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
34 #include "opt_inet.h"
35 #include "opt_rss.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/socket.h>
41 #include <sys/socketvar.h>
42 #include <sys/thread.h>
43 #include <sys/sysctl.h>
44 #include <sys/globaldata.h>
45
46 #include <net/if.h>
47 #include <net/netisr2.h>
48 #include <net/toeplitz2.h>
49
50 #include <netinet/in_systm.h>
51 #include <netinet/in.h>
52 #include <netinet/in_var.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip.h>
55 #include <netinet/ip_var.h>
56 #include <netinet/tcp.h>
57 #include <netinet/tcpip.h>
58 #include <netinet/tcp_var.h>
59 #include <netinet/udp.h>
60 #include <netinet/udp_var.h>
61
62 struct initport_index {
63 uint32_t port_index;
64 } __cachealign;
65 static struct initport_index initport_indices[MAXCPU];
66
67 /*
68 * Toeplitz hash functions - the idea is to match the hardware.
69 */
70 static __inline int
INP_MPORT_HASH_UDP(in_addr_t faddr,in_addr_t laddr,in_port_t fport,in_port_t lport)71 INP_MPORT_HASH_UDP(in_addr_t faddr, in_addr_t laddr,
72 in_port_t fport, in_port_t lport)
73 {
74 /*
75 * NOTE: laddr could be multicast, since UDP socket could be
76 * bound to multicast address.
77 */
78 if (IN_MULTICAST(ntohl(faddr)) || IN_MULTICAST(ntohl(laddr))) {
79 /* XXX handle multicast on CPU0 for now */
80 return 0;
81 }
82 return toeplitz_hash(toeplitz_rawhash_addr(faddr, laddr));
83 }
84
85 static __inline int
INP_MPORT_HASH_TCP(in_addr_t faddr,in_addr_t laddr,in_port_t fport,in_port_t lport)86 INP_MPORT_HASH_TCP(in_addr_t faddr, in_addr_t laddr,
87 in_port_t fport, in_port_t lport)
88 {
89 return toeplitz_hash(
90 toeplitz_rawhash_addrport(faddr, laddr, fport, lport));
91 }
92
93 /*
94 * Hash for the network address.
95 */
96 int
tcp_addrhash(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)97 tcp_addrhash(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
98 {
99 return (INP_MPORT_HASH_TCP(faddr, laddr, fport, lport));
100 }
101
102 int
udp_addrhash(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)103 udp_addrhash(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
104 {
105 return (INP_MPORT_HASH_UDP(faddr, laddr, fport, lport));
106 }
107
108 /*
109 * Map a network address to a processor.
110 */
111 int
tcp_addrcpu(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)112 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
113 {
114 return (netisr_hashcpu(INP_MPORT_HASH_TCP(faddr, laddr, fport, lport)));
115 }
116
117 int
udp_addrcpu(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)118 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
119 {
120 return (netisr_hashcpu(INP_MPORT_HASH_UDP(faddr, laddr, fport, lport)));
121 }
122
123 /*
124 * If the packet is a valid IP datagram, upon returning of this function
125 * following things are promised:
126 *
127 * o IP header (including any possible IP options) and any data preceding
128 * IP header (usually linker layer header) are in one mbuf (m_len).
129 * o IP header length is not less than the minimum (sizeof(struct ip)).
130 * o IP total length is not less than IP header length.
131 * o IP datagram resides completely in the mbuf chain,
132 * i.e. pkthdr.len >= IP total length.
133 *
134 * If the packet is a UDP datagram,
135 * o IP header (including any possible IP options) and UDP header are in
136 * one mbuf (m_len).
137 * o IP total length is not less than (IP header length + UDP header length).
138 *
139 * If the packet is a TCP segment,
140 * o IP header (including any possible IP options) and TCP header (including
141 * any possible TCP options) are in one mbuf (m_len).
142 * o TCP header length is not less than the minimum (sizeof(struct tcphdr)).
143 * o IP total length is not less than (IP header length + TCP header length).
144 */
145 boolean_t
ip_lengthcheck(struct mbuf ** mp,int hoff)146 ip_lengthcheck(struct mbuf **mp, int hoff)
147 {
148 struct mbuf *m = *mp;
149 struct ip *ip;
150 int len, iphlen, iplen;
151 struct tcphdr *th;
152 int thoff; /* TCP data offset */
153
154 len = hoff + sizeof(struct ip);
155
156 /* The packet must be at least the size of an IP header. */
157 if (m->m_pkthdr.len < len) {
158 ipstat.ips_tooshort++;
159 goto fail;
160 }
161
162 /* The fixed IP header must reside completely in the first mbuf. */
163 if (m->m_len < len) {
164 m = m_pullup(m, len);
165 if (m == NULL) {
166 ipstat.ips_toosmall++;
167 goto fail;
168 }
169 }
170
171 ip = mtodoff(m, struct ip *, hoff);
172
173 /* Bound check the packet's stated IP header length. */
174 iphlen = ip->ip_hl << 2;
175 if (iphlen < sizeof(struct ip)) { /* minimum header length */
176 ipstat.ips_badhlen++;
177 goto fail;
178 }
179
180 /* The full IP header must reside completely in the one mbuf. */
181 if (m->m_len < hoff + iphlen) {
182 m = m_pullup(m, hoff + iphlen);
183 if (m == NULL) {
184 ipstat.ips_badhlen++;
185 goto fail;
186 }
187 ip = mtodoff(m, struct ip *, hoff);
188 }
189
190 iplen = ntohs(ip->ip_len);
191
192 /*
193 * Check that the amount of data in the buffers is as
194 * at least much as the IP header would have us expect.
195 */
196 if (m->m_pkthdr.len < hoff + iplen) {
197 ipstat.ips_tooshort++;
198 goto fail;
199 }
200
201 /*
202 * Fragments other than the first fragment don't have much
203 * length information.
204 */
205 if (ip->ip_off & htons(IP_OFFMASK))
206 goto ipcheckonly;
207
208 /*
209 * The TCP/IP or UDP/IP header must be entirely contained within
210 * the first fragment of a packet. Packet filters will break if they
211 * aren't.
212 *
213 * Since the packet will be trimmed to ip_len we must also make sure
214 * the potentially trimmed down length is still sufficient to hold
215 * the header(s).
216 */
217 switch (ip->ip_p) {
218 case IPPROTO_TCP:
219 if (iplen < iphlen + sizeof(struct tcphdr)) {
220 ++tcpstat.tcps_rcvshort;
221 goto fail;
222 }
223 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) {
224 m = m_pullup(m, hoff + iphlen + sizeof(struct tcphdr));
225 if (m == NULL) {
226 tcpstat.tcps_rcvshort++;
227 goto fail;
228 }
229 ip = mtodoff(m, struct ip *, hoff);
230 }
231 th = (struct tcphdr *)((caddr_t)ip + iphlen);
232 thoff = th->th_off << 2;
233 if (thoff < sizeof(struct tcphdr) ||
234 thoff + iphlen > ntohs(ip->ip_len)) {
235 tcpstat.tcps_rcvbadoff++;
236 goto fail;
237 }
238 if (m->m_len < hoff + iphlen + thoff) {
239 m = m_pullup(m, hoff + iphlen + thoff);
240 if (m == NULL) {
241 tcpstat.tcps_rcvshort++;
242 goto fail;
243 }
244 }
245 break;
246 case IPPROTO_UDP:
247 if (iplen < iphlen + sizeof(struct udphdr)) {
248 ++udp_stat.udps_hdrops;
249 goto fail;
250 }
251 if (m->m_len < hoff + iphlen + sizeof(struct udphdr)) {
252 m = m_pullup(m, hoff + iphlen + sizeof(struct udphdr));
253 if (m == NULL) {
254 udp_stat.udps_hdrops++;
255 goto fail;
256 }
257 }
258 break;
259 default:
260 ipcheckonly:
261 if (iplen < iphlen) {
262 ++ipstat.ips_badlen;
263 goto fail;
264 }
265 break;
266 }
267
268 m->m_flags |= M_LENCHECKED;
269 *mp = m;
270 return TRUE;
271
272 fail:
273 if (m != NULL)
274 m_freem(m);
275 *mp = NULL;
276 return FALSE;
277 }
278
279 /*
280 * Assign a protocol processing thread to a packet. The IP header is at
281 * offset (hoff) in the packet (i.e. the mac header might still be intact).
282 *
283 * This function can blow away the mbuf if the packet is malformed.
284 */
285 void
ip_hashfn(struct mbuf ** mptr,int hoff)286 ip_hashfn(struct mbuf **mptr, int hoff)
287 {
288 struct ip *ip;
289 int iphlen;
290 struct tcphdr *th;
291 struct udphdr *uh;
292 struct mbuf *m;
293 int hash;
294
295 if (((*mptr)->m_flags & M_LENCHECKED) == 0) {
296 if (!ip_lengthcheck(mptr, hoff))
297 return;
298 }
299
300 m = *mptr;
301 ip = mtodoff(m, struct ip *, hoff);
302 iphlen = ip->ip_hl << 2;
303
304 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
305 hash = toeplitz_hash(toeplitz_rawhash_addr(
306 ip->ip_src.s_addr, ip->ip_dst.s_addr));
307 goto back;
308 }
309
310 switch (ip->ip_p) {
311 case IPPROTO_TCP:
312 th = (struct tcphdr *)((caddr_t)ip + iphlen);
313 hash = INP_MPORT_HASH_TCP(ip->ip_src.s_addr, ip->ip_dst.s_addr,
314 th->th_sport, th->th_dport);
315 break;
316
317 case IPPROTO_UDP:
318 uh = (struct udphdr *)((caddr_t)ip + iphlen);
319 hash = INP_MPORT_HASH_UDP(ip->ip_src.s_addr, ip->ip_dst.s_addr,
320 uh->uh_sport, uh->uh_dport);
321 break;
322
323 default:
324 hash = 0;
325 break;
326 }
327 back:
328 m_sethash(m, hash);
329 }
330
331 /*
332 * Verify and adjust the hash value of the packet.
333 *
334 * Unlike ip_hashfn(), the packet content is not accessed. The packet info
335 * (pi) and the hash of the packet (m_pkthdr.hash) is used instead.
336 *
337 * Caller has already made sure that m_pkthdr.hash is valid, i.e. m_flags
338 * has M_HASH set.
339 */
340 void
ip_hashcheck(struct mbuf * m,const struct pktinfo * pi)341 ip_hashcheck(struct mbuf *m, const struct pktinfo *pi)
342 {
343 KASSERT((m->m_flags & M_HASH), ("no valid packet hash"));
344
345 switch (pi->pi_l3proto) {
346 case IPPROTO_TCP:
347 case IPPROTO_UDP:
348 break;
349
350 default:
351 /* Let software calculate the hash */
352 m->m_flags &= ~M_HASH;
353 break;
354 }
355 }
356
357 /*
358 * This is used to map a socket to a message port for sendmsg() and friends.
359 * It is not called for any other purpose. In the case of TCP we just return
360 * the port already installed in the socket.
361 */
362 lwkt_port_t
tcp_soport(struct socket * so,struct sockaddr * nam,struct mbuf ** dummy __unused)363 tcp_soport(struct socket *so, struct sockaddr *nam,
364 struct mbuf **dummy __unused)
365 {
366 return(so->so_port);
367 }
368
369 /*
370 * Used to route icmp messages to the proper protocol thread for ctlinput
371 * operation.
372 */
373 lwkt_port_t
tcp_ctlport(int cmd,struct sockaddr * sa,void * vip,int * cpuid)374 tcp_ctlport(int cmd, struct sockaddr *sa, void *vip, int *cpuid)
375 {
376 struct ip *ip = vip;
377 inp_notify_t notify;
378 int arg;
379
380 notify = tcp_get_inpnotify(cmd, sa, &arg, &ip, cpuid);
381 if (notify == NULL)
382 return NULL;
383
384 if (*cpuid == netisr_ncpus) {
385 /*
386 * Go through all effective netisr CPUs.
387 *
388 * A new message will be allocated later to save necessary
389 * information and will be forwarded to all network protocol
390 * threads in the following way:
391 *
392 * (the the thread owns the msgport that we return here)
393 * netisr0 <--+
394 * | |
395 * | |
396 * | |
397 * +-------+
398 * sendmsg
399 * [msg is kmalloc()ed]
400 *
401 *
402 * Later on, when the msg is received by netisr0:
403 *
404 * forwardmsg forwardmsg
405 * netisr0 ---------> netisr1 ---------> netisrN
406 * [msg is kfree()ed]
407 */
408 return netisr_cpuport(0);
409 } else {
410 return netisr_cpuport(*cpuid);
411 }
412 }
413
414 lwkt_port_t
tcp_addrport(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)415 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
416 {
417 return(netisr_cpuport(tcp_addrcpu(faddr, fport, laddr, lport)));
418 }
419
420 lwkt_port_t
tcp_addrport0(void)421 tcp_addrport0(void)
422 {
423 return(netisr_cpuport(0));
424 }
425
426 lwkt_port_t
udp_addrport(in_addr_t faddr,in_port_t fport,in_addr_t laddr,in_port_t lport)427 udp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
428 {
429 return(netisr_cpuport(udp_addrcpu(faddr, fport, laddr, lport)));
430 }
431
432 /*
433 * Used to route icmp messages to the proper protocol thread for ctlinput
434 * operation.
435 */
436 lwkt_port_t
udp_ctlport(int cmd,struct sockaddr * sa,void * vip,int * cpuid)437 udp_ctlport(int cmd, struct sockaddr *sa, void *vip, int *cpuid)
438 {
439 struct ip *ip = vip;
440 inp_notify_t notify;
441
442 notify = udp_get_inpnotify(cmd, sa, &ip, cpuid);
443 if (notify == NULL)
444 return NULL;
445
446 if (*cpuid == netisr_ncpus) {
447 /*
448 * Go through all effective netisr CPUs.
449 *
450 * See the comment in tcp_ctlport.
451 */
452 return netisr_cpuport(0);
453 } else {
454 return netisr_cpuport(*cpuid);
455 }
456 }
457
458 static __inline struct lwkt_port *
inp_initport(void)459 inp_initport(void)
460 {
461 int cpu = mycpuid;
462
463 if (cpu < netisr_ncpus) {
464 return netisr_cpuport(cpu);
465 } else {
466 return netisr_cpuport(
467 ((initport_indices[cpu].port_index++) + (uint32_t)cpu) %
468 netisr_ncpus);
469 }
470 }
471
472 struct lwkt_port *
tcp_initport(void)473 tcp_initport(void)
474 {
475 return inp_initport();
476 }
477
478 struct lwkt_port *
udp_initport(void)479 udp_initport(void)
480 {
481 return inp_initport();
482 }
483