xref: /dragonfly/sys/netinet/ip_demux.c (revision 2ee85085) 
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
19  *
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
31  * SUCH DAMAGE.
32  *
33  * $DragonFly: src/sys/netinet/ip_demux.c,v 1.33 2005/03/23 07:47:40 hsu Exp $
34  */
35 
36 /*
37  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
38  *
39  * License terms: all terms for the DragonFly license above plus the following:
40  *
41  * 4. All advertising materials mentioning features or use of this software
42  *    must display the following acknowledgement:
43  *
44  *	This product includes software developed by Jeffrey M. Hsu
45  *	for the DragonFly Project.
46  *
47  *    This requirement may be waived with permission from Jeffrey Hsu.
48  *    This requirement will sunset and may be removed on July 8 2005,
49  *    after which the standard DragonFly license (as shown above) will
50  *    apply.
51  */
52 
53 #include "opt_inet.h"
54 
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/kernel.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/thread.h>
61 #include <sys/sysctl.h>
62 #include <sys/globaldata.h>
63 
64 #include <net/if.h>
65 #include <net/netisr.h>
66 
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/in_var.h>
70 #include <netinet/in_pcb.h>
71 #include <netinet/ip.h>
72 #include <netinet/ip_var.h>
73 #include <netinet/tcp.h>
74 #include <netinet/tcpip.h>
75 #include <netinet/tcp_var.h>
76 #include <netinet/udp.h>
77 #include <netinet/udp_var.h>
78 
79 extern struct thread netisr_cpu[];
80 
81 static struct thread tcp_thread[MAXCPU];
82 static struct thread udp_thread[MAXCPU];
83 
84 static __inline int
85 INP_MPORT_HASH(in_addr_t faddr, in_addr_t laddr,
86 	       in_port_t fport, in_port_t lport)
87 {
88 	/*
89 	 * Use low order bytes.
90 	 */
91 
92 #if (BYTE_ORDER == LITTLE_ENDIAN)
93 	KASSERT(ncpus2 < 256, ("need different hash function"));  /* XXX JH */
94 	return (((faddr >> 24) ^ (fport >> 8) ^ (laddr >> 24) ^ (lport >> 8)) &
95 		ncpus2_mask);
96 #else
97 	return ((faddr ^ fport ^ laddr ^ lport) & ncpus2_mask);
98 #endif
99 }
100 
101 boolean_t
102 ip_lengthcheck(struct mbuf **mp)
103 {
104 	struct mbuf *m = *mp;
105 	struct ip *ip;
106 	int iphlen, iplen;
107 	struct tcphdr *th;
108 	int thoff;				/* TCP data offset */
109 
110 	/* The packet must be at least the size of an IP header. */
111 	if (m->m_pkthdr.len < sizeof(struct ip)) {
112 		ipstat.ips_tooshort++;
113 		m_free(m);
114 		return FALSE;
115 	}
116 
117 	/* The fixed IP header must reside completely in the first mbuf. */
118 	if (m->m_len < sizeof(struct ip)) {
119 		m = m_pullup(m, sizeof(struct ip));
120 		if (m == NULL) {
121 			ipstat.ips_toosmall++;
122 			return FALSE;
123 		}
124 	}
125 
126 	ip = mtod(m, struct ip *);
127 
128 	/* Bound check the packet's stated IP header length. */
129 	iphlen = ip->ip_hl << 2;
130 	if (iphlen < sizeof(struct ip)) {	/* minimum header length */
131 		ipstat.ips_badhlen++;
132 		m_free(m);
133 		return FALSE;
134 	}
135 
136 	/* The full IP header must reside completely in the one mbuf. */
137 	if (m->m_len < iphlen) {
138 		m = m_pullup(m, iphlen);
139 		if (m == NULL) {
140 			ipstat.ips_badhlen++;
141 			return FALSE;
142 		}
143 		ip = mtod(m, struct ip *);
144 	}
145 
146 	iplen = ntohs(ip->ip_len);
147 
148 	/*
149 	 * Fragments other than the first fragment don't have much
150 	 * length information.
151 	 */
152 	if (ntohs(ip->ip_off) & IP_OFFMASK)
153 		goto ipcheckonly;
154 
155 	/*
156 	 * The TCP/IP or UDP/IP header must be entirely contained within
157 	 * the first fragment of a packet.  Packet filters will break if they
158 	 * aren't.
159 	 *
160 	 * Since the packet will be trimmed to ip_len we must also make sure
161 	 * the potentially trimmed down length is still sufficient to hold
162 	 * the header(s).
163 	 */
164 	switch (ip->ip_p) {
165 	case IPPROTO_TCP:
166 		if (iplen < iphlen + sizeof(struct tcphdr)) {
167 			++tcpstat.tcps_rcvshort;
168 			m_free(m);
169 			return FALSE;
170 		}
171 		if (m->m_len < iphlen + sizeof(struct tcphdr)) {
172 			m = m_pullup(m, iphlen + sizeof(struct tcphdr));
173 			if (m == NULL) {
174 				tcpstat.tcps_rcvshort++;
175 				return FALSE;
176 			}
177 			ip = mtod(m, struct ip *);
178 		}
179 		th = (struct tcphdr *)((caddr_t)ip + iphlen);
180 		thoff = th->th_off << 2;
181 		if (thoff < sizeof(struct tcphdr) ||
182 		    thoff + iphlen > ntohs(ip->ip_len)) {
183 			tcpstat.tcps_rcvbadoff++;
184 			m_free(m);
185 			return FALSE;
186 		}
187 		if (m->m_len < iphlen + thoff) {
188 			m = m_pullup(m, iphlen + thoff);
189 			if (m == NULL) {
190 				tcpstat.tcps_rcvshort++;
191 				return FALSE;
192 			}
193 		}
194 		break;
195 	case IPPROTO_UDP:
196 		if (iplen < iphlen + sizeof(struct udphdr)) {
197 			++udpstat.udps_hdrops;
198 			m_free(m);
199 			return FALSE;
200 		}
201 		if (m->m_len < iphlen + sizeof(struct udphdr)) {
202 			m = m_pullup(m, iphlen + sizeof(struct udphdr));
203 			if (m == NULL) {
204 				udpstat.udps_hdrops++;
205 				return FALSE;
206 			}
207 		}
208 		break;
209 	default:
210 ipcheckonly:
211 		if (iplen < iphlen) {
212 			++ipstat.ips_badlen;
213 			m_free(m);
214 			return FALSE;
215 		}
216 		break;
217 	}
218 
219 	*mp = m;
220 	return TRUE;
221 }
222 
223 /*
224  * Map a packet to a protocol processing thread and return the thread's port.
225  * If an error occurs, the passed mbuf will be freed, *mptr will be set
226  * to NULL, and NULL will be returned.  If no error occurs, the passed mbuf
227  * may be modified and a port pointer will be returned.
228  */
229 lwkt_port_t
230 ip_mport(struct mbuf **mptr)
231 {
232 	struct ip *ip;
233 	int iphlen;
234 	struct tcphdr *th;
235 	struct udphdr *uh;
236 	struct mbuf *m;
237 	int thoff;				/* TCP data offset */
238 	lwkt_port_t port;
239 	int cpu;
240 
241 	if (!ip_lengthcheck(mptr)) {
242 		*mptr = NULL;
243 		return (NULL);
244 	}
245 
246 	m = *mptr;
247 	ip = mtod(m, struct ip *);
248 	iphlen = ip->ip_hl << 2;
249 
250 	/*
251 	 * XXX generic packet handling defrag on CPU 0 for now.
252 	 */
253 	if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
254 		return (&netisr_cpu[0].td_msgport);
255 
256 	switch (ip->ip_p) {
257 	case IPPROTO_TCP:
258 		th = (struct tcphdr *)((caddr_t)ip + iphlen);
259 		thoff = th->th_off << 2;
260 		cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
261 		    th->th_sport, th->th_dport);
262 		port = &tcp_thread[cpu].td_msgport;
263 		break;
264 	case IPPROTO_UDP:
265 		uh = (struct udphdr *)((caddr_t)ip + iphlen);
266 
267 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
268 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
269 			cpu = 0;
270 		} else {
271 			cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
272 			    ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
273 		}
274 		port = &udp_thread[cpu].td_msgport;
275 		break;
276 	default:
277 		port = &netisr_cpu[0].td_msgport;
278 		break;
279 	}
280 
281 	return (port);
282 }
283 
284 /*
285  * Map a TCP socket to a protocol processing thread.
286  */
287 lwkt_port_t
288 tcp_soport(struct socket *so, struct sockaddr *nam, int req)
289 {
290 	struct inpcb *inp;
291 
292 	/* The following processing all take place on Protocol Thread 0. */
293 	if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
294 	    req == PRU_LISTEN)
295 		return (&tcp_thread[0].td_msgport);
296 
297 	inp = so->so_pcb;
298 	if (!inp)		/* connection reset by peer */
299 		return (&tcp_thread[0].td_msgport);
300 
301 	/*
302 	 * Already bound and connected or listening.  For TCP connections,
303 	 * the (faddr, fport, laddr, lport) association cannot change now.
304 	 *
305 	 * Note: T/TCP code needs some reorganization to fit into
306 	 * this model.  XXX JH
307 	 *
308 	 * Rely on type-stable memory and check in protocol handler
309 	 * to fix race condition here w/ deallocation of inp.  XXX JH
310 	 */
311 	return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
312 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
313 }
314 
315 lwkt_port_t
316 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
317 {
318 	return (&tcp_thread[tcp_addrcpu(faddr, fport,
319 					laddr, lport)].td_msgport);
320 }
321 
322 /*
323  * Map a UDP socket to a protocol processing thread.
324  */
325 lwkt_port_t
326 udp_soport(struct socket *so, struct sockaddr *nam, int req)
327 {
328 	struct inpcb *inp;
329 
330 	/*
331 	 * The following processing all take place on Protocol Thread 0:
332 	 *   only bind() and connect() have a non-null nam parameter
333 	 *   attach() has a null socket parameter
334 	 *   Fast and slow timeouts pass in two NULLs
335 	 */
336 	if (nam != NULL || so == NULL)
337 		return (&udp_thread[0].td_msgport);
338 
339 	inp = so->so_pcb;
340 
341 	if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
342 		return (&udp_thread[0].td_msgport);
343 
344 	/*
345 	 * Rely on type-stable memory and check in protocol handler
346 	 * to fix race condition here w/ deallocation of inp.  XXX JH
347 	 */
348 
349 	return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
350 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
351 }
352 
353 /*
354  * Map a network address to a processor.
355  */
356 int
357 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
358 {
359 	return (INP_MPORT_HASH(faddr, laddr, fport, lport));
360 }
361 
362 int
363 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
364 {
365 	if (IN_MULTICAST(ntohl(laddr)))
366 		return (0);
367 	else
368 		return (INP_MPORT_HASH(faddr, laddr, fport, lport));
369 }
370 
371 /*
372  * Return LWKT port for cpu.
373  */
374 lwkt_port_t
375 tcp_cport(int cpu)
376 {
377 	return (&tcp_thread[cpu].td_msgport);
378 }
379 
380 void
381 tcp_thread_init(void)
382 {
383 	int cpu;
384 
385 	for (cpu = 0; cpu < ncpus2; cpu++) {
386 		lwkt_create(tcpmsg_service_loop, NULL, NULL,
387 			&tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
388 		netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
389 	}
390 }
391 
392 void
393 udp_thread_init(void)
394 {
395 	int cpu;
396 
397 	for (cpu = 0; cpu < ncpus2; cpu++) {
398 		lwkt_create(netmsg_service_loop, NULL, NULL,
399 			&udp_thread[cpu], 0, cpu, "udp_thread %d", cpu);
400 		netmsg_service_port_init(&udp_thread[cpu].td_msgport);
401 	}
402 }
403