]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet/tcp_sack.c
xnu-792.12.6.tar.gz
[apple/xnu.git] / bsd / netinet / tcp_sack.c
diff --git a/bsd/netinet/tcp_sack.c b/bsd/netinet/tcp_sack.c
new file mode 100644 (file)
index 0000000..897f149
--- /dev/null
@@ -0,0 +1,645 @@
+/*
+ * Copyright (c) 2006 Apple Computer, Inc. All Rights Reserved.
+ * 
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
+ * 
+ * This file contains Original Code and/or Modifications of Original Code 
+ * as defined in and that are subject to the Apple Public Source License 
+ * Version 2.0 (the 'License'). You may not use this file except in 
+ * compliance with the License.  The rights granted to you under the 
+ * License may not be used to create, or enable the creation or 
+ * redistribution of, unlawful or unlicensed copies of an Apple operating 
+ * system, or to circumvent, violate, or enable the circumvention or 
+ * violation of, any terms of an Apple operating system software license 
+ * agreement.
+ *
+ * Please obtain a copy of the License at 
+ * http://www.opensource.apple.com/apsl/ and read it before using this 
+ * file.
+ *
+ * The Original Code and all software distributed under the License are 
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
+ * Please see the License for the specific language governing rights and 
+ * limitations under the License.
+ *
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
+ */
+/*
+ * Copyright (c) 1982, 1986, 1988, 1990, 1993, 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. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 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.
+ *
+ */
+
+#define        _IP_VHL
+
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/sysctl.h>
+#include <sys/mbuf.h>
+#include <sys/domain.h>
+#include <sys/protosw.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+
+#include <net/route.h>
+
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/ip.h>
+#include <netinet/in_pcb.h>
+#include <netinet/ip_var.h>
+#if INET6
+#include <netinet6/in6_pcb.h>
+#include <netinet/ip6.h>
+#include <netinet6/ip6_var.h>
+#endif
+#include <netinet/tcp.h>
+//#define      TCPOUTFLAGS
+#include <netinet/tcp_fsm.h>
+#include <netinet/tcp_seq.h>
+#include <netinet/tcp_timer.h>
+#include <netinet/tcp_var.h>
+#include <netinet/tcpip.h>
+#if TCPDEBUG
+#include <netinet/tcp_debug.h>
+#endif
+#include <sys/kdebug.h>
+
+#if IPSEC
+#include <netinet6/ipsec.h>
+#endif /*IPSEC*/
+
+int    tcp_do_sack = 1;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, &tcp_do_sack, 0,
+       "Enable/Disable TCP SACK support");
+static int tcp_sack_maxholes = 128;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_maxholes, CTLFLAG_RW,
+       &tcp_sack_maxholes, 0, 
+    "Maximum number of TCP SACK holes allowed per connection");
+
+static int tcp_sack_globalmaxholes = 65536;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_globalmaxholes, CTLFLAG_RW,
+       &tcp_sack_globalmaxholes, 0, 
+    "Global maximum number of TCP SACK holes");
+
+static int tcp_sack_globalholes = 0;
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_globalholes, CTLFLAG_RD,
+    &tcp_sack_globalholes, 0,
+    "Global number of TCP SACK holes currently allocated");
+
+extern struct zone *sack_hole_zone;
+
+/*
+ * 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;
+
+       /* 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;
+
+       tp->rcv_numsacks = 0;
+       for (i = 0; i < MAX_SACK_BLKS; i++)
+               tp->sackblks[i].start = tp->sackblks[i].end=0;
+*/
+       bzero(&tp->sackblks[0], sizeof (struct sackblk) * MAX_SACK_BLKS);
+}
+
+/*
+ * 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 >= tcp_sack_maxholes ||
+           tcp_sack_globalholes >= tcp_sack_globalmaxholes) {
+               tcpstat.tcps_sack_sboverflow++;
+               return NULL;
+       }
+
+       hole = (struct sackhole *)zalloc_noblock(sack_hole_zone);
+       if (hole == NULL)
+               return NULL;
+
+       hole->start = start;
+       hole->end = end;
+       hole->rxmit = start;
+
+       tp->snd_numholes++;
+       tcp_sack_globalholes++;
+
+       return hole;
+}
+
+/*
+ * Free struct sackhole.
+ */
+static void
+tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole)
+{
+       zfree(sack_hole_zone, hole);
+
+       tp->snd_numholes--;
+       tcp_sack_globalholes--;
+}
+
+/*
+ * 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;
+
+       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[].
+        */
+       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_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 */
+       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 */
+       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);
+               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;
+
+       while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL)
+               tcp_sackhole_remove(tp, q);
+       tp->sackhint.sack_bytes_rexmit = 0;
+
+}
+
+/*
+ * 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(tp, th)
+       struct tcpcb *tp;
+       struct tcphdr *th;
+{
+       int num_segs = 1;
+
+       tp->t_timer[TCPT_REXMT] = 0;
+       tp->t_rtttime = 0;
+       /* send one or 2 segments based on how much new data was acked */
+       if (((th->th_ack - tp->snd_una) / 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);
+}
+
+/*
+ * 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;
+
+       *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);
+}
+
+/*
+ * 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, *dbg_hole = NULL;
+       int dbg_bytes_rexmt;
+
+       dbg_hole = tcp_sack_output_debug(tp, &dbg_bytes_rexmt);
+       *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:
+       if (dbg_hole != hole) {
+               printf("%s: Computed sack hole not the same as cached value\n", __func__);
+               hole = dbg_hole;
+       }
+       if (*sack_bytes_rexmt != dbg_bytes_rexmt) {
+               printf("%s: Computed sack_bytes_retransmitted (%d) not "
+                      "the same as cached value (%d)\n",
+                      __func__, dbg_bytes_rexmt, *sack_bytes_rexmt);
+               *sack_bytes_rexmt = dbg_bytes_rexmt;
+       }
+       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);
+
+       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;
+       return;
+}