2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * $DragonFly: src/sys/netinet/ip_demux.c,v 1.45 2008/11/11 10:46:58 sephe Exp $
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/thread.h>
45 #include <sys/sysctl.h>
46 #include <sys/globaldata.h>
49 #include <net/netisr.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/in.h>
53 #include <netinet/in_var.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/ip.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcpip.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/udp.h>
61 #include <netinet/udp_var.h>
63 extern struct thread netisr_cpu[];
64 extern int udp_mpsafe_thread;
66 static struct thread tcp_thread[MAXCPU];
67 static struct thread udp_thread[MAXCPU];
70 INP_MPORT_HASH(in_addr_t faddr, in_addr_t laddr,
71 in_port_t fport, in_port_t lport)
74 * Use low order bytes.
77 #if (BYTE_ORDER == LITTLE_ENDIAN)
78 KASSERT(ncpus2 < 256, ("need different hash function")); /* XXX JH */
79 return (((faddr >> 24) ^ (fport >> 8) ^ (laddr >> 24) ^ (lport >> 8)) &
82 return ((faddr ^ fport ^ laddr ^ lport) & ncpus2_mask);
87 INP_MPORT_HASH_UDP(in_addr_t faddr, in_addr_t laddr,
88 in_port_t fport, in_port_t lport)
90 return INP_MPORT_HASH(faddr, laddr, fport, lport);
94 INP_MPORT_HASH_TCP(in_addr_t faddr, in_addr_t laddr,
95 in_port_t fport, in_port_t lport)
97 return INP_MPORT_HASH(faddr, laddr, fport, lport);
101 * If the packet is a valid IP datagram, upon returning of this function
102 * following things are promised:
104 * o IP header (including any possible IP options) is in one mbuf (m_len).
105 * o IP header length is not less than the minimum (sizeof(struct ip)).
106 * o IP total length is not less than IP header length.
107 * o IP datagram resides completely in the mbuf chain,
108 * i.e. pkthdr.len >= IP total length.
110 * If the packet is a UDP datagram,
111 * o IP header (including any possible IP options) and UDP header are in
113 * o IP total length is not less than (IP header length + UDP header length).
115 * If the packet is a TCP segment,
116 * o IP header (including any possible IP options) and TCP header (including
117 * any possible TCP options) are in one mbuf (m_len).
118 * o TCP header length is not less than the minimum (sizeof(struct tcphdr)).
119 * o IP total length is not less than (IP header length + TCP header length).
122 ip_lengthcheck(struct mbuf **mp)
124 struct mbuf *m = *mp;
128 int thoff; /* TCP data offset */
130 /* The packet must be at least the size of an IP header. */
131 if (m->m_pkthdr.len < sizeof(struct ip)) {
132 ipstat.ips_tooshort++;
136 /* The fixed IP header must reside completely in the first mbuf. */
137 if (m->m_len < sizeof(struct ip)) {
138 m = m_pullup(m, sizeof(struct ip));
140 ipstat.ips_toosmall++;
145 ip = mtod(m, struct ip *);
147 /* Bound check the packet's stated IP header length. */
148 iphlen = ip->ip_hl << 2;
149 if (iphlen < sizeof(struct ip)) { /* minimum header length */
150 ipstat.ips_badhlen++;
154 /* The full IP header must reside completely in the one mbuf. */
155 if (m->m_len < iphlen) {
156 m = m_pullup(m, iphlen);
158 ipstat.ips_badhlen++;
161 ip = mtod(m, struct ip *);
164 iplen = ntohs(ip->ip_len);
167 * Check that the amount of data in the buffers is as
168 * at least much as the IP header would have us expect.
170 if (m->m_pkthdr.len < iplen) {
171 ipstat.ips_tooshort++;
176 * Fragments other than the first fragment don't have much
177 * length information.
179 if (ntohs(ip->ip_off) & IP_OFFMASK)
183 * The TCP/IP or UDP/IP header must be entirely contained within
184 * the first fragment of a packet. Packet filters will break if they
187 * Since the packet will be trimmed to ip_len we must also make sure
188 * the potentially trimmed down length is still sufficient to hold
193 if (iplen < iphlen + sizeof(struct tcphdr)) {
194 ++tcpstat.tcps_rcvshort;
197 if (m->m_len < iphlen + sizeof(struct tcphdr)) {
198 m = m_pullup(m, iphlen + sizeof(struct tcphdr));
200 tcpstat.tcps_rcvshort++;
203 ip = mtod(m, struct ip *);
205 th = (struct tcphdr *)((caddr_t)ip + iphlen);
206 thoff = th->th_off << 2;
207 if (thoff < sizeof(struct tcphdr) ||
208 thoff + iphlen > ntohs(ip->ip_len)) {
209 tcpstat.tcps_rcvbadoff++;
212 if (m->m_len < iphlen + thoff) {
213 m = m_pullup(m, iphlen + thoff);
215 tcpstat.tcps_rcvshort++;
221 if (iplen < iphlen + sizeof(struct udphdr)) {
222 ++udpstat.udps_hdrops;
225 if (m->m_len < iphlen + sizeof(struct udphdr)) {
226 m = m_pullup(m, iphlen + sizeof(struct udphdr));
228 udpstat.udps_hdrops++;
235 if (iplen < iphlen) {
253 * Map a packet to a protocol processing thread and return the thread's port.
254 * If an error occurs, the passed mbuf will be freed, *mptr will be set
255 * to NULL, and NULL will be returned. If no error occurs, the passed mbuf
256 * may be modified and a port pointer will be returned.
259 ip_mport(struct mbuf **mptr, int dir)
266 int thoff; /* TCP data offset */
270 if (!ip_lengthcheck(mptr))
274 ip = mtod(m, struct ip *);
275 iphlen = ip->ip_hl << 2;
278 * XXX generic packet handling defrag on CPU 0 for now.
280 if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
281 return (&netisr_cpu[0].td_msgport);
285 th = (struct tcphdr *)((caddr_t)ip + iphlen);
286 thoff = th->th_off << 2;
287 cpu = INP_MPORT_HASH_TCP(ip->ip_src.s_addr, ip->ip_dst.s_addr,
288 th->th_sport, th->th_dport);
289 port = &tcp_thread[cpu].td_msgport;
293 uh = (struct udphdr *)((caddr_t)ip + iphlen);
296 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
297 (dir == IP_MPORT_IN &&
298 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))) {
303 cpu = INP_MPORT_HASH_UDP(ip->ip_src.s_addr,
304 ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
306 port = &udp_thread[cpu].td_msgport;
310 port = &netisr_cpu[0].td_msgport;
318 ip_mport_in(struct mbuf **mptr)
320 return ip_mport(mptr, IP_MPORT_IN);
324 * Map a TCP socket to a protocol processing thread.
327 tcp_soport(struct socket *so, struct sockaddr *nam __unused,
328 struct mbuf **dummy __unused, int req)
332 /* The following processing all take place on Protocol Thread 0. */
333 if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
335 return (&tcp_thread[0].td_msgport);
338 if (!inp) /* connection reset by peer */
339 return (&tcp_thread[0].td_msgport);
342 * Already bound and connected or listening. For TCP connections,
343 * the (faddr, fport, laddr, lport) association cannot change now.
345 * Note: T/TCP code needs some reorganization to fit into
348 * Rely on type-stable memory and check in protocol handler
349 * to fix race condition here w/ deallocation of inp. XXX JH
351 return (&tcp_thread[INP_MPORT_HASH_TCP(inp->inp_faddr.s_addr,
352 inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
356 * Used to route icmp messages to the proper protocol thread for ctlinput
360 tcp_ctlport(int cmd, struct sockaddr *sa, void *vip)
364 struct in_addr faddr;
367 faddr = ((struct sockaddr_in *)sa)->sin_addr;
368 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
370 if (ip == NULL || PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) {
372 * Message will be forwarded to all TCP protocol threads
375 * netisr0 (the msgport we return here)
378 * | domsg <----------------------------+
382 * V forwardmsg forwardmsg |
383 * tcp0 ------------> tcp1 ------------> tcpN
385 return cpu0_ctlport(cmd, sa, vip);
387 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
388 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
389 ip->ip_src.s_addr, th->th_sport);
391 return(&tcp_thread[cpu].td_msgport);
395 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
397 return (&tcp_thread[tcp_addrcpu(faddr, fport,
398 laddr, lport)].td_msgport);
402 * Map a UDP socket to a protocol processing thread.
405 udp_soport(struct socket *so, struct sockaddr *nam __unused,
406 struct mbuf **dummy __unused, int req)
411 * The following processing all take place on Protocol Thread 0:
413 * attach() has a null socket parameter
414 * Fast and slow timeouts pass in null socket parameter
416 if (req == PRU_BIND || so == NULL)
417 return (&udp_thread[0].td_msgport);
422 if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
423 return (&udp_thread[0].td_msgport);
427 * Rely on type-stable memory and check in protocol handler
428 * to fix race condition here w/ deallocation of inp. XXX JH
431 return (&udp_thread[INP_MPORT_HASH_UDP(inp->inp_faddr.s_addr,
432 inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
436 * Used to route icmp messages to the proper protocol thread for ctlinput
440 udp_ctlport(int cmd, struct sockaddr *sa, void *vip)
444 struct in_addr faddr;
447 faddr = ((struct sockaddr_in *)sa)->sin_addr;
448 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
450 if (PRC_IS_REDIRECT(cmd)) {
452 * See the comment in tcp_ctlport; the only difference
453 * is that message is forwarded to UDP protocol theads.
455 return cpu0_ctlport(cmd, sa, vip);
456 } else if (ip == NULL || cmd == PRC_HOSTDEAD) {
459 * Once UDP inpcbs are CPU localized, we should do
460 * the same forwarding as PRC_IS_REDIRECT(cmd)
464 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
466 cpu = INP_MPORT_HASH_UDP(faddr.s_addr, ip->ip_src.s_addr,
467 uh->uh_dport, uh->uh_sport);
469 return (&udp_thread[cpu].td_msgport);
473 * Map a network address to a processor.
476 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
478 return (INP_MPORT_HASH_TCP(faddr, laddr, fport, lport));
482 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
485 if (IN_MULTICAST(ntohl(laddr)))
489 return (INP_MPORT_HASH_UDP(faddr, laddr, fport, lport));
493 * Return LWKT port for cpu.
498 return (&tcp_thread[cpu].td_msgport);
504 return (&udp_thread[cpu].td_msgport);
508 tcp_thread_init(void)
512 for (cpu = 0; cpu < ncpus2; cpu++) {
513 lwkt_create(tcpmsg_service_loop, NULL, NULL,
514 &tcp_thread[cpu], TDF_NETWORK | TDF_MPSAFE, cpu,
515 "tcp_thread %d", cpu);
516 netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
521 udp_thread_init(void)
525 for (cpu = 0; cpu < ncpus2; cpu++) {
526 lwkt_create(netmsg_service_loop, &udp_mpsafe_thread, NULL,
527 &udp_thread[cpu], TDF_NETWORK | TDF_MPSAFE, cpu,
528 "udp_thread %d", cpu);
529 netmsg_service_port_init(&udp_thread[cpu].td_msgport);