xref: /freebsd/sys/netinet/tcp_reass.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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/route.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_pcb.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/ip_options.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/in6_pcb.h>
63 #include <netinet6/ip6_var.h>
64 #include <netinet6/nd6.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet6/tcp6_var.h>
71 #include <netinet/tcpip.h>
72 #ifdef TCPDEBUG
73 #include <netinet/tcp_debug.h>
74 #endif /* TCPDEBUG */
75 
76 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
77     "TCP Segment Reassembly Queue");
78 
79 static int tcp_reass_maxseg = 0;
80 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
81     &tcp_reass_maxseg, 0,
82     "Global maximum number of TCP Segments in Reassembly Queue");
83 
84 int tcp_reass_qsize = 0;
85 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
86     &tcp_reass_qsize, 0,
87     "Global number of TCP Segments currently in Reassembly Queue");
88 
89 static int tcp_reass_maxqlen = 48;
90 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
91     &tcp_reass_maxqlen, 0,
92     "Maximum number of TCP Segments per individual Reassembly Queue");
93 
94 static int tcp_reass_overflows = 0;
95 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
96     &tcp_reass_overflows, 0,
97     "Global number of TCP Segment Reassembly Queue Overflows");
98 
99 /* Initialize TCP reassembly queue */
100 static void
101 tcp_reass_zone_change(void *tag)
102 {
103 
104 	tcp_reass_maxseg = nmbclusters / 16;
105 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
106 }
107 
108 uma_zone_t	tcp_reass_zone;
109 
110 void
111 tcp_reass_init(void)
112 {
113 
114 	tcp_reass_maxseg = nmbclusters / 16;
115 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
116 	    &tcp_reass_maxseg);
117 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
118 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
119 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
120 	EVENTHANDLER_REGISTER(nmbclusters_change,
121 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
122 }
123 
124 int
125 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
126 {
127 	struct tseg_qent *q;
128 	struct tseg_qent *p = NULL;
129 	struct tseg_qent *nq;
130 	struct tseg_qent *te = NULL;
131 	struct socket *so = tp->t_inpcb->inp_socket;
132 	int flags;
133 
134 	INP_LOCK_ASSERT(tp->t_inpcb);
135 
136 	/*
137 	 * XXX: tcp_reass() is rather inefficient with its data structures
138 	 * and should be rewritten (see NetBSD for optimizations).  While
139 	 * doing that it should move to its own file tcp_reass.c.
140 	 */
141 
142 	/*
143 	 * Call with th==NULL after become established to
144 	 * force pre-ESTABLISHED data up to user socket.
145 	 */
146 	if (th == NULL)
147 		goto present;
148 
149 	/*
150 	 * Limit the number of segments in the reassembly queue to prevent
151 	 * holding on to too many segments (and thus running out of mbufs).
152 	 * Make sure to let the missing segment through which caused this
153 	 * queue.  Always keep one global queue entry spare to be able to
154 	 * process the missing segment.
155 	 */
156 	if (th->th_seq != tp->rcv_nxt &&
157 	    (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
158 	     tp->t_segqlen >= tcp_reass_maxqlen)) {
159 		tcp_reass_overflows++;
160 		tcpstat.tcps_rcvmemdrop++;
161 		m_freem(m);
162 		*tlenp = 0;
163 		return (0);
164 	}
165 
166 	/*
167 	 * Allocate a new queue entry. If we can't, or hit the zone limit
168 	 * just drop the pkt.
169 	 */
170 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
171 	if (te == NULL) {
172 		tcpstat.tcps_rcvmemdrop++;
173 		m_freem(m);
174 		*tlenp = 0;
175 		return (0);
176 	}
177 	tp->t_segqlen++;
178 	tcp_reass_qsize++;
179 
180 	/*
181 	 * Find a segment which begins after this one does.
182 	 */
183 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
184 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
185 			break;
186 		p = q;
187 	}
188 
189 	/*
190 	 * If there is a preceding segment, it may provide some of
191 	 * our data already.  If so, drop the data from the incoming
192 	 * segment.  If it provides all of our data, drop us.
193 	 */
194 	if (p != NULL) {
195 		int i;
196 		/* conversion to int (in i) handles seq wraparound */
197 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
198 		if (i > 0) {
199 			if (i >= *tlenp) {
200 				tcpstat.tcps_rcvduppack++;
201 				tcpstat.tcps_rcvdupbyte += *tlenp;
202 				m_freem(m);
203 				uma_zfree(tcp_reass_zone, te);
204 				tp->t_segqlen--;
205 				tcp_reass_qsize--;
206 				/*
207 				 * Try to present any queued data
208 				 * at the left window edge to the user.
209 				 * This is needed after the 3-WHS
210 				 * completes.
211 				 */
212 				goto present;	/* ??? */
213 			}
214 			m_adj(m, i);
215 			*tlenp -= i;
216 			th->th_seq += i;
217 		}
218 	}
219 	tcpstat.tcps_rcvoopack++;
220 	tcpstat.tcps_rcvoobyte += *tlenp;
221 
222 	/*
223 	 * While we overlap succeeding segments trim them or,
224 	 * if they are completely covered, dequeue them.
225 	 */
226 	while (q) {
227 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
228 		if (i <= 0)
229 			break;
230 		if (i < q->tqe_len) {
231 			q->tqe_th->th_seq += i;
232 			q->tqe_len -= i;
233 			m_adj(q->tqe_m, i);
234 			break;
235 		}
236 
237 		nq = LIST_NEXT(q, tqe_q);
238 		LIST_REMOVE(q, tqe_q);
239 		m_freem(q->tqe_m);
240 		uma_zfree(tcp_reass_zone, q);
241 		tp->t_segqlen--;
242 		tcp_reass_qsize--;
243 		q = nq;
244 	}
245 
246 	/* Insert the new segment queue entry into place. */
247 	te->tqe_m = m;
248 	te->tqe_th = th;
249 	te->tqe_len = *tlenp;
250 
251 	if (p == NULL) {
252 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
253 	} else {
254 		LIST_INSERT_AFTER(p, te, tqe_q);
255 	}
256 
257 present:
258 	/*
259 	 * Present data to user, advancing rcv_nxt through
260 	 * completed sequence space.
261 	 */
262 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
263 		return (0);
264 	q = LIST_FIRST(&tp->t_segq);
265 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
266 		return (0);
267 	SOCKBUF_LOCK(&so->so_rcv);
268 	do {
269 		tp->rcv_nxt += q->tqe_len;
270 		flags = q->tqe_th->th_flags & TH_FIN;
271 		nq = LIST_NEXT(q, tqe_q);
272 		LIST_REMOVE(q, tqe_q);
273 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
274 			m_freem(q->tqe_m);
275 		else
276 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
277 		uma_zfree(tcp_reass_zone, q);
278 		tp->t_segqlen--;
279 		tcp_reass_qsize--;
280 		q = nq;
281 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
282 	ND6_HINT(tp);
283 	sorwakeup_locked(so);
284 	return (flags);
285 }
286