xref: /dragonfly/sys/netinet/ip_demux.c (revision 113f6df6) 
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 1.28 2004/10/20 05:00:36 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 	/*
111 	 * The packet must be at least the size of an IP header
112 	 */
113 	if (m->m_pkthdr.len < sizeof(struct ip)) {
114 		ipstat.ips_tooshort++;
115 		m_free(m);
116 		return FALSE;
117 	}
118 
119 	/*
120 	 * The first mbuf must entirely contain the IP header
121 	 */
122 	if (m->m_len < sizeof(struct ip) &&
123 	    (m = m_pullup(m, sizeof(struct ip))) == NULL) {
124 		ipstat.ips_toosmall++;
125 		return FALSE;
126 	}
127 	ip = mtod(m, struct ip *);
128 
129 	/*
130 	 * Extract the actual IP header length and do a bounds check.  The
131 	 * first mbuf must entirely contain the extended IP header.
132 	 */
133 	iphlen = ip->ip_hl << 2;
134 	iplen = ntohs(ip->ip_len);
135 	if (iphlen < sizeof(struct ip)) {	/* minimum header length */
136 		ipstat.ips_badhlen++;
137 		m_free(m);
138 		return FALSE;
139 	}
140 	if (m->m_len < iphlen) {
141 		m = m_pullup(m, iphlen);
142 		if (m == NULL) {
143 			ipstat.ips_badhlen++;
144 			return FALSE;
145 		}
146 		ip = mtod(m, struct ip *);
147 	}
148 
149 	/*
150 	 * The TCP/IP or UDP/IP header must be entirely contained within
151 	 * the first fragment of a packet.  Packet filters will break if they
152 	 * aren't.
153 	 *
154 	 * Since the packet will be trimmed to ip_len we must also make sure
155 	 * the potentially trimmed down length is still sufficient to hold
156 	 * the header(s).
157 	 */
158 	if ((ntohs(ip->ip_off) & IP_OFFMASK) == 0) {
159 		switch (ip->ip_p) {
160 		case IPPROTO_TCP:
161 			if (iplen < iphlen + sizeof(struct tcphdr)) {
162 				++tcpstat.tcps_rcvshort;
163 				m_free(m);
164 				return FALSE;
165 			}
166 			if (m->m_len < iphlen + sizeof(struct tcphdr)) {
167 				m = m_pullup(m, iphlen + sizeof(struct tcphdr));
168 				if (m == NULL) {
169 					tcpstat.tcps_rcvshort++;
170 					return FALSE;
171 				}
172 				ip = mtod(m, struct ip *);
173 			}
174 			break;
175 		case IPPROTO_UDP:
176 			if (iplen < iphlen + sizeof(struct udphdr)) {
177 				++udpstat.udps_hdrops;
178 				m_free(m);
179 				return FALSE;
180 			}
181 			if (m->m_len < iphlen + sizeof(struct udphdr)) {
182 				m = m_pullup(m, iphlen + sizeof(struct udphdr));
183 				if (m == NULL) {
184 					udpstat.udps_hdrops++;
185 					return FALSE;
186 				}
187 				ip = mtod(m, struct ip *);
188 			}
189 			break;
190 		default:
191 			if (iplen < iphlen) {
192 				++ipstat.ips_badlen;
193 				m_free(m);
194 				return FALSE;
195 			}
196 			break;
197 		}
198 	}
199 
200 	switch (ip->ip_p) {
201 	case IPPROTO_TCP:
202 		th = (struct tcphdr *)((caddr_t)ip + iphlen);
203 		thoff = th->th_off << 2;
204 		if (thoff < sizeof(struct tcphdr) ||
205 		    thoff > ntohs(ip->ip_len)) {
206 			tcpstat.tcps_rcvbadoff++;
207 			m_free(m);
208 			return FALSE;
209 		}
210 		if (m->m_len < iphlen + thoff) {
211 			m = m_pullup(m, iphlen + thoff);
212 			if (m == NULL) {
213 				tcpstat.tcps_rcvshort++;
214 				return FALSE;
215 			}
216 		}
217 		break;
218 	}
219 
220 	*mp = m;
221 	return TRUE;
222 }
223 
224 /*
225  * Map a packet to a protocol processing thread and return the thread's port.
226  * If an error occurs, the passed mbuf will be freed, *mptr will be set
227  * to NULL, and NULL will be returned.  If no error occurs, the passed mbuf
228  * may be modified and a port pointer will be returned.
229  */
230 lwkt_port_t
231 ip_mport(struct mbuf **mptr)
232 {
233 	struct ip *ip;
234 	int iphlen;
235 	struct tcphdr *th;
236 	struct udphdr *uh;
237 	struct mbuf *m;
238 	int thoff;				/* TCP data offset */
239 	lwkt_port_t port;
240 	int cpu;
241 
242 	if (!ip_lengthcheck(mptr)) {
243 		*mptr = NULL;
244 		return (NULL);
245 	}
246 
247 	m = *mptr;
248 	ip = mtod(m, struct ip *);
249 	iphlen = ip->ip_hl << 2;
250 
251 	/*
252 	 * XXX generic packet handling defrag on CPU 0 for now.
253 	 */
254 	if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
255 		return (&netisr_cpu[0].td_msgport);
256 
257 	switch (ip->ip_p) {
258 	case IPPROTO_TCP:
259 		th = (struct tcphdr *)((caddr_t)ip + iphlen);
260 		thoff = th->th_off << 2;
261 		cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
262 		    th->th_sport, th->th_dport);
263 		port = &tcp_thread[cpu].td_msgport;
264 		break;
265 	case IPPROTO_UDP:
266 		uh = (struct udphdr *)((caddr_t)ip + iphlen);
267 
268 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
269 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
270 			cpu = 0;
271 		} else {
272 			cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
273 			    ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
274 		}
275 		port = &udp_thread[cpu].td_msgport;
276 		break;
277 	default:
278 		port = &netisr_cpu[0].td_msgport;
279 		break;
280 	}
281 
282 	return (port);
283 }
284 
285 /*
286  * Map a TCP socket to a protocol processing thread.
287  */
288 lwkt_port_t
289 tcp_soport(struct socket *so, struct sockaddr *nam, int req)
290 {
291 	struct inpcb *inp;
292 
293 	/* The following processing all take place on Protocol Thread 0. */
294 	if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
295 	    req == PRU_LISTEN)
296 		return (&tcp_thread[0].td_msgport);
297 
298 	inp = sotoinpcb(so);
299 	if (!inp)		/* connection reset by peer */
300 		return (&tcp_thread[0].td_msgport);
301 
302 	/*
303 	 * Already bound and connected or listening.  For TCP connections,
304 	 * the (faddr, fport, laddr, lport) association cannot change now.
305 	 *
306 	 * Note: T/TCP code needs some reorganization to fit into
307 	 * this model.  XXX JH
308 	 *
309 	 * Rely on type-stable memory and check in protocol handler
310 	 * to fix race condition here w/ deallocation of inp.  XXX JH
311 	 */
312 	return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
313 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
314 }
315 
316 lwkt_port_t
317 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
318 {
319 	return (&tcp_thread[tcp_addrcpu(faddr, fport,
320 					laddr, lport)].td_msgport);
321 }
322 
323 /*
324  * Map a UDP socket to a protocol processing thread.
325  */
326 lwkt_port_t
327 udp_soport(struct socket *so, struct sockaddr *nam, int req)
328 {
329 	struct inpcb *inp;
330 
331 	/*
332 	 * The following processing all take place on Protocol Thread 0:
333 	 *   only bind() and connect() have a non-null nam parameter
334 	 *   attach() has a null socket parameter
335 	 *   Fast and slow timeouts pass in two NULLs
336 	 */
337 	if (nam != NULL || so == NULL)
338 		return (&udp_thread[0].td_msgport);
339 
340 	inp = sotoinpcb(so);
341 
342 	if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
343 		return (&udp_thread[0].td_msgport);
344 
345 	/*
346 	 * Rely on type-stable memory and check in protocol handler
347 	 * to fix race condition here w/ deallocation of inp.  XXX JH
348 	 */
349 
350 	return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
351 	    inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
352 }
353 
354 /*
355  * Map a network address to a processor.
356  */
357 int
358 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
359 {
360 	return (INP_MPORT_HASH(faddr, laddr, fport, lport));
361 }
362 
363 int
364 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
365 {
366 	if (IN_MULTICAST(ntohl(laddr)))
367 		return (0);
368 	else
369 		return (INP_MPORT_HASH(faddr, laddr, fport, lport));
370 }
371 
372 /*
373  * Return LWKT port for cpu.
374  */
375 lwkt_port_t
376 tcp_cport(int cpu)
377 {
378 	return (&tcp_thread[cpu].td_msgport);
379 }
380 
381 void
382 tcp_thread_init(void)
383 {
384 	int cpu;
385 
386 	for (cpu = 0; cpu < ncpus2; cpu++) {
387 		lwkt_create(tcpmsg_service_loop, NULL, NULL,
388 			&tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
389 		tcp_thread[cpu].td_msgport.mp_putport = netmsg_put_port;
390 	}
391 }
392 
393 void
394 udp_thread_init(void)
395 {
396 	int cpu;
397 
398 	for (cpu = 0; cpu < ncpus2; cpu++) {
399 		lwkt_create(netmsg_service_loop, NULL, NULL,
400 			&udp_thread[cpu], 0, cpu, "udp_thread %d", cpu);
401 		udp_thread[cpu].td_msgport.mp_putport = netmsg_put_port;
402 	}
403 }
404