xref: /freebsd/sys/netinet/tcp_reass.c (revision 2f02600abfddfc4e9f20dd384a2e729b451e16bd)
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/if_var.h>
53 #include <net/route.h>
54 #include <net/vnet.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/ip_options.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/in6_pcb.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/nd6.h>
67 #include <netinet/tcp.h>
68 #include <netinet/tcp_fsm.h>
69 #include <netinet/tcp_seq.h>
70 #include <netinet/tcp_timer.h>
71 #include <netinet/tcp_var.h>
72 #include <netinet6/tcp6_var.h>
73 #include <netinet/tcpip.h>
74 #ifdef TCPDEBUG
75 #include <netinet/tcp_debug.h>
76 #endif /* TCPDEBUG */
77 
78 static int tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS);
79 
80 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
81     "TCP Segment Reassembly Queue");
82 
83 static VNET_DEFINE(int, tcp_reass_maxseg) = 0;
84 #define	V_tcp_reass_maxseg		VNET(tcp_reass_maxseg)
85 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
86     &VNET_NAME(tcp_reass_maxseg), 0,
87     "Global maximum number of TCP Segments in Reassembly Queue");
88 
89 SYSCTL_VNET_PROC(_net_inet_tcp_reass, OID_AUTO, cursegments,
90     (CTLTYPE_INT | CTLFLAG_RD), NULL, 0, &tcp_reass_sysctl_qsize, "I",
91     "Global number of TCP Segments currently in Reassembly Queue");
92 
93 static VNET_DEFINE(int, tcp_reass_overflows) = 0;
94 #define	V_tcp_reass_overflows		VNET(tcp_reass_overflows)
95 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, overflows,
96     CTLTYPE_INT | CTLFLAG_RD,
97     &VNET_NAME(tcp_reass_overflows), 0,
98     "Global number of TCP Segment Reassembly Queue Overflows");
99 
100 static VNET_DEFINE(uma_zone_t, tcp_reass_zone);
101 #define	V_tcp_reass_zone		VNET(tcp_reass_zone)
102 
103 /* Initialize TCP reassembly queue */
104 static void
105 tcp_reass_zone_change(void *tag)
106 {
107 
108 	/* Set the zone limit and read back the effective value. */
109 	V_tcp_reass_maxseg = nmbclusters / 16;
110 	V_tcp_reass_maxseg = uma_zone_set_max(V_tcp_reass_zone,
111 	    V_tcp_reass_maxseg);
112 }
113 
114 void
115 tcp_reass_init(void)
116 {
117 
118 	V_tcp_reass_maxseg = nmbclusters / 16;
119 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
120 	    &V_tcp_reass_maxseg);
121 	V_tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
122 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
123 	/* Set the zone limit and read back the effective value. */
124 	V_tcp_reass_maxseg = uma_zone_set_max(V_tcp_reass_zone,
125 	    V_tcp_reass_maxseg);
126 	EVENTHANDLER_REGISTER(nmbclusters_change,
127 	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
128 }
129 
130 #ifdef VIMAGE
131 void
132 tcp_reass_destroy(void)
133 {
134 
135 	uma_zdestroy(V_tcp_reass_zone);
136 }
137 #endif
138 
139 void
140 tcp_reass_flush(struct tcpcb *tp)
141 {
142 	struct tseg_qent *qe;
143 
144 	INP_WLOCK_ASSERT(tp->t_inpcb);
145 
146 	while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) {
147 		LIST_REMOVE(qe, tqe_q);
148 		m_freem(qe->tqe_m);
149 		uma_zfree(V_tcp_reass_zone, qe);
150 		tp->t_segqlen--;
151 	}
152 
153 	KASSERT((tp->t_segqlen == 0),
154 	    ("TCP reass queue %p segment count is %d instead of 0 after flush.",
155 	    tp, tp->t_segqlen));
156 }
157 
158 static int
159 tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS)
160 {
161 	int qsize;
162 
163 	qsize = uma_zone_get_cur(V_tcp_reass_zone);
164 	return (sysctl_handle_int(oidp, &qsize, 0, req));
165 }
166 
167 int
168 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
169 {
170 	struct tseg_qent *q;
171 	struct tseg_qent *p = NULL;
172 	struct tseg_qent *nq;
173 	struct tseg_qent *te = NULL;
174 	struct socket *so = tp->t_inpcb->inp_socket;
175 	char *s = NULL;
176 	int flags;
177 	struct tseg_qent tqs;
178 
179 	INP_WLOCK_ASSERT(tp->t_inpcb);
180 
181 	/*
182 	 * XXX: tcp_reass() is rather inefficient with its data structures
183 	 * and should be rewritten (see NetBSD for optimizations).
184 	 */
185 
186 	/*
187 	 * Call with th==NULL after become established to
188 	 * force pre-ESTABLISHED data up to user socket.
189 	 */
190 	if (th == NULL)
191 		goto present;
192 
193 	/*
194 	 * Limit the number of segments that can be queued to reduce the
195 	 * potential for mbuf exhaustion. For best performance, we want to be
196 	 * able to queue a full window's worth of segments. The size of the
197 	 * socket receive buffer determines our advertised window and grows
198 	 * automatically when socket buffer autotuning is enabled. Use it as the
199 	 * basis for our queue limit.
200 	 * Always let the missing segment through which caused this queue.
201 	 * NB: Access to the socket buffer is left intentionally unlocked as we
202 	 * can tolerate stale information here.
203 	 *
204 	 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
205 	 * should work but causes packets to be dropped when they shouldn't.
206 	 * Investigate why and re-evaluate the below limit after the behaviour
207 	 * is understood.
208 	 */
209 	if (th->th_seq != tp->rcv_nxt &&
210 	    tp->t_segqlen >= (so->so_rcv.sb_hiwat / tp->t_maxseg) + 1) {
211 		V_tcp_reass_overflows++;
212 		TCPSTAT_INC(tcps_rcvmemdrop);
213 		m_freem(m);
214 		*tlenp = 0;
215 		if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
216 			log(LOG_DEBUG, "%s; %s: queue limit reached, "
217 			    "segment dropped\n", s, __func__);
218 			free(s, M_TCPLOG);
219 		}
220 		return (0);
221 	}
222 
223 	/*
224 	 * Allocate a new queue entry. If we can't, or hit the zone limit
225 	 * just drop the pkt.
226 	 *
227 	 * Use a temporary structure on the stack for the missing segment
228 	 * when the zone is exhausted. Otherwise we may get stuck.
229 	 */
230 	te = uma_zalloc(V_tcp_reass_zone, M_NOWAIT);
231 	if (te == NULL) {
232 		if (th->th_seq != tp->rcv_nxt) {
233 			TCPSTAT_INC(tcps_rcvmemdrop);
234 			m_freem(m);
235 			*tlenp = 0;
236 			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
237 			    NULL))) {
238 				log(LOG_DEBUG, "%s; %s: global zone limit "
239 				    "reached, segment dropped\n", s, __func__);
240 				free(s, M_TCPLOG);
241 			}
242 			return (0);
243 		} else {
244 			bzero(&tqs, sizeof(struct tseg_qent));
245 			te = &tqs;
246 			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
247 			    NULL))) {
248 				log(LOG_DEBUG,
249 				    "%s; %s: global zone limit reached, using "
250 				    "stack for missing segment\n", s, __func__);
251 				free(s, M_TCPLOG);
252 			}
253 		}
254 	}
255 	tp->t_segqlen++;
256 
257 	/*
258 	 * Find a segment which begins after this one does.
259 	 */
260 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
261 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
262 			break;
263 		p = q;
264 	}
265 
266 	/*
267 	 * If there is a preceding segment, it may provide some of
268 	 * our data already.  If so, drop the data from the incoming
269 	 * segment.  If it provides all of our data, drop us.
270 	 */
271 	if (p != NULL) {
272 		int i;
273 		/* conversion to int (in i) handles seq wraparound */
274 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
275 		if (i > 0) {
276 			if (i >= *tlenp) {
277 				TCPSTAT_INC(tcps_rcvduppack);
278 				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
279 				m_freem(m);
280 				uma_zfree(V_tcp_reass_zone, te);
281 				tp->t_segqlen--;
282 				/*
283 				 * Try to present any queued data
284 				 * at the left window edge to the user.
285 				 * This is needed after the 3-WHS
286 				 * completes.
287 				 */
288 				goto present;	/* ??? */
289 			}
290 			m_adj(m, i);
291 			*tlenp -= i;
292 			th->th_seq += i;
293 		}
294 	}
295 	tp->t_rcvoopack++;
296 	TCPSTAT_INC(tcps_rcvoopack);
297 	TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
298 
299 	/*
300 	 * While we overlap succeeding segments trim them or,
301 	 * if they are completely covered, dequeue them.
302 	 */
303 	while (q) {
304 		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
305 		if (i <= 0)
306 			break;
307 		if (i < q->tqe_len) {
308 			q->tqe_th->th_seq += i;
309 			q->tqe_len -= i;
310 			m_adj(q->tqe_m, i);
311 			break;
312 		}
313 
314 		nq = LIST_NEXT(q, tqe_q);
315 		LIST_REMOVE(q, tqe_q);
316 		m_freem(q->tqe_m);
317 		uma_zfree(V_tcp_reass_zone, q);
318 		tp->t_segqlen--;
319 		q = nq;
320 	}
321 
322 	/* Insert the new segment queue entry into place. */
323 	te->tqe_m = m;
324 	te->tqe_th = th;
325 	te->tqe_len = *tlenp;
326 
327 	if (p == NULL) {
328 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
329 	} else {
330 		KASSERT(te != &tqs, ("%s: temporary stack based entry not "
331 		    "first element in queue", __func__));
332 		LIST_INSERT_AFTER(p, te, tqe_q);
333 	}
334 
335 present:
336 	/*
337 	 * Present data to user, advancing rcv_nxt through
338 	 * completed sequence space.
339 	 */
340 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
341 		return (0);
342 	q = LIST_FIRST(&tp->t_segq);
343 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
344 		return (0);
345 	SOCKBUF_LOCK(&so->so_rcv);
346 	do {
347 		tp->rcv_nxt += q->tqe_len;
348 		flags = q->tqe_th->th_flags & TH_FIN;
349 		nq = LIST_NEXT(q, tqe_q);
350 		LIST_REMOVE(q, tqe_q);
351 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
352 			m_freem(q->tqe_m);
353 		else
354 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
355 		if (q != &tqs)
356 			uma_zfree(V_tcp_reass_zone, q);
357 		tp->t_segqlen--;
358 		q = nq;
359 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
360 	ND6_HINT(tp);
361 	sorwakeup_locked(so);
362 	return (flags);
363 }
364