* 55 */
{42707, 57, 13}, /* 56 */
{44694, 58, 13}, /* 57 */
{46776, 59, 13}, /* 58 */
{48961, 60, 13}, /* 59 */
{51258, 61, 13}, /* 60 */
{53677, 62, 12}, /* 61 */
{56230, 63, 12}, /* 62 */
{58932, 64, 12}, /* 63 */
{61799, 65, 12}, /* 64 */
{64851, 66, 11}, /* 65 */
{68113, 67, 11}, /* 66 */
{71617, 68, 11}, /* 67 */
{75401, 69, 10}, /* 68 */
{79517, 70, 10}, /* 69 */
{84035, 71, 10}, /* 70 */
{89053, 72, 10}, /* 71 */
{94717, 73, 9} /* 72 */
};
static void
sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net)
{
int cur_val, i, indx, incr;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
int old_cwnd = net->cwnd;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
cur_val = net->cwnd >> 10;
indx = SCTP_HS_TABLE_SIZE - 1;
Use SCTP_PRINTF() instead of printf() in all SCTP sources. MFC after: 3 days
2012-05-04 09:27:00 +00:00
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
if (cur_val < sctp_cwnd_adjust[0].cwnd) {
/* normal mode */
if (net->net_ack > net->mtu) {
net->cwnd += net->mtu;
} else {
net->cwnd += net->net_ack;
}
} else {
for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) {
if (cur_val < sctp_cwnd_adjust[i].cwnd) {
indx = i;
break;
}
}
net->last_hs_used = indx;
incr = ((sctp_cwnd_adjust[indx].increase) << 10);
net->cwnd += incr;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
}
sctp_enforce_cwnd_limit(&stcb->asoc, net);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SS);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
}
static void
sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net)
{
int cur_val, i, indx;
int old_cwnd = net->cwnd;
cur_val = net->cwnd >> 10;
if (cur_val < sctp_cwnd_adjust[0].cwnd) {
/* normal mode */
net->ssthresh = net->cwnd / 2;
if (net->ssthresh < (net->mtu * 2)) {
net->ssthresh = 2 * net->mtu;
}
net->cwnd = net->ssthresh;
} else {
/* drop by the proper amount */
net->ssthresh = net->cwnd - (int)((net->cwnd / 100) *
sctp_cwnd_adjust[net->last_hs_used].drop_percent);
net->cwnd = net->ssthresh;
/* now where are we */
indx = net->last_hs_used;
cur_val = net->cwnd >> 10;
/* reset where we are in the table */
if (cur_val < sctp_cwnd_adjust[0].cwnd) {
/* feel out of hs */
net->last_hs_used = 0;
} else {
for (i = indx; i >= 1; i--) {
if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) {
break;
}
}
net->last_hs_used = indx;
}
}
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR);
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb,
struct sctp_association *asoc)
{
struct sctp_nets *net;
/*
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* (net->fast_retran_loss_recovery == 0)))
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if ((asoc->fast_retran_loss_recovery == 0) ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
(asoc->sctp_cmt_on_off > 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* out of a RFC2582 Fast recovery window? */
if (net->net_ack > 0) {
/*
* per section 7.2.3, are there any
* destinations that had a fast retransmit
* to them. If so what we need to do is
* adjust ssthresh and cwnd.
*/
struct sctp_tmit_chunk *lchk;
sctp_hs_cwnd_decrease(stcb, net);
lchk = TAILQ_FIRST(&asoc->send_queue);
net->partial_bytes_acked = 0;
/* Turn on fast recovery window */
asoc->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
asoc->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
/*
* CMT fast recovery -- per destination
* recovery variable.
*/
net->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
net->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
Fix and cleanup the debug information. This has no user-visible changes. Thanks to Irene Ruengeler for proving a patch. MFC after: 3 days
2015-05-28 16:00:23 +00:00
stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_2);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_timer_start(SCTP_TIMER_TYPE_SEND,
stcb->sctp_ep, stcb, net);
}
} else if (net->net_ack > 0) {
/*
* Mark a peg that we WOULD have done a cwnd
* reduction but RFC2582 prevented this action.
*/
SCTP_STAT_INCR(sctps_fastretransinrtt);
}
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb,
struct sctp_association *asoc,
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
int accum_moved, int reneged_all SCTP_UNUSED, int will_exit)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
struct sctp_nets *net;
/******************************/
/* update cwnd and Early FR */
/******************************/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code. Need to debug.
*/
if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->will_exit_fast_recovery = 1;
}
}
#endif
/* if nothing was acked on this destination skip it */
if (net->net_ack == 0) {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
}
continue;
}
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code
*/
/*
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery
* && net->will_exit_fast_recovery == 0) { @@@ Do something
* } else if (sctp_cmt_on_off == 0 &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* asoc->fast_retran_loss_recovery && will_exit == 0) {
*/
#endif
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (asoc->fast_retran_loss_recovery &&
(will_exit == 0) &&
(asoc->sctp_cmt_on_off == 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* If we are in loss recovery we skip any cwnd
* update
*/
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
return;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
/*
* CMT: CUC algorithm. Update cwnd if pseudo-cumack has
* moved.
*/
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (accum_moved ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* If the cumulative ack moved we can proceed */
if (net->cwnd <= net->ssthresh) {
/* We are in slow start */
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
if (net->flight_size + net->net_ack >= net->cwnd) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_hs_cwnd_increase(stcb, net);
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_SS);
}
}
} else {
/* We are in congestion avoidance */
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
net->partial_bytes_acked += net->net_ack;
if ((net->flight_size + net->net_ack >= net->cwnd) &&
(net->partial_bytes_acked >= net->cwnd)) {
net->partial_bytes_acked -= net->cwnd;
net->cwnd += net->mtu;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(asoc, net);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_FROM_CA);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_CA);
}
}
}
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_NO_CUMACK);
}
}
}
}
/*
* H-TCP congestion control. The algorithm is detailed in:
* R.N.Shorten, D.J.Leith:
* "H-TCP: TCP for high-speed and long-distance networks"
* Proc. PFLDnet, Argonne, 2004.
* http://www.hamilton.ie/net/htcp3.pdf
*/
static int use_rtt_scaling = 1;
static int use_bandwidth_switch = 1;
static inline int
between(uint32_t seq1, uint32_t seq2, uint32_t seq3)
{
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (seq3 - seq2 >= seq1 - seq2);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
static inline uint32_t
htcp_cong_time(struct htcp *ca)
{
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (sctp_get_tick_count() - ca->last_cong);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
static inline uint32_t
htcp_ccount(struct htcp *ca)
{
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (htcp_cong_time(ca) / ca->minRTT);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
static inline void
htcp_reset(struct htcp *ca)
{
ca->undo_last_cong = ca->last_cong;
ca->undo_maxRTT = ca->maxRTT;
ca->undo_old_maxB = ca->old_maxB;
- added pre-checks to the bindx call. - use proper tick gathering macro instead of ticks directly. - Placed reasonable boundaries on sets that a user can do that are converted to ticks from ms. - Fix CMT_PF to always check to be sure CMT is on. - Fix ticks use of CMT_PF. - put back code to allow asconfs to be queued while INITs are in flight and before the assoc is established. - During window probes, an ack'd packet might be left with the window probe mark on it causing it to be retransmitted. Change so that the flight decrease macro clears the window_probe mark. - Additional logging flight size/reading and ASOC LOG. This is only enabled if you manually insert things into opt_sctp.h since its a set of debug code only. - Found an interesting SMP race in the way data was appended which could cause a reader to lose a part of a message, had to reorder when we marked the message was complete to after the data was appended. - bug in ADD-IP for the subset bound socket case when the peer has only one address - fix ASCONF implicit success/error handling case - proper support of jails in Freebsd 6> - copy out the timeval for the 64 bit sparc world on cookie-echo alignment error crashes without this). Approved by: re(Ken Smith)
2007-07-17 20:58:26 +00:00
ca->last_cong = sctp_get_tick_count();
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
#ifdef SCTP_NOT_USED
static uint32_t
htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net)
{
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong;
net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT;
net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB;
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->cc_mod.htcp_ca.beta) * net->mtu));
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
#endif
static inline void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
measure_rtt(struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
* Fix several bugs where the scaled versions of srtt and rttvar where used incorrectly. * Use appropriate variable names for RTO instead of RTT. MFC after: 3 months.
2011-02-24 22:58:15 +00:00
uint32_t srtt = net->lastsa >> SCTP_RTT_SHIFT;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* keep track of minimum RTT seen so far, minRTT is zero at first */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT)
net->cc_mod.htcp_ca.minRTT = srtt;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* max RTT */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) {
if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT)
net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT;
if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT + MSEC_TO_TICKS(20))
net->cc_mod.htcp_ca.maxRTT = srtt;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
}
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
measure_achieved_throughput(struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
- added pre-checks to the bindx call. - use proper tick gathering macro instead of ticks directly. - Placed reasonable boundaries on sets that a user can do that are converted to ticks from ms. - Fix CMT_PF to always check to be sure CMT is on. - Fix ticks use of CMT_PF. - put back code to allow asconfs to be queued while INITs are in flight and before the assoc is established. - During window probes, an ack'd packet might be left with the window probe mark on it causing it to be retransmitted. Change so that the flight decrease macro clears the window_probe mark. - Additional logging flight size/reading and ASOC LOG. This is only enabled if you manually insert things into opt_sctp.h since its a set of debug code only. - Found an interesting SMP race in the way data was appended which could cause a reader to lose a part of a message, had to reorder when we marked the message was complete to after the data was appended. - bug in ADD-IP for the subset bound socket case when the peer has only one address - fix ASCONF implicit success/error handling case - proper support of jails in Freebsd 6> - copy out the timeval for the 64 bit sparc world on cookie-echo alignment error crashes without this). Approved by: re(Ken Smith)
2007-07-17 20:58:26 +00:00
uint32_t now = sctp_get_tick_count();
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
if (net->fast_retran_ip == 0)
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.bytes_acked = net->net_ack;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
if (!use_bandwidth_switch)
return;
/* achieved throughput calculations */
/* JRS - not 100% sure of this statement */
if (net->fast_retran_ip == 1) {
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.bytecount = 0;
net->cc_mod.htcp_ca.lasttime = now;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
return;
}
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.bytecount += net->net_ack;
Get rid of a gcc'ism. MFC after: 10 days
2012-09-06 07:03:56 +00:00
if ((net->cc_mod.htcp_ca.bytecount >= net->cwnd - (((net->cc_mod.htcp_ca.alpha >> 7) ? (net->cc_mod.htcp_ca.alpha >> 7) : 1) * net->mtu)) &&
(now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) &&
(net->cc_mod.htcp_ca.minRTT > 0)) {
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount / net->mtu * hz / (now - net->cc_mod.htcp_ca.lasttime);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* just after backoff */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
} else {
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.Bi = (3 * net->cc_mod.htcp_ca.Bi + cur_Bi) / 4;
if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB)
net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi;
if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB)
net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.bytecount = 0;
net->cc_mod.htcp_ca.lasttime = now;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
}
static inline void
htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT)
{
if (use_bandwidth_switch) {
uint32_t maxB = ca->maxB;
uint32_t old_maxB = ca->old_maxB;
ca->old_maxB = ca->maxB;
if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) {
ca->beta = BETA_MIN;
ca->modeswitch = 0;
return;
}
}
if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) {
ca->beta = (minRTT << 7) / maxRTT;
if (ca->beta < BETA_MIN)
ca->beta = BETA_MIN;
else if (ca->beta > BETA_MAX)
ca->beta = BETA_MAX;
} else {
ca->beta = BETA_MIN;
ca->modeswitch = 1;
}
}
static inline void
htcp_alpha_update(struct htcp *ca)
{
uint32_t minRTT = ca->minRTT;
uint32_t factor = 1;
uint32_t diff = htcp_cong_time(ca);
if (diff > (uint32_t) hz) {
diff -= hz;
factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz;
}
if (use_rtt_scaling && minRTT) {
uint32_t scale = (hz << 3) / (10 * minRTT);
scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to
* interval [0.5,10]<<3 */
factor = (factor << 3) / scale;
if (!factor)
factor = 1;
}
ca->alpha = 2 * factor * ((1 << 7) - ca->beta);
if (!ca->alpha)
ca->alpha = ALPHA_BASE;
}
/* After we have the rtt data to calculate beta, we'd still prefer to wait one
* rtt before we adjust our beta to ensure we are working from a consistent
* data.
*
* This function should be called when we hit a congestion event since only at
* that point do we really have a real sense of maxRTT (the queues en route
* were getting just too full now).
*/
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
htcp_param_update(struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
uint32_t minRTT = net->cc_mod.htcp_ca.minRTT;
uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT);
htcp_alpha_update(&net->cc_mod.htcp_ca);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* add slowly fading memory for maxRTT to accommodate routing
* changes etc
*/
if (minRTT > 0 && maxRTT > minRTT)
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
static uint32_t
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
htcp_recalc_ssthresh(struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
htcp_param_update(net);
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (max(((net->cwnd / net->mtu * net->cc_mod.htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu));
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
static void
htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net)
{
/*-
* How to handle these functions?
* if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question.
* return;
*/
if (net->cwnd <= net->ssthresh) {
/* We are in slow start */
if (net->flight_size + net->net_ack >= net->cwnd) {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) {
net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable));
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_FROM_SS);
}
} else {
net->cwnd += net->net_ack;
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_FROM_SS);
}
}
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_SS);
}
}
} else {
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
measure_rtt(net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* In dangerous area, increase slowly. In theory this is
* net->cwnd += alpha / net->cwnd
*/
/* What is snd_cwnd_cnt?? */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
if (((net->partial_bytes_acked / net->mtu * net->cc_mod.htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*-
* Does SCTP have a cwnd clamp?
* if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS).
*/
net->cwnd += net->mtu;
net->partial_bytes_acked = 0;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
htcp_alpha_update(&net->cc_mod.htcp_ca);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_FROM_CA);
}
} else {
net->partial_bytes_acked += net->net_ack;
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->net_ack,
SCTP_CWND_LOG_NOADV_CA);
}
}
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
net->cc_mod.htcp_ca.bytes_acked = net->mtu;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
}
#ifdef SCTP_NOT_USED
/* Lower bound on congestion window. */
static uint32_t
htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net)
{
Address issues found by clang. While there, fix also some style issues. MFC after: 3 months.
2011-12-27 10:16:24 +00:00
return (net->ssthresh);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
#endif
static void
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
htcp_init(struct sctp_nets *net)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp));
net->cc_mod.htcp_ca.alpha = ALPHA_BASE;
net->cc_mod.htcp_ca.beta = BETA_MIN;
net->cc_mod.htcp_ca.bytes_acked = net->mtu;
net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count();
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
{
/*
* We take the max of the burst limit times a MTU or the
* INITIAL_CWND. We then limit this to 4 MTU's of sending.
*/
net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
net->ssthresh = stcb->asoc.peers_rwnd;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
htcp_init(net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb,
struct sctp_association *asoc,
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
int accum_moved, int reneged_all SCTP_UNUSED, int will_exit)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
struct sctp_nets *net;
/******************************/
/* update cwnd and Early FR */
/******************************/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code. Need to debug.
*/
if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
Add support for resource pooling to CMT. An original version of the patch was developed by Martin Becke and Thomas Dreibholz. MFC after: 3 months
2011-01-16 10:02:46 +00:00
if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
SCTP_TSN_GE(net->pseudo_cumack, net->fast_recovery_tsn)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->will_exit_fast_recovery = 1;
}
}
#endif
/* if nothing was acked on this destination skip it */
if (net->net_ack == 0) {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
}
continue;
}
#ifdef JANA_CMT_FAST_RECOVERY
/*
* CMT fast recovery code
*/
/*
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery
* && net->will_exit_fast_recovery == 0) { @@@ Do something
* } else if (sctp_cmt_on_off == 0 &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* asoc->fast_retran_loss_recovery && will_exit == 0) {
*/
#endif
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (asoc->fast_retran_loss_recovery &&
will_exit == 0 &&
(asoc->sctp_cmt_on_off == 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/*
* If we are in loss recovery we skip any cwnd
* update
*/
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
return;
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
/*
* CMT: CUC algorithm. Update cwnd if pseudo-cumack has
* moved.
*/
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if (accum_moved ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
htcp_cong_avoid(stcb, net);
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
measure_achieved_throughput(net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
} else {
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->mtu,
SCTP_CWND_LOG_NO_CUMACK);
}
}
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb,
struct sctp_association *asoc)
{
struct sctp_nets *net;
/*
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
* CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
* (net->fast_retran_loss_recovery == 0)))
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
Fix the switching on/off of CMT using sysctl and socket option. Fix the switching on/off of PF and NR-SACKs using sysctl. Add minor improvement in handling malloc failures. Improve the address checks when sending. MFC after: 4 weeks
2010-08-28 17:59:51 +00:00
if ((asoc->fast_retran_loss_recovery == 0) ||
Improve plausibility check in sctp_handle_sack(). Allow cmt_on_off to support values 0 (no CMT), 1 (CMT), and 2 (CMT/RP). MFC after: 3 months.
2010-12-22 17:59:38 +00:00
(asoc->sctp_cmt_on_off > 0)) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
/* out of a RFC2582 Fast recovery window? */
if (net->net_ack > 0) {
/*
* per section 7.2.3, are there any
* destinations that had a fast retransmit
* to them. If so what we need to do is
* adjust ssthresh and cwnd.
*/
struct sctp_tmit_chunk *lchk;
int old_cwnd = net->cwnd;
/* JRS - reset as if state were changed */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
htcp_reset(&net->cc_mod.htcp_ca);
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
net->ssthresh = htcp_recalc_ssthresh(net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->cwnd = net->ssthresh;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(asoc, net);
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
SCTP_CWND_LOG_FROM_FR);
}
lchk = TAILQ_FIRST(&asoc->send_queue);
net->partial_bytes_acked = 0;
/* Turn on fast recovery window */
asoc->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
asoc->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
/*
* CMT fast recovery -- per destination
* recovery variable.
*/
net->fast_retran_loss_recovery = 1;
if (lchk == NULL) {
/* Mark end of the window */
net->fast_recovery_tsn = asoc->sending_seq - 1;
} else {
net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
}
sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
Fix and cleanup the debug information. This has no user-visible changes. Thanks to Irene Ruengeler for proving a patch. MFC after: 3 days
2015-05-28 16:00:23 +00:00
stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_CC_FUNCTIONS + SCTP_LOC_3);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_timer_start(SCTP_TIMER_TYPE_SEND,
stcb->sctp_ep, stcb, net);
}
} else if (net->net_ack > 0) {
/*
* Mark a peg that we WOULD have done a cwnd
* reduction but RFC2582 prevented this action.
*/
SCTP_STAT_INCR(sctps_fastretransinrtt);
}
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb,
struct sctp_nets *net)
{
int old_cwnd = net->cwnd;
/* JRS - reset as if the state were being changed to timeout */
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
htcp_reset(&net->cc_mod.htcp_ca);
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
net->ssthresh = htcp_recalc_ssthresh(net);
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
net->cwnd = net->mtu;
- Adds support for the multi-asconf (From Kozuka-san) - Adds some prepwork (Not all yet) for vimage in particular support the delete the sctppcbinfo.xx structs. There is still a leak in here if it were to be called plus we stil need the regrouping (From Me and Michael Tuexen) - Adds support for UDP tunneling. For BSD there is no socket yet setup so its disabled, but major argument changes are in here to emcompass the passing of the port number (zero when you don't have a udp tunnel, the default for BSD). Will add some hooks in UDP here shortly (discussed with Robert) that will allow easy tunneling. (Mainly from Peter Lei and Michael Tuexen with some BSD work from me :-D) - Some ease for windows, evidently leave is reserved by their compile move label leave: -> out: MFC after: 1 week
2008-05-20 13:47:46 +00:00
net->partial_bytes_acked = 0;
- Macro-izes the packed declaration in all headers. - Vimage prep - these are major restructures to move all global variables to be accessed via a macro or two. The variables all go into a single structure. - Asconf address addition tweaks (add_or_del Interfaces) - Fix rwnd calcualtion to be more conservative. - Support SACK_IMMEDIATE flag to skip delayed sack by demand of peer. - Comment updates in the sack mapping calculations - Invarients panic added. - Pre-support for UDP tunneling (we can do this on MAC but will need added support from UDP to get a "pipe" of UDP packets in. - clear trace buffer sysctl added when local tracing on. Note the majority of this huge patch is all the vimage prep stuff :-)
2008-06-14 07:58:05 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
static void
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb,
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED)
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
{
int old_cwnd;
old_cwnd = net->cwnd;
/* JRS - reset hctp as if state changed */
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
if (in_window == 0) {
Improvements to CC modules: 1) Add four new points that allow you to get more information to cc algo's 2) Fix the case where user changes module on a existing TCB, in such a case, the initialization module needs to be called on all nets. 3) Move htcp_cc structure to a union that other modules can use. 4) Add 5th point for get/set socket options for cc_module specific options MFC after: 2 months
2011-02-26 15:23:46 +00:00
htcp_reset(&net->cc_mod.htcp_ca);
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
SCTP_STAT_INCR(sctps_ecnereducedcwnd);
Fix unused parameter warnings. While there, fix some whitespace issues. MFC after: 3 months.
2011-12-17 19:21:40 +00:00
net->ssthresh = htcp_recalc_ssthresh(net);
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
if (net->ssthresh < net->mtu) {
net->ssthresh = net->mtu;
/* here back off the timer as well, to slow us down */
net->RTO <<= 1;
}
net->cwnd = net->ssthresh;
Add a SCTP socket option to limit the cwnd for each path. MFC after: 1 month
2015-03-10 19:49:25 +00:00
sctp_enforce_cwnd_limit(&stcb->asoc, net);
More ECN fixes: 1) We now remove ECN-Nonce since it will no longer continue as a I-D 2) Eliminate last_tsn_echo, this tied us to an assoc not the net and thus we were not doing m-homing on the ECN-Echo senders side right. 3) Increment the count going out even if the TSN in lower in the pending ECN-Echo, this way the receiver knows exactly how many packets were marked even with network re-ordering 4) Fix so we DO NOT stop doing delayed sack if a ECN Echo is in queue MFC after: 1 month
2011-01-31 11:50:11 +00:00
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
}
- Modular congestion control, with RFC2581 being the default. - CMT_PF states added (w/sysctl to turn the PF version on) - sctp_input.c had a missing incr of cookie case when the auth was bad. This meant a free was called without an increment to refcnt, added increment like rest of code. - There was a case, unlikely, when the scope of the destination changed (this is a TSNH case). In that case, it would not free the alloc'ed asoc (in sctp_input.c). - When listed addresses found a colliding cookie/Init, then the collided upon tcb was not unlocked in sctp_pcb.c - Add error checking on arguments of sctp_sendx(3) to prevent it from referencing a NULL pointer. - Fix an error return of sctp_sendx(3), it was returing ENOMEM not -1. - Get assoc id was changed to use the sanctified socket api method for getting a assoc id (PEER_ADDR_INFO instead of PEER_ADDR_PARAMS). - Fix it so a peeled off socket will get a proper error return if it trys to send to a different address then it is connected to. - Fix so that select_a_stream can avoid an endless loop that could hang a caller. - time_entered (state set time) was not being set in all cases to the time we went established. Approved by: re(ken smith)
2007-07-14 09:36:28 +00:00
}
}
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
struct sctp_cc_functions sctp_cc_functions[] = {
{
.sctp_set_initial_cc_param = sctp_set_initial_cc_param,
.sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack,
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
.sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
.sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr,
.sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
.sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo,
.sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
.sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
},
{
.sctp_set_initial_cc_param = sctp_set_initial_cc_param,
.sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack,
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
.sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
.sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr,
.sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
.sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo,
.sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
.sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
},
{
.sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param,
.sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack,
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
.sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
.sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr,
.sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout,
.sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo,
.sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
.sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
},
{
.sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param,
.sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack,
The result of a joint work between rrs@ and myself at the IETF: * Decouple the path supervision using a separate HB timer per path. * Add support for potentially failed state. * Bring back RTO.min to 1 second. * Accept packets on IP-addresses already announced via an ASCONF * While there: do some cleanups. Approved by: re@ MFC after: 2 months.
2011-08-03 20:21:00 +00:00
.sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
Adds a new Congestion Control that helps reduce the RTT that a flow will build up in buffers in transit. It is a slight modification to RFC2581 but is more friendly i.e. less aggressive. MFC after: 3 months
2011-03-01 00:37:46 +00:00
.sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr,
.sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
.sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo,
.sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
.sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
.sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted,
.sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged,
.sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins,
.sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack,
Tunes and fixes the new DC-CC to seem to hit the right mix. Still may need some tweaks but it appears to almost not give away too much to an RFC2581 flow, but can really minimize the amount of buffers used in the net. MFC after: 3 months
2011-03-08 11:58:25 +00:00
.sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option,
.sctp_rtt_calculated = sctp_rtt_rtcc_calculated
Cleanup the management of CC functions. MFC after: 3 months.
2011-01-19 22:10:35 +00:00
}
};
Reference in New Issue Copy Permalink