xref: /freebsd/sys/netinet/tcp_reass.c (revision b70d2a2aa5003027b422e62435ab5bb9390d543c)
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  * 3. 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/eventhandler.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48 #include <sys/systm.h>
49 
50 #include <vm/uma.h>
51 
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/route.h>
55 #include <net/vnet.h>
56 
57 #include <netinet/in.h>
58 #include <netinet/in_pcb.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/ip_options.h>
64 #include <netinet/ip6.h>
65 #include <netinet6/in6_pcb.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/tcp.h>
69 #include <netinet/tcp_fsm.h>
70 #include <netinet/tcp_seq.h>
71 #include <netinet/tcp_timer.h>
72 #include <netinet/tcp_var.h>
73 #include <netinet6/tcp6_var.h>
74 #include <netinet/tcpip.h>
75 #ifdef TCPDEBUG
76 #include <netinet/tcp_debug.h>
77 #endif /* TCPDEBUG */
78 
79 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
80     "TCP Segment Reassembly Queue");
81 
82 static int tcp_reass_maxseg = 0;
83 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
84     &tcp_reass_maxseg, 0,
85     "Global maximum number of TCP Segments in Reassembly Queue");
86 
87 static uma_zone_t tcp_reass_zone;
88 SYSCTL_UMA_CUR(_net_inet_tcp_reass, OID_AUTO, cursegments, 0,
89     &tcp_reass_zone,
90     "Global number of TCP Segments currently in Reassembly Queue");
91 
92 /* Initialize TCP reassembly queue */
93 static void
94 tcp_reass_zone_change(void *tag)
95 {
96 
97 	/* Set the zone limit and read back the effective value. */
98 	tcp_reass_maxseg = nmbclusters / 16;
99 	tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
100 	    tcp_reass_maxseg);
101 }
102 
103 void
104 tcp_reass_global_init(void)
105 {
106 
107 	tcp_reass_maxseg = nmbclusters / 16;
108 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
109 	    &tcp_reass_maxseg);
110 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
111 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
112 	/* Set the zone limit and read back the effective value. */
113 	tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
114 	    tcp_reass_maxseg);
115 	EVENTHANDLER_REGISTER(nmbclusters_change,
116 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
117 }
118 
119 void
120 tcp_reass_flush(struct tcpcb *tp)
121 {
122 	struct tseg_qent *qe;
123 
124 	INP_WLOCK_ASSERT(tp->t_inpcb);
125 
126 	while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) {
127 		LIST_REMOVE(qe, tqe_q);
128 		m_freem(qe->tqe_m);
129 		uma_zfree(tcp_reass_zone, qe);
130 		tp->t_segqlen--;
131 	}
132 
133 	KASSERT((tp->t_segqlen == 0),
134 	    ("TCP reass queue %p segment count is %d instead of 0 after flush.",
135 	    tp, tp->t_segqlen));
136 }
137 
138 int
139 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
140 {
141 	struct tseg_qent *q;
142 	struct tseg_qent *p = NULL;
143 	struct tseg_qent *nq;
144 	struct tseg_qent *te = NULL;
145 	struct socket *so = tp->t_inpcb->inp_socket;
146 	char *s = NULL;
147 	int flags;
148 	struct tseg_qent tqs;
149 
150 	INP_WLOCK_ASSERT(tp->t_inpcb);
151 
152 	/*
153 	 * XXX: tcp_reass() is rather inefficient with its data structures
154 	 * and should be rewritten (see NetBSD for optimizations).
155 	 */
156 
157 	/*
158 	 * Call with th==NULL after become established to
159 	 * force pre-ESTABLISHED data up to user socket.
160 	 */
161 	if (th == NULL)
162 		goto present;
163 
164 	/*
165 	 * Limit the number of segments that can be queued to reduce the
166 	 * potential for mbuf exhaustion. For best performance, we want to be
167 	 * able to queue a full window's worth of segments. The size of the
168 	 * socket receive buffer determines our advertised window and grows
169 	 * automatically when socket buffer autotuning is enabled. Use it as the
170 	 * basis for our queue limit.
171 	 * Always let the missing segment through which caused this queue.
172 	 * NB: Access to the socket buffer is left intentionally unlocked as we
173 	 * can tolerate stale information here.
174 	 *
175 	 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
176 	 * should work but causes packets to be dropped when they shouldn't.
177 	 * Investigate why and re-evaluate the below limit after the behaviour
178 	 * is understood.
179 	 */
180 	if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
181 	    tp->t_segqlen >= (so->so_rcv.sb_hiwat / tp->t_maxseg) + 1) {
182 		TCPSTAT_INC(tcps_rcvreassfull);
183 		*tlenp = 0;
184 		if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
185 			log(LOG_DEBUG, "%s; %s: queue limit reached, "
186 			    "segment dropped\n", s, __func__);
187 			free(s, M_TCPLOG);
188 		}
189 		m_freem(m);
190 		return (0);
191 	}
192 
193 	/*
194 	 * Allocate a new queue entry. If we can't, or hit the zone limit
195 	 * just drop the pkt.
196 	 *
197 	 * Use a temporary structure on the stack for the missing segment
198 	 * when the zone is exhausted. Otherwise we may get stuck.
199 	 */
200 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
201 	if (te == NULL) {
202 		if (th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) {
203 			TCPSTAT_INC(tcps_rcvmemdrop);
204 			m_freem(m);
205 			*tlenp = 0;
206 			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
207 			    NULL))) {
208 				log(LOG_DEBUG, "%s; %s: global zone limit "
209 				    "reached, segment dropped\n", s, __func__);
210 				free(s, M_TCPLOG);
211 			}
212 			return (0);
213 		} else {
214 			bzero(&tqs, sizeof(struct tseg_qent));
215 			te = &tqs;
216 			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
217 			    NULL))) {
218 				log(LOG_DEBUG,
219 				    "%s; %s: global zone limit reached, using "
220 				    "stack for missing segment\n", s, __func__);
221 				free(s, M_TCPLOG);
222 			}
223 		}
224 	}
225 	tp->t_segqlen++;
226 
227 	/*
228 	 * Find a segment which begins after this one does.
229 	 */
230 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
231 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
232 			break;
233 		p = q;
234 	}
235 
236 	/*
237 	 * If there is a preceding segment, it may provide some of
238 	 * our data already.  If so, drop the data from the incoming
239 	 * segment.  If it provides all of our data, drop us.
240 	 */
241 	if (p != NULL) {
242 		int i;
243 		/* conversion to int (in i) handles seq wraparound */
244 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
245 		if (i > 0) {
246 			if (i >= *tlenp) {
247 				TCPSTAT_INC(tcps_rcvduppack);
248 				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
249 				m_freem(m);
250 				if (te != &tqs)
251 					uma_zfree(tcp_reass_zone, te);
252 				tp->t_segqlen--;
253 				/*
254 				 * Try to present any queued data
255 				 * at the left window edge to the user.
256 				 * This is needed after the 3-WHS
257 				 * completes.
258 				 */
259 				goto present;	/* ??? */
260 			}
261 			m_adj(m, i);
262 			*tlenp -= i;
263 			th->th_seq += i;
264 		}
265 	}
266 	tp->t_rcvoopack++;
267 	TCPSTAT_INC(tcps_rcvoopack);
268 	TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
269 
270 	/*
271 	 * While we overlap succeeding segments trim them or,
272 	 * if they are completely covered, dequeue them.
273 	 */
274 	while (q) {
275 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
276 		if (i <= 0)
277 			break;
278 		if (i < q->tqe_len) {
279 			q->tqe_th->th_seq += i;
280 			q->tqe_len -= i;
281 			m_adj(q->tqe_m, i);
282 			break;
283 		}
284 
285 		nq = LIST_NEXT(q, tqe_q);
286 		LIST_REMOVE(q, tqe_q);
287 		m_freem(q->tqe_m);
288 		uma_zfree(tcp_reass_zone, q);
289 		tp->t_segqlen--;
290 		q = nq;
291 	}
292 
293 	/* Insert the new segment queue entry into place. */
294 	te->tqe_m = m;
295 	te->tqe_th = th;
296 	te->tqe_len = *tlenp;
297 
298 	if (p == NULL) {
299 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
300 	} else {
301 		KASSERT(te != &tqs, ("%s: temporary stack based entry not "
302 		    "first element in queue", __func__));
303 		LIST_INSERT_AFTER(p, te, tqe_q);
304 	}
305 
306 present:
307 	/*
308 	 * Present data to user, advancing rcv_nxt through
309 	 * completed sequence space.
310 	 */
311 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
312 		return (0);
313 	q = LIST_FIRST(&tp->t_segq);
314 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
315 		return (0);
316 	SOCKBUF_LOCK(&so->so_rcv);
317 	do {
318 		tp->rcv_nxt += q->tqe_len;
319 		flags = q->tqe_th->th_flags & TH_FIN;
320 		nq = LIST_NEXT(q, tqe_q);
321 		LIST_REMOVE(q, tqe_q);
322 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
323 			m_freem(q->tqe_m);
324 		else
325 			sbappendstream_locked(&so->so_rcv, q->tqe_m, 0);
326 		if (q != &tqs)
327 			uma_zfree(tcp_reass_zone, q);
328 		tp->t_segqlen--;
329 		q = nq;
330 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
331 	sorwakeup_locked(so);
332 	return (flags);
333 }
334