xref: /freebsd/sys/netinet/tcp_reass.c (revision b28624fde638caadd4a89f50c9b7e7da0f98c4d2)
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  * $FreeBSD$
31  */
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_tcpdebug.h"
36 
37 #include <sys/param.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/sysctl.h>
44 #include <sys/syslog.h>
45 #include <sys/systm.h>
46 
47 #include <vm/uma.h>
48 
49 #include <net/if.h>
50 #include <net/route.h>
51 
52 #include <netinet/in.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/in_var.h>
56 #include <netinet/ip.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/ip_options.h>
59 #include <netinet/ip6.h>
60 #include <netinet6/in6_pcb.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet6/nd6.h>
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_fsm.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/tcp_timer.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet6/tcp6_var.h>
69 #include <netinet/tcpip.h>
70 #ifdef TCPDEBUG
71 #include <netinet/tcp_debug.h>
72 #endif /* TCPDEBUG */
73 
74 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
75     "TCP Segment Reassembly Queue");
76 
77 static int tcp_reass_maxseg = 0;
78 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
79     &tcp_reass_maxseg, 0,
80     "Global maximum number of TCP Segments in Reassembly Queue");
81 
82 int tcp_reass_qsize = 0;
83 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
84     &tcp_reass_qsize, 0,
85     "Global number of TCP Segments currently in Reassembly Queue");
86 
87 static int tcp_reass_maxqlen = 48;
88 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
89     &tcp_reass_maxqlen, 0,
90     "Maximum number of TCP Segments per individual Reassembly Queue");
91 
92 static int tcp_reass_overflows = 0;
93 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
94     &tcp_reass_overflows, 0,
95     "Global number of TCP Segment Reassembly Queue Overflows");
96 
97 /* Initialize TCP reassembly queue */
98 static void
99 tcp_reass_zone_change(void *tag)
100 {
101 
102 	tcp_reass_maxseg = nmbclusters / 16;
103 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
104 }
105 
106 uma_zone_t	tcp_reass_zone;
107 
108 void
109 tcp_reass_init(void)
110 {
111 
112 	tcp_reass_maxseg = nmbclusters / 16;
113 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
114 	    &tcp_reass_maxseg);
115 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
116 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
117 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
118 	EVENTHANDLER_REGISTER(nmbclusters_change,
119 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
120 }
121 
122 int
123 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
124 {
125 	struct tseg_qent *q;
126 	struct tseg_qent *p = NULL;
127 	struct tseg_qent *nq;
128 	struct tseg_qent *te = NULL;
129 	struct socket *so = tp->t_inpcb->inp_socket;
130 	int flags;
131 
132 	INP_LOCK_ASSERT(tp->t_inpcb);
133 
134 	/*
135 	 * XXX: tcp_reass() is rather inefficient with its data structures
136 	 * and should be rewritten (see NetBSD for optimizations).  While
137 	 * doing that it should move to its own file tcp_reass.c.
138 	 */
139 
140 	/*
141 	 * Call with th==NULL after become established to
142 	 * force pre-ESTABLISHED data up to user socket.
143 	 */
144 	if (th == NULL)
145 		goto present;
146 
147 	/*
148 	 * Limit the number of segments in the reassembly queue to prevent
149 	 * holding on to too many segments (and thus running out of mbufs).
150 	 * Make sure to let the missing segment through which caused this
151 	 * queue.  Always keep one global queue entry spare to be able to
152 	 * process the missing segment.
153 	 */
154 	if (th->th_seq != tp->rcv_nxt &&
155 	    (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
156 	     tp->t_segqlen >= tcp_reass_maxqlen)) {
157 		tcp_reass_overflows++;
158 		tcpstat.tcps_rcvmemdrop++;
159 		m_freem(m);
160 		*tlenp = 0;
161 		return (0);
162 	}
163 
164 	/*
165 	 * Allocate a new queue entry. If we can't, or hit the zone limit
166 	 * just drop the pkt.
167 	 */
168 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
169 	if (te == NULL) {
170 		tcpstat.tcps_rcvmemdrop++;
171 		m_freem(m);
172 		*tlenp = 0;
173 		return (0);
174 	}
175 	tp->t_segqlen++;
176 	tcp_reass_qsize++;
177 
178 	/*
179 	 * Find a segment which begins after this one does.
180 	 */
181 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
182 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
183 			break;
184 		p = q;
185 	}
186 
187 	/*
188 	 * If there is a preceding segment, it may provide some of
189 	 * our data already.  If so, drop the data from the incoming
190 	 * segment.  If it provides all of our data, drop us.
191 	 */
192 	if (p != NULL) {
193 		int i;
194 		/* conversion to int (in i) handles seq wraparound */
195 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
196 		if (i > 0) {
197 			if (i >= *tlenp) {
198 				tcpstat.tcps_rcvduppack++;
199 				tcpstat.tcps_rcvdupbyte += *tlenp;
200 				m_freem(m);
201 				uma_zfree(tcp_reass_zone, te);
202 				tp->t_segqlen--;
203 				tcp_reass_qsize--;
204 				/*
205 				 * Try to present any queued data
206 				 * at the left window edge to the user.
207 				 * This is needed after the 3-WHS
208 				 * completes.
209 				 */
210 				goto present;	/* ??? */
211 			}
212 			m_adj(m, i);
213 			*tlenp -= i;
214 			th->th_seq += i;
215 		}
216 	}
217 	tcpstat.tcps_rcvoopack++;
218 	tcpstat.tcps_rcvoobyte += *tlenp;
219 
220 	/*
221 	 * While we overlap succeeding segments trim them or,
222 	 * if they are completely covered, dequeue them.
223 	 */
224 	while (q) {
225 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
226 		if (i <= 0)
227 			break;
228 		if (i < q->tqe_len) {
229 			q->tqe_th->th_seq += i;
230 			q->tqe_len -= i;
231 			m_adj(q->tqe_m, i);
232 			break;
233 		}
234 
235 		nq = LIST_NEXT(q, tqe_q);
236 		LIST_REMOVE(q, tqe_q);
237 		m_freem(q->tqe_m);
238 		uma_zfree(tcp_reass_zone, q);
239 		tp->t_segqlen--;
240 		tcp_reass_qsize--;
241 		q = nq;
242 	}
243 
244 	/* Insert the new segment queue entry into place. */
245 	te->tqe_m = m;
246 	te->tqe_th = th;
247 	te->tqe_len = *tlenp;
248 
249 	if (p == NULL) {
250 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
251 	} else {
252 		LIST_INSERT_AFTER(p, te, tqe_q);
253 	}
254 
255 present:
256 	/*
257 	 * Present data to user, advancing rcv_nxt through
258 	 * completed sequence space.
259 	 */
260 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
261 		return (0);
262 	q = LIST_FIRST(&tp->t_segq);
263 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
264 		return (0);
265 	SOCKBUF_LOCK(&so->so_rcv);
266 	do {
267 		tp->rcv_nxt += q->tqe_len;
268 		flags = q->tqe_th->th_flags & TH_FIN;
269 		nq = LIST_NEXT(q, tqe_q);
270 		LIST_REMOVE(q, tqe_q);
271 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
272 			m_freem(q->tqe_m);
273 		else
274 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
275 		uma_zfree(tcp_reass_zone, q);
276 		tp->t_segqlen--;
277 		tcp_reass_qsize--;
278 		q = nq;
279 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
280 	ND6_HINT(tp);
281 	sorwakeup_locked(so);
282 	return (flags);
283 }
284