]> CyberLeo.Net >> Repos - FreeBSD/stable/8.git/blob - sys/netinet/tcp_reass.c
MFC r204838:
[FreeBSD/stable/8.git] / sys / netinet / tcp_reass.c
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/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
77 static VNET_DEFINE(int, tcp_reass_maxseg);
78 VNET_DEFINE(int, tcp_reass_qsize);
79 static VNET_DEFINE(int, tcp_reass_maxqlen);
80 static VNET_DEFINE(int, tcp_reass_overflows);
81
82 #define V_tcp_reass_maxseg              VNET(tcp_reass_maxseg)
83 #define V_tcp_reass_maxqlen             VNET(tcp_reass_maxqlen)
84 #define V_tcp_reass_overflows           VNET(tcp_reass_overflows)
85
86 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
87     "TCP Segment Reassembly Queue");
88
89 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
90     &VNET_NAME(tcp_reass_maxseg), 0,
91     "Global maximum number of TCP Segments in Reassembly Queue");
92
93 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
94     &VNET_NAME(tcp_reass_qsize), 0,
95     "Global number of TCP Segments currently in Reassembly Queue");
96
97 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
98     &VNET_NAME(tcp_reass_maxqlen), 0,
99     "Maximum number of TCP Segments per individual Reassembly Queue");
100
101 SYSCTL_VNET_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
102     &VNET_NAME(tcp_reass_overflows), 0,
103     "Global number of TCP Segment Reassembly Queue Overflows");
104
105 /* Initialize TCP reassembly queue */
106 static void
107 tcp_reass_zone_change(void *tag)
108 {
109
110         V_tcp_reass_maxseg = nmbclusters / 16;
111         uma_zone_set_max(V_tcp_reass_zone, V_tcp_reass_maxseg);
112 }
113
114 VNET_DEFINE(uma_zone_t, tcp_reass_zone);
115
116 void
117 tcp_reass_init(void)
118 {
119
120         V_tcp_reass_maxseg = 0;
121         V_tcp_reass_qsize = 0;
122         V_tcp_reass_maxqlen = 48;
123         V_tcp_reass_overflows = 0;
124
125         V_tcp_reass_maxseg = nmbclusters / 16;
126         TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
127             &V_tcp_reass_maxseg);
128         V_tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
129             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
130         uma_zone_set_max(V_tcp_reass_zone, V_tcp_reass_maxseg);
131         EVENTHANDLER_REGISTER(nmbclusters_change,
132             tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
133 }
134
135 #ifdef VIMAGE
136 void
137 tcp_reass_destroy(void)
138 {
139
140         uma_zdestroy(V_tcp_reass_zone);
141 }
142 #endif
143
144 int
145 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
146 {
147         struct tseg_qent *q;
148         struct tseg_qent *p = NULL;
149         struct tseg_qent *nq;
150         struct tseg_qent *te = NULL;
151         struct socket *so = tp->t_inpcb->inp_socket;
152         int flags;
153
154         INP_WLOCK_ASSERT(tp->t_inpcb);
155
156         /*
157          * XXX: tcp_reass() is rather inefficient with its data structures
158          * and should be rewritten (see NetBSD for optimizations).
159          */
160
161         /*
162          * Call with th==NULL after become established to
163          * force pre-ESTABLISHED data up to user socket.
164          */
165         if (th == NULL)
166                 goto present;
167
168         /*
169          * Limit the number of segments in the reassembly queue to prevent
170          * holding on to too many segments (and thus running out of mbufs).
171          * Make sure to let the missing segment through which caused this
172          * queue.  Always keep one global queue entry spare to be able to
173          * process the missing segment.
174          */
175         if (th->th_seq != tp->rcv_nxt &&
176             (V_tcp_reass_qsize + 1 >= V_tcp_reass_maxseg ||
177              tp->t_segqlen >= V_tcp_reass_maxqlen)) {
178                 V_tcp_reass_overflows++;
179                 TCPSTAT_INC(tcps_rcvmemdrop);
180                 m_freem(m);
181                 *tlenp = 0;
182                 return (0);
183         }
184
185         /*
186          * Allocate a new queue entry. If we can't, or hit the zone limit
187          * just drop the pkt.
188          */
189         te = uma_zalloc(V_tcp_reass_zone, M_NOWAIT);
190         if (te == NULL) {
191                 TCPSTAT_INC(tcps_rcvmemdrop);
192                 m_freem(m);
193                 *tlenp = 0;
194                 return (0);
195         }
196         tp->t_segqlen++;
197         V_tcp_reass_qsize++;
198
199         /*
200          * Find a segment which begins after this one does.
201          */
202         LIST_FOREACH(q, &tp->t_segq, tqe_q) {
203                 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
204                         break;
205                 p = q;
206         }
207
208         /*
209          * If there is a preceding segment, it may provide some of
210          * our data already.  If so, drop the data from the incoming
211          * segment.  If it provides all of our data, drop us.
212          */
213         if (p != NULL) {
214                 int i;
215                 /* conversion to int (in i) handles seq wraparound */
216                 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
217                 if (i > 0) {
218                         if (i >= *tlenp) {
219                                 TCPSTAT_INC(tcps_rcvduppack);
220                                 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
221                                 m_freem(m);
222                                 uma_zfree(V_tcp_reass_zone, te);
223                                 tp->t_segqlen--;
224                                 V_tcp_reass_qsize--;
225                                 /*
226                                  * Try to present any queued data
227                                  * at the left window edge to the user.
228                                  * This is needed after the 3-WHS
229                                  * completes.
230                                  */
231                                 goto present;   /* ??? */
232                         }
233                         m_adj(m, i);
234                         *tlenp -= i;
235                         th->th_seq += i;
236                 }
237         }
238         TCPSTAT_INC(tcps_rcvoopack);
239         TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
240
241         /*
242          * While we overlap succeeding segments trim them or,
243          * if they are completely covered, dequeue them.
244          */
245         while (q) {
246                 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
247                 if (i <= 0)
248                         break;
249                 if (i < q->tqe_len) {
250                         q->tqe_th->th_seq += i;
251                         q->tqe_len -= i;
252                         m_adj(q->tqe_m, i);
253                         break;
254                 }
255
256                 nq = LIST_NEXT(q, tqe_q);
257                 LIST_REMOVE(q, tqe_q);
258                 m_freem(q->tqe_m);
259                 uma_zfree(V_tcp_reass_zone, q);
260                 tp->t_segqlen--;
261                 V_tcp_reass_qsize--;
262                 q = nq;
263         }
264
265         /* Insert the new segment queue entry into place. */
266         te->tqe_m = m;
267         te->tqe_th = th;
268         te->tqe_len = *tlenp;
269
270         if (p == NULL) {
271                 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
272         } else {
273                 LIST_INSERT_AFTER(p, te, tqe_q);
274         }
275
276 present:
277         /*
278          * Present data to user, advancing rcv_nxt through
279          * completed sequence space.
280          */
281         if (!TCPS_HAVEESTABLISHED(tp->t_state))
282                 return (0);
283         q = LIST_FIRST(&tp->t_segq);
284         if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
285                 return (0);
286         SOCKBUF_LOCK(&so->so_rcv);
287         do {
288                 tp->rcv_nxt += q->tqe_len;
289                 flags = q->tqe_th->th_flags & TH_FIN;
290                 nq = LIST_NEXT(q, tqe_q);
291                 LIST_REMOVE(q, tqe_q);
292                 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
293                         m_freem(q->tqe_m);
294                 else
295                         sbappendstream_locked(&so->so_rcv, q->tqe_m);
296                 uma_zfree(V_tcp_reass_zone, q);
297                 tp->t_segqlen--;
298                 V_tcp_reass_qsize--;
299                 q = nq;
300         } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
301         ND6_HINT(tp);
302         sorwakeup_locked(so);
303         return (flags);
304 }