xref: /dragonfly/sys/netinet/ip_demux.c (revision 1ab20d67)
1 /*
2  * Copyright (c) 2003 Jeffrey Hsu
3  * All rights reserved.
4  *
5  * $DragonFly: src/sys/netinet/ip_demux.c,v 1.22 2004/04/24 06:55:57 hsu Exp $
6  */
7 
8 #include "opt_inet.h"
9 
10 #include <sys/param.h>
11 #include <sys/systm.h>
12 #include <sys/kernel.h>
13 #include <sys/socket.h>
14 #include <sys/socketvar.h>
15 #include <sys/thread.h>
16 #include <sys/sysctl.h>
17 #include <sys/globaldata.h>
18 
19 #include <net/if.h>
20 #include <net/netisr.h>
21 
22 #include <netinet/in_systm.h>
23 #include <netinet/in.h>
24 #include <netinet/in_var.h>
25 #include <netinet/in_pcb.h>
26 #include <netinet/ip.h>
27 #include <netinet/ip_var.h>
28 #include <netinet/tcp.h>
29 #include <netinet/tcpip.h>
30 #include <netinet/tcp_var.h>
31 #include <netinet/udp.h>
32 #include <netinet/udp_var.h>
33 
34 extern struct thread netisr_cpu[];
35 
36 static struct thread tcp_thread[MAXCPU];
37 static struct thread udp_thread[MAXCPU];
38 
39 static __inline int
40 INP_MPORT_HASH(in_addr_t faddr, in_addr_t laddr,
41 	       in_port_t fport, in_port_t lport)
42 {
43 	/*
44 	 * Use low order bytes.
45 	 */
46 
47 #if (BYTE_ORDER == LITTLE_ENDIAN)
48 	KASSERT(ncpus2 < 256, ("need different hash function"));  /* XXX JH */
49 	return (((faddr >> 24) ^ (fport >> 8) ^ (laddr >> 24) ^ (lport >> 8)) &
50 		ncpus2_mask);
51 #else
52 	return ((faddr ^ fport ^ laddr ^ lport) & ncpus2_mask);
53 #endif
54 }
55 
56 /*
57  * Map a packet to a protocol processing thread.
58  */
59 lwkt_port_t
60 ip_mport(struct mbuf *m)
61 {
62 	struct ip *ip;
63 	int iphlen;
64 	struct tcphdr *th;
65 	struct udphdr *uh;
66 	int thoff;				/* TCP data offset */
67 	lwkt_port_t port;
68 	int cpu;
69 
70 	if (m->m_pkthdr.len < sizeof(struct ip)) {
71 		ipstat.ips_tooshort++;
72 		return (NULL);
73 	}
74 
75 	if (m->m_len < sizeof(struct ip) &&
76 	    (m = m_pullup(m, sizeof(struct ip))) == NULL) {
77 		ipstat.ips_toosmall++;
78 		return (NULL);
79 	}
80 
81 	ip = mtod(m, struct ip *);
82 
83 	/*
84 	 * XXX generic packet handling defrag on CPU 0 for now.
85 	 */
86 	if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
87 		return (&netisr_cpu[0].td_msgport);
88 
89 	iphlen = ip->ip_hl << 2;
90 	if (iphlen < sizeof(struct ip)) {	/* minimum header length */
91 		ipstat.ips_badhlen++;
92 		return (NULL);
93 	}
94 
95 	switch (ip->ip_p) {
96 	case IPPROTO_TCP:
97 		if (m->m_len < iphlen + sizeof(struct tcphdr) &&
98 		    (m = m_pullup(m, iphlen + sizeof(struct tcphdr))) == NULL) {
99 			tcpstat.tcps_rcvshort++;
100 			return (NULL);
101 		}
102 		th = (struct tcphdr *)((caddr_t)ip + iphlen);
103 		thoff = th->th_off << 2;
104 		if (thoff < sizeof(struct tcphdr) ||
105 		    thoff > ntohs(ip->ip_len)) {
106 			tcpstat.tcps_rcvbadoff++;
107 			return (NULL);
108 		}
109 		if (m->m_len < iphlen + thoff) {
110 			m = m_pullup(m, iphlen + thoff);
111 			if (m == NULL) {
112 				tcpstat.tcps_rcvshort++;
113 				return (NULL);
114 			}
115 			ip = mtod(m, struct ip *);
116 			th = (struct tcphdr *)((caddr_t)ip + iphlen);
117 		}
118 
119 		cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
120 		    th->th_sport, th->th_dport);
121 		port = &tcp_thread[cpu].td_msgport;
122 		break;
123 	case IPPROTO_UDP:
124 		if (m->m_len < iphlen + sizeof(struct udphdr)) {
125 			m = m_pullup(m, iphlen + sizeof(struct udphdr));
126 			if (m == NULL) {
127 				udpstat.udps_hdrops++;
128 				return (NULL);
129 			}
130 			ip = mtod(m, struct ip *);
131 		}
132 		uh = (struct udphdr *)((caddr_t)ip + iphlen);
133 
134 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
135 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
136 			cpu = 0;
137 		} else {
138 			cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
139 			    ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
140 		}
141 		port = &udp_thread[cpu].td_msgport;
142 		break;
143 	default:
144 		if (m->m_len < iphlen && (m = m_pullup(m, iphlen)) == NULL) {
145 			ipstat.ips_badhlen++;
146 			return (NULL);
147 		}
148 		port = &netisr_cpu[0].td_msgport;
149 		break;
150 	}
151 	KKASSERT(port->mp_putport != NULL);
152 
153 	return (port);
154 }
155 
156 /*
157  * Map a TCP socket to a protocol processing thread.
158  */
159 lwkt_port_t
160 tcp_soport(struct socket *so, struct sockaddr *nam, int req)
161 {
162 	struct inpcb *inp;
163 
164 	/* The following processing all take place on Protocol Thread 0. */
165 	if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
166 	    req == PRU_LISTEN)
167 		return (&tcp_thread[0].td_msgport);
168 
169 	inp = sotoinpcb(so);
170 	if (!inp)		/* connection reset by peer */
171 		return (&tcp_thread[0].td_msgport);
172 
173 	/*
174 	 * Already bound and connected or listening.  For TCP connections,
175 	 * the (faddr, fport, laddr, lport) association cannot change now.
176 	 *
177 	 * Note: T/TCP code needs some reorganization to fit into
178 	 * this model.  XXX JH
179 	 *
180 	 * Rely on type-stable memory and check in protocol handler
181 	 * to fix race condition here w/ deallocation of inp.  XXX JH
182 	 */
183 	return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
184 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
185 }
186 
187 lwkt_port_t
188 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
189 {
190 	return (&tcp_thread[tcp_addrcpu(faddr, fport,
191 					laddr, lport)].td_msgport);
192 }
193 
194 /*
195  * Map a UDP socket to a protocol processing thread.
196  */
197 lwkt_port_t
198 udp_soport(struct socket *so, struct sockaddr *nam, int req)
199 {
200 	struct inpcb *inp;
201 
202 	/*
203 	 * The following processing all take place on Protocol Thread 0:
204 	 *   only bind() and connect() have a non-null nam parameter
205 	 *   attach() has a null socket parameter
206 	 *   Fast and slow timeouts pass in two NULLs
207 	 */
208 	if (nam != NULL || so == NULL)
209 		return (&udp_thread[0].td_msgport);
210 
211 	inp = sotoinpcb(so);
212 
213 	if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
214 		return (&udp_thread[0].td_msgport);
215 
216 	/*
217 	 * Rely on type-stable memory and check in protocol handler
218 	 * to fix race condition here w/ deallocation of inp.  XXX JH
219 	 */
220 
221 	return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
222 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
223 }
224 
225 /*
226  * Map a network address to a processor.
227  */
228 int
229 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
230 {
231 	return (INP_MPORT_HASH(faddr, laddr, fport, lport));
232 }
233 
234 int
235 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
236 {
237 	if (IN_MULTICAST(ntohl(laddr)))
238 		return (0);
239 	else
240 		return (INP_MPORT_HASH(faddr, laddr, fport, lport));
241 }
242 
243 /*
244  * Return LWKT port for cpu.
245  */
246 lwkt_port_t
247 tcp_cport(int cpu)
248 {
249 	return (&tcp_thread[cpu].td_msgport);
250 }
251 
252 void
253 tcp_thread_init(void)
254 {
255 	int cpu;
256 
257 	for (cpu = 0; cpu < ncpus2; cpu++) {
258 		lwkt_create(netmsg_service_loop, NULL, NULL,
259 			&tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
260 		tcp_thread[cpu].td_msgport.mp_putport = netmsg_put_port;
261 	}
262 }
263 
264 void
265 udp_thread_init(void)
266 {
267 	int cpu;
268 
269 	for (cpu = 0; cpu < ncpus2; cpu++) {
270 		lwkt_create(netmsg_service_loop, NULL, NULL,
271 			&udp_thread[cpu], 0, cpu, "udp_thread %d", cpu);
272 		udp_thread[cpu].td_msgport.mp_putport = netmsg_put_port;
273 	}
274 }
275