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/7.4/sys/netinet/tcp_reass.c 214890 2010-11-06 15:49:59Z lstewart $");
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
54 #include <netinet/in.h>
55 #include <netinet/in_pcb.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/ip_options.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/in6_pcb.h>
63 #include <netinet6/ip6_var.h>
64 #include <netinet6/nd6.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet6/tcp6_var.h>
71 #include <netinet/tcpip.h>
72 #ifdef TCPDEBUG
73 #include <netinet/tcp_debug.h>
74 #endif /* TCPDEBUG */
75
76 static int tcp_reass_sysctl_maxseg(SYSCTL_HANDLER_ARGS);
77 static int tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS);
78
79 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_PROC(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
84 &tcp_reass_maxseg, 0, &tcp_reass_sysctl_maxseg, "I",
85 "Global maximum number of TCP Segments in Reassembly Queue");
86
87 static int tcp_reass_qsize = 0;
88 SYSCTL_PROC(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
89 &tcp_reass_qsize, 0, &tcp_reass_sysctl_qsize, "I",
90 "Global number of TCP Segments currently in Reassembly Queue");
91
92 static int tcp_reass_overflows = 0;
93 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
94 &tcp_reass_overflows, 0,
95 "Global number of TCP Segment Reassembly Queue Overflows");
96
97 static uma_zone_t tcp_reass_zone;
98
99 /* Initialize TCP reassembly queue */
100 static void
101 tcp_reass_zone_change(void *tag)
102 {
103
104 tcp_reass_maxseg = nmbclusters / 16;
105 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
106 }
107
108 void
109 tcp_reass_init(void)
110 {
111
112 tcp_reass_maxseg = nmbclusters / 16;
113 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
114 &tcp_reass_maxseg);
115 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
116 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
117 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
118 EVENTHANDLER_REGISTER(nmbclusters_change,
119 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
120 }
121
122 void
123 tcp_reass_flush(struct tcpcb *tp)
124 {
125 struct tseg_qent *qe;
126
127 INP_WLOCK_ASSERT(tp->t_inpcb);
128
129 while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) {
130 LIST_REMOVE(qe, tqe_q);
131 m_freem(qe->tqe_m);
132 uma_zfree(tcp_reass_zone, qe);
133 tp->t_segqlen--;
134 }
135
136 KASSERT((tp->t_segqlen == 0),
137 ("TCP reass queue %p segment count is %d instead of 0 after flush.",
138 tp, tp->t_segqlen));
139 }
140
141 static int
142 tcp_reass_sysctl_maxseg(SYSCTL_HANDLER_ARGS)
143 {
144 tcp_reass_maxseg = uma_zone_get_max(tcp_reass_zone);
145 return (sysctl_handle_int(oidp, arg1, arg2, req));
146 }
147
148 static int
149 tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS)
150 {
151 tcp_reass_qsize = uma_zone_get_cur(tcp_reass_zone);
152 return (sysctl_handle_int(oidp, arg1, arg2, req));
153 }
154
155 int
156 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
157 {
158 struct tseg_qent *q;
159 struct tseg_qent *p = NULL;
160 struct tseg_qent *nq;
161 struct tseg_qent *te = NULL;
162 struct socket *so = tp->t_inpcb->inp_socket;
163 int flags;
164
165 INP_WLOCK_ASSERT(tp->t_inpcb);
166
167 /*
168 * XXX: tcp_reass() is rather inefficient with its data structures
169 * and should be rewritten (see NetBSD for optimizations). While
170 * doing that it should move to its own file tcp_reass.c.
171 */
172
173 /*
174 * Call with th==NULL after become established to
175 * force pre-ESTABLISHED data up to user socket.
176 */
177 if (th == NULL)
178 goto present;
179
180 /*
181 * Limit the number of segments that can be queued to reduce the
182 * potential for mbuf exhaustion. For best performance, we want to be
183 * able to queue a full window's worth of segments. The size of the
184 * socket receive buffer determines our advertised window and grows
185 * automatically when socket buffer autotuning is enabled. Use it as the
186 * basis for our queue limit.
187 * Always let the missing segment through which caused this queue.
188 * NB: Access to the socket buffer is left intentionally unlocked as we
189 * can tolerate stale information here.
190 *
191 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
192 * should work but causes packets to be dropped when they shouldn't.
193 * Investigate why and re-evaluate the below limit after the behaviour
194 * is understood.
195 */
196 if (th->th_seq != tp->rcv_nxt &&
197 tp->t_segqlen >= (so->so_rcv.sb_hiwat / tp->t_maxseg) + 1) {
198 tcp_reass_overflows++;
199 tcpstat.tcps_rcvmemdrop++;
200 m_freem(m);
201 *tlenp = 0;
202 return (0);
203 }
204
205 /*
206 * Allocate a new queue entry. If we can't, or hit the zone limit
207 * just drop the pkt.
208 */
209 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
210 if (te == NULL) {
211 tcpstat.tcps_rcvmemdrop++;
212 m_freem(m);
213 *tlenp = 0;
214 return (0);
215 }
216 tp->t_segqlen++;
217
218 /*
219 * Find a segment which begins after this one does.
220 */
221 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
222 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
223 break;
224 p = q;
225 }
226
227 /*
228 * If there is a preceding segment, it may provide some of
229 * our data already. If so, drop the data from the incoming
230 * segment. If it provides all of our data, drop us.
231 */
232 if (p != NULL) {
233 int i;
234 /* conversion to int (in i) handles seq wraparound */
235 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
236 if (i > 0) {
237 if (i >= *tlenp) {
238 tcpstat.tcps_rcvduppack++;
239 tcpstat.tcps_rcvdupbyte += *tlenp;
240 m_freem(m);
241 uma_zfree(tcp_reass_zone, te);
242 tp->t_segqlen--;
243 /*
244 * Try to present any queued data
245 * at the left window edge to the user.
246 * This is needed after the 3-WHS
247 * completes.
248 */
249 goto present; /* ??? */
250 }
251 m_adj(m, i);
252 *tlenp -= i;
253 th->th_seq += i;
254 }
255 }
256 tcpstat.tcps_rcvoopack++;
257 tcpstat.tcps_rcvoobyte += *tlenp;
258
259 /*
260 * While we overlap succeeding segments trim them or,
261 * if they are completely covered, dequeue them.
262 */
263 while (q) {
264 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
265 if (i <= 0)
266 break;
267 if (i < q->tqe_len) {
268 q->tqe_th->th_seq += i;
269 q->tqe_len -= i;
270 m_adj(q->tqe_m, i);
271 break;
272 }
273
274 nq = LIST_NEXT(q, tqe_q);
275 LIST_REMOVE(q, tqe_q);
276 m_freem(q->tqe_m);
277 uma_zfree(tcp_reass_zone, q);
278 tp->t_segqlen--;
279 q = nq;
280 }
281
282 /* Insert the new segment queue entry into place. */
283 te->tqe_m = m;
284 te->tqe_th = th;
285 te->tqe_len = *tlenp;
286
287 if (p == NULL) {
288 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
289 } else {
290 LIST_INSERT_AFTER(p, te, tqe_q);
291 }
292
293 present:
294 /*
295 * Present data to user, advancing rcv_nxt through
296 * completed sequence space.
297 */
298 if (!TCPS_HAVEESTABLISHED(tp->t_state))
299 return (0);
300 q = LIST_FIRST(&tp->t_segq);
301 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
302 return (0);
303 SOCKBUF_LOCK(&so->so_rcv);
304 do {
305 tp->rcv_nxt += q->tqe_len;
306 flags = q->tqe_th->th_flags & TH_FIN;
307 nq = LIST_NEXT(q, tqe_q);
308 LIST_REMOVE(q, tqe_q);
309 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
310 m_freem(q->tqe_m);
311 else
312 sbappendstream_locked(&so->so_rcv, q->tqe_m);
313 uma_zfree(tcp_reass_zone, q);
314 tp->t_segqlen--;
315 q = nq;
316 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
317 ND6_HINT(tp);
318 sorwakeup_locked(so);
319 return (flags);
320 }
Cache object: 3878e5968a95763e6f769fa3815bccd3
|