freebsd-dev/sys/netinet/tcp_sack.c

687 lines
21 KiB
C
Raw Normal View History

/*-
* 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.
* 4. 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 "opt_tcpdebug.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/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 <netinet6/tcp6_var.h>
#include <netinet/tcpip.h>
#ifdef TCPDEBUG
#include <netinet/tcp_debug.h>
#endif /* TCPDEBUG */
#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, 0, "TCP SACK");
VNET_DEFINE(int, tcp_do_sack) = 1;
#define V_tcp_do_sack VNET(tcp_do_sack)
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
SYSCTL_VNET_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW,
&VNET_NAME(tcp_do_sack), 0, "Enable/Disable TCP SACK support");
VNET_DEFINE(int, tcp_sack_maxholes) = 128;
#define V_tcp_sack_maxholes VNET(tcp_sack_maxholes)
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
SYSCTL_VNET_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_RW,
&VNET_NAME(tcp_sack_maxholes), 0,
"Maximum number of TCP SACK holes allowed per connection");
VNET_DEFINE(int, tcp_sack_globalmaxholes) = 65536;
#define V_tcp_sack_globalmaxholes VNET(tcp_sack_globalmaxholes)
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
SYSCTL_VNET_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_RW,
&VNET_NAME(tcp_sack_globalmaxholes), 0,
"Global maximum number of TCP SACK holes");
VNET_DEFINE(int, tcp_sack_globalholes) = 0;
#define V_tcp_sack_globalholes VNET(tcp_sack_globalholes)
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
SYSCTL_VNET_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_RD,
&VNET_NAME(tcp_sack_globalholes), 0,
"Global number of TCP SACK holes currently allocated");
/*
* 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(tp->t_inpcb);
/* Check arguments. */
KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end"));
/* 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.
*/
if (SEQ_GT(head_blk.start, start))
head_blk.start = start;
if (SEQ_LT(head_blk.end, end))
head_blk.end = end;
} else {
/*
* 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_GT(head_blk.start, tp->rcv_nxt)) {
/*
* 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 (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;
}
/*
* Delete all receiver-side SACK information.
*/
void
tcp_clean_sackreport(struct tcpcb *tp)
{
int i;
INP_WLOCK_ASSERT(tp->t_inpcb);
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).
*/
void
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;
int i, j, num_sack_blks;
INP_WLOCK_ASSERT(tp->t_inpcb);
num_sack_blks = 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.
*/
if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) {
sack_blocks[num_sack_blks].start = tp->snd_una;
sack_blocks[num_sack_blks++].end = th_ack;
}
/*
* 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) &&
SEQ_LEQ(sack.end, tp->snd_max))
sack_blocks[num_sack_blks++] = sack;
}
}
/*
* Return if SND.UNA is not advanced and no valid SACK block is
* received.
*/
if (num_sack_blks == 0)
return;
/*
* Sort the SACK blocks so we can update the scoreboard with just one
* pass. The overhead of sorting upto 4+1 elements is less than
* making upto 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);
/*
* 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. Append 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) {
tp->snd_fack = sblkp->end;
/* Go to the previous sack block. */
sblkp--;
} 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))
tp->snd_fack = sblkp->end;
}
} else if (SEQ_LT(tp->snd_fack, sblkp->end))
/* fack is advanced. */
tp->snd_fack = sblkp->end;
/* We must have at least one SACK hole in scoreboard. */
KASSERT(!TAILQ_EMPTY(&tp->snd_holes),
("SACK scoreboard must not be empty"));
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 -= (cur->rxmit - cur->start);
KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
("sackhint bytes rtx >= 0"));
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. */
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. */
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. */
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
} 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
+= (temp->rxmit
- temp->start);
}
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit,
cur->end);
}
}
}
tp->sackhint.sack_bytes_rexmit += (cur->rxmit - 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--;
}
}
/*
* Free all SACK holes to clear the scoreboard.
*/
void
tcp_free_sackholes(struct tcpcb *tp)
{
struct sackhole *q;
INP_WLOCK_ASSERT(tp->t_inpcb);
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;
INP_WLOCK_ASSERT(tp->t_inpcb);
tcp_timer_activate(tp, TT_REXMT, 0);
tp->t_rtttime = 0;
/* Send one or 2 segments based on how much new data was acked. */
This commit marks the first formal contribution of the "Five New TCP Congestion Control Algorithms for FreeBSD" FreeBSD Foundation funded project. More details about the project are available at: http://caia.swin.edu.au/freebsd/5cc/ - Add a KPI and supporting infrastructure to allow modular congestion control algorithms to be used in the net stack. Algorithms can maintain per-connection state if required, and connections maintain their own algorithm pointer, which allows different connections to concurrently use different algorithms. The TCP_CONGESTION socket option can be used with getsockopt()/setsockopt() to programmatically query or change the congestion control algorithm respectively from within an application at runtime. - Integrate the framework with the TCP stack in as least intrusive a manner as possible. Care was also taken to develop the framework in a way that should allow integration with other congestion aware transport protocols (e.g. SCTP) in the future. The hope is that we will one day be able to share a single set of congestion control algorithm modules between all congestion aware transport protocols. - Introduce a new congestion recovery (TF_CONGRECOVERY) state into the TCP stack and use it to decouple the meaning of recovery from a congestion event and recovery from packet loss (TF_FASTRECOVERY) a la RFC2581. ECN and delay based congestion control protocols don't generally need to recover from packet loss and need a different way to note a congestion recovery episode within the stack. - Remove the net.inet.tcp.newreno sysctl, which simplifies some portions of code and ensures the stack always uses the appropriate mechanisms for recovering from packet loss during a congestion recovery episode. - Extract the NewReno congestion control algorithm from the TCP stack and massage it into module form. NewReno is always built into the kernel and will remain the default algorithm for the forseeable future. Implementations of additional different algorithms will become available in the near future. - Bump __FreeBSD_version to 900025 and note in UPDATING that rebuilding code that relies on the size of "struct tcpcb" is required. Many thanks go to the Cisco University Research Program Fund at Community Foundation Silicon Valley and the FreeBSD Foundation. Their support of our work at the Centre for Advanced Internet Architectures, Swinburne University of Technology is greatly appreciated. In collaboration with: David Hayes <dahayes at swin edu au> and Grenville Armitage <garmitage at swin edu au> Sponsored by: Cisco URP, FreeBSD Foundation Reviewed by: rpaulo Tested by: David Hayes (and many others over the years) MFC after: 3 months
2010-11-12 06:41:55 +00:00
if ((BYTES_THIS_ACK(tp, th) / tp->t_maxseg) > 2)
num_segs = 2;
tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit +
(tp->snd_nxt - tp->sack_newdata) + num_segs * tp->t_maxseg);
if (tp->snd_cwnd > tp->snd_ssthresh)
tp->snd_cwnd = tp->snd_ssthresh;
tp->t_flags |= TF_ACKNOW;
(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(tp->t_inpcb);
*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 += (p->rxmit - 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(tp->t_inpcb);
*sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit;
hole = tp->sackhint.nexthole;
if (hole == NULL || SEQ_LT(hole->rxmit, hole->end))
goto out;
while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) {
if (SEQ_LT(hole->rxmit, hole->end)) {
tp->sackhint.nexthole = hole;
break;
}
}
out:
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(tp->t_inpcb);
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;
}