tcp_reass.c revision 285978
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/10.2/sys/netinet/tcp_reass.c 285978 2015-07-28 19:59:04Z 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 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 99/* Initialize TCP reassembly queue */ 100static void 101tcp_reass_zone_change(void *tag) 102{ 103 104 /* Set the zone limit and read back the effective value. */ 105 tcp_reass_maxseg = nmbclusters / 16; 106 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone, 107 tcp_reass_maxseg); 108} 109 110void 111tcp_reass_global_init(void) 112{ 113 114 tcp_reass_maxseg = nmbclusters / 16; 115 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 116 &tcp_reass_maxseg); 117 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 118 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 119 /* Set the zone limit and read back the effective value. */ 120 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone, 121 tcp_reass_maxseg); 122 EVENTHANDLER_REGISTER(nmbclusters_change, 123 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 124} 125 126void 127tcp_reass_flush(struct tcpcb *tp) 128{ 129 struct tseg_qent *qe; 130 131 INP_WLOCK_ASSERT(tp->t_inpcb); 132 133 while ((qe = LIST_FIRST(&tp->t_segq)) != NULL) { 134 LIST_REMOVE(qe, tqe_q); 135 m_freem(qe->tqe_m); 136 uma_zfree(tcp_reass_zone, qe); 137 tp->t_segqlen--; 138 } 139 140 KASSERT((tp->t_segqlen == 0), 141 ("TCP reass queue %p segment count is %d instead of 0 after flush.", 142 tp, tp->t_segqlen)); 143} 144 145static int 146tcp_reass_sysctl_qsize(SYSCTL_HANDLER_ARGS) 147{ 148 int qsize; 149 150 qsize = uma_zone_get_cur(tcp_reass_zone); 151 return (sysctl_handle_int(oidp, &qsize, 0, req)); 152} 153 154int 155tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 156{ 157 struct tseg_qent *q; 158 struct tseg_qent *p = NULL; 159 struct tseg_qent *nq; 160 struct tseg_qent *te = NULL; 161 struct socket *so = tp->t_inpcb->inp_socket; 162 char *s = NULL; 163 int flags; 164 struct tseg_qent tqs; 165 166 INP_WLOCK_ASSERT(tp->t_inpcb); 167 168 /* 169 * XXX: tcp_reass() is rather inefficient with its data structures 170 * and should be rewritten (see NetBSD for optimizations). 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 || !TCPS_HAVEESTABLISHED(tp->t_state)) && 197 tp->t_segqlen >= (so->so_rcv.sb_hiwat / tp->t_maxseg) + 1) { 198 tcp_reass_overflows++; 199 TCPSTAT_INC(tcps_rcvmemdrop); 200 m_freem(m); 201 *tlenp = 0; 202 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 203 log(LOG_DEBUG, "%s; %s: queue limit reached, " 204 "segment dropped\n", s, __func__); 205 free(s, M_TCPLOG); 206 } 207 return (0); 208 } 209 210 /* 211 * Allocate a new queue entry. If we can't, or hit the zone limit 212 * just drop the pkt. 213 * 214 * Use a temporary structure on the stack for the missing segment 215 * when the zone is exhausted. Otherwise we may get stuck. 216 */ 217 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 218 if (te == NULL) { 219 if (th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) { 220 TCPSTAT_INC(tcps_rcvmemdrop); 221 m_freem(m); 222 *tlenp = 0; 223 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, 224 NULL))) { 225 log(LOG_DEBUG, "%s; %s: global zone limit " 226 "reached, segment dropped\n", s, __func__); 227 free(s, M_TCPLOG); 228 } 229 return (0); 230 } else { 231 bzero(&tqs, sizeof(struct tseg_qent)); 232 te = &tqs; 233 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, 234 NULL))) { 235 log(LOG_DEBUG, 236 "%s; %s: global zone limit reached, using " 237 "stack for missing segment\n", s, __func__); 238 free(s, M_TCPLOG); 239 } 240 } 241 } 242 tp->t_segqlen++; 243 244 /* 245 * Find a segment which begins after this one does. 246 */ 247 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 248 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 249 break; 250 p = q; 251 } 252 253 /* 254 * If there is a preceding segment, it may provide some of 255 * our data already. If so, drop the data from the incoming 256 * segment. If it provides all of our data, drop us. 257 */ 258 if (p != NULL) { 259 int i; 260 /* conversion to int (in i) handles seq wraparound */ 261 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 262 if (i > 0) { 263 if (i >= *tlenp) { 264 TCPSTAT_INC(tcps_rcvduppack); 265 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp); 266 m_freem(m); 267 if (te != &tqs) 268 uma_zfree(tcp_reass_zone, te); 269 tp->t_segqlen--; 270 /* 271 * Try to present any queued data 272 * at the left window edge to the user. 273 * This is needed after the 3-WHS 274 * completes. 275 */ 276 goto present; /* ??? */ 277 } 278 m_adj(m, i); 279 *tlenp -= i; 280 th->th_seq += i; 281 } 282 } 283 tp->t_rcvoopack++; 284 TCPSTAT_INC(tcps_rcvoopack); 285 TCPSTAT_ADD(tcps_rcvoobyte, *tlenp); 286 287 /* 288 * While we overlap succeeding segments trim them or, 289 * if they are completely covered, dequeue them. 290 */ 291 while (q) { 292 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 293 if (i <= 0) 294 break; 295 if (i < q->tqe_len) { 296 q->tqe_th->th_seq += i; 297 q->tqe_len -= i; 298 m_adj(q->tqe_m, i); 299 break; 300 } 301 302 nq = LIST_NEXT(q, tqe_q); 303 LIST_REMOVE(q, tqe_q); 304 m_freem(q->tqe_m); 305 uma_zfree(tcp_reass_zone, q); 306 tp->t_segqlen--; 307 q = nq; 308 } 309 310 /* Insert the new segment queue entry into place. */ 311 te->tqe_m = m; 312 te->tqe_th = th; 313 te->tqe_len = *tlenp; 314 315 if (p == NULL) { 316 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 317 } else { 318 KASSERT(te != &tqs, ("%s: temporary stack based entry not " 319 "first element in queue", __func__)); 320 LIST_INSERT_AFTER(p, te, tqe_q); 321 } 322 323present: 324 /* 325 * Present data to user, advancing rcv_nxt through 326 * completed sequence space. 327 */ 328 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 329 return (0); 330 q = LIST_FIRST(&tp->t_segq); 331 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 332 return (0); 333 SOCKBUF_LOCK(&so->so_rcv); 334 do { 335 tp->rcv_nxt += q->tqe_len; 336 flags = q->tqe_th->th_flags & TH_FIN; 337 nq = LIST_NEXT(q, tqe_q); 338 LIST_REMOVE(q, tqe_q); 339 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 340 m_freem(q->tqe_m); 341 else 342 sbappendstream_locked(&so->so_rcv, q->tqe_m); 343 if (q != &tqs) 344 uma_zfree(tcp_reass_zone, q); 345 tp->t_segqlen--; 346 q = nq; 347 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 348 ND6_HINT(tp); 349 sorwakeup_locked(so); 350 return (flags); 351} 352