31 */
32
33#include "opt_bootp.h"
34#include "opt_ipfw.h"
35#include "opt_ipstealth.h"
36#include "opt_ipsec.h"
37#include "opt_mac.h"
38#include "opt_carp.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/callout.h>
43#include <sys/mbuf.h>
44#include <sys/malloc.h>
45#include <sys/domain.h>
46#include <sys/protosw.h>
47#include <sys/socket.h>
48#include <sys/time.h>
49#include <sys/kernel.h>
50#include <sys/syslog.h>
51#include <sys/sysctl.h>
52
53#include <net/pfil.h>
54#include <net/if.h>
55#include <net/if_types.h>
56#include <net/if_var.h>
57#include <net/if_dl.h>
58#include <net/route.h>
59#include <net/netisr.h>
60
61#include <netinet/in.h>
62#include <netinet/in_systm.h>
63#include <netinet/in_var.h>
64#include <netinet/ip.h>
65#include <netinet/in_pcb.h>
66#include <netinet/ip_var.h>
67#include <netinet/ip_icmp.h>
68#include <netinet/ip_options.h>
69#include <machine/in_cksum.h>
70#ifdef DEV_CARP
71#include <netinet/ip_carp.h>
72#endif
73#ifdef IPSEC
74#include <netinet/ip_ipsec.h>
75#endif /* IPSEC */
76
77#include <sys/socketvar.h>
78
79/* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
80#include <netinet/ip_fw.h>
81#include <netinet/ip_dummynet.h>
82
83#include <security/mac/mac_framework.h>
84
85int rsvp_on = 0;
86
87int ipforwarding = 0;
88SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
89 &ipforwarding, 0, "Enable IP forwarding between interfaces");
90
91static int ipsendredirects = 1; /* XXX */
92SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
93 &ipsendredirects, 0, "Enable sending IP redirects");
94
95int ip_defttl = IPDEFTTL;
96SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
97 &ip_defttl, 0, "Maximum TTL on IP packets");
98
99static int ip_keepfaith = 0;
100SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
101 &ip_keepfaith, 0,
102 "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
103
104static int ip_sendsourcequench = 0;
105SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
106 &ip_sendsourcequench, 0,
107 "Enable the transmission of source quench packets");
108
109int ip_do_randomid = 0;
110SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
111 &ip_do_randomid, 0,
112 "Assign random ip_id values");
113
114/*
115 * XXX - Setting ip_checkinterface mostly implements the receive side of
116 * the Strong ES model described in RFC 1122, but since the routing table
117 * and transmit implementation do not implement the Strong ES model,
118 * setting this to 1 results in an odd hybrid.
119 *
120 * XXX - ip_checkinterface currently must be disabled if you use ipnat
121 * to translate the destination address to another local interface.
122 *
123 * XXX - ip_checkinterface must be disabled if you add IP aliases
124 * to the loopback interface instead of the interface where the
125 * packets for those addresses are received.
126 */
127static int ip_checkinterface = 0;
128SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
129 &ip_checkinterface, 0, "Verify packet arrives on correct interface");
130
131struct pfil_head inet_pfil_hook; /* Packet filter hooks */
132
133static struct ifqueue ipintrq;
134static int ipqmaxlen = IFQ_MAXLEN;
135
136extern struct domain inetdomain;
137extern struct protosw inetsw[];
138u_char ip_protox[IPPROTO_MAX];
139struct in_ifaddrhead in_ifaddrhead; /* first inet address */
140struct in_ifaddrhashhead *in_ifaddrhashtbl; /* inet addr hash table */
141u_long in_ifaddrhmask; /* mask for hash table */
142
143SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
144 &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
145SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
146 &ipintrq.ifq_drops, 0,
147 "Number of packets dropped from the IP input queue");
148
149struct ipstat ipstat;
150SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
151 &ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
152
153/*
154 * IP datagram reassembly.
155 */
156#define IPREASS_NHASH_LOG2 6
157#define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2)
158#define IPREASS_HMASK (IPREASS_NHASH - 1)
159#define IPREASS_HASH(x,y) \
160 (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
161
162static uma_zone_t ipq_zone;
163static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
164static struct mtx ipqlock;
165
166#define IPQ_LOCK() mtx_lock(&ipqlock)
167#define IPQ_UNLOCK() mtx_unlock(&ipqlock)
168#define IPQ_LOCK_INIT() mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
169#define IPQ_LOCK_ASSERT() mtx_assert(&ipqlock, MA_OWNED)
170
171static void maxnipq_update(void);
172static void ipq_zone_change(void *);
173
174static int maxnipq; /* Administrative limit on # reass queues. */
175static int nipq = 0; /* Total # of reass queues */
176SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
177 &nipq, 0, "Current number of IPv4 fragment reassembly queue entries");
178
179static int maxfragsperpacket;
180SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
181 &maxfragsperpacket, 0,
182 "Maximum number of IPv4 fragments allowed per packet");
183
184struct callout ipport_tick_callout;
185
186#ifdef IPCTL_DEFMTU
187SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
188 &ip_mtu, 0, "Default MTU");
189#endif
190
191#ifdef IPSTEALTH
192int ipstealth = 0;
193SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
194 &ipstealth, 0, "IP stealth mode, no TTL decrementation on forwarding");
195#endif
196
197/*
198 * ipfw_ether and ipfw_bridge hooks.
199 * XXX: Temporary until those are converted to pfil_hooks as well.
200 */
201ip_fw_chk_t *ip_fw_chk_ptr = NULL;
202ip_dn_io_t *ip_dn_io_ptr = NULL;
203int fw_one_pass = 1;
204
205static void ip_freef(struct ipqhead *, struct ipq *);
206
207/*
208 * IP initialization: fill in IP protocol switch table.
209 * All protocols not implemented in kernel go to raw IP protocol handler.
210 */
211void
212ip_init(void)
213{
214 struct protosw *pr;
215 int i;
216
217 TAILQ_INIT(&in_ifaddrhead);
218 in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
219 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
220 if (pr == NULL)
221 panic("ip_init: PF_INET not found");
222
223 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
224 for (i = 0; i < IPPROTO_MAX; i++)
225 ip_protox[i] = pr - inetsw;
226 /*
227 * Cycle through IP protocols and put them into the appropriate place
228 * in ip_protox[].
229 */
230 for (pr = inetdomain.dom_protosw;
231 pr < inetdomain.dom_protoswNPROTOSW; pr++)
232 if (pr->pr_domain->dom_family == PF_INET &&
233 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
234 /* Be careful to only index valid IP protocols. */
235 if (pr->pr_protocol < IPPROTO_MAX)
236 ip_protox[pr->pr_protocol] = pr - inetsw;
237 }
238
239 /* Initialize packet filter hooks. */
240 inet_pfil_hook.ph_type = PFIL_TYPE_AF;
241 inet_pfil_hook.ph_af = AF_INET;
242 if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
243 printf("%s: WARNING: unable to register pfil hook, "
244 "error %d\n", __func__, i);
245
246 /* Initialize IP reassembly queue. */
247 IPQ_LOCK_INIT();
248 for (i = 0; i < IPREASS_NHASH; i++)
249 TAILQ_INIT(&ipq[i]);
250 maxnipq = nmbclusters / 32;
251 maxfragsperpacket = 16;
252 ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
253 NULL, UMA_ALIGN_PTR, 0);
254 maxnipq_update();
255
256 /* Start ipport_tick. */
257 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
258 ipport_tick(NULL);
259 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
260 SHUTDOWN_PRI_DEFAULT);
261 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
262 NULL, EVENTHANDLER_PRI_ANY);
263
264 /* Initialize various other remaining things. */
265 ip_id = time_second & 0xffff;
266 ipintrq.ifq_maxlen = ipqmaxlen;
267 mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
268 netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
269}
270
271void
272ip_fini(void *xtp)
273{
274
275 callout_stop(&ipport_tick_callout);
276}
277
278/*
279 * Ip input routine. Checksum and byte swap header. If fragmented
280 * try to reassemble. Process options. Pass to next level.
281 */
282void
283ip_input(struct mbuf *m)
284{
285 struct ip *ip = NULL;
286 struct in_ifaddr *ia = NULL;
287 struct ifaddr *ifa;
288 int checkif, hlen = 0;
289 u_short sum;
290 int dchg = 0; /* dest changed after fw */
291 struct in_addr odst; /* original dst address */
292
293 M_ASSERTPKTHDR(m);
294
295 if (m->m_flags & M_FASTFWD_OURS) {
296 /*
297 * Firewall or NAT changed destination to local.
298 * We expect ip_len and ip_off to be in host byte order.
299 */
300 m->m_flags &= ~M_FASTFWD_OURS;
301 /* Set up some basics that will be used later. */
302 ip = mtod(m, struct ip *);
303 hlen = ip->ip_hl << 2;
304 goto ours;
305 }
306
307 ipstat.ips_total++;
308
309 if (m->m_pkthdr.len < sizeof(struct ip))
310 goto tooshort;
311
312 if (m->m_len < sizeof (struct ip) &&
313 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
314 ipstat.ips_toosmall++;
315 return;
316 }
317 ip = mtod(m, struct ip *);
318
319 if (ip->ip_v != IPVERSION) {
320 ipstat.ips_badvers++;
321 goto bad;
322 }
323
324 hlen = ip->ip_hl << 2;
325 if (hlen < sizeof(struct ip)) { /* minimum header length */
326 ipstat.ips_badhlen++;
327 goto bad;
328 }
329 if (hlen > m->m_len) {
330 if ((m = m_pullup(m, hlen)) == NULL) {
331 ipstat.ips_badhlen++;
332 return;
333 }
334 ip = mtod(m, struct ip *);
335 }
336
337 /* 127/8 must not appear on wire - RFC1122 */
338 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
339 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
340 if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
341 ipstat.ips_badaddr++;
342 goto bad;
343 }
344 }
345
346 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
347 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
348 } else {
349 if (hlen == sizeof(struct ip)) {
350 sum = in_cksum_hdr(ip);
351 } else {
352 sum = in_cksum(m, hlen);
353 }
354 }
355 if (sum) {
356 ipstat.ips_badsum++;
357 goto bad;
358 }
359
360#ifdef ALTQ
361 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
362 /* packet is dropped by traffic conditioner */
363 return;
364#endif
365
366 /*
367 * Convert fields to host representation.
368 */
369 ip->ip_len = ntohs(ip->ip_len);
370 if (ip->ip_len < hlen) {
371 ipstat.ips_badlen++;
372 goto bad;
373 }
374 ip->ip_off = ntohs(ip->ip_off);
375
376 /*
377 * Check that the amount of data in the buffers
378 * is as at least much as the IP header would have us expect.
379 * Trim mbufs if longer than we expect.
380 * Drop packet if shorter than we expect.
381 */
382 if (m->m_pkthdr.len < ip->ip_len) {
383tooshort:
384 ipstat.ips_tooshort++;
385 goto bad;
386 }
387 if (m->m_pkthdr.len > ip->ip_len) {
388 if (m->m_len == m->m_pkthdr.len) {
389 m->m_len = ip->ip_len;
390 m->m_pkthdr.len = ip->ip_len;
391 } else
392 m_adj(m, ip->ip_len - m->m_pkthdr.len);
393 }
394#ifdef IPSEC
395 /*
396 * Bypass packet filtering for packets from a tunnel (gif).
397 */
| 31 */
32
33#include "opt_bootp.h"
34#include "opt_ipfw.h"
35#include "opt_ipstealth.h"
36#include "opt_ipsec.h"
37#include "opt_mac.h"
38#include "opt_carp.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/callout.h>
43#include <sys/mbuf.h>
44#include <sys/malloc.h>
45#include <sys/domain.h>
46#include <sys/protosw.h>
47#include <sys/socket.h>
48#include <sys/time.h>
49#include <sys/kernel.h>
50#include <sys/syslog.h>
51#include <sys/sysctl.h>
52
53#include <net/pfil.h>
54#include <net/if.h>
55#include <net/if_types.h>
56#include <net/if_var.h>
57#include <net/if_dl.h>
58#include <net/route.h>
59#include <net/netisr.h>
60
61#include <netinet/in.h>
62#include <netinet/in_systm.h>
63#include <netinet/in_var.h>
64#include <netinet/ip.h>
65#include <netinet/in_pcb.h>
66#include <netinet/ip_var.h>
67#include <netinet/ip_icmp.h>
68#include <netinet/ip_options.h>
69#include <machine/in_cksum.h>
70#ifdef DEV_CARP
71#include <netinet/ip_carp.h>
72#endif
73#ifdef IPSEC
74#include <netinet/ip_ipsec.h>
75#endif /* IPSEC */
76
77#include <sys/socketvar.h>
78
79/* XXX: Temporary until ipfw_ether and ipfw_bridge are converted. */
80#include <netinet/ip_fw.h>
81#include <netinet/ip_dummynet.h>
82
83#include <security/mac/mac_framework.h>
84
85int rsvp_on = 0;
86
87int ipforwarding = 0;
88SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_RW,
89 &ipforwarding, 0, "Enable IP forwarding between interfaces");
90
91static int ipsendredirects = 1; /* XXX */
92SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_RW,
93 &ipsendredirects, 0, "Enable sending IP redirects");
94
95int ip_defttl = IPDEFTTL;
96SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW,
97 &ip_defttl, 0, "Maximum TTL on IP packets");
98
99static int ip_keepfaith = 0;
100SYSCTL_INT(_net_inet_ip, IPCTL_KEEPFAITH, keepfaith, CTLFLAG_RW,
101 &ip_keepfaith, 0,
102 "Enable packet capture for FAITH IPv4->IPv6 translater daemon");
103
104static int ip_sendsourcequench = 0;
105SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench, CTLFLAG_RW,
106 &ip_sendsourcequench, 0,
107 "Enable the transmission of source quench packets");
108
109int ip_do_randomid = 0;
110SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id, CTLFLAG_RW,
111 &ip_do_randomid, 0,
112 "Assign random ip_id values");
113
114/*
115 * XXX - Setting ip_checkinterface mostly implements the receive side of
116 * the Strong ES model described in RFC 1122, but since the routing table
117 * and transmit implementation do not implement the Strong ES model,
118 * setting this to 1 results in an odd hybrid.
119 *
120 * XXX - ip_checkinterface currently must be disabled if you use ipnat
121 * to translate the destination address to another local interface.
122 *
123 * XXX - ip_checkinterface must be disabled if you add IP aliases
124 * to the loopback interface instead of the interface where the
125 * packets for those addresses are received.
126 */
127static int ip_checkinterface = 0;
128SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_RW,
129 &ip_checkinterface, 0, "Verify packet arrives on correct interface");
130
131struct pfil_head inet_pfil_hook; /* Packet filter hooks */
132
133static struct ifqueue ipintrq;
134static int ipqmaxlen = IFQ_MAXLEN;
135
136extern struct domain inetdomain;
137extern struct protosw inetsw[];
138u_char ip_protox[IPPROTO_MAX];
139struct in_ifaddrhead in_ifaddrhead; /* first inet address */
140struct in_ifaddrhashhead *in_ifaddrhashtbl; /* inet addr hash table */
141u_long in_ifaddrhmask; /* mask for hash table */
142
143SYSCTL_INT(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLFLAG_RW,
144 &ipintrq.ifq_maxlen, 0, "Maximum size of the IP input queue");
145SYSCTL_INT(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLFLAG_RD,
146 &ipintrq.ifq_drops, 0,
147 "Number of packets dropped from the IP input queue");
148
149struct ipstat ipstat;
150SYSCTL_STRUCT(_net_inet_ip, IPCTL_STATS, stats, CTLFLAG_RW,
151 &ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)");
152
153/*
154 * IP datagram reassembly.
155 */
156#define IPREASS_NHASH_LOG2 6
157#define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2)
158#define IPREASS_HMASK (IPREASS_NHASH - 1)
159#define IPREASS_HASH(x,y) \
160 (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
161
162static uma_zone_t ipq_zone;
163static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH];
164static struct mtx ipqlock;
165
166#define IPQ_LOCK() mtx_lock(&ipqlock)
167#define IPQ_UNLOCK() mtx_unlock(&ipqlock)
168#define IPQ_LOCK_INIT() mtx_init(&ipqlock, "ipqlock", NULL, MTX_DEF)
169#define IPQ_LOCK_ASSERT() mtx_assert(&ipqlock, MA_OWNED)
170
171static void maxnipq_update(void);
172static void ipq_zone_change(void *);
173
174static int maxnipq; /* Administrative limit on # reass queues. */
175static int nipq = 0; /* Total # of reass queues */
176SYSCTL_INT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD,
177 &nipq, 0, "Current number of IPv4 fragment reassembly queue entries");
178
179static int maxfragsperpacket;
180SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_RW,
181 &maxfragsperpacket, 0,
182 "Maximum number of IPv4 fragments allowed per packet");
183
184struct callout ipport_tick_callout;
185
186#ifdef IPCTL_DEFMTU
187SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW,
188 &ip_mtu, 0, "Default MTU");
189#endif
190
191#ifdef IPSTEALTH
192int ipstealth = 0;
193SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW,
194 &ipstealth, 0, "IP stealth mode, no TTL decrementation on forwarding");
195#endif
196
197/*
198 * ipfw_ether and ipfw_bridge hooks.
199 * XXX: Temporary until those are converted to pfil_hooks as well.
200 */
201ip_fw_chk_t *ip_fw_chk_ptr = NULL;
202ip_dn_io_t *ip_dn_io_ptr = NULL;
203int fw_one_pass = 1;
204
205static void ip_freef(struct ipqhead *, struct ipq *);
206
207/*
208 * IP initialization: fill in IP protocol switch table.
209 * All protocols not implemented in kernel go to raw IP protocol handler.
210 */
211void
212ip_init(void)
213{
214 struct protosw *pr;
215 int i;
216
217 TAILQ_INIT(&in_ifaddrhead);
218 in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &in_ifaddrhmask);
219 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
220 if (pr == NULL)
221 panic("ip_init: PF_INET not found");
222
223 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
224 for (i = 0; i < IPPROTO_MAX; i++)
225 ip_protox[i] = pr - inetsw;
226 /*
227 * Cycle through IP protocols and put them into the appropriate place
228 * in ip_protox[].
229 */
230 for (pr = inetdomain.dom_protosw;
231 pr < inetdomain.dom_protoswNPROTOSW; pr++)
232 if (pr->pr_domain->dom_family == PF_INET &&
233 pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) {
234 /* Be careful to only index valid IP protocols. */
235 if (pr->pr_protocol < IPPROTO_MAX)
236 ip_protox[pr->pr_protocol] = pr - inetsw;
237 }
238
239 /* Initialize packet filter hooks. */
240 inet_pfil_hook.ph_type = PFIL_TYPE_AF;
241 inet_pfil_hook.ph_af = AF_INET;
242 if ((i = pfil_head_register(&inet_pfil_hook)) != 0)
243 printf("%s: WARNING: unable to register pfil hook, "
244 "error %d\n", __func__, i);
245
246 /* Initialize IP reassembly queue. */
247 IPQ_LOCK_INIT();
248 for (i = 0; i < IPREASS_NHASH; i++)
249 TAILQ_INIT(&ipq[i]);
250 maxnipq = nmbclusters / 32;
251 maxfragsperpacket = 16;
252 ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL,
253 NULL, UMA_ALIGN_PTR, 0);
254 maxnipq_update();
255
256 /* Start ipport_tick. */
257 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE);
258 ipport_tick(NULL);
259 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
260 SHUTDOWN_PRI_DEFAULT);
261 EVENTHANDLER_REGISTER(nmbclusters_change, ipq_zone_change,
262 NULL, EVENTHANDLER_PRI_ANY);
263
264 /* Initialize various other remaining things. */
265 ip_id = time_second & 0xffff;
266 ipintrq.ifq_maxlen = ipqmaxlen;
267 mtx_init(&ipintrq.ifq_mtx, "ip_inq", NULL, MTX_DEF);
268 netisr_register(NETISR_IP, ip_input, &ipintrq, NETISR_MPSAFE);
269}
270
271void
272ip_fini(void *xtp)
273{
274
275 callout_stop(&ipport_tick_callout);
276}
277
278/*
279 * Ip input routine. Checksum and byte swap header. If fragmented
280 * try to reassemble. Process options. Pass to next level.
281 */
282void
283ip_input(struct mbuf *m)
284{
285 struct ip *ip = NULL;
286 struct in_ifaddr *ia = NULL;
287 struct ifaddr *ifa;
288 int checkif, hlen = 0;
289 u_short sum;
290 int dchg = 0; /* dest changed after fw */
291 struct in_addr odst; /* original dst address */
292
293 M_ASSERTPKTHDR(m);
294
295 if (m->m_flags & M_FASTFWD_OURS) {
296 /*
297 * Firewall or NAT changed destination to local.
298 * We expect ip_len and ip_off to be in host byte order.
299 */
300 m->m_flags &= ~M_FASTFWD_OURS;
301 /* Set up some basics that will be used later. */
302 ip = mtod(m, struct ip *);
303 hlen = ip->ip_hl << 2;
304 goto ours;
305 }
306
307 ipstat.ips_total++;
308
309 if (m->m_pkthdr.len < sizeof(struct ip))
310 goto tooshort;
311
312 if (m->m_len < sizeof (struct ip) &&
313 (m = m_pullup(m, sizeof (struct ip))) == NULL) {
314 ipstat.ips_toosmall++;
315 return;
316 }
317 ip = mtod(m, struct ip *);
318
319 if (ip->ip_v != IPVERSION) {
320 ipstat.ips_badvers++;
321 goto bad;
322 }
323
324 hlen = ip->ip_hl << 2;
325 if (hlen < sizeof(struct ip)) { /* minimum header length */
326 ipstat.ips_badhlen++;
327 goto bad;
328 }
329 if (hlen > m->m_len) {
330 if ((m = m_pullup(m, hlen)) == NULL) {
331 ipstat.ips_badhlen++;
332 return;
333 }
334 ip = mtod(m, struct ip *);
335 }
336
337 /* 127/8 must not appear on wire - RFC1122 */
338 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
339 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
340 if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) {
341 ipstat.ips_badaddr++;
342 goto bad;
343 }
344 }
345
346 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
347 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
348 } else {
349 if (hlen == sizeof(struct ip)) {
350 sum = in_cksum_hdr(ip);
351 } else {
352 sum = in_cksum(m, hlen);
353 }
354 }
355 if (sum) {
356 ipstat.ips_badsum++;
357 goto bad;
358 }
359
360#ifdef ALTQ
361 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0)
362 /* packet is dropped by traffic conditioner */
363 return;
364#endif
365
366 /*
367 * Convert fields to host representation.
368 */
369 ip->ip_len = ntohs(ip->ip_len);
370 if (ip->ip_len < hlen) {
371 ipstat.ips_badlen++;
372 goto bad;
373 }
374 ip->ip_off = ntohs(ip->ip_off);
375
376 /*
377 * Check that the amount of data in the buffers
378 * is as at least much as the IP header would have us expect.
379 * Trim mbufs if longer than we expect.
380 * Drop packet if shorter than we expect.
381 */
382 if (m->m_pkthdr.len < ip->ip_len) {
383tooshort:
384 ipstat.ips_tooshort++;
385 goto bad;
386 }
387 if (m->m_pkthdr.len > ip->ip_len) {
388 if (m->m_len == m->m_pkthdr.len) {
389 m->m_len = ip->ip_len;
390 m->m_pkthdr.len = ip->ip_len;
391 } else
392 m_adj(m, ip->ip_len - m->m_pkthdr.len);
393 }
394#ifdef IPSEC
395 /*
396 * Bypass packet filtering for packets from a tunnel (gif).
397 */
|
399 goto passin;
400#endif /* IPSEC */
401
402 /*
403 * Run through list of hooks for input packets.
404 *
405 * NB: Beware of the destination address changing (e.g.
406 * by NAT rewriting). When this happens, tell
407 * ip_forward to do the right thing.
408 */
409
410 /* Jump over all PFIL processing if hooks are not active. */
411 if (!PFIL_HOOKED(&inet_pfil_hook))
412 goto passin;
413
414 odst = ip->ip_dst;
415 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
416 PFIL_IN, NULL) != 0)
417 return;
418 if (m == NULL) /* consumed by filter */
419 return;
420
421 ip = mtod(m, struct ip *);
422 dchg = (odst.s_addr != ip->ip_dst.s_addr);
423
424#ifdef IPFIREWALL_FORWARD
425 if (m->m_flags & M_FASTFWD_OURS) {
426 m->m_flags &= ~M_FASTFWD_OURS;
427 goto ours;
428 }
429 if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
430 /*
431 * Directly ship on the packet. This allows to forward packets
432 * that were destined for us to some other directly connected
433 * host.
434 */
435 ip_forward(m, dchg);
436 return;
437 }
438#endif /* IPFIREWALL_FORWARD */
439
440passin:
441 /*
442 * Process options and, if not destined for us,
443 * ship it on. ip_dooptions returns 1 when an
444 * error was detected (causing an icmp message
445 * to be sent and the original packet to be freed).
446 */
447 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
448 return;
449
450 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
451 * matter if it is destined to another node, or whether it is
452 * a multicast one, RSVP wants it! and prevents it from being forwarded
453 * anywhere else. Also checks if the rsvp daemon is running before
454 * grabbing the packet.
455 */
456 if (rsvp_on && ip->ip_p==IPPROTO_RSVP)
457 goto ours;
458
459 /*
460 * Check our list of addresses, to see if the packet is for us.
461 * If we don't have any addresses, assume any unicast packet
462 * we receive might be for us (and let the upper layers deal
463 * with it).
464 */
465 if (TAILQ_EMPTY(&in_ifaddrhead) &&
466 (m->m_flags & (M_MCAST|M_BCAST)) == 0)
467 goto ours;
468
469 /*
470 * Enable a consistency check between the destination address
471 * and the arrival interface for a unicast packet (the RFC 1122
472 * strong ES model) if IP forwarding is disabled and the packet
473 * is not locally generated and the packet is not subject to
474 * 'ipfw fwd'.
475 *
476 * XXX - Checking also should be disabled if the destination
477 * address is ipnat'ed to a different interface.
478 *
479 * XXX - Checking is incompatible with IP aliases added
480 * to the loopback interface instead of the interface where
481 * the packets are received.
482 *
483 * XXX - This is the case for carp vhost IPs as well so we
484 * insert a workaround. If the packet got here, we already
485 * checked with carp_iamatch() and carp_forus().
486 */
487 checkif = ip_checkinterface && (ipforwarding == 0) &&
488 m->m_pkthdr.rcvif != NULL &&
489 ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
490#ifdef DEV_CARP
491 !m->m_pkthdr.rcvif->if_carp &&
492#endif
493 (dchg == 0);
494
495 /*
496 * Check for exact addresses in the hash bucket.
497 */
498 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
499 /*
500 * If the address matches, verify that the packet
501 * arrived via the correct interface if checking is
502 * enabled.
503 */
504 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
505 (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
506 goto ours;
507 }
508 /*
509 * Check for broadcast addresses.
510 *
511 * Only accept broadcast packets that arrive via the matching
512 * interface. Reception of forwarded directed broadcasts would
513 * be handled via ip_forward() and ether_output() with the loopback
514 * into the stack for SIMPLEX interfaces handled by ether_output().
515 */
516 if (m->m_pkthdr.rcvif != NULL &&
517 m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
518 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
519 if (ifa->ifa_addr->sa_family != AF_INET)
520 continue;
521 ia = ifatoia(ifa);
522 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
523 ip->ip_dst.s_addr)
524 goto ours;
525 if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
526 goto ours;
527#ifdef BOOTP_COMPAT
528 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
529 goto ours;
530#endif
531 }
532 }
533 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
534 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
535 ipstat.ips_cantforward++;
536 m_freem(m);
537 return;
538 }
539 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
540 struct in_multi *inm;
541 if (ip_mrouter) {
542 /*
543 * If we are acting as a multicast router, all
544 * incoming multicast packets are passed to the
545 * kernel-level multicast forwarding function.
546 * The packet is returned (relatively) intact; if
547 * ip_mforward() returns a non-zero value, the packet
548 * must be discarded, else it may be accepted below.
549 */
550 if (ip_mforward &&
551 ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
552 ipstat.ips_cantforward++;
553 m_freem(m);
554 return;
555 }
556
557 /*
558 * The process-level routing daemon needs to receive
559 * all multicast IGMP packets, whether or not this
560 * host belongs to their destination groups.
561 */
562 if (ip->ip_p == IPPROTO_IGMP)
563 goto ours;
564 ipstat.ips_forward++;
565 }
566 /*
567 * See if we belong to the destination multicast group on the
568 * arrival interface.
569 */
570 IN_MULTI_LOCK();
571 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
572 IN_MULTI_UNLOCK();
573 if (inm == NULL) {
574 ipstat.ips_notmember++;
575 m_freem(m);
576 return;
577 }
578 goto ours;
579 }
580 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
581 goto ours;
582 if (ip->ip_dst.s_addr == INADDR_ANY)
583 goto ours;
584
585 /*
586 * FAITH(Firewall Aided Internet Translator)
587 */
588 if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
589 if (ip_keepfaith) {
590 if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
591 goto ours;
592 }
593 m_freem(m);
594 return;
595 }
596
597 /*
598 * Not for us; forward if possible and desirable.
599 */
600 if (ipforwarding == 0) {
601 ipstat.ips_cantforward++;
602 m_freem(m);
603 } else {
604#ifdef IPSEC
605 if (ip_ipsec_fwd(m))
606 goto bad;
607#endif /* IPSEC */
608 ip_forward(m, dchg);
609 }
610 return;
611
612ours:
613#ifdef IPSTEALTH
614 /*
615 * IPSTEALTH: Process non-routing options only
616 * if the packet is destined for us.
617 */
618 if (ipstealth && hlen > sizeof (struct ip) &&
619 ip_dooptions(m, 1))
620 return;
621#endif /* IPSTEALTH */
622
623 /* Count the packet in the ip address stats */
624 if (ia != NULL) {
625 ia->ia_ifa.if_ipackets++;
626 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
627 }
628
629 /*
630 * Attempt reassembly; if it succeeds, proceed.
631 * ip_reass() will return a different mbuf.
632 */
633 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
634 m = ip_reass(m);
635 if (m == NULL)
636 return;
637 ip = mtod(m, struct ip *);
638 /* Get the header length of the reassembled packet */
639 hlen = ip->ip_hl << 2;
640 }
641
642 /*
643 * Further protocols expect the packet length to be w/o the
644 * IP header.
645 */
646 ip->ip_len -= hlen;
647
648#ifdef IPSEC
649 /*
650 * enforce IPsec policy checking if we are seeing last header.
651 * note that we do not visit this with protocols with pcb layer
652 * code - like udp/tcp/raw ip.
653 */
654 if (ip_ipsec_input(m))
655 goto bad;
656#endif /* IPSEC */
657
658 /*
659 * Switch out to protocol's input routine.
660 */
661 ipstat.ips_delivered++;
662
663 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
664 return;
665bad:
666 m_freem(m);
667}
668
669/*
670 * After maxnipq has been updated, propagate the change to UMA. The UMA zone
671 * max has slightly different semantics than the sysctl, for historical
672 * reasons.
673 */
674static void
675maxnipq_update(void)
676{
677
678 /*
679 * -1 for unlimited allocation.
680 */
681 if (maxnipq < 0)
682 uma_zone_set_max(ipq_zone, 0);
683 /*
684 * Positive number for specific bound.
685 */
686 if (maxnipq > 0)
687 uma_zone_set_max(ipq_zone, maxnipq);
688 /*
689 * Zero specifies no further fragment queue allocation -- set the
690 * bound very low, but rely on implementation elsewhere to actually
691 * prevent allocation and reclaim current queues.
692 */
693 if (maxnipq == 0)
694 uma_zone_set_max(ipq_zone, 1);
695}
696
697static void
698ipq_zone_change(void *tag)
699{
700
701 if (maxnipq > 0 && maxnipq < (nmbclusters / 32)) {
702 maxnipq = nmbclusters / 32;
703 maxnipq_update();
704 }
705}
706
707static int
708sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
709{
710 int error, i;
711
712 i = maxnipq;
713 error = sysctl_handle_int(oidp, &i, 0, req);
714 if (error || !req->newptr)
715 return (error);
716
717 /*
718 * XXXRW: Might be a good idea to sanity check the argument and place
719 * an extreme upper bound.
720 */
721 if (i < -1)
722 return (EINVAL);
723 maxnipq = i;
724 maxnipq_update();
725 return (0);
726}
727
728SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
729 NULL, 0, sysctl_maxnipq, "I",
730 "Maximum number of IPv4 fragment reassembly queue entries");
731
732/*
733 * Take incoming datagram fragment and try to reassemble it into
734 * whole datagram. If the argument is the first fragment or one
735 * in between the function will return NULL and store the mbuf
736 * in the fragment chain. If the argument is the last fragment
737 * the packet will be reassembled and the pointer to the new
738 * mbuf returned for further processing. Only m_tags attached
739 * to the first packet/fragment are preserved.
740 * The IP header is *NOT* adjusted out of iplen.
741 */
742struct mbuf *
743ip_reass(struct mbuf *m)
744{
745 struct ip *ip;
746 struct mbuf *p, *q, *nq, *t;
747 struct ipq *fp = NULL;
748 struct ipqhead *head;
749 int i, hlen, next;
750 u_int8_t ecn, ecn0;
751 u_short hash;
752
753 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
754 if (maxnipq == 0 || maxfragsperpacket == 0) {
755 ipstat.ips_fragments++;
756 ipstat.ips_fragdropped++;
757 m_freem(m);
758 return (NULL);
759 }
760
761 ip = mtod(m, struct ip *);
762 hlen = ip->ip_hl << 2;
763
764 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
765 head = &ipq[hash];
766 IPQ_LOCK();
767
768 /*
769 * Look for queue of fragments
770 * of this datagram.
771 */
772 TAILQ_FOREACH(fp, head, ipq_list)
773 if (ip->ip_id == fp->ipq_id &&
774 ip->ip_src.s_addr == fp->ipq_src.s_addr &&
775 ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
776#ifdef MAC
777 mac_fragment_match(m, fp) &&
778#endif
779 ip->ip_p == fp->ipq_p)
780 goto found;
781
782 fp = NULL;
783
784 /*
785 * Attempt to trim the number of allocated fragment queues if it
786 * exceeds the administrative limit.
787 */
788 if ((nipq > maxnipq) && (maxnipq > 0)) {
789 /*
790 * drop something from the tail of the current queue
791 * before proceeding further
792 */
793 struct ipq *q = TAILQ_LAST(head, ipqhead);
794 if (q == NULL) { /* gak */
795 for (i = 0; i < IPREASS_NHASH; i++) {
796 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
797 if (r) {
798 ipstat.ips_fragtimeout += r->ipq_nfrags;
799 ip_freef(&ipq[i], r);
800 break;
801 }
802 }
803 } else {
804 ipstat.ips_fragtimeout += q->ipq_nfrags;
805 ip_freef(head, q);
806 }
807 }
808
809found:
810 /*
811 * Adjust ip_len to not reflect header,
812 * convert offset of this to bytes.
813 */
814 ip->ip_len -= hlen;
815 if (ip->ip_off & IP_MF) {
816 /*
817 * Make sure that fragments have a data length
818 * that's a non-zero multiple of 8 bytes.
819 */
820 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
821 ipstat.ips_toosmall++; /* XXX */
822 goto dropfrag;
823 }
824 m->m_flags |= M_FRAG;
825 } else
826 m->m_flags &= ~M_FRAG;
827 ip->ip_off <<= 3;
828
829
830 /*
831 * Attempt reassembly; if it succeeds, proceed.
832 * ip_reass() will return a different mbuf.
833 */
834 ipstat.ips_fragments++;
835 m->m_pkthdr.header = ip;
836
837 /* Previous ip_reass() started here. */
838 /*
839 * Presence of header sizes in mbufs
840 * would confuse code below.
841 */
842 m->m_data += hlen;
843 m->m_len -= hlen;
844
845 /*
846 * If first fragment to arrive, create a reassembly queue.
847 */
848 if (fp == NULL) {
849 fp = uma_zalloc(ipq_zone, M_NOWAIT);
850 if (fp == NULL)
851 goto dropfrag;
852#ifdef MAC
853 if (mac_init_ipq(fp, M_NOWAIT) != 0) {
854 uma_zfree(ipq_zone, fp);
855 fp = NULL;
856 goto dropfrag;
857 }
858 mac_create_ipq(m, fp);
859#endif
860 TAILQ_INSERT_HEAD(head, fp, ipq_list);
861 nipq++;
862 fp->ipq_nfrags = 1;
863 fp->ipq_ttl = IPFRAGTTL;
864 fp->ipq_p = ip->ip_p;
865 fp->ipq_id = ip->ip_id;
866 fp->ipq_src = ip->ip_src;
867 fp->ipq_dst = ip->ip_dst;
868 fp->ipq_frags = m;
869 m->m_nextpkt = NULL;
870 goto done;
871 } else {
872 fp->ipq_nfrags++;
873#ifdef MAC
874 mac_update_ipq(m, fp);
875#endif
876 }
877
878#define GETIP(m) ((struct ip*)((m)->m_pkthdr.header))
879
880 /*
881 * Handle ECN by comparing this segment with the first one;
882 * if CE is set, do not lose CE.
883 * drop if CE and not-ECT are mixed for the same packet.
884 */
885 ecn = ip->ip_tos & IPTOS_ECN_MASK;
886 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
887 if (ecn == IPTOS_ECN_CE) {
888 if (ecn0 == IPTOS_ECN_NOTECT)
889 goto dropfrag;
890 if (ecn0 != IPTOS_ECN_CE)
891 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
892 }
893 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
894 goto dropfrag;
895
896 /*
897 * Find a segment which begins after this one does.
898 */
899 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
900 if (GETIP(q)->ip_off > ip->ip_off)
901 break;
902
903 /*
904 * If there is a preceding segment, it may provide some of
905 * our data already. If so, drop the data from the incoming
906 * segment. If it provides all of our data, drop us, otherwise
907 * stick new segment in the proper place.
908 *
909 * If some of the data is dropped from the the preceding
910 * segment, then it's checksum is invalidated.
911 */
912 if (p) {
913 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
914 if (i > 0) {
915 if (i >= ip->ip_len)
916 goto dropfrag;
917 m_adj(m, i);
918 m->m_pkthdr.csum_flags = 0;
919 ip->ip_off += i;
920 ip->ip_len -= i;
921 }
922 m->m_nextpkt = p->m_nextpkt;
923 p->m_nextpkt = m;
924 } else {
925 m->m_nextpkt = fp->ipq_frags;
926 fp->ipq_frags = m;
927 }
928
929 /*
930 * While we overlap succeeding segments trim them or,
931 * if they are completely covered, dequeue them.
932 */
933 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
934 q = nq) {
935 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
936 if (i < GETIP(q)->ip_len) {
937 GETIP(q)->ip_len -= i;
938 GETIP(q)->ip_off += i;
939 m_adj(q, i);
940 q->m_pkthdr.csum_flags = 0;
941 break;
942 }
943 nq = q->m_nextpkt;
944 m->m_nextpkt = nq;
945 ipstat.ips_fragdropped++;
946 fp->ipq_nfrags--;
947 m_freem(q);
948 }
949
950 /*
951 * Check for complete reassembly and perform frag per packet
952 * limiting.
953 *
954 * Frag limiting is performed here so that the nth frag has
955 * a chance to complete the packet before we drop the packet.
956 * As a result, n+1 frags are actually allowed per packet, but
957 * only n will ever be stored. (n = maxfragsperpacket.)
958 *
959 */
960 next = 0;
961 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
962 if (GETIP(q)->ip_off != next) {
963 if (fp->ipq_nfrags > maxfragsperpacket) {
964 ipstat.ips_fragdropped += fp->ipq_nfrags;
965 ip_freef(head, fp);
966 }
967 goto done;
968 }
969 next += GETIP(q)->ip_len;
970 }
971 /* Make sure the last packet didn't have the IP_MF flag */
972 if (p->m_flags & M_FRAG) {
973 if (fp->ipq_nfrags > maxfragsperpacket) {
974 ipstat.ips_fragdropped += fp->ipq_nfrags;
975 ip_freef(head, fp);
976 }
977 goto done;
978 }
979
980 /*
981 * Reassembly is complete. Make sure the packet is a sane size.
982 */
983 q = fp->ipq_frags;
984 ip = GETIP(q);
985 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
986 ipstat.ips_toolong++;
987 ipstat.ips_fragdropped += fp->ipq_nfrags;
988 ip_freef(head, fp);
989 goto done;
990 }
991
992 /*
993 * Concatenate fragments.
994 */
995 m = q;
996 t = m->m_next;
997 m->m_next = NULL;
998 m_cat(m, t);
999 nq = q->m_nextpkt;
1000 q->m_nextpkt = NULL;
1001 for (q = nq; q != NULL; q = nq) {
1002 nq = q->m_nextpkt;
1003 q->m_nextpkt = NULL;
1004 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1005 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1006 m_cat(m, q);
1007 }
1008 /*
1009 * In order to do checksumming faster we do 'end-around carry' here
1010 * (and not in for{} loop), though it implies we are not going to
1011 * reassemble more than 64k fragments.
1012 */
1013 m->m_pkthdr.csum_data =
1014 (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1015#ifdef MAC
1016 mac_create_datagram_from_ipq(fp, m);
1017 mac_destroy_ipq(fp);
1018#endif
1019
1020 /*
1021 * Create header for new ip packet by modifying header of first
1022 * packet; dequeue and discard fragment reassembly header.
1023 * Make header visible.
1024 */
1025 ip->ip_len = (ip->ip_hl << 2) + next;
1026 ip->ip_src = fp->ipq_src;
1027 ip->ip_dst = fp->ipq_dst;
1028 TAILQ_REMOVE(head, fp, ipq_list);
1029 nipq--;
1030 uma_zfree(ipq_zone, fp);
1031 m->m_len += (ip->ip_hl << 2);
1032 m->m_data -= (ip->ip_hl << 2);
1033 /* some debugging cruft by sklower, below, will go away soon */
1034 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */
1035 m_fixhdr(m);
1036 ipstat.ips_reassembled++;
1037 IPQ_UNLOCK();
1038 return (m);
1039
1040dropfrag:
1041 ipstat.ips_fragdropped++;
1042 if (fp != NULL)
1043 fp->ipq_nfrags--;
1044 m_freem(m);
1045done:
1046 IPQ_UNLOCK();
1047 return (NULL);
1048
1049#undef GETIP
1050}
1051
1052/*
1053 * Free a fragment reassembly header and all
1054 * associated datagrams.
1055 */
1056static void
1057ip_freef(struct ipqhead *fhp, struct ipq *fp)
1058{
1059 struct mbuf *q;
1060
1061 IPQ_LOCK_ASSERT();
1062
1063 while (fp->ipq_frags) {
1064 q = fp->ipq_frags;
1065 fp->ipq_frags = q->m_nextpkt;
1066 m_freem(q);
1067 }
1068 TAILQ_REMOVE(fhp, fp, ipq_list);
1069 uma_zfree(ipq_zone, fp);
1070 nipq--;
1071}
1072
1073/*
1074 * IP timer processing;
1075 * if a timer expires on a reassembly
1076 * queue, discard it.
1077 */
1078void
1079ip_slowtimo(void)
1080{
1081 struct ipq *fp;
1082 int i;
1083
1084 IPQ_LOCK();
1085 for (i = 0; i < IPREASS_NHASH; i++) {
1086 for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1087 struct ipq *fpp;
1088
1089 fpp = fp;
1090 fp = TAILQ_NEXT(fp, ipq_list);
1091 if(--fpp->ipq_ttl == 0) {
1092 ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1093 ip_freef(&ipq[i], fpp);
1094 }
1095 }
1096 }
1097 /*
1098 * If we are over the maximum number of fragments
1099 * (due to the limit being lowered), drain off
1100 * enough to get down to the new limit.
1101 */
1102 if (maxnipq >= 0 && nipq > maxnipq) {
1103 for (i = 0; i < IPREASS_NHASH; i++) {
1104 while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1105 ipstat.ips_fragdropped +=
1106 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1107 ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1108 }
1109 }
1110 }
1111 IPQ_UNLOCK();
1112}
1113
1114/*
1115 * Drain off all datagram fragments.
1116 */
1117void
1118ip_drain(void)
1119{
1120 int i;
1121
1122 IPQ_LOCK();
1123 for (i = 0; i < IPREASS_NHASH; i++) {
1124 while(!TAILQ_EMPTY(&ipq[i])) {
1125 ipstat.ips_fragdropped +=
1126 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1127 ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1128 }
1129 }
1130 IPQ_UNLOCK();
1131 in_rtqdrain();
1132}
1133
1134/*
1135 * The protocol to be inserted into ip_protox[] must be already registered
1136 * in inetsw[], either statically or through pf_proto_register().
1137 */
1138int
1139ipproto_register(u_char ipproto)
1140{
1141 struct protosw *pr;
1142
1143 /* Sanity checks. */
1144 if (ipproto == 0)
1145 return (EPROTONOSUPPORT);
1146
1147 /*
1148 * The protocol slot must not be occupied by another protocol
1149 * already. An index pointing to IPPROTO_RAW is unused.
1150 */
1151 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1152 if (pr == NULL)
1153 return (EPFNOSUPPORT);
1154 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */
1155 return (EEXIST);
1156
1157 /* Find the protocol position in inetsw[] and set the index. */
1158 for (pr = inetdomain.dom_protosw;
1159 pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1160 if (pr->pr_domain->dom_family == PF_INET &&
1161 pr->pr_protocol && pr->pr_protocol == ipproto) {
1162 /* Be careful to only index valid IP protocols. */
1163 if (pr->pr_protocol < IPPROTO_MAX) {
1164 ip_protox[pr->pr_protocol] = pr - inetsw;
1165 return (0);
1166 } else
1167 return (EINVAL);
1168 }
1169 }
1170 return (EPROTONOSUPPORT);
1171}
1172
1173int
1174ipproto_unregister(u_char ipproto)
1175{
1176 struct protosw *pr;
1177
1178 /* Sanity checks. */
1179 if (ipproto == 0)
1180 return (EPROTONOSUPPORT);
1181
1182 /* Check if the protocol was indeed registered. */
1183 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1184 if (pr == NULL)
1185 return (EPFNOSUPPORT);
1186 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */
1187 return (ENOENT);
1188
1189 /* Reset the protocol slot to IPPROTO_RAW. */
1190 ip_protox[ipproto] = pr - inetsw;
1191 return (0);
1192}
1193
1194/*
1195 * Given address of next destination (final or next hop),
1196 * return internet address info of interface to be used to get there.
1197 */
1198struct in_ifaddr *
1199ip_rtaddr(struct in_addr dst)
1200{
1201 struct route sro;
1202 struct sockaddr_in *sin;
1203 struct in_ifaddr *ifa;
1204
1205 bzero(&sro, sizeof(sro));
1206 sin = (struct sockaddr_in *)&sro.ro_dst;
1207 sin->sin_family = AF_INET;
1208 sin->sin_len = sizeof(*sin);
1209 sin->sin_addr = dst;
1210 rtalloc_ign(&sro, RTF_CLONING);
1211
1212 if (sro.ro_rt == NULL)
1213 return (NULL);
1214
1215 ifa = ifatoia(sro.ro_rt->rt_ifa);
1216 RTFREE(sro.ro_rt);
1217 return (ifa);
1218}
1219
1220u_char inetctlerrmap[PRC_NCMDS] = {
1221 0, 0, 0, 0,
1222 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH,
1223 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED,
1224 EMSGSIZE, EHOSTUNREACH, 0, 0,
1225 0, 0, EHOSTUNREACH, 0,
1226 ENOPROTOOPT, ECONNREFUSED
1227};
1228
1229/*
1230 * Forward a packet. If some error occurs return the sender
1231 * an icmp packet. Note we can't always generate a meaningful
1232 * icmp message because icmp doesn't have a large enough repertoire
1233 * of codes and types.
1234 *
1235 * If not forwarding, just drop the packet. This could be confusing
1236 * if ipforwarding was zero but some routing protocol was advancing
1237 * us as a gateway to somewhere. However, we must let the routing
1238 * protocol deal with that.
1239 *
1240 * The srcrt parameter indicates whether the packet is being forwarded
1241 * via a source route.
1242 */
1243void
1244ip_forward(struct mbuf *m, int srcrt)
1245{
1246 struct ip *ip = mtod(m, struct ip *);
1247 struct in_ifaddr *ia = NULL;
1248 struct mbuf *mcopy;
1249 struct in_addr dest;
1250 int error, type = 0, code = 0, mtu = 0;
1251
1252 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1253 ipstat.ips_cantforward++;
1254 m_freem(m);
1255 return;
1256 }
1257#ifdef IPSTEALTH
1258 if (!ipstealth) {
1259#endif
1260 if (ip->ip_ttl <= IPTTLDEC) {
1261 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1262 0, 0);
1263 return;
1264 }
1265#ifdef IPSTEALTH
1266 }
1267#endif
1268
1269 if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1270 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1271 return;
1272 }
1273
1274 /*
1275 * Save the IP header and at most 8 bytes of the payload,
1276 * in case we need to generate an ICMP message to the src.
1277 *
1278 * XXX this can be optimized a lot by saving the data in a local
1279 * buffer on the stack (72 bytes at most), and only allocating the
1280 * mbuf if really necessary. The vast majority of the packets
1281 * are forwarded without having to send an ICMP back (either
1282 * because unnecessary, or because rate limited), so we are
1283 * really we are wasting a lot of work here.
1284 *
1285 * We don't use m_copy() because it might return a reference
1286 * to a shared cluster. Both this function and ip_output()
1287 * assume exclusive access to the IP header in `m', so any
1288 * data in a cluster may change before we reach icmp_error().
1289 */
1290 MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1291 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1292 /*
1293 * It's probably ok if the pkthdr dup fails (because
1294 * the deep copy of the tag chain failed), but for now
1295 * be conservative and just discard the copy since
1296 * code below may some day want the tags.
1297 */
1298 m_free(mcopy);
1299 mcopy = NULL;
1300 }
1301 if (mcopy != NULL) {
1302 mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1303 mcopy->m_pkthdr.len = mcopy->m_len;
1304 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1305 }
1306
1307#ifdef IPSTEALTH
1308 if (!ipstealth) {
1309#endif
1310 ip->ip_ttl -= IPTTLDEC;
1311#ifdef IPSTEALTH
1312 }
1313#endif
1314
1315 /*
1316 * If forwarding packet using same interface that it came in on,
1317 * perhaps should send a redirect to sender to shortcut a hop.
1318 * Only send redirect if source is sending directly to us,
1319 * and if packet was not source routed (or has any options).
1320 * Also, don't send redirect if forwarding using a default route
1321 * or a route modified by a redirect.
1322 */
1323 dest.s_addr = 0;
1324 if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1325 struct sockaddr_in *sin;
1326 struct route ro;
1327 struct rtentry *rt;
1328
1329 bzero(&ro, sizeof(ro));
1330 sin = (struct sockaddr_in *)&ro.ro_dst;
1331 sin->sin_family = AF_INET;
1332 sin->sin_len = sizeof(*sin);
1333 sin->sin_addr = ip->ip_dst;
1334 rtalloc_ign(&ro, RTF_CLONING);
1335
1336 rt = ro.ro_rt;
1337
1338 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1339 satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1340#define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa))
1341 u_long src = ntohl(ip->ip_src.s_addr);
1342
1343 if (RTA(rt) &&
1344 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1345 if (rt->rt_flags & RTF_GATEWAY)
1346 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1347 else
1348 dest.s_addr = ip->ip_dst.s_addr;
1349 /* Router requirements says to only send host redirects */
1350 type = ICMP_REDIRECT;
1351 code = ICMP_REDIRECT_HOST;
1352 }
1353 }
1354 if (rt)
1355 RTFREE(rt);
1356 }
1357
1358 error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1359 if (error)
1360 ipstat.ips_cantforward++;
1361 else {
1362 ipstat.ips_forward++;
1363 if (type)
1364 ipstat.ips_redirectsent++;
1365 else {
1366 if (mcopy)
1367 m_freem(mcopy);
1368 return;
1369 }
1370 }
1371 if (mcopy == NULL)
1372 return;
1373
1374 switch (error) {
1375
1376 case 0: /* forwarded, but need redirect */
1377 /* type, code set above */
1378 break;
1379
1380 case ENETUNREACH: /* shouldn't happen, checked above */
1381 case EHOSTUNREACH:
1382 case ENETDOWN:
1383 case EHOSTDOWN:
1384 default:
1385 type = ICMP_UNREACH;
1386 code = ICMP_UNREACH_HOST;
1387 break;
1388
1389 case EMSGSIZE:
1390 type = ICMP_UNREACH;
1391 code = ICMP_UNREACH_NEEDFRAG;
1392
1393#ifdef IPSEC
1394 mtu = ip_ipsec_mtu(m);
1395#endif /* IPSEC */
1396 /*
1397 * If the MTU wasn't set before use the interface mtu or
1398 * fall back to the next smaller mtu step compared to the
1399 * current packet size.
1400 */
1401 if (mtu == 0) {
1402 if (ia != NULL)
1403 mtu = ia->ia_ifp->if_mtu;
1404 else
1405 mtu = ip_next_mtu(ip->ip_len, 0);
1406 }
1407 ipstat.ips_cantfrag++;
1408 break;
1409
1410 case ENOBUFS:
1411 /*
1412 * A router should not generate ICMP_SOURCEQUENCH as
1413 * required in RFC1812 Requirements for IP Version 4 Routers.
1414 * Source quench could be a big problem under DoS attacks,
1415 * or if the underlying interface is rate-limited.
1416 * Those who need source quench packets may re-enable them
1417 * via the net.inet.ip.sendsourcequench sysctl.
1418 */
1419 if (ip_sendsourcequench == 0) {
1420 m_freem(mcopy);
1421 return;
1422 } else {
1423 type = ICMP_SOURCEQUENCH;
1424 code = 0;
1425 }
1426 break;
1427
1428 case EACCES: /* ipfw denied packet */
1429 m_freem(mcopy);
1430 return;
1431 }
1432 icmp_error(mcopy, type, code, dest.s_addr, mtu);
1433}
1434
1435void
1436ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1437 struct mbuf *m)
1438{
1439 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1440 struct bintime bt;
1441
1442 bintime(&bt);
1443 if (inp->inp_socket->so_options & SO_BINTIME) {
1444 *mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1445 SCM_BINTIME, SOL_SOCKET);
1446 if (*mp)
1447 mp = &(*mp)->m_next;
1448 }
1449 if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1450 struct timeval tv;
1451
1452 bintime2timeval(&bt, &tv);
1453 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1454 SCM_TIMESTAMP, SOL_SOCKET);
1455 if (*mp)
1456 mp = &(*mp)->m_next;
1457 }
1458 }
1459 if (inp->inp_flags & INP_RECVDSTADDR) {
1460 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1461 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1462 if (*mp)
1463 mp = &(*mp)->m_next;
1464 }
1465 if (inp->inp_flags & INP_RECVTTL) {
1466 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1467 sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1468 if (*mp)
1469 mp = &(*mp)->m_next;
1470 }
1471#ifdef notyet
1472 /* XXX
1473 * Moving these out of udp_input() made them even more broken
1474 * than they already were.
1475 */
1476 /* options were tossed already */
1477 if (inp->inp_flags & INP_RECVOPTS) {
1478 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1479 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1480 if (*mp)
1481 mp = &(*mp)->m_next;
1482 }
1483 /* ip_srcroute doesn't do what we want here, need to fix */
1484 if (inp->inp_flags & INP_RECVRETOPTS) {
1485 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1486 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1487 if (*mp)
1488 mp = &(*mp)->m_next;
1489 }
1490#endif
1491 if (inp->inp_flags & INP_RECVIF) {
1492 struct ifnet *ifp;
1493 struct sdlbuf {
1494 struct sockaddr_dl sdl;
1495 u_char pad[32];
1496 } sdlbuf;
1497 struct sockaddr_dl *sdp;
1498 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1499
1500 if (((ifp = m->m_pkthdr.rcvif))
1501 && ( ifp->if_index && (ifp->if_index <= if_index))) {
1502 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1503 /*
1504 * Change our mind and don't try copy.
1505 */
1506 if ((sdp->sdl_family != AF_LINK)
1507 || (sdp->sdl_len > sizeof(sdlbuf))) {
1508 goto makedummy;
1509 }
1510 bcopy(sdp, sdl2, sdp->sdl_len);
1511 } else {
1512makedummy:
1513 sdl2->sdl_len
1514 = offsetof(struct sockaddr_dl, sdl_data[0]);
1515 sdl2->sdl_family = AF_LINK;
1516 sdl2->sdl_index = 0;
1517 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1518 }
1519 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1520 IP_RECVIF, IPPROTO_IP);
1521 if (*mp)
1522 mp = &(*mp)->m_next;
1523 }
1524}
1525
1526/*
1527 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1528 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1529 * locking. This code remains in ip_input.c as ip_mroute.c is optionally
1530 * compiled.
1531 */
1532static int ip_rsvp_on;
1533struct socket *ip_rsvpd;
1534int
1535ip_rsvp_init(struct socket *so)
1536{
1537 if (so->so_type != SOCK_RAW ||
1538 so->so_proto->pr_protocol != IPPROTO_RSVP)
1539 return EOPNOTSUPP;
1540
1541 if (ip_rsvpd != NULL)
1542 return EADDRINUSE;
1543
1544 ip_rsvpd = so;
1545 /*
1546 * This may seem silly, but we need to be sure we don't over-increment
1547 * the RSVP counter, in case something slips up.
1548 */
1549 if (!ip_rsvp_on) {
1550 ip_rsvp_on = 1;
1551 rsvp_on++;
1552 }
1553
1554 return 0;
1555}
1556
1557int
1558ip_rsvp_done(void)
1559{
1560 ip_rsvpd = NULL;
1561 /*
1562 * This may seem silly, but we need to be sure we don't over-decrement
1563 * the RSVP counter, in case something slips up.
1564 */
1565 if (ip_rsvp_on) {
1566 ip_rsvp_on = 0;
1567 rsvp_on--;
1568 }
1569 return 0;
1570}
1571
1572void
1573rsvp_input(struct mbuf *m, int off) /* XXX must fixup manually */
1574{
1575 if (rsvp_input_p) { /* call the real one if loaded */
1576 rsvp_input_p(m, off);
1577 return;
1578 }
1579
1580 /* Can still get packets with rsvp_on = 0 if there is a local member
1581 * of the group to which the RSVP packet is addressed. But in this
1582 * case we want to throw the packet away.
1583 */
1584
1585 if (!rsvp_on) {
1586 m_freem(m);
1587 return;
1588 }
1589
1590 if (ip_rsvpd != NULL) {
1591 rip_input(m, off);
1592 return;
1593 }
1594 /* Drop the packet */
1595 m_freem(m);
1596}
| 399 goto passin;
400#endif /* IPSEC */
401
402 /*
403 * Run through list of hooks for input packets.
404 *
405 * NB: Beware of the destination address changing (e.g.
406 * by NAT rewriting). When this happens, tell
407 * ip_forward to do the right thing.
408 */
409
410 /* Jump over all PFIL processing if hooks are not active. */
411 if (!PFIL_HOOKED(&inet_pfil_hook))
412 goto passin;
413
414 odst = ip->ip_dst;
415 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif,
416 PFIL_IN, NULL) != 0)
417 return;
418 if (m == NULL) /* consumed by filter */
419 return;
420
421 ip = mtod(m, struct ip *);
422 dchg = (odst.s_addr != ip->ip_dst.s_addr);
423
424#ifdef IPFIREWALL_FORWARD
425 if (m->m_flags & M_FASTFWD_OURS) {
426 m->m_flags &= ~M_FASTFWD_OURS;
427 goto ours;
428 }
429 if ((dchg = (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL)) != 0) {
430 /*
431 * Directly ship on the packet. This allows to forward packets
432 * that were destined for us to some other directly connected
433 * host.
434 */
435 ip_forward(m, dchg);
436 return;
437 }
438#endif /* IPFIREWALL_FORWARD */
439
440passin:
441 /*
442 * Process options and, if not destined for us,
443 * ship it on. ip_dooptions returns 1 when an
444 * error was detected (causing an icmp message
445 * to be sent and the original packet to be freed).
446 */
447 if (hlen > sizeof (struct ip) && ip_dooptions(m, 0))
448 return;
449
450 /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no
451 * matter if it is destined to another node, or whether it is
452 * a multicast one, RSVP wants it! and prevents it from being forwarded
453 * anywhere else. Also checks if the rsvp daemon is running before
454 * grabbing the packet.
455 */
456 if (rsvp_on && ip->ip_p==IPPROTO_RSVP)
457 goto ours;
458
459 /*
460 * Check our list of addresses, to see if the packet is for us.
461 * If we don't have any addresses, assume any unicast packet
462 * we receive might be for us (and let the upper layers deal
463 * with it).
464 */
465 if (TAILQ_EMPTY(&in_ifaddrhead) &&
466 (m->m_flags & (M_MCAST|M_BCAST)) == 0)
467 goto ours;
468
469 /*
470 * Enable a consistency check between the destination address
471 * and the arrival interface for a unicast packet (the RFC 1122
472 * strong ES model) if IP forwarding is disabled and the packet
473 * is not locally generated and the packet is not subject to
474 * 'ipfw fwd'.
475 *
476 * XXX - Checking also should be disabled if the destination
477 * address is ipnat'ed to a different interface.
478 *
479 * XXX - Checking is incompatible with IP aliases added
480 * to the loopback interface instead of the interface where
481 * the packets are received.
482 *
483 * XXX - This is the case for carp vhost IPs as well so we
484 * insert a workaround. If the packet got here, we already
485 * checked with carp_iamatch() and carp_forus().
486 */
487 checkif = ip_checkinterface && (ipforwarding == 0) &&
488 m->m_pkthdr.rcvif != NULL &&
489 ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) == 0) &&
490#ifdef DEV_CARP
491 !m->m_pkthdr.rcvif->if_carp &&
492#endif
493 (dchg == 0);
494
495 /*
496 * Check for exact addresses in the hash bucket.
497 */
498 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
499 /*
500 * If the address matches, verify that the packet
501 * arrived via the correct interface if checking is
502 * enabled.
503 */
504 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr &&
505 (!checkif || ia->ia_ifp == m->m_pkthdr.rcvif))
506 goto ours;
507 }
508 /*
509 * Check for broadcast addresses.
510 *
511 * Only accept broadcast packets that arrive via the matching
512 * interface. Reception of forwarded directed broadcasts would
513 * be handled via ip_forward() and ether_output() with the loopback
514 * into the stack for SIMPLEX interfaces handled by ether_output().
515 */
516 if (m->m_pkthdr.rcvif != NULL &&
517 m->m_pkthdr.rcvif->if_flags & IFF_BROADCAST) {
518 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) {
519 if (ifa->ifa_addr->sa_family != AF_INET)
520 continue;
521 ia = ifatoia(ifa);
522 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr ==
523 ip->ip_dst.s_addr)
524 goto ours;
525 if (ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr)
526 goto ours;
527#ifdef BOOTP_COMPAT
528 if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY)
529 goto ours;
530#endif
531 }
532 }
533 /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */
534 if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) {
535 ipstat.ips_cantforward++;
536 m_freem(m);
537 return;
538 }
539 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
540 struct in_multi *inm;
541 if (ip_mrouter) {
542 /*
543 * If we are acting as a multicast router, all
544 * incoming multicast packets are passed to the
545 * kernel-level multicast forwarding function.
546 * The packet is returned (relatively) intact; if
547 * ip_mforward() returns a non-zero value, the packet
548 * must be discarded, else it may be accepted below.
549 */
550 if (ip_mforward &&
551 ip_mforward(ip, m->m_pkthdr.rcvif, m, 0) != 0) {
552 ipstat.ips_cantforward++;
553 m_freem(m);
554 return;
555 }
556
557 /*
558 * The process-level routing daemon needs to receive
559 * all multicast IGMP packets, whether or not this
560 * host belongs to their destination groups.
561 */
562 if (ip->ip_p == IPPROTO_IGMP)
563 goto ours;
564 ipstat.ips_forward++;
565 }
566 /*
567 * See if we belong to the destination multicast group on the
568 * arrival interface.
569 */
570 IN_MULTI_LOCK();
571 IN_LOOKUP_MULTI(ip->ip_dst, m->m_pkthdr.rcvif, inm);
572 IN_MULTI_UNLOCK();
573 if (inm == NULL) {
574 ipstat.ips_notmember++;
575 m_freem(m);
576 return;
577 }
578 goto ours;
579 }
580 if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST)
581 goto ours;
582 if (ip->ip_dst.s_addr == INADDR_ANY)
583 goto ours;
584
585 /*
586 * FAITH(Firewall Aided Internet Translator)
587 */
588 if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
589 if (ip_keepfaith) {
590 if (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_ICMP)
591 goto ours;
592 }
593 m_freem(m);
594 return;
595 }
596
597 /*
598 * Not for us; forward if possible and desirable.
599 */
600 if (ipforwarding == 0) {
601 ipstat.ips_cantforward++;
602 m_freem(m);
603 } else {
604#ifdef IPSEC
605 if (ip_ipsec_fwd(m))
606 goto bad;
607#endif /* IPSEC */
608 ip_forward(m, dchg);
609 }
610 return;
611
612ours:
613#ifdef IPSTEALTH
614 /*
615 * IPSTEALTH: Process non-routing options only
616 * if the packet is destined for us.
617 */
618 if (ipstealth && hlen > sizeof (struct ip) &&
619 ip_dooptions(m, 1))
620 return;
621#endif /* IPSTEALTH */
622
623 /* Count the packet in the ip address stats */
624 if (ia != NULL) {
625 ia->ia_ifa.if_ipackets++;
626 ia->ia_ifa.if_ibytes += m->m_pkthdr.len;
627 }
628
629 /*
630 * Attempt reassembly; if it succeeds, proceed.
631 * ip_reass() will return a different mbuf.
632 */
633 if (ip->ip_off & (IP_MF | IP_OFFMASK)) {
634 m = ip_reass(m);
635 if (m == NULL)
636 return;
637 ip = mtod(m, struct ip *);
638 /* Get the header length of the reassembled packet */
639 hlen = ip->ip_hl << 2;
640 }
641
642 /*
643 * Further protocols expect the packet length to be w/o the
644 * IP header.
645 */
646 ip->ip_len -= hlen;
647
648#ifdef IPSEC
649 /*
650 * enforce IPsec policy checking if we are seeing last header.
651 * note that we do not visit this with protocols with pcb layer
652 * code - like udp/tcp/raw ip.
653 */
654 if (ip_ipsec_input(m))
655 goto bad;
656#endif /* IPSEC */
657
658 /*
659 * Switch out to protocol's input routine.
660 */
661 ipstat.ips_delivered++;
662
663 (*inetsw[ip_protox[ip->ip_p]].pr_input)(m, hlen);
664 return;
665bad:
666 m_freem(m);
667}
668
669/*
670 * After maxnipq has been updated, propagate the change to UMA. The UMA zone
671 * max has slightly different semantics than the sysctl, for historical
672 * reasons.
673 */
674static void
675maxnipq_update(void)
676{
677
678 /*
679 * -1 for unlimited allocation.
680 */
681 if (maxnipq < 0)
682 uma_zone_set_max(ipq_zone, 0);
683 /*
684 * Positive number for specific bound.
685 */
686 if (maxnipq > 0)
687 uma_zone_set_max(ipq_zone, maxnipq);
688 /*
689 * Zero specifies no further fragment queue allocation -- set the
690 * bound very low, but rely on implementation elsewhere to actually
691 * prevent allocation and reclaim current queues.
692 */
693 if (maxnipq == 0)
694 uma_zone_set_max(ipq_zone, 1);
695}
696
697static void
698ipq_zone_change(void *tag)
699{
700
701 if (maxnipq > 0 && maxnipq < (nmbclusters / 32)) {
702 maxnipq = nmbclusters / 32;
703 maxnipq_update();
704 }
705}
706
707static int
708sysctl_maxnipq(SYSCTL_HANDLER_ARGS)
709{
710 int error, i;
711
712 i = maxnipq;
713 error = sysctl_handle_int(oidp, &i, 0, req);
714 if (error || !req->newptr)
715 return (error);
716
717 /*
718 * XXXRW: Might be a good idea to sanity check the argument and place
719 * an extreme upper bound.
720 */
721 if (i < -1)
722 return (EINVAL);
723 maxnipq = i;
724 maxnipq_update();
725 return (0);
726}
727
728SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLTYPE_INT|CTLFLAG_RW,
729 NULL, 0, sysctl_maxnipq, "I",
730 "Maximum number of IPv4 fragment reassembly queue entries");
731
732/*
733 * Take incoming datagram fragment and try to reassemble it into
734 * whole datagram. If the argument is the first fragment or one
735 * in between the function will return NULL and store the mbuf
736 * in the fragment chain. If the argument is the last fragment
737 * the packet will be reassembled and the pointer to the new
738 * mbuf returned for further processing. Only m_tags attached
739 * to the first packet/fragment are preserved.
740 * The IP header is *NOT* adjusted out of iplen.
741 */
742struct mbuf *
743ip_reass(struct mbuf *m)
744{
745 struct ip *ip;
746 struct mbuf *p, *q, *nq, *t;
747 struct ipq *fp = NULL;
748 struct ipqhead *head;
749 int i, hlen, next;
750 u_int8_t ecn, ecn0;
751 u_short hash;
752
753 /* If maxnipq or maxfragsperpacket are 0, never accept fragments. */
754 if (maxnipq == 0 || maxfragsperpacket == 0) {
755 ipstat.ips_fragments++;
756 ipstat.ips_fragdropped++;
757 m_freem(m);
758 return (NULL);
759 }
760
761 ip = mtod(m, struct ip *);
762 hlen = ip->ip_hl << 2;
763
764 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
765 head = &ipq[hash];
766 IPQ_LOCK();
767
768 /*
769 * Look for queue of fragments
770 * of this datagram.
771 */
772 TAILQ_FOREACH(fp, head, ipq_list)
773 if (ip->ip_id == fp->ipq_id &&
774 ip->ip_src.s_addr == fp->ipq_src.s_addr &&
775 ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
776#ifdef MAC
777 mac_fragment_match(m, fp) &&
778#endif
779 ip->ip_p == fp->ipq_p)
780 goto found;
781
782 fp = NULL;
783
784 /*
785 * Attempt to trim the number of allocated fragment queues if it
786 * exceeds the administrative limit.
787 */
788 if ((nipq > maxnipq) && (maxnipq > 0)) {
789 /*
790 * drop something from the tail of the current queue
791 * before proceeding further
792 */
793 struct ipq *q = TAILQ_LAST(head, ipqhead);
794 if (q == NULL) { /* gak */
795 for (i = 0; i < IPREASS_NHASH; i++) {
796 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
797 if (r) {
798 ipstat.ips_fragtimeout += r->ipq_nfrags;
799 ip_freef(&ipq[i], r);
800 break;
801 }
802 }
803 } else {
804 ipstat.ips_fragtimeout += q->ipq_nfrags;
805 ip_freef(head, q);
806 }
807 }
808
809found:
810 /*
811 * Adjust ip_len to not reflect header,
812 * convert offset of this to bytes.
813 */
814 ip->ip_len -= hlen;
815 if (ip->ip_off & IP_MF) {
816 /*
817 * Make sure that fragments have a data length
818 * that's a non-zero multiple of 8 bytes.
819 */
820 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
821 ipstat.ips_toosmall++; /* XXX */
822 goto dropfrag;
823 }
824 m->m_flags |= M_FRAG;
825 } else
826 m->m_flags &= ~M_FRAG;
827 ip->ip_off <<= 3;
828
829
830 /*
831 * Attempt reassembly; if it succeeds, proceed.
832 * ip_reass() will return a different mbuf.
833 */
834 ipstat.ips_fragments++;
835 m->m_pkthdr.header = ip;
836
837 /* Previous ip_reass() started here. */
838 /*
839 * Presence of header sizes in mbufs
840 * would confuse code below.
841 */
842 m->m_data += hlen;
843 m->m_len -= hlen;
844
845 /*
846 * If first fragment to arrive, create a reassembly queue.
847 */
848 if (fp == NULL) {
849 fp = uma_zalloc(ipq_zone, M_NOWAIT);
850 if (fp == NULL)
851 goto dropfrag;
852#ifdef MAC
853 if (mac_init_ipq(fp, M_NOWAIT) != 0) {
854 uma_zfree(ipq_zone, fp);
855 fp = NULL;
856 goto dropfrag;
857 }
858 mac_create_ipq(m, fp);
859#endif
860 TAILQ_INSERT_HEAD(head, fp, ipq_list);
861 nipq++;
862 fp->ipq_nfrags = 1;
863 fp->ipq_ttl = IPFRAGTTL;
864 fp->ipq_p = ip->ip_p;
865 fp->ipq_id = ip->ip_id;
866 fp->ipq_src = ip->ip_src;
867 fp->ipq_dst = ip->ip_dst;
868 fp->ipq_frags = m;
869 m->m_nextpkt = NULL;
870 goto done;
871 } else {
872 fp->ipq_nfrags++;
873#ifdef MAC
874 mac_update_ipq(m, fp);
875#endif
876 }
877
878#define GETIP(m) ((struct ip*)((m)->m_pkthdr.header))
879
880 /*
881 * Handle ECN by comparing this segment with the first one;
882 * if CE is set, do not lose CE.
883 * drop if CE and not-ECT are mixed for the same packet.
884 */
885 ecn = ip->ip_tos & IPTOS_ECN_MASK;
886 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
887 if (ecn == IPTOS_ECN_CE) {
888 if (ecn0 == IPTOS_ECN_NOTECT)
889 goto dropfrag;
890 if (ecn0 != IPTOS_ECN_CE)
891 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
892 }
893 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT)
894 goto dropfrag;
895
896 /*
897 * Find a segment which begins after this one does.
898 */
899 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt)
900 if (GETIP(q)->ip_off > ip->ip_off)
901 break;
902
903 /*
904 * If there is a preceding segment, it may provide some of
905 * our data already. If so, drop the data from the incoming
906 * segment. If it provides all of our data, drop us, otherwise
907 * stick new segment in the proper place.
908 *
909 * If some of the data is dropped from the the preceding
910 * segment, then it's checksum is invalidated.
911 */
912 if (p) {
913 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
914 if (i > 0) {
915 if (i >= ip->ip_len)
916 goto dropfrag;
917 m_adj(m, i);
918 m->m_pkthdr.csum_flags = 0;
919 ip->ip_off += i;
920 ip->ip_len -= i;
921 }
922 m->m_nextpkt = p->m_nextpkt;
923 p->m_nextpkt = m;
924 } else {
925 m->m_nextpkt = fp->ipq_frags;
926 fp->ipq_frags = m;
927 }
928
929 /*
930 * While we overlap succeeding segments trim them or,
931 * if they are completely covered, dequeue them.
932 */
933 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
934 q = nq) {
935 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
936 if (i < GETIP(q)->ip_len) {
937 GETIP(q)->ip_len -= i;
938 GETIP(q)->ip_off += i;
939 m_adj(q, i);
940 q->m_pkthdr.csum_flags = 0;
941 break;
942 }
943 nq = q->m_nextpkt;
944 m->m_nextpkt = nq;
945 ipstat.ips_fragdropped++;
946 fp->ipq_nfrags--;
947 m_freem(q);
948 }
949
950 /*
951 * Check for complete reassembly and perform frag per packet
952 * limiting.
953 *
954 * Frag limiting is performed here so that the nth frag has
955 * a chance to complete the packet before we drop the packet.
956 * As a result, n+1 frags are actually allowed per packet, but
957 * only n will ever be stored. (n = maxfragsperpacket.)
958 *
959 */
960 next = 0;
961 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
962 if (GETIP(q)->ip_off != next) {
963 if (fp->ipq_nfrags > maxfragsperpacket) {
964 ipstat.ips_fragdropped += fp->ipq_nfrags;
965 ip_freef(head, fp);
966 }
967 goto done;
968 }
969 next += GETIP(q)->ip_len;
970 }
971 /* Make sure the last packet didn't have the IP_MF flag */
972 if (p->m_flags & M_FRAG) {
973 if (fp->ipq_nfrags > maxfragsperpacket) {
974 ipstat.ips_fragdropped += fp->ipq_nfrags;
975 ip_freef(head, fp);
976 }
977 goto done;
978 }
979
980 /*
981 * Reassembly is complete. Make sure the packet is a sane size.
982 */
983 q = fp->ipq_frags;
984 ip = GETIP(q);
985 if (next + (ip->ip_hl << 2) > IP_MAXPACKET) {
986 ipstat.ips_toolong++;
987 ipstat.ips_fragdropped += fp->ipq_nfrags;
988 ip_freef(head, fp);
989 goto done;
990 }
991
992 /*
993 * Concatenate fragments.
994 */
995 m = q;
996 t = m->m_next;
997 m->m_next = NULL;
998 m_cat(m, t);
999 nq = q->m_nextpkt;
1000 q->m_nextpkt = NULL;
1001 for (q = nq; q != NULL; q = nq) {
1002 nq = q->m_nextpkt;
1003 q->m_nextpkt = NULL;
1004 m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags;
1005 m->m_pkthdr.csum_data += q->m_pkthdr.csum_data;
1006 m_cat(m, q);
1007 }
1008 /*
1009 * In order to do checksumming faster we do 'end-around carry' here
1010 * (and not in for{} loop), though it implies we are not going to
1011 * reassemble more than 64k fragments.
1012 */
1013 m->m_pkthdr.csum_data =
1014 (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16);
1015#ifdef MAC
1016 mac_create_datagram_from_ipq(fp, m);
1017 mac_destroy_ipq(fp);
1018#endif
1019
1020 /*
1021 * Create header for new ip packet by modifying header of first
1022 * packet; dequeue and discard fragment reassembly header.
1023 * Make header visible.
1024 */
1025 ip->ip_len = (ip->ip_hl << 2) + next;
1026 ip->ip_src = fp->ipq_src;
1027 ip->ip_dst = fp->ipq_dst;
1028 TAILQ_REMOVE(head, fp, ipq_list);
1029 nipq--;
1030 uma_zfree(ipq_zone, fp);
1031 m->m_len += (ip->ip_hl << 2);
1032 m->m_data -= (ip->ip_hl << 2);
1033 /* some debugging cruft by sklower, below, will go away soon */
1034 if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */
1035 m_fixhdr(m);
1036 ipstat.ips_reassembled++;
1037 IPQ_UNLOCK();
1038 return (m);
1039
1040dropfrag:
1041 ipstat.ips_fragdropped++;
1042 if (fp != NULL)
1043 fp->ipq_nfrags--;
1044 m_freem(m);
1045done:
1046 IPQ_UNLOCK();
1047 return (NULL);
1048
1049#undef GETIP
1050}
1051
1052/*
1053 * Free a fragment reassembly header and all
1054 * associated datagrams.
1055 */
1056static void
1057ip_freef(struct ipqhead *fhp, struct ipq *fp)
1058{
1059 struct mbuf *q;
1060
1061 IPQ_LOCK_ASSERT();
1062
1063 while (fp->ipq_frags) {
1064 q = fp->ipq_frags;
1065 fp->ipq_frags = q->m_nextpkt;
1066 m_freem(q);
1067 }
1068 TAILQ_REMOVE(fhp, fp, ipq_list);
1069 uma_zfree(ipq_zone, fp);
1070 nipq--;
1071}
1072
1073/*
1074 * IP timer processing;
1075 * if a timer expires on a reassembly
1076 * queue, discard it.
1077 */
1078void
1079ip_slowtimo(void)
1080{
1081 struct ipq *fp;
1082 int i;
1083
1084 IPQ_LOCK();
1085 for (i = 0; i < IPREASS_NHASH; i++) {
1086 for(fp = TAILQ_FIRST(&ipq[i]); fp;) {
1087 struct ipq *fpp;
1088
1089 fpp = fp;
1090 fp = TAILQ_NEXT(fp, ipq_list);
1091 if(--fpp->ipq_ttl == 0) {
1092 ipstat.ips_fragtimeout += fpp->ipq_nfrags;
1093 ip_freef(&ipq[i], fpp);
1094 }
1095 }
1096 }
1097 /*
1098 * If we are over the maximum number of fragments
1099 * (due to the limit being lowered), drain off
1100 * enough to get down to the new limit.
1101 */
1102 if (maxnipq >= 0 && nipq > maxnipq) {
1103 for (i = 0; i < IPREASS_NHASH; i++) {
1104 while (nipq > maxnipq && !TAILQ_EMPTY(&ipq[i])) {
1105 ipstat.ips_fragdropped +=
1106 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1107 ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1108 }
1109 }
1110 }
1111 IPQ_UNLOCK();
1112}
1113
1114/*
1115 * Drain off all datagram fragments.
1116 */
1117void
1118ip_drain(void)
1119{
1120 int i;
1121
1122 IPQ_LOCK();
1123 for (i = 0; i < IPREASS_NHASH; i++) {
1124 while(!TAILQ_EMPTY(&ipq[i])) {
1125 ipstat.ips_fragdropped +=
1126 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
1127 ip_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
1128 }
1129 }
1130 IPQ_UNLOCK();
1131 in_rtqdrain();
1132}
1133
1134/*
1135 * The protocol to be inserted into ip_protox[] must be already registered
1136 * in inetsw[], either statically or through pf_proto_register().
1137 */
1138int
1139ipproto_register(u_char ipproto)
1140{
1141 struct protosw *pr;
1142
1143 /* Sanity checks. */
1144 if (ipproto == 0)
1145 return (EPROTONOSUPPORT);
1146
1147 /*
1148 * The protocol slot must not be occupied by another protocol
1149 * already. An index pointing to IPPROTO_RAW is unused.
1150 */
1151 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1152 if (pr == NULL)
1153 return (EPFNOSUPPORT);
1154 if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */
1155 return (EEXIST);
1156
1157 /* Find the protocol position in inetsw[] and set the index. */
1158 for (pr = inetdomain.dom_protosw;
1159 pr < inetdomain.dom_protoswNPROTOSW; pr++) {
1160 if (pr->pr_domain->dom_family == PF_INET &&
1161 pr->pr_protocol && pr->pr_protocol == ipproto) {
1162 /* Be careful to only index valid IP protocols. */
1163 if (pr->pr_protocol < IPPROTO_MAX) {
1164 ip_protox[pr->pr_protocol] = pr - inetsw;
1165 return (0);
1166 } else
1167 return (EINVAL);
1168 }
1169 }
1170 return (EPROTONOSUPPORT);
1171}
1172
1173int
1174ipproto_unregister(u_char ipproto)
1175{
1176 struct protosw *pr;
1177
1178 /* Sanity checks. */
1179 if (ipproto == 0)
1180 return (EPROTONOSUPPORT);
1181
1182 /* Check if the protocol was indeed registered. */
1183 pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW);
1184 if (pr == NULL)
1185 return (EPFNOSUPPORT);
1186 if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */
1187 return (ENOENT);
1188
1189 /* Reset the protocol slot to IPPROTO_RAW. */
1190 ip_protox[ipproto] = pr - inetsw;
1191 return (0);
1192}
1193
1194/*
1195 * Given address of next destination (final or next hop),
1196 * return internet address info of interface to be used to get there.
1197 */
1198struct in_ifaddr *
1199ip_rtaddr(struct in_addr dst)
1200{
1201 struct route sro;
1202 struct sockaddr_in *sin;
1203 struct in_ifaddr *ifa;
1204
1205 bzero(&sro, sizeof(sro));
1206 sin = (struct sockaddr_in *)&sro.ro_dst;
1207 sin->sin_family = AF_INET;
1208 sin->sin_len = sizeof(*sin);
1209 sin->sin_addr = dst;
1210 rtalloc_ign(&sro, RTF_CLONING);
1211
1212 if (sro.ro_rt == NULL)
1213 return (NULL);
1214
1215 ifa = ifatoia(sro.ro_rt->rt_ifa);
1216 RTFREE(sro.ro_rt);
1217 return (ifa);
1218}
1219
1220u_char inetctlerrmap[PRC_NCMDS] = {
1221 0, 0, 0, 0,
1222 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH,
1223 EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED,
1224 EMSGSIZE, EHOSTUNREACH, 0, 0,
1225 0, 0, EHOSTUNREACH, 0,
1226 ENOPROTOOPT, ECONNREFUSED
1227};
1228
1229/*
1230 * Forward a packet. If some error occurs return the sender
1231 * an icmp packet. Note we can't always generate a meaningful
1232 * icmp message because icmp doesn't have a large enough repertoire
1233 * of codes and types.
1234 *
1235 * If not forwarding, just drop the packet. This could be confusing
1236 * if ipforwarding was zero but some routing protocol was advancing
1237 * us as a gateway to somewhere. However, we must let the routing
1238 * protocol deal with that.
1239 *
1240 * The srcrt parameter indicates whether the packet is being forwarded
1241 * via a source route.
1242 */
1243void
1244ip_forward(struct mbuf *m, int srcrt)
1245{
1246 struct ip *ip = mtod(m, struct ip *);
1247 struct in_ifaddr *ia = NULL;
1248 struct mbuf *mcopy;
1249 struct in_addr dest;
1250 int error, type = 0, code = 0, mtu = 0;
1251
1252 if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) {
1253 ipstat.ips_cantforward++;
1254 m_freem(m);
1255 return;
1256 }
1257#ifdef IPSTEALTH
1258 if (!ipstealth) {
1259#endif
1260 if (ip->ip_ttl <= IPTTLDEC) {
1261 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
1262 0, 0);
1263 return;
1264 }
1265#ifdef IPSTEALTH
1266 }
1267#endif
1268
1269 if (!srcrt && (ia = ip_rtaddr(ip->ip_dst)) == NULL) {
1270 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
1271 return;
1272 }
1273
1274 /*
1275 * Save the IP header and at most 8 bytes of the payload,
1276 * in case we need to generate an ICMP message to the src.
1277 *
1278 * XXX this can be optimized a lot by saving the data in a local
1279 * buffer on the stack (72 bytes at most), and only allocating the
1280 * mbuf if really necessary. The vast majority of the packets
1281 * are forwarded without having to send an ICMP back (either
1282 * because unnecessary, or because rate limited), so we are
1283 * really we are wasting a lot of work here.
1284 *
1285 * We don't use m_copy() because it might return a reference
1286 * to a shared cluster. Both this function and ip_output()
1287 * assume exclusive access to the IP header in `m', so any
1288 * data in a cluster may change before we reach icmp_error().
1289 */
1290 MGETHDR(mcopy, M_DONTWAIT, m->m_type);
1291 if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_DONTWAIT)) {
1292 /*
1293 * It's probably ok if the pkthdr dup fails (because
1294 * the deep copy of the tag chain failed), but for now
1295 * be conservative and just discard the copy since
1296 * code below may some day want the tags.
1297 */
1298 m_free(mcopy);
1299 mcopy = NULL;
1300 }
1301 if (mcopy != NULL) {
1302 mcopy->m_len = min(ip->ip_len, M_TRAILINGSPACE(mcopy));
1303 mcopy->m_pkthdr.len = mcopy->m_len;
1304 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
1305 }
1306
1307#ifdef IPSTEALTH
1308 if (!ipstealth) {
1309#endif
1310 ip->ip_ttl -= IPTTLDEC;
1311#ifdef IPSTEALTH
1312 }
1313#endif
1314
1315 /*
1316 * If forwarding packet using same interface that it came in on,
1317 * perhaps should send a redirect to sender to shortcut a hop.
1318 * Only send redirect if source is sending directly to us,
1319 * and if packet was not source routed (or has any options).
1320 * Also, don't send redirect if forwarding using a default route
1321 * or a route modified by a redirect.
1322 */
1323 dest.s_addr = 0;
1324 if (!srcrt && ipsendredirects && ia->ia_ifp == m->m_pkthdr.rcvif) {
1325 struct sockaddr_in *sin;
1326 struct route ro;
1327 struct rtentry *rt;
1328
1329 bzero(&ro, sizeof(ro));
1330 sin = (struct sockaddr_in *)&ro.ro_dst;
1331 sin->sin_family = AF_INET;
1332 sin->sin_len = sizeof(*sin);
1333 sin->sin_addr = ip->ip_dst;
1334 rtalloc_ign(&ro, RTF_CLONING);
1335
1336 rt = ro.ro_rt;
1337
1338 if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
1339 satosin(rt_key(rt))->sin_addr.s_addr != 0) {
1340#define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa))
1341 u_long src = ntohl(ip->ip_src.s_addr);
1342
1343 if (RTA(rt) &&
1344 (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) {
1345 if (rt->rt_flags & RTF_GATEWAY)
1346 dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr;
1347 else
1348 dest.s_addr = ip->ip_dst.s_addr;
1349 /* Router requirements says to only send host redirects */
1350 type = ICMP_REDIRECT;
1351 code = ICMP_REDIRECT_HOST;
1352 }
1353 }
1354 if (rt)
1355 RTFREE(rt);
1356 }
1357
1358 error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL);
1359 if (error)
1360 ipstat.ips_cantforward++;
1361 else {
1362 ipstat.ips_forward++;
1363 if (type)
1364 ipstat.ips_redirectsent++;
1365 else {
1366 if (mcopy)
1367 m_freem(mcopy);
1368 return;
1369 }
1370 }
1371 if (mcopy == NULL)
1372 return;
1373
1374 switch (error) {
1375
1376 case 0: /* forwarded, but need redirect */
1377 /* type, code set above */
1378 break;
1379
1380 case ENETUNREACH: /* shouldn't happen, checked above */
1381 case EHOSTUNREACH:
1382 case ENETDOWN:
1383 case EHOSTDOWN:
1384 default:
1385 type = ICMP_UNREACH;
1386 code = ICMP_UNREACH_HOST;
1387 break;
1388
1389 case EMSGSIZE:
1390 type = ICMP_UNREACH;
1391 code = ICMP_UNREACH_NEEDFRAG;
1392
1393#ifdef IPSEC
1394 mtu = ip_ipsec_mtu(m);
1395#endif /* IPSEC */
1396 /*
1397 * If the MTU wasn't set before use the interface mtu or
1398 * fall back to the next smaller mtu step compared to the
1399 * current packet size.
1400 */
1401 if (mtu == 0) {
1402 if (ia != NULL)
1403 mtu = ia->ia_ifp->if_mtu;
1404 else
1405 mtu = ip_next_mtu(ip->ip_len, 0);
1406 }
1407 ipstat.ips_cantfrag++;
1408 break;
1409
1410 case ENOBUFS:
1411 /*
1412 * A router should not generate ICMP_SOURCEQUENCH as
1413 * required in RFC1812 Requirements for IP Version 4 Routers.
1414 * Source quench could be a big problem under DoS attacks,
1415 * or if the underlying interface is rate-limited.
1416 * Those who need source quench packets may re-enable them
1417 * via the net.inet.ip.sendsourcequench sysctl.
1418 */
1419 if (ip_sendsourcequench == 0) {
1420 m_freem(mcopy);
1421 return;
1422 } else {
1423 type = ICMP_SOURCEQUENCH;
1424 code = 0;
1425 }
1426 break;
1427
1428 case EACCES: /* ipfw denied packet */
1429 m_freem(mcopy);
1430 return;
1431 }
1432 icmp_error(mcopy, type, code, dest.s_addr, mtu);
1433}
1434
1435void
1436ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
1437 struct mbuf *m)
1438{
1439 if (inp->inp_socket->so_options & (SO_BINTIME | SO_TIMESTAMP)) {
1440 struct bintime bt;
1441
1442 bintime(&bt);
1443 if (inp->inp_socket->so_options & SO_BINTIME) {
1444 *mp = sbcreatecontrol((caddr_t) &bt, sizeof(bt),
1445 SCM_BINTIME, SOL_SOCKET);
1446 if (*mp)
1447 mp = &(*mp)->m_next;
1448 }
1449 if (inp->inp_socket->so_options & SO_TIMESTAMP) {
1450 struct timeval tv;
1451
1452 bintime2timeval(&bt, &tv);
1453 *mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
1454 SCM_TIMESTAMP, SOL_SOCKET);
1455 if (*mp)
1456 mp = &(*mp)->m_next;
1457 }
1458 }
1459 if (inp->inp_flags & INP_RECVDSTADDR) {
1460 *mp = sbcreatecontrol((caddr_t) &ip->ip_dst,
1461 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP);
1462 if (*mp)
1463 mp = &(*mp)->m_next;
1464 }
1465 if (inp->inp_flags & INP_RECVTTL) {
1466 *mp = sbcreatecontrol((caddr_t) &ip->ip_ttl,
1467 sizeof(u_char), IP_RECVTTL, IPPROTO_IP);
1468 if (*mp)
1469 mp = &(*mp)->m_next;
1470 }
1471#ifdef notyet
1472 /* XXX
1473 * Moving these out of udp_input() made them even more broken
1474 * than they already were.
1475 */
1476 /* options were tossed already */
1477 if (inp->inp_flags & INP_RECVOPTS) {
1478 *mp = sbcreatecontrol((caddr_t) opts_deleted_above,
1479 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP);
1480 if (*mp)
1481 mp = &(*mp)->m_next;
1482 }
1483 /* ip_srcroute doesn't do what we want here, need to fix */
1484 if (inp->inp_flags & INP_RECVRETOPTS) {
1485 *mp = sbcreatecontrol((caddr_t) ip_srcroute(m),
1486 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP);
1487 if (*mp)
1488 mp = &(*mp)->m_next;
1489 }
1490#endif
1491 if (inp->inp_flags & INP_RECVIF) {
1492 struct ifnet *ifp;
1493 struct sdlbuf {
1494 struct sockaddr_dl sdl;
1495 u_char pad[32];
1496 } sdlbuf;
1497 struct sockaddr_dl *sdp;
1498 struct sockaddr_dl *sdl2 = &sdlbuf.sdl;
1499
1500 if (((ifp = m->m_pkthdr.rcvif))
1501 && ( ifp->if_index && (ifp->if_index <= if_index))) {
1502 sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr;
1503 /*
1504 * Change our mind and don't try copy.
1505 */
1506 if ((sdp->sdl_family != AF_LINK)
1507 || (sdp->sdl_len > sizeof(sdlbuf))) {
1508 goto makedummy;
1509 }
1510 bcopy(sdp, sdl2, sdp->sdl_len);
1511 } else {
1512makedummy:
1513 sdl2->sdl_len
1514 = offsetof(struct sockaddr_dl, sdl_data[0]);
1515 sdl2->sdl_family = AF_LINK;
1516 sdl2->sdl_index = 0;
1517 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
1518 }
1519 *mp = sbcreatecontrol((caddr_t) sdl2, sdl2->sdl_len,
1520 IP_RECVIF, IPPROTO_IP);
1521 if (*mp)
1522 mp = &(*mp)->m_next;
1523 }
1524}
1525
1526/*
1527 * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the
1528 * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on
1529 * locking. This code remains in ip_input.c as ip_mroute.c is optionally
1530 * compiled.
1531 */
1532static int ip_rsvp_on;
1533struct socket *ip_rsvpd;
1534int
1535ip_rsvp_init(struct socket *so)
1536{
1537 if (so->so_type != SOCK_RAW ||
1538 so->so_proto->pr_protocol != IPPROTO_RSVP)
1539 return EOPNOTSUPP;
1540
1541 if (ip_rsvpd != NULL)
1542 return EADDRINUSE;
1543
1544 ip_rsvpd = so;
1545 /*
1546 * This may seem silly, but we need to be sure we don't over-increment
1547 * the RSVP counter, in case something slips up.
1548 */
1549 if (!ip_rsvp_on) {
1550 ip_rsvp_on = 1;
1551 rsvp_on++;
1552 }
1553
1554 return 0;
1555}
1556
1557int
1558ip_rsvp_done(void)
1559{
1560 ip_rsvpd = NULL;
1561 /*
1562 * This may seem silly, but we need to be sure we don't over-decrement
1563 * the RSVP counter, in case something slips up.
1564 */
1565 if (ip_rsvp_on) {
1566 ip_rsvp_on = 0;
1567 rsvp_on--;
1568 }
1569 return 0;
1570}
1571
1572void
1573rsvp_input(struct mbuf *m, int off) /* XXX must fixup manually */
1574{
1575 if (rsvp_input_p) { /* call the real one if loaded */
1576 rsvp_input_p(m, off);
1577 return;
1578 }
1579
1580 /* Can still get packets with rsvp_on = 0 if there is a local member
1581 * of the group to which the RSVP packet is addressed. But in this
1582 * case we want to throw the packet away.
1583 */
1584
1585 if (!rsvp_on) {
1586 m_freem(m);
1587 return;
1588 }
1589
1590 if (ip_rsvpd != NULL) {
1591 rip_input(m, off);
1592 return;
1593 }
1594 /* Drop the packet */
1595 m_freem(m);
1596}
|