tcp_reass.c revision 265124 
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: releng/10.0/sys/netinet/tcp_reass.c 265124 2014-04-30 04:04:42Z delphij $");
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
77static int tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS);
78
79static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
80    "TCP Segment Reassembly Queue");
81
82static VNET_DEFINE(int, tcp_reass_maxseg) = 0;
83#define	V_tcp_reass_maxseg		VNET(tcp_reass_maxseg)
84SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
85    &VNET_NAME(tcp_reass_maxseg), 0,
86    "Global maximum number of TCP Segments in Reassembly Queue");
87
88SYSCTL_VNET_PROC(_net_inet_tcp_reass, OID_AUTO, cursegments,
89    (CTLTYPE_INT | CTLFLAG_RD), NULL, 0, &tcp_reass_sysctl_qsize, "I",
90    "Global number of TCP Segments currently in Reassembly Queue");
91
92static VNET_DEFINE(int, tcp_reass_overflows) = 0;
93#define	V_tcp_reass_overflows		VNET(tcp_reass_overflows)
94SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, overflows,
95    CTLTYPE_INT | CTLFLAG_RD,
96    &VNET_NAME(tcp_reass_overflows), 0,
97    "Global number of TCP Segment Reassembly Queue Overflows");
98
99static VNET_DEFINE(uma_zone_t, tcp_reass_zone);
100#define	V_tcp_reass_zone		VNET(tcp_reass_zone)
101
102/* Initialize TCP reassembly queue */
103static void
104tcp_reass_zone_change(void *tag)
105{
106
107	/* Set the zone limit and read back the effective value. */
108	V_tcp_reass_maxseg = nmbclusters / 16;
109	V_tcp_reass_maxseg = uma_zone_set_max(V_tcp_reass_zone,
110	    V_tcp_reass_maxseg);
111}
112
113void
114tcp_reass_init(void)
115{
116
117	V_tcp_reass_maxseg = nmbclusters / 16;
118	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
119	    &V_tcp_reass_maxseg);
120	V_tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
121	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
122	/* Set the zone limit and read back the effective value. */
123	V_tcp_reass_maxseg = uma_zone_set_max(V_tcp_reass_zone,
124	    V_tcp_reass_maxseg);
125	EVENTHANDLER_REGISTER(nmbclusters_change,
126	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
127}
128
129#ifdef VIMAGE
130void
131tcp_reass_destroy(void)
132{
133
134	uma_zdestroy(V_tcp_reass_zone);
135}
136#endif
137
138void
139tcp_reass_flush(struct tcpcb *tp)
140{
141	struct tseg_qent *qe;
142
143	INP_WLOCK_ASSERT(tp->t_inpcb);
144
145	while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) {
146		LIST_REMOVE(qe, tqe_q);
147		m_freem(qe->tqe_m);
148		uma_zfree(V_tcp_reass_zone, qe);
149		tp->t_segqlen--;
150	}
151
152	KASSERT((tp->t_segqlen == 0),
153	    ("TCP reass queue %p segment count is %d instead of 0 after flush.",
154	    tp, tp->t_segqlen));
155}
156
157static int
158tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS)
159{
160	int qsize;
161
162	qsize = uma_zone_get_cur(V_tcp_reass_zone);
163	return (sysctl_handle_int(oidp, &qsize, 0, req));
164}
165
166int
167tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
168{
169	struct tseg_qent *q;
170	struct tseg_qent *p = NULL;
171	struct tseg_qent *nq;
172	struct tseg_qent *te = NULL;
173	struct socket *so = tp->t_inpcb->inp_socket;
174	char *s = NULL;
175	int flags;
176	struct tseg_qent tqs;
177
178	INP_WLOCK_ASSERT(tp->t_inpcb);
179
180	/*
181	 * XXX: tcp_reass() is rather inefficient with its data structures
182	 * and should be rewritten (see NetBSD for optimizations).
183	 */
184
185	/*
186	 * Call with th==NULL after become established to
187	 * force pre-ESTABLISHED data up to user socket.
188	 */
189	if (th == NULL)
190		goto present;
191
192	/*
193	 * Limit the number of segments that can be queued to reduce the
194	 * potential for mbuf exhaustion. For best performance, we want to be
195	 * able to queue a full window's worth of segments. The size of the
196	 * socket receive buffer determines our advertised window and grows
197	 * automatically when socket buffer autotuning is enabled. Use it as the
198	 * basis for our queue limit.
199	 * Always let the missing segment through which caused this queue.
200	 * NB: Access to the socket buffer is left intentionally unlocked as we
201	 * can tolerate stale information here.
202	 *
203	 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
204	 * should work but causes packets to be dropped when they shouldn't.
205	 * Investigate why and re-evaluate the below limit after the behaviour
206	 * is understood.
207	 */
208	if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
209	    tp->t_segqlen >= (so->so_rcv.sb_hiwat / tp->t_maxseg) + 1) {
210		V_tcp_reass_overflows++;
211		TCPSTAT_INC(tcps_rcvmemdrop);
212		m_freem(m);
213		*tlenp = 0;
214		if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
215			log(LOG_DEBUG, "%s; %s: queue limit reached, "
216			    "segment dropped\n", s, __func__);
217			free(s, M_TCPLOG);
218		}
219		return (0);
220	}
221
222	/*
223	 * Allocate a new queue entry. If we can't, or hit the zone limit
224	 * just drop the pkt.
225	 *
226	 * Use a temporary structure on the stack for the missing segment
227	 * when the zone is exhausted. Otherwise we may get stuck.
228	 */
229	te = uma_zalloc(V_tcp_reass_zone, M_NOWAIT);
230	if (te == NULL) {
231		if (th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) {
232			TCPSTAT_INC(tcps_rcvmemdrop);
233			m_freem(m);
234			*tlenp = 0;
235			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
236			    NULL))) {
237				log(LOG_DEBUG, "%s; %s: global zone limit "
238				    "reached, segment dropped\n", s, __func__);
239				free(s, M_TCPLOG);
240			}
241			return (0);
242		} else {
243			bzero(&tqs, sizeof(struct tseg_qent));
244			te = &tqs;
245			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
246			    NULL))) {
247				log(LOG_DEBUG,
248				    "%s; %s: global zone limit reached, using "
249				    "stack for missing segment\n", s, __func__);
250				free(s, M_TCPLOG);
251			}
252		}
253	}
254	tp->t_segqlen++;
255
256	/*
257	 * Find a segment which begins after this one does.
258	 */
259	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
260		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
261			break;
262		p = q;
263	}
264
265	/*
266	 * If there is a preceding segment, it may provide some of
267	 * our data already.  If so, drop the data from the incoming
268	 * segment.  If it provides all of our data, drop us.
269	 */
270	if (p != NULL) {
271		int i;
272		/* conversion to int (in i) handles seq wraparound */
273		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
274		if (i > 0) {
275			if (i >= *tlenp) {
276				TCPSTAT_INC(tcps_rcvduppack);
277				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
278				m_freem(m);
279				if (te != &tqs)
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
335present:
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