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>
50 #include <net/toeplitz2.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/in.h>
54 #include <netinet/in_var.h>
55 #include <netinet/in_pcb.h>
56 #include <netinet/ip.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcpip.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/udp.h>
62 #include <netinet/udp_var.h>
64 extern struct thread netisr_cpu[];
65 extern int udp_mpsafe_thread;
67 static struct thread tcp_thread[MAXCPU];
68 static struct thread udp_thread[MAXCPU];
71 INP_MPORT_HASH_UDP(in_addr_t faddr, in_addr_t laddr,
72 in_port_t fport, in_port_t lport)
74 return toeplitz_hash(toeplitz_rawhash_addr(faddr, laddr));
78 INP_MPORT_HASH_TCP(in_addr_t faddr, in_addr_t laddr,
79 in_port_t fport, in_port_t lport)
82 toeplitz_rawhash_addrport(faddr, laddr, fport, lport));
86 * If the packet is a valid IP datagram, upon returning of this function
87 * following things are promised:
89 * o IP header (including any possible IP options) is in one mbuf (m_len).
90 * o IP header length is not less than the minimum (sizeof(struct ip)).
91 * o IP total length is not less than IP header length.
92 * o IP datagram resides completely in the mbuf chain,
93 * i.e. pkthdr.len >= IP total length.
95 * If the packet is a UDP datagram,
96 * o IP header (including any possible IP options) and UDP header are in
98 * o IP total length is not less than (IP header length + UDP header length).
100 * If the packet is a TCP segment,
101 * o IP header (including any possible IP options) and TCP header (including
102 * any possible TCP options) are in one mbuf (m_len).
103 * o TCP header length is not less than the minimum (sizeof(struct tcphdr)).
104 * o IP total length is not less than (IP header length + TCP header length).
107 ip_lengthcheck(struct mbuf **mp)
109 struct mbuf *m = *mp;
113 int thoff; /* TCP data offset */
115 /* The packet must be at least the size of an IP header. */
116 if (m->m_pkthdr.len < sizeof(struct ip)) {
117 ipstat.ips_tooshort++;
121 /* The fixed IP header must reside completely in the first mbuf. */
122 if (m->m_len < sizeof(struct ip)) {
123 m = m_pullup(m, sizeof(struct ip));
125 ipstat.ips_toosmall++;
130 ip = mtod(m, struct ip *);
132 /* Bound check the packet's stated IP header length. */
133 iphlen = ip->ip_hl << 2;
134 if (iphlen < sizeof(struct ip)) { /* minimum header length */
135 ipstat.ips_badhlen++;
139 /* The full IP header must reside completely in the one mbuf. */
140 if (m->m_len < iphlen) {
141 m = m_pullup(m, iphlen);
143 ipstat.ips_badhlen++;
146 ip = mtod(m, struct ip *);
149 iplen = ntohs(ip->ip_len);
152 * Check that the amount of data in the buffers is as
153 * at least much as the IP header would have us expect.
155 if (m->m_pkthdr.len < iplen) {
156 ipstat.ips_tooshort++;
161 * Fragments other than the first fragment don't have much
162 * length information.
164 if (ntohs(ip->ip_off) & IP_OFFMASK)
168 * The TCP/IP or UDP/IP header must be entirely contained within
169 * the first fragment of a packet. Packet filters will break if they
172 * Since the packet will be trimmed to ip_len we must also make sure
173 * the potentially trimmed down length is still sufficient to hold
178 if (iplen < iphlen + sizeof(struct tcphdr)) {
179 ++tcpstat.tcps_rcvshort;
182 if (m->m_len < iphlen + sizeof(struct tcphdr)) {
183 m = m_pullup(m, iphlen + sizeof(struct tcphdr));
185 tcpstat.tcps_rcvshort++;
188 ip = mtod(m, struct ip *);
190 th = (struct tcphdr *)((caddr_t)ip + iphlen);
191 thoff = th->th_off << 2;
192 if (thoff < sizeof(struct tcphdr) ||
193 thoff + iphlen > ntohs(ip->ip_len)) {
194 tcpstat.tcps_rcvbadoff++;
197 if (m->m_len < iphlen + thoff) {
198 m = m_pullup(m, iphlen + thoff);
200 tcpstat.tcps_rcvshort++;
206 if (iplen < iphlen + sizeof(struct udphdr)) {
207 ++udpstat.udps_hdrops;
210 if (m->m_len < iphlen + sizeof(struct udphdr)) {
211 m = m_pullup(m, iphlen + sizeof(struct udphdr));
213 udpstat.udps_hdrops++;
220 if (iplen < iphlen) {
227 m->m_flags |= M_LENCHECKED;
239 * Map a packet to a protocol processing thread and return the thread's port.
240 * If an error occurs, the passed mbuf will be freed, *mptr will be set
241 * to NULL, and NULL will be returned. If no error occurs, the passed mbuf
242 * may be modified and a port pointer will be returned.
245 ip_mport(struct mbuf **mptr, int dir)
252 int thoff; /* TCP data offset */
256 if (!ip_lengthcheck(mptr))
260 ip = mtod(m, struct ip *);
261 iphlen = ip->ip_hl << 2;
264 * XXX generic packet handling defrag on CPU 0 for now.
266 if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) {
268 port = &netisr_cpu[cpu].td_msgport;
274 th = (struct tcphdr *)((caddr_t)ip + iphlen);
275 thoff = th->th_off << 2;
276 cpu = INP_MPORT_HASH_TCP(ip->ip_src.s_addr,
280 port = &tcp_thread[cpu].td_msgport;
284 uh = (struct udphdr *)((caddr_t)ip + iphlen);
286 cpu = INP_MPORT_HASH_UDP(ip->ip_src.s_addr,
290 port = &udp_thread[cpu].td_msgport;
295 port = &netisr_cpu[cpu].td_msgport;
299 m->m_flags |= M_HASH;
300 m->m_pkthdr.hash = cpu;
305 ip_mport_in(struct mbuf **mptr)
307 return ip_mport(mptr, IP_MPORT_IN);
311 * Map a packet to a protocol processing thread and return the thread's port.
312 * Unlike ip_mport(), the packet content is not accessed. The packet info
313 * (pi) and the hash of the packet (m_pkthdr.hash) is used instead. NULL is
314 * returned if the packet info does not contain enough information.
316 * Caller has already made sure that m_pkthdr.hash is valid, i.e. m_flags
320 ip_mport_pktinfo(const struct pktinfo *pi, struct mbuf *m)
324 KASSERT(m->m_pkthdr.hash < ncpus2,
325 ("invalid packet hash %#x\n", m->m_pkthdr.hash));
328 * XXX generic packet handling defrag on CPU 0 for now.
330 if (pi->pi_flags & PKTINFO_FLAG_FRAG) {
331 m->m_pkthdr.hash = 0;
332 return &netisr_cpu[0].td_msgport;
335 switch (pi->pi_l3proto) {
337 port = &tcp_thread[m->m_pkthdr.hash].td_msgport;
341 port = &udp_thread[m->m_pkthdr.hash].td_msgport;
352 * Initital port when creating the socket, generally before
353 * binding or connect.
356 tcp_soport_attach(struct socket *so)
358 return(&tcp_thread[0].td_msgport);
362 * This is used to map a socket to a message port for sendmsg() and friends.
363 * It is not called for any other purpose. In the case of TCP we just return
364 * the port already installed in the socket.
367 tcp_soport(struct socket *so, struct sockaddr *nam,
368 struct mbuf **dummy __unused)
374 * Used to route icmp messages to the proper protocol thread for ctlinput
378 tcp_ctlport(int cmd, struct sockaddr *sa, void *vip)
382 struct in_addr faddr;
385 faddr = ((struct sockaddr_in *)sa)->sin_addr;
386 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
388 if (ip == NULL || PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) {
390 * Message will be forwarded to all TCP protocol threads
393 * netisr0 (the msgport we return here)
396 * | domsg <----------------------------+
400 * V forwardmsg forwardmsg |
401 * tcp0 ------------> tcp1 ------------> tcpN
403 return cpu0_ctlport(cmd, sa, vip);
405 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
406 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
407 ip->ip_src.s_addr, th->th_sport);
409 return(&tcp_thread[cpu].td_msgport);
413 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
415 return (&tcp_thread[tcp_addrcpu(faddr, fport,
416 laddr, lport)].td_msgport);
422 return (&tcp_thread[0].td_msgport);
426 udp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
428 return (&udp_thread[udp_addrcpu(faddr, fport,
429 laddr, lport)].td_msgport);
433 * Initital port when creating the socket, generally before
434 * binding or connect.
437 udp_soport_attach(struct socket *so)
439 return(&udp_thread[0].td_msgport);
443 * This is used to map a socket to a message port for sendmsg() and friends.
444 * It is not called for any other purpose.
446 * In the case of UDP we just return the port already installed in the socket,
447 * regardless of what (nam) is.
450 udp_soport(struct socket *so, struct sockaddr *nam,
451 struct mbuf **dummy __unused)
457 * Used to route icmp messages to the proper protocol thread for ctlinput
461 udp_ctlport(int cmd, struct sockaddr *sa, void *vip)
465 struct in_addr faddr;
468 faddr = ((struct sockaddr_in *)sa)->sin_addr;
469 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
471 if (PRC_IS_REDIRECT(cmd)) {
473 * See the comment in tcp_ctlport; the only difference
474 * is that message is forwarded to UDP protocol theads.
476 return cpu0_ctlport(cmd, sa, vip);
477 } else if (ip == NULL || cmd == PRC_HOSTDEAD) {
480 * Once UDP inpcbs are CPU localized, we should do
481 * the same forwarding as PRC_IS_REDIRECT(cmd)
485 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
487 cpu = INP_MPORT_HASH_UDP(faddr.s_addr, ip->ip_src.s_addr,
488 uh->uh_dport, uh->uh_sport);
490 return (&udp_thread[cpu].td_msgport);
494 * Map a network address to a processor.
497 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
499 return (INP_MPORT_HASH_TCP(faddr, laddr, fport, lport));
503 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
505 return (INP_MPORT_HASH_UDP(faddr, laddr, fport, lport));
509 * Return LWKT port for cpu.
514 return (&tcp_thread[cpu].td_msgport);
520 return (&udp_thread[cpu].td_msgport);
524 tcp_thread_init(void)
528 for (cpu = 0; cpu < ncpus2; cpu++) {
529 lwkt_create(tcpmsg_service_loop, NULL, NULL,
530 &tcp_thread[cpu], TDF_NETWORK, cpu,
531 "tcp_thread %d", cpu);
532 netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
537 udp_thread_init(void)
541 for (cpu = 0; cpu < ncpus2; cpu++) {
542 lwkt_create(netmsg_service_loop, &udp_mpsafe_thread, NULL,
543 &udp_thread[cpu], TDF_NETWORK, cpu,
544 "udp_thread %d", cpu);
545 netmsg_service_port_init(&udp_thread[cpu].td_msgport);