xref: /freebsd/sys/netinet/tcp_reass.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 #include <net/vnet.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 static VNET_DEFINE(int, tcp_reass_maxseg);
78 VNET_DEFINE(int, tcp_reass_qsize);
79 static VNET_DEFINE(int, tcp_reass_maxqlen);
80 static VNET_DEFINE(int, tcp_reass_overflows);
81 
82 #define	V_tcp_reass_maxseg		VNET(tcp_reass_maxseg)
83 #define	V_tcp_reass_maxqlen		VNET(tcp_reass_maxqlen)
84 #define	V_tcp_reass_overflows		VNET(tcp_reass_overflows)
85 
86 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
87     "TCP Segment Reassembly Queue");
88 
89 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
90     &VNET_NAME(tcp_reass_maxseg), 0,
91     "Global maximum number of TCP Segments in Reassembly Queue");
92 
93 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
94     &VNET_NAME(tcp_reass_qsize), 0,
95     "Global number of TCP Segments currently in Reassembly Queue");
96 
97 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
98     &VNET_NAME(tcp_reass_maxqlen), 0,
99     "Maximum number of TCP Segments per individual Reassembly Queue");
100 
101 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
102     &VNET_NAME(tcp_reass_overflows), 0,
103     "Global number of TCP Segment Reassembly Queue Overflows");
104 
105 /* Initialize TCP reassembly queue */
106 static void
107 tcp_reass_zone_change(void *tag)
108 {
109 
110 	V_tcp_reass_maxseg = nmbclusters / 16;
111 	uma_zone_set_max(V_tcp_reass_zone, V_tcp_reass_maxseg);
112 }
113 
114 VNET_DEFINE(uma_zone_t, tcp_reass_zone);
115 
116 void
117 tcp_reass_init(void)
118 {
119 
120 	V_tcp_reass_maxseg = 0;
121 	V_tcp_reass_qsize = 0;
122 	V_tcp_reass_maxqlen = 48;
123 	V_tcp_reass_overflows = 0;
124 
125 	V_tcp_reass_maxseg = nmbclusters / 16;
126 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
127 	    &V_tcp_reass_maxseg);
128 	V_tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
129 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
130 	uma_zone_set_max(V_tcp_reass_zone, V_tcp_reass_maxseg);
131 	EVENTHANDLER_REGISTER(nmbclusters_change,
132 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
133 }
134 
135 int
136 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
137 {
138 	struct tseg_qent *q;
139 	struct tseg_qent *p = NULL;
140 	struct tseg_qent *nq;
141 	struct tseg_qent *te = NULL;
142 	struct socket *so = tp->t_inpcb->inp_socket;
143 	int flags;
144 
145 	INP_WLOCK_ASSERT(tp->t_inpcb);
146 
147 	/*
148 	 * XXX: tcp_reass() is rather inefficient with its data structures
149 	 * and should be rewritten (see NetBSD for optimizations).
150 	 */
151 
152 	/*
153 	 * Call with th==NULL after become established to
154 	 * force pre-ESTABLISHED data up to user socket.
155 	 */
156 	if (th == NULL)
157 		goto present;
158 
159 	/*
160 	 * Limit the number of segments in the reassembly queue to prevent
161 	 * holding on to too many segments (and thus running out of mbufs).
162 	 * Make sure to let the missing segment through which caused this
163 	 * queue.  Always keep one global queue entry spare to be able to
164 	 * process the missing segment.
165 	 */
166 	if (th->th_seq != tp->rcv_nxt &&
167 	    (V_tcp_reass_qsize + 1 >= V_tcp_reass_maxseg ||
168 	     tp->t_segqlen >= V_tcp_reass_maxqlen)) {
169 		V_tcp_reass_overflows++;
170 		TCPSTAT_INC(tcps_rcvmemdrop);
171 		m_freem(m);
172 		*tlenp = 0;
173 		return (0);
174 	}
175 
176 	/*
177 	 * Allocate a new queue entry. If we can't, or hit the zone limit
178 	 * just drop the pkt.
179 	 */
180 	te = uma_zalloc(V_tcp_reass_zone, M_NOWAIT);
181 	if (te == NULL) {
182 		TCPSTAT_INC(tcps_rcvmemdrop);
183 		m_freem(m);
184 		*tlenp = 0;
185 		return (0);
186 	}
187 	tp->t_segqlen++;
188 	V_tcp_reass_qsize++;
189 
190 	/*
191 	 * Find a segment which begins after this one does.
192 	 */
193 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
194 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
195 			break;
196 		p = q;
197 	}
198 
199 	/*
200 	 * If there is a preceding segment, it may provide some of
201 	 * our data already.  If so, drop the data from the incoming
202 	 * segment.  If it provides all of our data, drop us.
203 	 */
204 	if (p != NULL) {
205 		int i;
206 		/* conversion to int (in i) handles seq wraparound */
207 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
208 		if (i > 0) {
209 			if (i >= *tlenp) {
210 				TCPSTAT_INC(tcps_rcvduppack);
211 				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
212 				m_freem(m);
213 				uma_zfree(V_tcp_reass_zone, te);
214 				tp->t_segqlen--;
215 				V_tcp_reass_qsize--;
216 				/*
217 				 * Try to present any queued data
218 				 * at the left window edge to the user.
219 				 * This is needed after the 3-WHS
220 				 * completes.
221 				 */
222 				goto present;	/* ??? */
223 			}
224 			m_adj(m, i);
225 			*tlenp -= i;
226 			th->th_seq += i;
227 		}
228 	}
229 	TCPSTAT_INC(tcps_rcvoopack);
230 	TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
231 
232 	/*
233 	 * While we overlap succeeding segments trim them or,
234 	 * if they are completely covered, dequeue them.
235 	 */
236 	while (q) {
237 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
238 		if (i <= 0)
239 			break;
240 		if (i < q->tqe_len) {
241 			q->tqe_th->th_seq += i;
242 			q->tqe_len -= i;
243 			m_adj(q->tqe_m, i);
244 			break;
245 		}
246 
247 		nq = LIST_NEXT(q, tqe_q);
248 		LIST_REMOVE(q, tqe_q);
249 		m_freem(q->tqe_m);
250 		uma_zfree(V_tcp_reass_zone, q);
251 		tp->t_segqlen--;
252 		V_tcp_reass_qsize--;
253 		q = nq;
254 	}
255 
256 	/* Insert the new segment queue entry into place. */
257 	te->tqe_m = m;
258 	te->tqe_th = th;
259 	te->tqe_len = *tlenp;
260 
261 	if (p == NULL) {
262 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
263 	} else {
264 		LIST_INSERT_AFTER(p, te, tqe_q);
265 	}
266 
267 present:
268 	/*
269 	 * Present data to user, advancing rcv_nxt through
270 	 * completed sequence space.
271 	 */
272 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
273 		return (0);
274 	q = LIST_FIRST(&tp->t_segq);
275 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
276 		return (0);
277 	SOCKBUF_LOCK(&so->so_rcv);
278 	do {
279 		tp->rcv_nxt += q->tqe_len;
280 		flags = q->tqe_th->th_flags & TH_FIN;
281 		nq = LIST_NEXT(q, tqe_q);
282 		LIST_REMOVE(q, tqe_q);
283 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
284 			m_freem(q->tqe_m);
285 		else
286 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
287 		uma_zfree(V_tcp_reass_zone, q);
288 		tp->t_segqlen--;
289 		V_tcp_reass_qsize--;
290 		q = nq;
291 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
292 	ND6_HINT(tp);
293 	sorwakeup_locked(so);
294 	return (flags);
295 }
296