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: stable/10/sys/netinet/tcp_reass.c 337386 2018-08-06 17:46:28Z jtl $");
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 int tcp_reass_maxseg = 0;
83SYSCTL_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
87SYSCTL_PROC(_net_inet_tcp_reass, OID_AUTO, cursegments,
88    (CTLTYPE_INT | CTLFLAG_RD), NULL, 0, &tcp_reass_sysctl_qsize, "I",
89    "Global number of TCP Segments currently in Reassembly Queue");
90
91static int tcp_reass_overflows = 0;
92SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows,
93    CTLFLAG_RD,
94    &tcp_reass_overflows, 0,
95    "Global number of TCP Segment Reassembly Queue Overflows");
96
97static uma_zone_t tcp_reass_zone;
98
99static u_int tcp_reass_maxqueuelen = 100;
100SYSCTL_UINT(_net_inet_tcp_reass, OID_AUTO, maxqueuelen, CTLFLAG_RWTUN,
101    &tcp_reass_maxqueuelen, 0,
102    "Maximum number of TCP Segments per Reassembly Queue");
103
104/* Initialize TCP reassembly queue */
105static void
106tcp_reass_zone_change(void *tag)
107{
108
109	/* Set the zone limit and read back the effective value. */
110	tcp_reass_maxseg = nmbclusters / 16;
111	tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
112	    tcp_reass_maxseg);
113}
114
115void
116tcp_reass_global_init(void)
117{
118
119	tcp_reass_maxseg = nmbclusters / 16;
120	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
121	    &tcp_reass_maxseg);
122	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
123	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
124	/* Set the zone limit and read back the effective value. */
125	tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
126	    tcp_reass_maxseg);
127	EVENTHANDLER_REGISTER(nmbclusters_change,
128	    tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
129}
130
131void
132tcp_reass_flush(struct tcpcb *tp)
133{
134	struct tseg_qent *qe;
135
136	INP_WLOCK_ASSERT(tp->t_inpcb);
137
138	while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) {
139		LIST_REMOVE(qe, tqe_q);
140		m_freem(qe->tqe_m);
141		uma_zfree(tcp_reass_zone, qe);
142		tp->t_segqlen--;
143	}
144
145	KASSERT((tp->t_segqlen == 0),
146	    ("TCP reass queue %p segment count is %d instead of 0 after flush.",
147	    tp, tp->t_segqlen));
148}
149
150static int
151tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS)
152{
153	int qsize;
154
155	qsize = uma_zone_get_cur(tcp_reass_zone);
156	return (sysctl_handle_int(oidp, &qsize, 0, req));
157}
158
159int
160tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
161{
162	struct tseg_qent *q;
163	struct tseg_qent *p = NULL;
164	struct tseg_qent *nq;
165	struct tseg_qent *te = NULL;
166	struct socket *so = tp->t_inpcb->inp_socket;
167	char *s = NULL;
168	int flags;
169	struct tseg_qent tqs;
170
171	INP_WLOCK_ASSERT(tp->t_inpcb);
172
173	/*
174	 * XXX: tcp_reass() is rather inefficient with its data structures
175	 * and should be rewritten (see NetBSD for optimizations).
176	 */
177
178	/*
179	 * Call with th==NULL after become established to
180	 * force pre-ESTABLISHED data up to user socket.
181	 */
182	if (th == NULL)
183		goto present;
184
185	/*
186	 * Limit the number of segments that can be queued to reduce the
187	 * potential for mbuf exhaustion. For best performance, we want to be
188	 * able to queue a full window's worth of segments. The size of the
189	 * socket receive buffer determines our advertised window and grows
190	 * automatically when socket buffer autotuning is enabled. Use it as the
191	 * basis for our queue limit.
192	 *
193	 * However, allow the user to specify a ceiling for the number of
194	 * segments in each queue.
195	 *
196	 * Always let the missing segment through which caused this queue.
197	 * NB: Access to the socket buffer is left intentionally unlocked as we
198	 * can tolerate stale information here.
199	 *
200	 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
201	 * should work but causes packets to be dropped when they shouldn't.
202	 * Investigate why and re-evaluate the below limit after the behaviour
203	 * is understood.
204	 */
205	if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
206	    tp->t_segqlen >= min((so->so_rcv.sb_hiwat / tp->t_maxseg) + 1,
207	    tcp_reass_maxqueuelen)) {
208		tcp_reass_overflows++;
209		TCPSTAT_INC(tcps_rcvmemdrop);
210		m_freem(m);
211		*tlenp = 0;
212		if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
213			log(LOG_DEBUG, "%s; %s: queue limit reached, "
214			    "segment dropped\n", s, __func__);
215			free(s, M_TCPLOG);
216		}
217		return (0);
218	}
219
220	/*
221	 * Allocate a new queue entry. If we can't, or hit the zone limit
222	 * just drop the pkt.
223	 *
224	 * Use a temporary structure on the stack for the missing segment
225	 * when the zone is exhausted. Otherwise we may get stuck.
226	 */
227	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
228	if (te == NULL) {
229		if (th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) {
230			TCPSTAT_INC(tcps_rcvmemdrop);
231			m_freem(m);
232			*tlenp = 0;
233			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
234			    NULL))) {
235				log(LOG_DEBUG, "%s; %s: global zone limit "
236				    "reached, segment dropped\n", s, __func__);
237				free(s, M_TCPLOG);
238			}
239			return (0);
240		} else {
241			bzero(&tqs, sizeof(struct tseg_qent));
242			te = &tqs;
243			if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
244			    NULL))) {
245				log(LOG_DEBUG,
246				    "%s; %s: global zone limit reached, using "
247				    "stack for missing segment\n", s, __func__);
248				free(s, M_TCPLOG);
249			}
250		}
251	}
252	tp->t_segqlen++;
253
254	/*
255	 * Find a segment which begins after this one does.
256	 */
257	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
258		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
259			break;
260		p = q;
261	}
262
263	/*
264	 * If there is a preceding segment, it may provide some of
265	 * our data already.  If so, drop the data from the incoming
266	 * segment.  If it provides all of our data, drop us.
267	 */
268	if (p != NULL) {
269		int i;
270		/* conversion to int (in i) handles seq wraparound */
271		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
272		if (i > 0) {
273			if (i >= *tlenp) {
274				TCPSTAT_INC(tcps_rcvduppack);
275				TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
276				m_freem(m);
277				if (te != &tqs)
278					uma_zfree(tcp_reass_zone, te);
279				tp->t_segqlen--;
280				/*
281				 * Try to present any queued data
282				 * at the left window edge to the user.
283				 * This is needed after the 3-WHS
284				 * completes.
285				 */
286				goto present;	/* ??? */
287			}
288			m_adj(m, i);
289			*tlenp -= i;
290			th->th_seq += i;
291		}
292	}
293	tp->t_rcvoopack++;
294	TCPSTAT_INC(tcps_rcvoopack);
295	TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
296
297	/*
298	 * While we overlap succeeding segments trim them or,
299	 * if they are completely covered, dequeue them.
300	 */
301	while (q) {
302		int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
303		if (i <= 0)
304			break;
305		if (i < q->tqe_len) {
306			q->tqe_th->th_seq += i;
307			q->tqe_len -= i;
308			m_adj(q->tqe_m, i);
309			break;
310		}
311
312		nq = LIST_NEXT(q, tqe_q);
313		LIST_REMOVE(q, tqe_q);
314		m_freem(q->tqe_m);
315		uma_zfree(tcp_reass_zone, q);
316		tp->t_segqlen--;
317		q = nq;
318	}
319
320	/* Insert the new segment queue entry into place. */
321	te->tqe_m = m;
322	te->tqe_th = th;
323	te->tqe_len = *tlenp;
324
325	if (p == NULL) {
326		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
327	} else {
328		KASSERT(te != &tqs, ("%s: temporary stack based entry not "
329		    "first element in queue", __func__));
330		LIST_INSERT_AFTER(p, te, tqe_q);
331	}
332
333present:
334	/*
335	 * Present data to user, advancing rcv_nxt through
336	 * completed sequence space.
337	 */
338	if (!TCPS_HAVEESTABLISHED(tp->t_state))
339		return (0);
340	q = LIST_FIRST(&tp->t_segq);
341	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
342		return (0);
343	SOCKBUF_LOCK(&so->so_rcv);
344	do {
345		tp->rcv_nxt += q->tqe_len;
346		flags = q->tqe_th->th_flags & TH_FIN;
347		nq = LIST_NEXT(q, tqe_q);
348		LIST_REMOVE(q, tqe_q);
349		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
350			m_freem(q->tqe_m);
351		else
352			sbappendstream_locked(&so->so_rcv, q->tqe_m);
353		if (q != &tqs)
354			uma_zfree(tcp_reass_zone, q);
355		tp->t_segqlen--;
356		q = nq;
357	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
358	ND6_HINT(tp);
359	sorwakeup_locked(so);
360	return (flags);
361}
362