xref: /freebsd/sys/netinet/tcp_reass.c (revision 4a5216a6dc0c3ce4cf5f2d3ee8af0c3ff3402c4f)
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 #include <sys/vimage.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/if.h>
53 #include <net/route.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_pcb.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/ip_options.h>
62 #include <netinet/ip6.h>
63 #include <netinet6/in6_pcb.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet6/nd6.h>
66 #include <netinet/tcp.h>
67 #include <netinet/tcp_fsm.h>
68 #include <netinet/tcp_seq.h>
69 #include <netinet/tcp_timer.h>
70 #include <netinet/tcp_var.h>
71 #include <netinet6/tcp6_var.h>
72 #include <netinet/tcpip.h>
73 #ifdef TCPDEBUG
74 #include <netinet/tcp_debug.h>
75 #endif /* TCPDEBUG */
76 
77 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
78     "TCP Segment Reassembly Queue");
79 
80 static int tcp_reass_maxseg = 0;
81 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_reass, OID_AUTO, maxsegments,
82     CTLFLAG_RDTUN, tcp_reass_maxseg, 0,
83     "Global maximum number of TCP Segments in Reassembly Queue");
84 
85 int tcp_reass_qsize = 0;
86 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_reass, OID_AUTO, cursegments,
87     CTLFLAG_RD, tcp_reass_qsize, 0,
88     "Global number of TCP Segments currently in Reassembly Queue");
89 
90 static int tcp_reass_maxqlen = 48;
91 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_reass, OID_AUTO, maxqlen,
92     CTLFLAG_RW, tcp_reass_maxqlen, 0,
93     "Maximum number of TCP Segments per individual Reassembly Queue");
94 
95 static int tcp_reass_overflows = 0;
96 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_reass, OID_AUTO, overflows,
97     CTLFLAG_RD, tcp_reass_overflows, 0,
98     "Global number of TCP Segment Reassembly Queue Overflows");
99 
100 /* Initialize TCP reassembly queue */
101 static void
102 tcp_reass_zone_change(void *tag)
103 {
104 	INIT_VNET_INET(curvnet);
105 
106 	V_tcp_reass_maxseg = nmbclusters / 16;
107 	uma_zone_set_max(tcp_reass_zone, V_tcp_reass_maxseg);
108 }
109 
110 uma_zone_t	tcp_reass_zone;
111 
112 void
113 tcp_reass_init(void)
114 {
115 	INIT_VNET_INET(curvnet);
116 
117 	V_tcp_reass_maxseg = nmbclusters / 16;
118 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
119 	    &V_tcp_reass_maxseg);
120 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
121 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
122 	uma_zone_set_max(tcp_reass_zone, V_tcp_reass_maxseg);
123 	EVENTHANDLER_REGISTER(nmbclusters_change,
124 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
125 }
126 
127 int
128 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
129 {
130 	INIT_VNET_INET(curvnet);
131 	struct tseg_qent *q;
132 	struct tseg_qent *p = NULL;
133 	struct tseg_qent *nq;
134 	struct tseg_qent *te = NULL;
135 	struct socket *so = tp->t_inpcb->inp_socket;
136 	int flags;
137 
138 	INP_WLOCK_ASSERT(tp->t_inpcb);
139 
140 	/*
141 	 * XXX: tcp_reass() is rather inefficient with its data structures
142 	 * and should be rewritten (see NetBSD for optimizations).  While
143 	 * doing that it should move to its own file tcp_reass.c.
144 	 */
145 
146 	/*
147 	 * Call with th==NULL after become established to
148 	 * force pre-ESTABLISHED data up to user socket.
149 	 */
150 	if (th == NULL)
151 		goto present;
152 
153 	/*
154 	 * Limit the number of segments in the reassembly queue to prevent
155 	 * holding on to too many segments (and thus running out of mbufs).
156 	 * Make sure to let the missing segment through which caused this
157 	 * queue.  Always keep one global queue entry spare to be able to
158 	 * process the missing segment.
159 	 */
160 	if (th->th_seq != tp->rcv_nxt &&
161 	    (V_tcp_reass_qsize + 1 >= V_tcp_reass_maxseg ||
162 	     tp->t_segqlen >= V_tcp_reass_maxqlen)) {
163 		V_tcp_reass_overflows++;
164 		V_tcpstat.tcps_rcvmemdrop++;
165 		m_freem(m);
166 		*tlenp = 0;
167 		return (0);
168 	}
169 
170 	/*
171 	 * Allocate a new queue entry. If we can't, or hit the zone limit
172 	 * just drop the pkt.
173 	 */
174 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
175 	if (te == NULL) {
176 		V_tcpstat.tcps_rcvmemdrop++;
177 		m_freem(m);
178 		*tlenp = 0;
179 		return (0);
180 	}
181 	tp->t_segqlen++;
182 	V_tcp_reass_qsize++;
183 
184 	/*
185 	 * Find a segment which begins after this one does.
186 	 */
187 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
188 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
189 			break;
190 		p = q;
191 	}
192 
193 	/*
194 	 * If there is a preceding segment, it may provide some of
195 	 * our data already.  If so, drop the data from the incoming
196 	 * segment.  If it provides all of our data, drop us.
197 	 */
198 	if (p != NULL) {
199 		int i;
200 		/* conversion to int (in i) handles seq wraparound */
201 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
202 		if (i > 0) {
203 			if (i >= *tlenp) {
204 				V_tcpstat.tcps_rcvduppack++;
205 				V_tcpstat.tcps_rcvdupbyte += *tlenp;
206 				m_freem(m);
207 				uma_zfree(tcp_reass_zone, te);
208 				tp->t_segqlen--;
209 				V_tcp_reass_qsize--;
210 				/*
211 				 * Try to present any queued data
212 				 * at the left window edge to the user.
213 				 * This is needed after the 3-WHS
214 				 * completes.
215 				 */
216 				goto present;	/* ??? */
217 			}
218 			m_adj(m, i);
219 			*tlenp -= i;
220 			th->th_seq += i;
221 		}
222 	}
223 	V_tcpstat.tcps_rcvoopack++;
224 	V_tcpstat.tcps_rcvoobyte += *tlenp;
225 
226 	/*
227 	 * While we overlap succeeding segments trim them or,
228 	 * if they are completely covered, dequeue them.
229 	 */
230 	while (q) {
231 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
232 		if (i <= 0)
233 			break;
234 		if (i < q->tqe_len) {
235 			q->tqe_th->th_seq += i;
236 			q->tqe_len -= i;
237 			m_adj(q->tqe_m, i);
238 			break;
239 		}
240 
241 		nq = LIST_NEXT(q, tqe_q);
242 		LIST_REMOVE(q, tqe_q);
243 		m_freem(q->tqe_m);
244 		uma_zfree(tcp_reass_zone, q);
245 		tp->t_segqlen--;
246 		V_tcp_reass_qsize--;
247 		q = nq;
248 	}
249 
250 	/* Insert the new segment queue entry into place. */
251 	te->tqe_m = m;
252 	te->tqe_th = th;
253 	te->tqe_len = *tlenp;
254 
255 	if (p == NULL) {
256 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
257 	} else {
258 		LIST_INSERT_AFTER(p, te, tqe_q);
259 	}
260 
261 present:
262 	/*
263 	 * Present data to user, advancing rcv_nxt through
264 	 * completed sequence space.
265 	 */
266 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
267 		return (0);
268 	q = LIST_FIRST(&tp->t_segq);
269 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
270 		return (0);
271 	SOCKBUF_LOCK(&so->so_rcv);
272 	do {
273 		tp->rcv_nxt += q->tqe_len;
274 		flags = q->tqe_th->th_flags & TH_FIN;
275 		nq = LIST_NEXT(q, tqe_q);
276 		LIST_REMOVE(q, tqe_q);
277 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
278 			m_freem(q->tqe_m);
279 		else
280 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
281 		uma_zfree(tcp_reass_zone, q);
282 		tp->t_segqlen--;
283 		V_tcp_reass_qsize--;
284 		q = nq;
285 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
286 	ND6_HINT(tp);
287 	sorwakeup_locked(so);
288 	return (flags);
289 }
290