freebsd-dev/sys/netinet/tcp_sack.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1071 lines
32 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
* The Regents of the University of California.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University 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 REGENTS 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 REGENTS 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.
*
* @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95
*/
/*-
* @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995
*
* NRL grants permission for redistribution and use in source and binary
* forms, with or without modification, of the software and documentation
* created at NRL provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgements:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* This product includes software developed at the Information
* Technology Division, US Naval Research Laboratory.
* 4. Neither the name of the NRL nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL 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 NRL 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.
*
* The views and conclusions contained in the software and documentation
* are those of the authors and should not be interpreted as representing
* official policies, either expressed or implied, of the US Naval
* Research Laboratory (NRL).
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet.h"
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/proc.h> /* for proc0 declaration */
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/syslog.h>
#include <sys/systm.h>
#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
#include <vm/uma.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/route.h>
#include <net/vnet.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/in_var.h>
#include <netinet/in_pcb.h>
#include <netinet/ip_var.h>
#include <netinet/ip6.h>
#include <netinet/icmp6.h>
#include <netinet6/nd6.h>
#include <netinet6/ip6_var.h>
#include <netinet6/in6_pcb.h>
#include <netinet/tcp.h>
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_seq.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_var.h>
#include <netinet/tcpip.h>
#include <netinet/cc/cc.h>
#include <machine/in_cksum.h>
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
VNET_DECLARE(struct uma_zone *, sack_hole_zone);
#define V_sack_hole_zone VNET(sack_hole_zone)
SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"TCP SACK");
VNET_DEFINE(int, tcp_do_sack) = 1;
SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(tcp_do_sack), 0,
"Enable/Disable TCP SACK support");
VNET_DEFINE(int, tcp_do_newsack) = 1;
SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, revised, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(tcp_do_newsack), 0,
"Use revised SACK loss recovery per RFC 6675");
VNET_DEFINE(int, tcp_sack_maxholes) = 128;
SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_VNET | CTLFLAG_RW,
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
&VNET_NAME(tcp_sack_maxholes), 0,
"Maximum number of TCP SACK holes allowed per connection");
VNET_DEFINE(int, tcp_sack_globalmaxholes) = 65536;
SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(tcp_sack_globalmaxholes), 0,
"Global maximum number of TCP SACK holes");
VNET_DEFINE(int, tcp_sack_globalholes) = 0;
SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_VNET | CTLFLAG_RD,
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
&VNET_NAME(tcp_sack_globalholes), 0,
"Global number of TCP SACK holes currently allocated");
int
tcp_dsack_block_exists(struct tcpcb *tp)
{
/* Return true if a DSACK block exists */
if (tp->rcv_numsacks == 0)
return (0);
if (SEQ_LEQ(tp->sackblks[0].end, tp->rcv_nxt))
return(1);
return (0);
}
/*
* This function will find overlaps with the currently stored sackblocks
* and add any overlap as a dsack block upfront
*/
void
tcp_update_dsack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
{
struct sackblk head_blk,mid_blk,saved_blks[MAX_SACK_BLKS];
int i, j, n, identical;
tcp_seq start, end;
INP_WLOCK_ASSERT(tptoinpcb(tp));
KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end"));
if (SEQ_LT(rcv_end, tp->rcv_nxt) ||
((rcv_end == tp->rcv_nxt) &&
(tp->rcv_numsacks > 0 ) &&
(tp->sackblks[0].end == tp->rcv_nxt))) {
saved_blks[0].start = rcv_start;
saved_blks[0].end = rcv_end;
} else {
saved_blks[0].start = saved_blks[0].end = 0;
}
head_blk.start = head_blk.end = 0;
mid_blk.start = rcv_start;
mid_blk.end = rcv_end;
identical = 0;
for (i = 0; i < tp->rcv_numsacks; i++) {
start = tp->sackblks[i].start;
end = tp->sackblks[i].end;
if (SEQ_LT(rcv_end, start)) {
/* pkt left to sack blk */
continue;
}
if (SEQ_GT(rcv_start, end)) {
/* pkt right to sack blk */
continue;
}
if (SEQ_GT(tp->rcv_nxt, end)) {
if ((SEQ_MAX(rcv_start, start) != SEQ_MIN(rcv_end, end)) &&
(SEQ_GT(head_blk.start, SEQ_MAX(rcv_start, start)) ||
(head_blk.start == head_blk.end))) {
head_blk.start = SEQ_MAX(rcv_start, start);
head_blk.end = SEQ_MIN(rcv_end, end);
}
continue;
}
if (((head_blk.start == head_blk.end) ||
SEQ_LT(start, head_blk.start)) &&
(SEQ_GT(end, rcv_start) &&
SEQ_LEQ(start, rcv_end))) {
head_blk.start = start;
head_blk.end = end;
}
mid_blk.start = SEQ_MIN(mid_blk.start, start);
mid_blk.end = SEQ_MAX(mid_blk.end, end);
if ((mid_blk.start == start) &&
(mid_blk.end == end))
identical = 1;
}
if (SEQ_LT(head_blk.start, head_blk.end)) {
/* store overlapping range */
saved_blks[0].start = SEQ_MAX(rcv_start, head_blk.start);
saved_blks[0].end = SEQ_MIN(rcv_end, head_blk.end);
}
n = 1;
/*
* Second, if not ACKed, store the SACK block that
* overlaps with the DSACK block unless it is identical
*/
if ((SEQ_LT(tp->rcv_nxt, mid_blk.end) &&
!((mid_blk.start == saved_blks[0].start) &&
(mid_blk.end == saved_blks[0].end))) ||
identical == 1) {
saved_blks[n].start = mid_blk.start;
saved_blks[n++].end = mid_blk.end;
}
for (j = 0; (j < tp->rcv_numsacks) && (n < MAX_SACK_BLKS); j++) {
if (((SEQ_LT(tp->sackblks[j].end, mid_blk.start) ||
SEQ_GT(tp->sackblks[j].start, mid_blk.end)) &&
(SEQ_GT(tp->sackblks[j].start, tp->rcv_nxt))))
saved_blks[n++] = tp->sackblks[j];
}
j = 0;
for (i = 0; i < n; i++) {
/* we can end up with a stale initial entry */
if (SEQ_LT(saved_blks[i].start, saved_blks[i].end)) {
tp->sackblks[j++] = saved_blks[i];
}
}
tp->rcv_numsacks = j;
}
/*
* This function is called upon receipt of new valid data (while not in
* header prediction mode), and it updates the ordered list of sacks.
*/
void
tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
{
/*
* First reported block MUST be the most recent one. Subsequent
* blocks SHOULD be in the order in which they arrived at the
* receiver. These two conditions make the implementation fully
* compliant with RFC 2018.
*/
struct sackblk head_blk, saved_blks[MAX_SACK_BLKS];
int num_head, num_saved, i;
INP_WLOCK_ASSERT(tptoinpcb(tp));
/* Check arguments. */
KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("rcv_start <= rcv_end"));
if ((rcv_start == rcv_end) &&
(tp->rcv_numsacks >= 1) &&
(rcv_end == tp->sackblks[0].end)) {
/* retaining DSACK block below rcv_nxt (todrop) */
head_blk = tp->sackblks[0];
} else {
/* SACK block for the received segment. */
head_blk.start = rcv_start;
head_blk.end = rcv_end;
}
/*
* Merge updated SACK blocks into head_blk, and save unchanged SACK
* blocks into saved_blks[]. num_saved will have the number of the
* saved SACK blocks.
*/
num_saved = 0;
for (i = 0; i < tp->rcv_numsacks; i++) {
tcp_seq start = tp->sackblks[i].start;
tcp_seq end = tp->sackblks[i].end;
if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) {
/*
* Discard this SACK block.
*/
} else if (SEQ_LEQ(head_blk.start, end) &&
SEQ_GEQ(head_blk.end, start)) {
/*
* Merge this SACK block into head_blk. This SACK
* block itself will be discarded.
*/
/*
* |-|
* |---| merge
*
* |-|
* |---| merge
*
* |-----|
* |-| DSACK smaller
*
* |-|
* |-----| DSACK smaller
*/
if (head_blk.start == end)
head_blk.start = start;
else if (head_blk.end == start)
head_blk.end = end;
else {
if (SEQ_LT(head_blk.start, start)) {
tcp_seq temp = start;
start = head_blk.start;
head_blk.start = temp;
}
if (SEQ_GT(head_blk.end, end)) {
tcp_seq temp = end;
end = head_blk.end;
head_blk.end = temp;
}
if ((head_blk.start != start) ||
(head_blk.end != end)) {
if ((num_saved >= 1) &&
SEQ_GEQ(saved_blks[num_saved-1].start, start) &&
SEQ_LEQ(saved_blks[num_saved-1].end, end))
num_saved--;
saved_blks[num_saved].start = start;
saved_blks[num_saved].end = end;
num_saved++;
}
}
} else {
/*
* This block supercedes the prior block
*/
if ((num_saved >= 1) &&
SEQ_GEQ(saved_blks[num_saved-1].start, start) &&
SEQ_LEQ(saved_blks[num_saved-1].end, end))
num_saved--;
/*
* Save this SACK block.
*/
saved_blks[num_saved].start = start;
saved_blks[num_saved].end = end;
num_saved++;
}
}
/*
* Update SACK list in tp->sackblks[].
*/
num_head = 0;
if (SEQ_LT(rcv_start, rcv_end)) {
/*
* The received data segment is an out-of-order segment. Put
* head_blk at the top of SACK list.
*/
tp->sackblks[0] = head_blk;
num_head = 1;
/*
* If the number of saved SACK blocks exceeds its limit,
* discard the last SACK block.
*/
if (num_saved >= MAX_SACK_BLKS)
num_saved--;
}
if ((rcv_start == rcv_end) &&
(rcv_start == tp->sackblks[0].end)) {
num_head = 1;
}
if (num_saved > 0) {
/*
* Copy the saved SACK blocks back.
*/
bcopy(saved_blks, &tp->sackblks[num_head],
sizeof(struct sackblk) * num_saved);
}
/* Save the number of SACK blocks. */
tp->rcv_numsacks = num_head + num_saved;
}
void
tcp_clean_dsack_blocks(struct tcpcb *tp)
{
struct sackblk saved_blks[MAX_SACK_BLKS];
int num_saved, i;
INP_WLOCK_ASSERT(tptoinpcb(tp));
/*
* Clean up any DSACK blocks that
* are in our queue of sack blocks.
*
*/
num_saved = 0;
for (i = 0; i < tp->rcv_numsacks; i++) {
tcp_seq start = tp->sackblks[i].start;
tcp_seq end = tp->sackblks[i].end;
if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) {
/*
* Discard this D-SACK block.
*/
continue;
}
/*
* Save this SACK block.
*/
saved_blks[num_saved].start = start;
saved_blks[num_saved].end = end;
num_saved++;
}
if (num_saved > 0) {
/*
* Copy the saved SACK blocks back.
*/
bcopy(saved_blks, &tp->sackblks[0],
sizeof(struct sackblk) * num_saved);
}
tp->rcv_numsacks = num_saved;
}
/*
* Delete all receiver-side SACK information.
*/
void
tcp_clean_sackreport(struct tcpcb *tp)
{
int i;
INP_WLOCK_ASSERT(tptoinpcb(tp));
tp->rcv_numsacks = 0;
for (i = 0; i < MAX_SACK_BLKS; i++)
tp->sackblks[i].start = tp->sackblks[i].end=0;
}
/*
* Allocate struct sackhole.
*/
static struct sackhole *
tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end)
{
struct sackhole *hole;
if (tp->snd_numholes >= V_tcp_sack_maxholes ||
V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) {
TCPSTAT_INC(tcps_sack_sboverflow);
return NULL;
}
hole = (struct sackhole *)uma_zalloc(V_sack_hole_zone, M_NOWAIT);
if (hole == NULL)
return NULL;
hole->start = start;
hole->end = end;
hole->rxmit = start;
tp->snd_numholes++;
atomic_add_int(&V_tcp_sack_globalholes, 1);
return hole;
}
/*
* Free struct sackhole.
*/
static void
tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole)
{
uma_zfree(V_sack_hole_zone, hole);
tp->snd_numholes--;
atomic_subtract_int(&V_tcp_sack_globalholes, 1);
KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0"));
KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0"));
}
/*
* Insert new SACK hole into scoreboard.
*/
static struct sackhole *
tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end,
struct sackhole *after)
{
struct sackhole *hole;
/* Allocate a new SACK hole. */
hole = tcp_sackhole_alloc(tp, start, end);
if (hole == NULL)
return NULL;
/* Insert the new SACK hole into scoreboard. */
if (after != NULL)
TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink);
else
TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink);
/* Update SACK hint. */
if (tp->sackhint.nexthole == NULL)
tp->sackhint.nexthole = hole;
return hole;
}
/*
* Remove SACK hole from scoreboard.
*/
static void
tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole)
{
/* Update SACK hint. */
if (tp->sackhint.nexthole == hole)
tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink);
/* Remove this SACK hole. */
TAILQ_REMOVE(&tp->snd_holes, hole, scblink);
/* Free this SACK hole. */
tcp_sackhole_free(tp, hole);
}
/*
* Process cumulative ACK and the TCP SACK option to update the scoreboard.
* tp->snd_holes is an ordered list of holes (oldest to newest, in terms of
* the sequence space).
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
* Returns 1 if incoming ACK has previously unknown SACK information,
* 0 otherwise.
*/
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
int
tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack)
{
struct sackhole *cur, *temp;
struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp;
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
int i, j, num_sack_blks, sack_changed;
int delivered_data, left_edge_delta;
INP_WLOCK_ASSERT(tptoinpcb(tp));
num_sack_blks = 0;
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
sack_changed = 0;
delivered_data = 0;
left_edge_delta = 0;
/*
* If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist,
* treat [SND.UNA, SEG.ACK) as if it is a SACK block.
* Account changes to SND.UNA always in delivered data.
*/
if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) {
left_edge_delta = th_ack - tp->snd_una;
sack_blocks[num_sack_blks].start = tp->snd_una;
sack_blocks[num_sack_blks++].end = th_ack;
/*
* Pulling snd_fack forward if we got here
* due to DSACK blocks
*/
if (SEQ_LT(tp->snd_fack, th_ack)) {
delivered_data += th_ack - tp->snd_una;
tp->snd_fack = th_ack;
sack_changed = 1;
}
}
/*
* Append received valid SACK blocks to sack_blocks[], but only if we
* received new blocks from the other side.
*/
if (to->to_flags & TOF_SACK) {
for (i = 0; i < to->to_nsacks; i++) {
bcopy((to->to_sacks + i * TCPOLEN_SACK),
&sack, sizeof(sack));
sack.start = ntohl(sack.start);
sack.end = ntohl(sack.end);
if (SEQ_GT(sack.end, sack.start) &&
SEQ_GT(sack.start, tp->snd_una) &&
SEQ_GT(sack.start, th_ack) &&
SEQ_LT(sack.start, tp->snd_max) &&
SEQ_GT(sack.end, tp->snd_una) &&
Calculate the correct amount of bytes that are in-flight for a connection as suggested by RFC 6675. Currently differnt places in the stack tries to guess this in suboptimal ways. The main problem is that current calculations don't take sacked bytes into account. Sacked bytes are the bytes receiver acked via SACK option. This is suboptimal because it assumes that network has more outstanding (unacked) bytes than the actual value and thus sends less data by setting congestion window lower than what's possible which in turn may cause slower recovery from losses. As an example, one of the current calculations looks something like this: snd_nxt - snd_fack + sackhint.sack_bytes_rexmit New proposal from RFC 6675 is: snd_max - snd_una - sackhint.sacked_bytes + sackhint.sack_bytes_rexmit which takes sacked bytes into account which is a new addition to the sackhint struct. Only thing we are missing from RFC 6675 is isLost() i.e. segment being considered lost and thus adjusting pipe based on that which makes this calculation a bit on conservative side. The approach is very simple. We already process each ack with sack info in tcp_sack_doack() and extract sack blocks/holes out of it. We'd now also track this new variable sacked_bytes which keeps track of total sacked bytes reported. One downside to this approach is that we may get incorrect count of sacked_bytes if the other end decides to drop sack info in the ack because of memory pressure or some other reasons. But in this (not very likely) case also the pipe calculation would be conservative which is okay as opposed to being aggressive in sending packets into the network. Next step is to use this more accurate pipe estimation to drive congestion window adjustments. In collaboration with: rrs Reviewed by: jason_eggnet dot com, rrs MFC after: 2 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D3971
2015-10-28 22:57:51 +00:00
SEQ_LEQ(sack.end, tp->snd_max)) {
sack_blocks[num_sack_blks++] = sack;
} else if (SEQ_LEQ(sack.start, th_ack) &&
SEQ_LEQ(sack.end, th_ack)) {
/*
* Its a D-SACK block.
*/
tcp_record_dsack(tp, sack.start, sack.end, 0);
Calculate the correct amount of bytes that are in-flight for a connection as suggested by RFC 6675. Currently differnt places in the stack tries to guess this in suboptimal ways. The main problem is that current calculations don't take sacked bytes into account. Sacked bytes are the bytes receiver acked via SACK option. This is suboptimal because it assumes that network has more outstanding (unacked) bytes than the actual value and thus sends less data by setting congestion window lower than what's possible which in turn may cause slower recovery from losses. As an example, one of the current calculations looks something like this: snd_nxt - snd_fack + sackhint.sack_bytes_rexmit New proposal from RFC 6675 is: snd_max - snd_una - sackhint.sacked_bytes + sackhint.sack_bytes_rexmit which takes sacked bytes into account which is a new addition to the sackhint struct. Only thing we are missing from RFC 6675 is isLost() i.e. segment being considered lost and thus adjusting pipe based on that which makes this calculation a bit on conservative side. The approach is very simple. We already process each ack with sack info in tcp_sack_doack() and extract sack blocks/holes out of it. We'd now also track this new variable sacked_bytes which keeps track of total sacked bytes reported. One downside to this approach is that we may get incorrect count of sacked_bytes if the other end decides to drop sack info in the ack because of memory pressure or some other reasons. But in this (not very likely) case also the pipe calculation would be conservative which is okay as opposed to being aggressive in sending packets into the network. Next step is to use this more accurate pipe estimation to drive congestion window adjustments. In collaboration with: rrs Reviewed by: jason_eggnet dot com, rrs MFC after: 2 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D3971
2015-10-28 22:57:51 +00:00
}
}
}
/*
* Return if SND.UNA is not advanced and no valid SACK block is
* received.
*/
if (num_sack_blks == 0)
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
return (sack_changed);
/*
* Sort the SACK blocks so we can update the scoreboard with just one
* pass. The overhead of sorting up to 4+1 elements is less than
* making up to 4+1 passes over the scoreboard.
*/
for (i = 0; i < num_sack_blks; i++) {
for (j = i + 1; j < num_sack_blks; j++) {
if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) {
sack = sack_blocks[i];
sack_blocks[i] = sack_blocks[j];
sack_blocks[j] = sack;
}
}
}
if (TAILQ_EMPTY(&tp->snd_holes)) {
/*
* Empty scoreboard. Need to initialize snd_fack (it may be
* uninitialized or have a bogus value). Scoreboard holes
* (from the sack blocks received) are created later below
* (in the logic that adds holes to the tail of the
* scoreboard).
*/
tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack);
tp->sackhint.sacked_bytes = 0; /* reset */
}
/*
* In the while-loop below, incoming SACK blocks (sack_blocks[]) and
* SACK holes (snd_holes) are traversed from their tails with just
* one pass in order to reduce the number of compares especially when
* the bandwidth-delay product is large.
*
* Note: Typically, in the first RTT of SACK recovery, the highest
* three or four SACK blocks with the same ack number are received.
* In the second RTT, if retransmitted data segments are not lost,
* the highest three or four SACK blocks with ack number advancing
* are received.
*/
sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */
tp->sackhint.last_sack_ack = sblkp->end;
if (SEQ_LT(tp->snd_fack, sblkp->start)) {
/*
* The highest SACK block is beyond fack. First,
* check if there was a successful Rescue Retransmission,
* and move this hole left. With normal holes, snd_fack
* is always to the right of the end.
*/
if (((temp = TAILQ_LAST(&tp->snd_holes, sackhole_head)) != NULL) &&
SEQ_LEQ(tp->snd_fack,temp->end)) {
temp->start = SEQ_MAX(tp->snd_fack, SEQ_MAX(tp->snd_una, th_ack));
temp->end = sblkp->start;
temp->rxmit = temp->start;
delivered_data += sblkp->end - sblkp->start;
tp->snd_fack = sblkp->end;
sblkp--;
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
sack_changed = 1;
} else {
/*
* Append a new SACK hole at the tail. If the
* second or later highest SACK blocks are also
* beyond the current fack, they will be inserted
* by way of hole splitting in the while-loop below.
*/
temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL);
if (temp != NULL) {
delivered_data += sblkp->end - sblkp->start;
tp->snd_fack = sblkp->end;
/* Go to the previous sack block. */
sblkp--;
sack_changed = 1;
} else {
/*
* We failed to add a new hole based on the current
* sack block. Skip over all the sack blocks that
* fall completely to the right of snd_fack and
* proceed to trim the scoreboard based on the
* remaining sack blocks. This also trims the
* scoreboard for th_ack (which is sack_blocks[0]).
*/
while (sblkp >= sack_blocks &&
SEQ_LT(tp->snd_fack, sblkp->start))
sblkp--;
if (sblkp >= sack_blocks &&
SEQ_LT(tp->snd_fack, sblkp->end)) {
delivered_data += sblkp->end - tp->snd_fack;
tp->snd_fack = sblkp->end;
sack_changed = 1;
}
}
}
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
} else if (SEQ_LT(tp->snd_fack, sblkp->end)) {
/* fack is advanced. */
delivered_data += sblkp->end - tp->snd_fack;
tp->snd_fack = sblkp->end;
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
sack_changed = 1;
}
cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */
/*
* Since the incoming sack blocks are sorted, we can process them
* making one sweep of the scoreboard.
*/
while (sblkp >= sack_blocks && cur != NULL) {
if (SEQ_GEQ(sblkp->start, cur->end)) {
/*
* SACKs data beyond the current hole. Go to the
* previous sack block.
*/
sblkp--;
continue;
}
if (SEQ_LEQ(sblkp->end, cur->start)) {
/*
* SACKs data before the current hole. Go to the
* previous hole.
*/
cur = TAILQ_PREV(cur, sackhole_head, scblink);
continue;
}
tp->sackhint.sack_bytes_rexmit -=
(SEQ_MIN(cur->rxmit, cur->end) - cur->start);
KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
("sackhint bytes rtx >= 0"));
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
sack_changed = 1;
if (SEQ_LEQ(sblkp->start, cur->start)) {
/* Data acks at least the beginning of hole. */
if (SEQ_GEQ(sblkp->end, cur->end)) {
/* Acks entire hole, so delete hole. */
delivered_data += (cur->end - cur->start);
temp = cur;
cur = TAILQ_PREV(cur, sackhole_head, scblink);
tcp_sackhole_remove(tp, temp);
/*
* The sack block may ack all or part of the
* next hole too, so continue onto the next
* hole.
*/
continue;
} else {
/* Move start of hole forward. */
delivered_data += (sblkp->end - cur->start);
cur->start = sblkp->end;
cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
}
} else {
/* Data acks at least the end of hole. */
if (SEQ_GEQ(sblkp->end, cur->end)) {
/* Move end of hole backward. */
delivered_data += (cur->end - sblkp->start);
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end))
cur->rxmit = tp->snd_recover;
} else {
/*
* ACKs some data in middle of a hole; need
* to split current hole
*/
temp = tcp_sackhole_insert(tp, sblkp->end,
cur->end, cur);
if (temp != NULL) {
if (SEQ_GT(cur->rxmit, temp->rxmit)) {
temp->rxmit = cur->rxmit;
tp->sackhint.sack_bytes_rexmit +=
(SEQ_MIN(temp->rxmit,
temp->end) - temp->start);
}
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit,
cur->end);
if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end))
cur->rxmit = tp->snd_recover;
delivered_data += (sblkp->end - sblkp->start);
}
}
}
tp->sackhint.sack_bytes_rexmit +=
(SEQ_MIN(cur->rxmit, cur->end) - cur->start);
/*
* Testing sblkp->start against cur->start tells us whether
* we're done with the sack block or the sack hole.
* Accordingly, we advance one or the other.
*/
if (SEQ_LEQ(sblkp->start, cur->start))
cur = TAILQ_PREV(cur, sackhole_head, scblink);
else
sblkp--;
}
if (!(to->to_flags & TOF_SACK))
/*
* If this ACK did not contain any
* SACK blocks, any only moved the
* left edge right, it is a pure
* cumulative ACK. Do not count
* DupAck for this. Also required
* for RFC6675 rescue retransmission.
*/
sack_changed = 0;
tp->sackhint.delivered_data = delivered_data;
tp->sackhint.sacked_bytes += delivered_data - left_edge_delta;
KASSERT((delivered_data >= 0), ("delivered_data < 0"));
KASSERT((tp->sackhint.sacked_bytes >= 0), ("sacked_bytes < 0"));
One of the ways to detect loss is to count duplicate acks coming back from the other end till it reaches predetermined threshold which is 3 for us right now. Once that happens, we trigger fast-retransmit to do loss recovery. Main problem with the current implementation is that we don't honor SACK information well to detect whether an incoming ack is a dupack or not. RFC6675 has latest recommendations for that. According to it, dupack is a segment that arrives carrying a SACK block that identifies previously unknown information between snd_una and snd_max even if it carries new data, changes the advertised window, or moves the cumulative acknowledgment point. With the prevalence of Selective ACK (SACK) these days, improper handling can lead to delayed loss recovery. With the fix, new behavior looks like following: 0) th_ack < snd_una --> ignore Old acks are ignored. 1) th_ack == snd_una, !sack_changed --> ignore Acks with SACK enabled but without any new SACK info in them are ignored. 2) th_ack == snd_una, window == old_window --> increment Increment on a good dupack. 3) th_ack == snd_una, window != old_window, sack_changed --> increment When SACK enabled, it's okay to have advertized window changed if the ack has new SACK info. 4) th_ack > snd_una --> reset to 0 Reset to 0 when left edge moves. 5) th_ack > snd_una, sack_changed --> increment Increment if left edge moves but there is new SACK info. Here, sack_changed is the indicator that incoming ack has previously unknown SACK info in it. Note: This fix is not fully compliant to RFC6675. That may require a few changes to current implementation in order to keep per-sackhole dupack counter and change to the way we mark/handle sack holes. PR: 203663 Reviewed by: jtl MFC after: 3 weeks Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D4225
2015-12-08 21:21:48 +00:00
return (sack_changed);
}
/*
* Free all SACK holes to clear the scoreboard.
*/
void
tcp_free_sackholes(struct tcpcb *tp)
{
struct sackhole *q;
INP_WLOCK_ASSERT(tptoinpcb(tp));
while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL)
tcp_sackhole_remove(tp, q);
tp->sackhint.sack_bytes_rexmit = 0;
KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0"));
KASSERT(tp->sackhint.nexthole == NULL,
("tp->sackhint.nexthole == NULL"));
}
/*
* Partial ack handling within a sack recovery episode. Keeping this very
* simple for now. When a partial ack is received, force snd_cwnd to a value
* that will allow the sender to transmit no more than 2 segments. If
* necessary, a better scheme can be adopted at a later point, but for now,
* the goal is to prevent the sender from bursting a large amount of data in
* the midst of sack recovery.
*/
void
tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th)
{
int num_segs = 1;
u_int maxseg = tcp_maxseg(tp);
INP_WLOCK_ASSERT(tptoinpcb(tp));
tcp_timer_activate(tp, TT_REXMT, 0);
tp->t_rtttime = 0;
/* Send one or 2 segments based on how much new data was acked. */
if ((BYTES_THIS_ACK(tp, th) / maxseg) >= 2)
num_segs = 2;
tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit +
(tp->snd_nxt - tp->snd_recover) + num_segs * maxseg);
if (tp->snd_cwnd > tp->snd_ssthresh)
tp->snd_cwnd = tp->snd_ssthresh;
tp->t_flags |= TF_ACKNOW;
/*
* RFC6675 rescue retransmission
* Add a hole between th_ack (snd_una is not yet set) and snd_max,
* if this was a pure cumulative ACK and no data was send beyond
* recovery point. Since the data in the socket has not been freed
* at this point, we check if the scoreboard is empty, and the ACK
* delivered some new data, indicating a full ACK. Also, if the
* recovery point is still at snd_max, we are probably application
* limited. However, this inference might not always be true. The
* rescue retransmission may rarely be slightly premature
* compared to RFC6675.
* The corresponding ACK+SACK will cause any further outstanding
* segments to be retransmitted. This addresses a corner case, when
* the trailing packets of a window are lost and no further data
* is available for sending.
*/
if ((V_tcp_do_newsack) &&
SEQ_LT(th->th_ack, tp->snd_recover) &&
TAILQ_EMPTY(&tp->snd_holes) &&
(tp->sackhint.delivered_data > 0)) {
/*
* Exclude FIN sequence space in
* the hole for the rescue retransmission,
* and also don't create a hole, if only
* the ACK for a FIN is outstanding.
*/
tcp_seq highdata = tp->snd_max;
if (tp->t_flags & TF_SENTFIN)
highdata--;
highdata = SEQ_MIN(highdata, tp->snd_recover);
if (th->th_ack != highdata) {
tp->snd_fack = th->th_ack;
(void)tcp_sackhole_insert(tp, SEQ_MAX(th->th_ack,
highdata - maxseg), highdata, NULL);
}
}
(void) tcp_output(tp);
}
#if 0
/*
* Debug version of tcp_sack_output() that walks the scoreboard. Used for
* now to sanity check the hint.
*/
static struct sackhole *
tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt)
{
struct sackhole *p;
INP_WLOCK_ASSERT(tptoinpcb(tp));
*sack_bytes_rexmt = 0;
TAILQ_FOREACH(p, &tp->snd_holes, scblink) {
if (SEQ_LT(p->rxmit, p->end)) {
if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */
continue;
}
*sack_bytes_rexmt += (p->rxmit - p->start);
break;
}
*sack_bytes_rexmt += (SEQ_MIN(p->rxmit, p->end) - p->start);
}
return (p);
}
#endif
/*
* Returns the next hole to retransmit and the number of retransmitted bytes
* from the scoreboard. We store both the next hole and the number of
* retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK
* reception). This avoids scoreboard traversals completely.
*
* The loop here will traverse *at most* one link. Here's the argument. For
* the loop to traverse more than 1 link before finding the next hole to
* retransmit, we would need to have at least 1 node following the current
* hint with (rxmit == end). But, for all holes following the current hint,
* (start == rxmit), since we have not yet retransmitted from them.
* Therefore, in order to traverse more 1 link in the loop below, we need to
* have at least one node following the current hint with (start == rxmit ==
* end). But that can't happen, (start == end) means that all the data in
* that hole has been sacked, in which case, the hole would have been removed
* from the scoreboard.
*/
struct sackhole *
tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt)
{
struct sackhole *hole = NULL;
INP_WLOCK_ASSERT(tptoinpcb(tp));
*sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit;
hole = tp->sackhint.nexthole;
if (hole == NULL)
return (hole);
if (SEQ_GEQ(hole->rxmit, hole->end)) {
for (;;) {
hole = TAILQ_NEXT(hole, scblink);
if (hole == NULL)
return (hole);
if (SEQ_LT(hole->rxmit, hole->end)) {
tp->sackhint.nexthole = hole;
break;
}
}
}
KASSERT(SEQ_LT(hole->start, hole->end), ("%s: hole.start >= hole.end", __func__));
if (!(V_tcp_do_newsack)) {
KASSERT(SEQ_LT(hole->start, tp->snd_fack), ("%s: hole.start >= snd.fack", __func__));
KASSERT(SEQ_LT(hole->end, tp->snd_fack), ("%s: hole.end >= snd.fack", __func__));
KASSERT(SEQ_LT(hole->rxmit, tp->snd_fack), ("%s: hole.rxmit >= snd.fack", __func__));
if (SEQ_GEQ(hole->start, hole->end) ||
SEQ_GEQ(hole->start, tp->snd_fack) ||
SEQ_GEQ(hole->end, tp->snd_fack) ||
SEQ_GEQ(hole->rxmit, tp->snd_fack)) {
log(LOG_CRIT,"tcp: invalid SACK hole (%u-%u,%u) vs fwd ack %u, ignoring.\n",
hole->start, hole->end, hole->rxmit, tp->snd_fack);
return (NULL);
}
}
return (hole);
}
/*
* After a timeout, the SACK list may be rebuilt. This SACK information
* should be used to avoid retransmitting SACKed data. This function
* traverses the SACK list to see if snd_nxt should be moved forward.
*/
void
tcp_sack_adjust(struct tcpcb *tp)
{
struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes);
INP_WLOCK_ASSERT(tptoinpcb(tp));
if (cur == NULL)
return; /* No holes */
if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack))
return; /* We're already beyond any SACKed blocks */
/*-
* Two cases for which we want to advance snd_nxt:
* i) snd_nxt lies between end of one hole and beginning of another
* ii) snd_nxt lies between end of last hole and snd_fack
*/
while ((p = TAILQ_NEXT(cur, scblink)) != NULL) {
if (SEQ_LT(tp->snd_nxt, cur->end))
return;
if (SEQ_GEQ(tp->snd_nxt, p->start))
cur = p;
else {
tp->snd_nxt = p->start;
return;
}
}
if (SEQ_LT(tp->snd_nxt, cur->end))
return;
tp->snd_nxt = tp->snd_fack;
}
/*
* Lost Retransmission Detection
* Check is FACK is beyond the rexmit of the leftmost hole.
* If yes, we restart sending from still existing holes,
* and adjust cwnd via the congestion control module.
*/
void
tcp_sack_lost_retransmission(struct tcpcb *tp, struct tcphdr *th)
{
struct sackhole *temp;
if (IN_RECOVERY(tp->t_flags) &&
SEQ_GT(tp->snd_fack, tp->snd_recover) &&
((temp = TAILQ_FIRST(&tp->snd_holes)) != NULL) &&
SEQ_GEQ(temp->rxmit, temp->end) &&
SEQ_GEQ(tp->snd_fack, temp->rxmit)) {
TCPSTAT_INC(tcps_sack_lostrexmt);
/*
* Start retransmissions from the first hole, and
* subsequently all other remaining holes, including
* those, which had been sent completely before.
*/
tp->sackhint.nexthole = temp;
TAILQ_FOREACH(temp, &tp->snd_holes, scblink) {
if (SEQ_GEQ(tp->snd_fack, temp->rxmit) &&
SEQ_GEQ(temp->rxmit, temp->end))
temp->rxmit = temp->start;
}
/*
* Remember the old ssthresh, to deduct the beta factor used
* by the CC module. Finally, set cwnd to ssthresh just
* prior to invoking another cwnd reduction by the CC
* module, to not shrink it excessively.
*/
tp->snd_cwnd = tp->snd_ssthresh;
/*
* Formally exit recovery, and let the CC module adjust
* ssthresh as intended.
*/
EXIT_RECOVERY(tp->t_flags);
cc_cong_signal(tp, th, CC_NDUPACK);
/*
* For PRR, adjust recover_fs as if this new reduction
* initialized this variable.
* cwnd will be adjusted by SACK or PRR processing
* subsequently, only set it to a safe value here.
*/
tp->snd_cwnd = tcp_maxseg(tp);
tp->sackhint.recover_fs = (tp->snd_max - tp->snd_una) -
tp->sackhint.recover_fs;
}
}