xref: /freebsd/sys/netinet/tcp_reass.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_tcpdebug.h"
38 
39 #include <sys/param.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/sysctl.h>
46 #include <sys/syslog.h>
47 #include <sys/systm.h>
48 
49 #include <vm/uma.h>
50 
51 #include <net/if.h>
52 #include <net/if_var.h>
53 #include <net/route.h>
54 #include <net/vnet.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/ip_options.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/in6_pcb.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/nd6.h>
67 #include <netinet/tcp.h>
68 #include <netinet/tcp_fsm.h>
69 #include <netinet/tcp_seq.h>
70 #include <netinet/tcp_timer.h>
71 #include <netinet/tcp_var.h>
72 #include <netinet6/tcp6_var.h>
73 #include <netinet/tcpip.h>
74 
75 void
76 tcp_reass_flush(struct tcpcb *tp)
77 {
78 	struct mbuf *m;
79 
80 	INP_WLOCK_ASSERT(tp->t_inpcb);
81 
82 	while ((m = tp->t_segq) != NULL) {
83 		tp->t_segq = m->m_nextpkt;
84 		tp->t_segqlen -= m->m_pkthdr.len;
85 		m_freem(m);
86 	}
87 
88 	KASSERT((tp->t_segqlen == 0),
89 	    ("TCP reass queue %p length is %d instead of 0 after flush.",
90 	    tp, tp->t_segqlen));
91 }
92 
93 #define	M_TCPHDR(m)	((struct tcphdr *)((m)->m_pkthdr.pkt_tcphdr))
94 
95 int
96 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
97 {
98 	struct socket *so = tp->t_inpcb->inp_socket;
99 	struct mbuf *mq, *mp;
100 	int flags, wakeup;
101 
102 	INP_WLOCK_ASSERT(tp->t_inpcb);
103 
104 	/*
105 	 * XXX: tcp_reass() is rather inefficient with its data structures
106 	 * and should be rewritten (see NetBSD for optimizations).
107 	 */
108 
109 	/*
110 	 * Call with th==NULL after become established to
111 	 * force pre-ESTABLISHED data up to user socket.
112 	 */
113 	if (th == NULL)
114 		goto present;
115 
116 	M_ASSERTPKTHDR(m);
117 	KASSERT(*tlenp == m->m_pkthdr.len, ("%s: tlenp %u len %u", __func__,
118 	    *tlenp, m->m_pkthdr.len));
119 
120 	/*
121 	 * Limit the number of segments that can be queued to reduce the
122 	 * potential for mbuf exhaustion. For best performance, we want to be
123 	 * able to queue a full window's worth of segments. The size of the
124 	 * socket receive buffer determines our advertised window and grows
125 	 * automatically when socket buffer autotuning is enabled. Use it as the
126 	 * basis for our queue limit.
127 	 * Always let the missing segment through which caused this queue.
128 	 * NB: Access to the socket buffer is left intentionally unlocked as we
129 	 * can tolerate stale information here.
130 	 */
131 	if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
132 	    tp->t_segqlen + m->m_pkthdr.len >= sbspace(&so->so_rcv)) {
133 		char *s;
134 
135 		TCPSTAT_INC(tcps_rcvreassfull);
136 		*tlenp = 0;
137 		if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
138 		    NULL))) {
139 			log(LOG_DEBUG, "%s; %s: queue limit reached, "
140 			    "segment dropped\n", s, __func__);
141 			free(s, M_TCPLOG);
142 		}
143 		m_freem(m);
144 		return (0);
145 	}
146 
147 	/*
148 	 * Find a segment which begins after this one does.
149 	 */
150 	mp = NULL;
151 	for (mq = tp->t_segq; mq != NULL; mq = mq->m_nextpkt) {
152 		if (SEQ_GT(M_TCPHDR(mq)->th_seq, th->th_seq))
153 			break;
154 		mp = mq;
155 	}
156 
157 	/*
158 	 * If there is a preceding segment, it may provide some of
159 	 * our data already.  If so, drop the data from the incoming
160 	 * segment.  If it provides all of our data, drop us.
161 	 */
162 	if (mp != NULL) {
163 		int i;
164 
165 		/* conversion to int (in i) handles seq wraparound */
166 		i = M_TCPHDR(mp)->th_seq + mp->m_pkthdr.len - th->th_seq;
167 		if (i > 0) {
168 			if (i >= *tlenp) {
169 				TCPSTAT_INC(tcps_rcvduppack);
170 				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
171 				m_freem(m);
172 				/*
173 				 * Try to present any queued data
174 				 * at the left window edge to the user.
175 				 * This is needed after the 3-WHS
176 				 * completes.
177 				 */
178 				goto present;	/* ??? */
179 			}
180 			m_adj(m, i);
181 			*tlenp -= i;
182 			th->th_seq += i;
183 		}
184 	}
185 	tp->t_rcvoopack++;
186 	TCPSTAT_INC(tcps_rcvoopack);
187 	TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
188 
189 	/*
190 	 * While we overlap succeeding segments trim them or,
191 	 * if they are completely covered, dequeue them.
192 	 */
193 	while (mq) {
194 		struct mbuf *nq;
195 		int i;
196 
197 		i = (th->th_seq + *tlenp) - M_TCPHDR(mq)->th_seq;
198 		if (i <= 0)
199 			break;
200 		if (i < mq->m_pkthdr.len) {
201 			M_TCPHDR(mq)->th_seq += i;
202 			m_adj(mq, i);
203 			tp->t_segqlen -= i;
204 			break;
205 		}
206 
207 		nq = mq->m_nextpkt;
208 		tp->t_segqlen -= mq->m_pkthdr.len;
209 		m_freem(mq);
210 		if (mp)
211 			mp->m_nextpkt = nq;
212 		else
213 			tp->t_segq = nq;
214 		mq = nq;
215 	}
216 
217 	/*
218 	 * Insert the new segment queue entry into place.  Try to collapse
219 	 * mbuf chains if segments are adjacent.
220 	 */
221 	if (mp) {
222 		if (M_TCPHDR(mp)->th_seq + mp->m_pkthdr.len == th->th_seq)
223 			m_catpkt(mp, m);
224 		else {
225 			m->m_nextpkt = mp->m_nextpkt;
226 			mp->m_nextpkt = m;
227 			m->m_pkthdr.pkt_tcphdr = th;
228 		}
229 	} else {
230 		mq = tp->t_segq;
231 		tp->t_segq = m;
232 		if (mq && th->th_seq + *tlenp == M_TCPHDR(mq)->th_seq) {
233 			m->m_nextpkt = mq->m_nextpkt;
234 			mq->m_nextpkt = NULL;
235 			m_catpkt(m, mq);
236 		} else
237 			m->m_nextpkt = mq;
238 		m->m_pkthdr.pkt_tcphdr = th;
239 	}
240 	tp->t_segqlen += *tlenp;
241 
242 present:
243 	/*
244 	 * Present data to user, advancing rcv_nxt through
245 	 * completed sequence space.
246 	 */
247 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
248 		return (0);
249 
250 	flags = 0;
251 	wakeup = 0;
252 	SOCKBUF_LOCK(&so->so_rcv);
253 	while ((mq = tp->t_segq) != NULL &&
254 	    M_TCPHDR(mq)->th_seq == tp->rcv_nxt) {
255 		tp->t_segq = mq->m_nextpkt;
256 
257 		tp->rcv_nxt += mq->m_pkthdr.len;
258 		tp->t_segqlen -= mq->m_pkthdr.len;
259 		flags = M_TCPHDR(mq)->th_flags & TH_FIN;
260 
261 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
262 			m_freem(mq);
263 		else {
264 			mq->m_nextpkt = NULL;
265 			sbappendstream_locked(&so->so_rcv, mq, 0);
266 			wakeup = 1;
267 		}
268 	}
269 	ND6_HINT(tp);
270 	if (wakeup)
271 		sorwakeup_locked(so);
272 	else
273 		SOCKBUF_UNLOCK(&so->so_rcv);
274 	return (flags);
275 }
276