1 /*
2 * Copyright (c) 2000-2021 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * Copyright (c) 1982, 1986, 1988, 1993
30 * The Regents of the University of California. All rights reserved.
31 *
32 * Redistribution and use in source and binary forms, with or without
33 * modification, are permitted provided that the following conditions
34 * are met:
35 * 1. Redistributions of source code must retain the above copyright
36 * notice, this list of conditions and the following disclaimer.
37 * 2. Redistributions in binary form must reproduce the above copyright
38 * notice, this list of conditions and the following disclaimer in the
39 * documentation and/or other materials provided with the distribution.
40 * 3. All advertising materials mentioning features or use of this software
41 * must display the following acknowledgement:
42 * This product includes software developed by the University of
43 * California, Berkeley and its contributors.
44 * 4. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)ip_input.c 8.2 (Berkeley) 1/4/94
61 */
62 /*
63 * NOTICE: This file was modified by SPARTA, Inc. in 2007 to introduce
64 * support for mandatory and extensible security protections. This notice
65 * is included in support of clause 2.2 (b) of the Apple Public License,
66 * Version 2.0.
67 */
68
69 #define _IP_VHL
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/mbuf.h>
74 #include <sys/malloc.h>
75 #include <sys/domain.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/time.h>
79 #include <sys/kernel.h>
80 #include <sys/syslog.h>
81 #include <sys/sysctl.h>
82 #include <sys/mcache.h>
83 #include <sys/socketvar.h>
84 #include <sys/kdebug.h>
85 #include <mach/mach_time.h>
86 #include <mach/sdt.h>
87
88 #include <machine/endian.h>
89 #include <dev/random/randomdev.h>
90
91 #include <kern/queue.h>
92 #include <kern/locks.h>
93 #include <libkern/OSAtomic.h>
94
95 #include <pexpert/pexpert.h>
96
97 #include <net/if.h>
98 #include <net/if_var.h>
99 #include <net/if_dl.h>
100 #include <net/route.h>
101 #include <net/kpi_protocol.h>
102 #include <net/ntstat.h>
103 #include <net/dlil.h>
104 #include <net/classq/classq.h>
105 #include <net/net_perf.h>
106 #include <net/init.h>
107 #if PF
108 #include <net/pfvar.h>
109 #endif /* PF */
110 #include <net/if_ports_used.h>
111
112 #include <netinet/in.h>
113 #include <netinet/in_systm.h>
114 #include <netinet/in_var.h>
115 #include <netinet/in_arp.h>
116 #include <netinet/ip.h>
117 #include <netinet/in_pcb.h>
118 #include <netinet/ip_var.h>
119 #include <netinet/ip_icmp.h>
120 #include <netinet/kpi_ipfilter_var.h>
121 #include <netinet/udp.h>
122 #include <netinet/udp_var.h>
123 #include <netinet/bootp.h>
124
125 #if DUMMYNET
126 #include <netinet/ip_dummynet.h>
127 #endif /* DUMMYNET */
128
129 #if IPSEC
130 #include <netinet6/ipsec.h>
131 #include <netkey/key.h>
132 #endif /* IPSEC */
133
134 #include <os/log.h>
135
136 #define DBG_LAYER_BEG NETDBG_CODE(DBG_NETIP, 0)
137 #define DBG_LAYER_END NETDBG_CODE(DBG_NETIP, 2)
138 #define DBG_FNC_IP_INPUT NETDBG_CODE(DBG_NETIP, (2 << 8))
139
140 #if IPSEC
141 extern int ipsec_bypass;
142 #endif /* IPSEC */
143
144 MBUFQ_HEAD(fq_head);
145
146 static int frag_timeout_run; /* frag timer is scheduled to run */
147 static void frag_timeout(void *);
148 static void frag_sched_timeout(void);
149
150 static struct ipq *ipq_alloc(int);
151 static void ipq_free(struct ipq *);
152 static void ipq_updateparams(void);
153 static void ip_input_second_pass(struct mbuf *, struct ifnet *,
154 int, int, struct ip_fw_in_args *);
155
156 static LCK_GRP_DECLARE(ipqlock_grp, "ipqlock");
157 static LCK_MTX_DECLARE(ipqlock, &ipqlock_grp);
158
159
160 /* Packet reassembly stuff */
161 #define IPREASS_NHASH_LOG2 6
162 #define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2)
163 #define IPREASS_HMASK (IPREASS_NHASH - 1)
164 #define IPREASS_HASH(x, y) \
165 (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
166
167 /* IP fragment reassembly queues (protected by ipqlock) */
168 static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH]; /* ip reassembly queues */
169 static int maxnipq; /* max packets in reass queues */
170 static u_int32_t maxfragsperpacket; /* max frags/packet in reass queues */
171 static u_int32_t nipq; /* # of packets in reass queues */
172 static u_int32_t ipq_limit; /* ipq allocation limit */
173 static u_int32_t ipq_count; /* current # of allocated ipq's */
174
175 static int sysctl_ipforwarding SYSCTL_HANDLER_ARGS;
176 static int sysctl_maxnipq SYSCTL_HANDLER_ARGS;
177 static int sysctl_maxfragsperpacket SYSCTL_HANDLER_ARGS;
178
179 #if (DEBUG || DEVELOPMENT)
180 static int sysctl_reset_ip_input_stats SYSCTL_HANDLER_ARGS;
181 static int sysctl_ip_input_measure_bins SYSCTL_HANDLER_ARGS;
182 static int sysctl_ip_input_getperf SYSCTL_HANDLER_ARGS;
183 #endif /* (DEBUG || DEVELOPMENT) */
184
185 int ipforwarding = 0;
186 SYSCTL_PROC(_net_inet_ip, IPCTL_FORWARDING, forwarding,
187 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ipforwarding, 0,
188 sysctl_ipforwarding, "I", "Enable IP forwarding between interfaces");
189
190 static int ipsendredirects = 1; /* XXX */
191 SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect,
192 CTLFLAG_RW | CTLFLAG_LOCKED, &ipsendredirects, 0,
193 "Enable sending IP redirects");
194
195 int ip_defttl = IPDEFTTL;
196 SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW | CTLFLAG_LOCKED,
197 &ip_defttl, 0, "Maximum TTL on IP packets");
198
199 static int ip_dosourceroute = 0;
200 SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute,
201 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_dosourceroute, 0,
202 "Enable forwarding source routed IP packets");
203
204 static int ip_acceptsourceroute = 0;
205 SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
206 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_acceptsourceroute, 0,
207 "Enable accepting source routed IP packets");
208
209 static int ip_sendsourcequench = 0;
210 SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench,
211 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_sendsourcequench, 0,
212 "Enable the transmission of source quench packets");
213
214 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets,
215 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &maxnipq, 0, sysctl_maxnipq,
216 "I", "Maximum number of IPv4 fragment reassembly queue entries");
217
218 SYSCTL_UINT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD | CTLFLAG_LOCKED,
219 &nipq, 0, "Current number of IPv4 fragment reassembly queue entries");
220
221 SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragsperpacket,
222 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &maxfragsperpacket, 0,
223 sysctl_maxfragsperpacket, "I",
224 "Maximum number of IPv4 fragments allowed per packet");
225
226 static uint32_t ip_adj_clear_hwcksum = 0;
227 SYSCTL_UINT(_net_inet_ip, OID_AUTO, adj_clear_hwcksum,
228 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_adj_clear_hwcksum, 0,
229 "Invalidate hwcksum info when adjusting length");
230
231 static uint32_t ip_adj_partial_sum = 1;
232 SYSCTL_UINT(_net_inet_ip, OID_AUTO, adj_partial_sum,
233 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_adj_partial_sum, 0,
234 "Perform partial sum adjustment of trailing bytes at IP layer");
235
236 /*
237 * ip_checkinterface controls the receive side of the models for multihoming
238 * that are discussed in RFC 1122.
239 *
240 * ip_checkinterface values are:
241 * IP_CHECKINTERFACE_WEAK_ES:
242 * This corresponds to the Weak End-System model where incoming packets from
243 * any interface are accepted provided the destination address of the incoming packet
244 * is assigned to some interface.
245 *
246 * IP_CHECKINTERFACE_HYBRID_ES:
247 * The Hybrid End-System model use the Strong End-System for tunnel interfaces
248 * (ipsec and utun) and the weak End-System model for other interfaces families.
249 * This prevents a rogue middle box to probe for signs of TCP connections
250 * that use the tunnel interface.
251 *
252 * IP_CHECKINTERFACE_STRONG_ES:
253 * The Strong model model requires the packet arrived on an interface that
254 * is assigned the destination address of the packet.
255 *
256 * Since the routing table and transmit implementation do not implement the Strong ES model,
257 * setting this to a value different from IP_CHECKINTERFACE_WEAK_ES may lead to unexpected results.
258 *
259 * When forwarding is enabled, the system reverts to the Weak ES model as a router
260 * is expected by design to receive packets from several interfaces to the same address.
261 *
262 * XXX - ip_checkinterface currently must be set to IP_CHECKINTERFACE_WEAK_ES if you use ipnat
263 * to translate the destination address to another local interface.
264 *
265 * XXX - ip_checkinterface must be set to IP_CHECKINTERFACE_WEAK_ES if you add IP aliases
266 * to the loopback interface instead of the interface where the
267 * packets for those addresses are received.
268 */
269 #define IP_CHECKINTERFACE_WEAK_ES 0
270 #define IP_CHECKINTERFACE_HYBRID_ES 1
271 #define IP_CHECKINTERFACE_STRONG_ES 2
272
273 static int ip_checkinterface = IP_CHECKINTERFACE_HYBRID_ES;
274
275 static int sysctl_ip_checkinterface SYSCTL_HANDLER_ARGS;
276 SYSCTL_PROC(_net_inet_ip, OID_AUTO, check_interface,
277 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
278 0, 0, sysctl_ip_checkinterface, "I", "Verify packet arrives on correct interface");
279
280 #if (DEBUG || DEVELOPMENT)
281 #define IP_CHECK_IF_DEBUG 1
282 #else
283 #define IP_CHECK_IF_DEBUG 0
284 #endif /* (DEBUG || DEVELOPMENT) */
285 static int ip_checkinterface_debug = IP_CHECK_IF_DEBUG;
286 SYSCTL_INT(_net_inet_ip, OID_AUTO, checkinterface_debug, CTLFLAG_RW | CTLFLAG_LOCKED,
287 &ip_checkinterface_debug, IP_CHECK_IF_DEBUG, "");
288
289 static int ip_chaining = 1;
290 SYSCTL_INT(_net_inet_ip, OID_AUTO, rx_chaining, CTLFLAG_RW | CTLFLAG_LOCKED,
291 &ip_chaining, 1, "Do receive side ip address based chaining");
292
293 static int ip_chainsz = 6;
294 SYSCTL_INT(_net_inet_ip, OID_AUTO, rx_chainsz, CTLFLAG_RW | CTLFLAG_LOCKED,
295 &ip_chainsz, 1, "IP receive side max chaining");
296
297 #if (DEBUG || DEVELOPMENT)
298 static int ip_input_measure = 0;
299 SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf,
300 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
301 &ip_input_measure, 0, sysctl_reset_ip_input_stats, "I", "Do time measurement");
302
303 static uint64_t ip_input_measure_bins = 0;
304 SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf_bins,
305 CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &ip_input_measure_bins, 0,
306 sysctl_ip_input_measure_bins, "I",
307 "bins for chaining performance data histogram");
308
309 static net_perf_t net_perf;
310 SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf_data,
311 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED,
312 0, 0, sysctl_ip_input_getperf, "S,net_perf",
313 "IP input performance data (struct net_perf, net/net_perf.h)");
314 #endif /* (DEBUG || DEVELOPMENT) */
315
316 #if DIAGNOSTIC
317 static int ipprintfs = 0;
318 #endif
319
320 struct protosw *ip_protox[IPPROTO_MAX];
321
322 static LCK_GRP_DECLARE(in_ifaddr_rwlock_grp, "in_ifaddr_rwlock");
323 LCK_RW_DECLARE(in_ifaddr_rwlock, &in_ifaddr_rwlock_grp);
324
325 /* Protected by in_ifaddr_rwlock */
326 struct in_ifaddrhead in_ifaddrhead; /* first inet address */
327 struct in_ifaddrhashhead *in_ifaddrhashtbl; /* inet addr hash table */
328
329 #define INADDR_NHASH 61
330 static u_int32_t inaddr_nhash; /* hash table size */
331 static u_int32_t inaddr_hashp; /* next largest prime */
332
333 static int ip_getstat SYSCTL_HANDLER_ARGS;
334 struct ipstat ipstat;
335 SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats,
336 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED,
337 0, 0, ip_getstat, "S,ipstat",
338 "IP statistics (struct ipstat, netinet/ip_var.h)");
339
340 #if IPCTL_DEFMTU
341 SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW | CTLFLAG_LOCKED,
342 &ip_mtu, 0, "Default MTU");
343 #endif /* IPCTL_DEFMTU */
344
345 #if IPSTEALTH
346 static int ipstealth = 0;
347 SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW | CTLFLAG_LOCKED,
348 &ipstealth, 0, "");
349 #endif /* IPSTEALTH */
350
351 #if DUMMYNET
352 ip_dn_io_t *ip_dn_io_ptr;
353 #endif /* DUMMYNET */
354
355 SYSCTL_NODE(_net_inet_ip, OID_AUTO, linklocal,
356 CTLFLAG_RW | CTLFLAG_LOCKED, 0, "link local");
357
358 struct ip_linklocal_stat ip_linklocal_stat;
359 SYSCTL_STRUCT(_net_inet_ip_linklocal, OID_AUTO, stat,
360 CTLFLAG_RD | CTLFLAG_LOCKED, &ip_linklocal_stat, ip_linklocal_stat,
361 "Number of link local packets with TTL less than 255");
362
363 SYSCTL_NODE(_net_inet_ip_linklocal, OID_AUTO, in,
364 CTLFLAG_RW | CTLFLAG_LOCKED, 0, "link local input");
365
366 int ip_linklocal_in_allowbadttl = 1;
367 SYSCTL_INT(_net_inet_ip_linklocal_in, OID_AUTO, allowbadttl,
368 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_linklocal_in_allowbadttl, 0,
369 "Allow incoming link local packets with TTL less than 255");
370
371
372 /*
373 * We need to save the IP options in case a protocol wants to respond
374 * to an incoming packet over the same route if the packet got here
375 * using IP source routing. This allows connection establishment and
376 * maintenance when the remote end is on a network that is not known
377 * to us.
378 */
379 static int ip_nhops = 0;
380 static struct ip_srcrt {
381 struct in_addr dst; /* final destination */
382 char nop; /* one NOP to align */
383 char srcopt[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN and OFFSET */
384 struct in_addr route[MAX_IPOPTLEN / sizeof(struct in_addr)];
385 } ip_srcrt;
386
387 static void in_ifaddrhashtbl_init(void);
388 static void save_rte(u_char *, struct in_addr);
389 static int ip_dooptions(struct mbuf *, int, struct sockaddr_in *);
390 static void ip_forward(struct mbuf *, int, struct sockaddr_in *);
391 static void frag_freef(struct ipqhead *, struct ipq *);
392 static struct mbuf *ip_reass(struct mbuf *);
393 static void ip_fwd_route_copyout(struct ifnet *, struct route *);
394 static void ip_fwd_route_copyin(struct ifnet *, struct route *);
395 static inline u_short ip_cksum(struct mbuf *, int);
396
397 /*
398 * On platforms which require strict alignment (currently for anything but
399 * i386 or x86_64), check if the IP header pointer is 32-bit aligned; if not,
400 * copy the contents of the mbuf chain into a new chain, and free the original
401 * one. Create some head room in the first mbuf of the new chain, in case
402 * it's needed later on.
403 */
404 #if defined(__i386__) || defined(__x86_64__)
405 #define IP_HDR_ALIGNMENT_FIXUP(_m, _ifp, _action) do { } while (0)
406 #else /* !__i386__ && !__x86_64__ */
407 #define IP_HDR_ALIGNMENT_FIXUP(_m, _ifp, _action) do { \
408 if (!IP_HDR_ALIGNED_P(mtod(_m, caddr_t))) { \
409 struct mbuf *_n; \
410 struct ifnet *__ifp = (_ifp); \
411 atomic_add_64(&(__ifp)->if_alignerrs, 1); \
412 if (((_m)->m_flags & M_PKTHDR) && \
413 (_m)->m_pkthdr.pkt_hdr != NULL) \
414 (_m)->m_pkthdr.pkt_hdr = NULL; \
415 _n = m_defrag_offset(_m, max_linkhdr, M_NOWAIT); \
416 if (_n == NULL) { \
417 atomic_add_32(&ipstat.ips_toosmall, 1); \
418 m_freem(_m); \
419 (_m) = NULL; \
420 _action; \
421 } else { \
422 VERIFY(_n != (_m)); \
423 (_m) = _n; \
424 } \
425 } \
426 } while (0)
427 #endif /* !__i386__ && !__x86_64__ */
428
429
430 typedef enum ip_check_if_result {
431 IP_CHECK_IF_NONE = 0,
432 IP_CHECK_IF_OURS = 1,
433 IP_CHECK_IF_DROP = 2,
434 IP_CHECK_IF_FORWARD = 3
435 } ip_check_if_result_t;
436
437 static ip_check_if_result_t ip_input_check_interface(struct mbuf **, struct ip *, struct ifnet *);
438
439 /*
440 * GRE input handler function, settable via ip_gre_register_input() for PPTP.
441 */
442 static gre_input_func_t gre_input_func;
443
444 static void
ip_init_delayed(void)445 ip_init_delayed(void)
446 {
447 struct ifreq ifr;
448 int error;
449 struct sockaddr_in *sin;
450
451 bzero(&ifr, sizeof(ifr));
452 strlcpy(ifr.ifr_name, "lo0", sizeof(ifr.ifr_name));
453 sin = (struct sockaddr_in *)(void *)&ifr.ifr_addr;
454 sin->sin_len = sizeof(struct sockaddr_in);
455 sin->sin_family = AF_INET;
456 sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
457 error = in_control(NULL, SIOCSIFADDR, (caddr_t)&ifr, lo_ifp, kernproc);
458 if (error) {
459 printf("%s: failed to initialise lo0's address, error=%d\n",
460 __func__, error);
461 }
462 }
463
464 /*
465 * IP initialization: fill in IP protocol switch table.
466 * All protocols not implemented in kernel go to raw IP protocol handler.
467 */
468 void
ip_init(struct protosw * pp,struct domain * dp)469 ip_init(struct protosw *pp, struct domain *dp)
470 {
471 static int ip_initialized = 0;
472 struct protosw *pr;
473 struct timeval tv;
474 int i;
475
476 domain_proto_mtx_lock_assert_held();
477 VERIFY((pp->pr_flags & (PR_INITIALIZED | PR_ATTACHED)) == PR_ATTACHED);
478
479 /* ipq_alloc() uses mbufs for IP fragment queue structures */
480 _CASSERT(sizeof(struct ipq) <= _MLEN);
481
482 /*
483 * Some ioctls (e.g. SIOCAIFADDR) use ifaliasreq struct, which is
484 * interchangeable with in_aliasreq; they must have the same size.
485 */
486 _CASSERT(sizeof(struct ifaliasreq) == sizeof(struct in_aliasreq));
487
488 if (ip_initialized) {
489 return;
490 }
491 ip_initialized = 1;
492
493 in_ifaddr_init();
494
495 TAILQ_INIT(&in_ifaddrhead);
496 in_ifaddrhashtbl_init();
497
498 ip_moptions_init();
499
500 pr = pffindproto_locked(PF_INET, IPPROTO_RAW, SOCK_RAW);
501 if (pr == NULL) {
502 panic("%s: Unable to find [PF_INET,IPPROTO_RAW,SOCK_RAW]",
503 __func__);
504 /* NOTREACHED */
505 }
506
507 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
508 for (i = 0; i < IPPROTO_MAX; i++) {
509 ip_protox[i] = pr;
510 }
511 /*
512 * Cycle through IP protocols and put them into the appropriate place
513 * in ip_protox[], skipping protocols IPPROTO_{IP,RAW}.
514 */
515 VERIFY(dp == inetdomain && dp->dom_family == PF_INET);
516 TAILQ_FOREACH(pr, &dp->dom_protosw, pr_entry) {
517 VERIFY(pr->pr_domain == dp);
518 if (pr->pr_protocol != 0 && pr->pr_protocol != IPPROTO_RAW) {
519 /* Be careful to only index valid IP protocols. */
520 if (pr->pr_protocol < IPPROTO_MAX) {
521 ip_protox[pr->pr_protocol] = pr;
522 }
523 }
524 }
525
526 lck_mtx_lock(&ipqlock);
527 /* Initialize IP reassembly queue. */
528 for (i = 0; i < IPREASS_NHASH; i++) {
529 TAILQ_INIT(&ipq[i]);
530 }
531
532 maxnipq = nmbclusters / 32;
533 maxfragsperpacket = 128; /* enough for 64k in 512 byte fragments */
534 ipq_updateparams();
535 lck_mtx_unlock(&ipqlock);
536
537 getmicrotime(&tv);
538 ip_id = (u_short)(RandomULong() ^ tv.tv_usec);
539
540 PE_parse_boot_argn("ip_checkinterface", &i, sizeof(i));
541 switch (i) {
542 case IP_CHECKINTERFACE_WEAK_ES:
543 case IP_CHECKINTERFACE_HYBRID_ES:
544 case IP_CHECKINTERFACE_STRONG_ES:
545 ip_checkinterface = i;
546 break;
547 default:
548 break;
549 }
550
551 arp_init();
552 net_init_add(ip_init_delayed);
553 }
554
555 /*
556 * Initialize IPv4 source address hash table.
557 */
558 static void
in_ifaddrhashtbl_init(void)559 in_ifaddrhashtbl_init(void)
560 {
561 int i, k, p;
562
563 if (in_ifaddrhashtbl != NULL) {
564 return;
565 }
566
567 PE_parse_boot_argn("inaddr_nhash", &inaddr_nhash,
568 sizeof(inaddr_nhash));
569 if (inaddr_nhash == 0) {
570 inaddr_nhash = INADDR_NHASH;
571 }
572
573 in_ifaddrhashtbl = zalloc_permanent(
574 inaddr_nhash * sizeof(*in_ifaddrhashtbl),
575 ZALIGN_PTR);
576
577 /*
578 * Generate the next largest prime greater than inaddr_nhash.
579 */
580 k = (inaddr_nhash % 2 == 0) ? inaddr_nhash + 1 : inaddr_nhash + 2;
581 for (;;) {
582 p = 1;
583 for (i = 3; i * i <= k; i += 2) {
584 if (k % i == 0) {
585 p = 0;
586 }
587 }
588 if (p == 1) {
589 break;
590 }
591 k += 2;
592 }
593 inaddr_hashp = k;
594 }
595
596 u_int32_t
inaddr_hashval(u_int32_t key)597 inaddr_hashval(u_int32_t key)
598 {
599 /*
600 * The hash index is the computed prime times the key modulo
601 * the hash size, as documented in "Introduction to Algorithms"
602 * (Cormen, Leiserson, Rivest).
603 */
604 if (inaddr_nhash > 1) {
605 return (key * inaddr_hashp) % inaddr_nhash;
606 } else {
607 return 0;
608 }
609 }
610
611 __private_extern__ void
ip_proto_dispatch_in(struct mbuf * m,int hlen,u_int8_t proto,ipfilter_t inject_ipfref)612 ip_proto_dispatch_in(struct mbuf *m, int hlen, u_int8_t proto,
613 ipfilter_t inject_ipfref)
614 {
615 struct ipfilter *filter;
616 int seen = (inject_ipfref == NULL);
617 int changed_header = 0;
618 struct ip *ip;
619 void (*pr_input)(struct mbuf *, int len);
620
621 if (!TAILQ_EMPTY(&ipv4_filters)) {
622 ipf_ref();
623 TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
624 if (seen == 0) {
625 if ((struct ipfilter *)inject_ipfref == filter) {
626 seen = 1;
627 }
628 } else if (filter->ipf_filter.ipf_input) {
629 errno_t result;
630
631 if (changed_header == 0) {
632 /*
633 * Perform IP header alignment fixup,
634 * if needed, before passing packet
635 * into filter(s).
636 */
637 IP_HDR_ALIGNMENT_FIXUP(m,
638 m->m_pkthdr.rcvif, ipf_unref());
639
640 /* ipf_unref() already called */
641 if (m == NULL) {
642 return;
643 }
644
645 changed_header = 1;
646 ip = mtod(m, struct ip *);
647 ip->ip_len = htons(ip->ip_len + (uint16_t)hlen);
648 ip->ip_off = htons(ip->ip_off);
649 ip->ip_sum = 0;
650 ip->ip_sum = ip_cksum_hdr_in(m, hlen);
651 }
652 result = filter->ipf_filter.ipf_input(
653 filter->ipf_filter.cookie, (mbuf_t *)&m,
654 hlen, proto);
655 if (result == EJUSTRETURN) {
656 ipf_unref();
657 return;
658 }
659 if (result != 0) {
660 ipf_unref();
661 m_freem(m);
662 return;
663 }
664 }
665 }
666 ipf_unref();
667 }
668
669 /* Perform IP header alignment fixup (post-filters), if needed */
670 IP_HDR_ALIGNMENT_FIXUP(m, m->m_pkthdr.rcvif, return );
671
672 ip = mtod(m, struct ip *);
673
674 if (changed_header) {
675 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
676 ip->ip_off = ntohs(ip->ip_off);
677 }
678
679 /*
680 * If there isn't a specific lock for the protocol
681 * we're about to call, use the generic lock for AF_INET.
682 * otherwise let the protocol deal with its own locking
683 */
684 if ((pr_input = ip_protox[ip->ip_p]->pr_input) == NULL) {
685 m_freem(m);
686 } else if (!(ip_protox[ip->ip_p]->pr_flags & PR_PROTOLOCK)) {
687 lck_mtx_lock(inet_domain_mutex);
688 pr_input(m, hlen);
689 lck_mtx_unlock(inet_domain_mutex);
690 } else {
691 pr_input(m, hlen);
692 }
693 }
694
695 struct pktchain_elm {
696 struct mbuf *pkte_head;
697 struct mbuf *pkte_tail;
698 struct in_addr pkte_saddr;
699 struct in_addr pkte_daddr;
700 uint16_t pkte_npkts;
701 uint16_t pkte_proto;
702 uint32_t pkte_nbytes;
703 };
704
705 typedef struct pktchain_elm pktchain_elm_t;
706
707 /* Store upto PKTTBL_SZ unique flows on the stack */
708 #define PKTTBL_SZ 7
709
710 static struct mbuf *
ip_chain_insert(struct mbuf * packet,pktchain_elm_t * tbl)711 ip_chain_insert(struct mbuf *packet, pktchain_elm_t *tbl)
712 {
713 struct ip* ip;
714 int pkttbl_idx = 0;
715
716 ip = mtod(packet, struct ip*);
717
718 /* reusing the hash function from inaddr_hashval */
719 pkttbl_idx = inaddr_hashval(ntohl(ip->ip_src.s_addr)) % PKTTBL_SZ;
720 if (tbl[pkttbl_idx].pkte_head == NULL) {
721 tbl[pkttbl_idx].pkte_head = packet;
722 tbl[pkttbl_idx].pkte_saddr.s_addr = ip->ip_src.s_addr;
723 tbl[pkttbl_idx].pkte_daddr.s_addr = ip->ip_dst.s_addr;
724 tbl[pkttbl_idx].pkte_proto = ip->ip_p;
725 } else {
726 if ((ip->ip_dst.s_addr == tbl[pkttbl_idx].pkte_daddr.s_addr) &&
727 (ip->ip_src.s_addr == tbl[pkttbl_idx].pkte_saddr.s_addr) &&
728 (ip->ip_p == tbl[pkttbl_idx].pkte_proto)) {
729 } else {
730 return packet;
731 }
732 }
733 if (tbl[pkttbl_idx].pkte_tail != NULL) {
734 mbuf_setnextpkt(tbl[pkttbl_idx].pkte_tail, packet);
735 }
736
737 tbl[pkttbl_idx].pkte_tail = packet;
738 tbl[pkttbl_idx].pkte_npkts += 1;
739 tbl[pkttbl_idx].pkte_nbytes += packet->m_pkthdr.len;
740 return NULL;
741 }
742
743 /* args is a dummy variable here for backward compatibility */
744 static void
ip_input_second_pass_loop_tbl(pktchain_elm_t * tbl,struct ip_fw_in_args * args)745 ip_input_second_pass_loop_tbl(pktchain_elm_t *tbl, struct ip_fw_in_args *args)
746 {
747 int i = 0;
748
749 for (i = 0; i < PKTTBL_SZ; i++) {
750 if (tbl[i].pkte_head != NULL) {
751 struct mbuf *m = tbl[i].pkte_head;
752 ip_input_second_pass(m, m->m_pkthdr.rcvif,
753 tbl[i].pkte_npkts, tbl[i].pkte_nbytes, args);
754
755 if (tbl[i].pkte_npkts > 2) {
756 ipstat.ips_rxc_chainsz_gt2++;
757 }
758 if (tbl[i].pkte_npkts > 4) {
759 ipstat.ips_rxc_chainsz_gt4++;
760 }
761 #if (DEBUG || DEVELOPMENT)
762 if (ip_input_measure) {
763 net_perf_histogram(&net_perf, tbl[i].pkte_npkts);
764 }
765 #endif /* (DEBUG || DEVELOPMENT) */
766 tbl[i].pkte_head = tbl[i].pkte_tail = NULL;
767 tbl[i].pkte_npkts = 0;
768 tbl[i].pkte_nbytes = 0;
769 /* no need to initialize address and protocol in tbl */
770 }
771 }
772 }
773
774 static void
ip_input_cpout_args(struct ip_fw_in_args * args,struct ip_fw_args * args1,boolean_t * done_init)775 ip_input_cpout_args(struct ip_fw_in_args *args, struct ip_fw_args *args1,
776 boolean_t *done_init)
777 {
778 if (*done_init == FALSE) {
779 bzero(args1, sizeof(struct ip_fw_args));
780 *done_init = TRUE;
781 }
782 args1->fwa_pf_rule = args->fwai_pf_rule;
783 }
784
785 static void
ip_input_cpin_args(struct ip_fw_args * args1,struct ip_fw_in_args * args)786 ip_input_cpin_args(struct ip_fw_args *args1, struct ip_fw_in_args *args)
787 {
788 args->fwai_pf_rule = args1->fwa_pf_rule;
789 }
790
791 typedef enum {
792 IPINPUT_DOCHAIN = 0,
793 IPINPUT_DONTCHAIN,
794 IPINPUT_FREED,
795 IPINPUT_DONE
796 } ipinput_chain_ret_t;
797
798 static void
ip_input_update_nstat(struct ifnet * ifp,struct in_addr src_ip,u_int32_t packets,u_int32_t bytes)799 ip_input_update_nstat(struct ifnet *ifp, struct in_addr src_ip,
800 u_int32_t packets, u_int32_t bytes)
801 {
802 if (nstat_collect) {
803 struct rtentry *rt = ifnet_cached_rtlookup_inet(ifp,
804 src_ip);
805 if (rt != NULL) {
806 nstat_route_rx(rt, packets, bytes, 0);
807 rtfree(rt);
808 }
809 }
810 }
811
812 static void
ip_input_dispatch_chain(struct mbuf * m)813 ip_input_dispatch_chain(struct mbuf *m)
814 {
815 struct mbuf *tmp_mbuf = m;
816 struct mbuf *nxt_mbuf = NULL;
817 struct ip *ip = NULL;
818 unsigned int hlen;
819
820 ip = mtod(tmp_mbuf, struct ip *);
821 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
822 while (tmp_mbuf != NULL) {
823 nxt_mbuf = mbuf_nextpkt(tmp_mbuf);
824 mbuf_setnextpkt(tmp_mbuf, NULL);
825 ip_proto_dispatch_in(tmp_mbuf, hlen, ip->ip_p, 0);
826 tmp_mbuf = nxt_mbuf;
827 if (tmp_mbuf) {
828 ip = mtod(tmp_mbuf, struct ip *);
829 /* first mbuf of chain already has adjusted ip_len */
830 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
831 ip->ip_len -= hlen;
832 }
833 }
834 }
835
836 static void
ip_input_setdst_chain(struct mbuf * m,uint16_t ifindex,struct in_ifaddr * ia)837 ip_input_setdst_chain(struct mbuf *m, uint16_t ifindex, struct in_ifaddr *ia)
838 {
839 struct mbuf *tmp_mbuf = m;
840
841 while (tmp_mbuf != NULL) {
842 ip_setdstifaddr_info(tmp_mbuf, ifindex, ia);
843 tmp_mbuf = mbuf_nextpkt(tmp_mbuf);
844 }
845 }
846
847 static void
ip_input_adjust(struct mbuf * m,struct ip * ip,struct ifnet * inifp)848 ip_input_adjust(struct mbuf *m, struct ip *ip, struct ifnet *inifp)
849 {
850 boolean_t adjust = TRUE;
851
852 ASSERT(m_pktlen(m) > ip->ip_len);
853
854 /*
855 * Invalidate hardware checksum info if ip_adj_clear_hwcksum
856 * is set; useful to handle buggy drivers. Note that this
857 * should not be enabled by default, as we may get here due
858 * to link-layer padding.
859 */
860 if (ip_adj_clear_hwcksum &&
861 (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
862 !(inifp->if_flags & IFF_LOOPBACK) &&
863 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
864 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
865 m->m_pkthdr.csum_data = 0;
866 ipstat.ips_adj_hwcsum_clr++;
867 }
868
869 /*
870 * If partial checksum information is available, subtract
871 * out the partial sum of postpended extraneous bytes, and
872 * update the checksum metadata accordingly. By doing it
873 * here, the upper layer transport only needs to adjust any
874 * prepended extraneous bytes (else it will do both.)
875 */
876 if (ip_adj_partial_sum &&
877 (m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
878 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
879 m->m_pkthdr.csum_rx_val = m_adj_sum16(m,
880 m->m_pkthdr.csum_rx_start, m->m_pkthdr.csum_rx_start,
881 (ip->ip_len - m->m_pkthdr.csum_rx_start),
882 m->m_pkthdr.csum_rx_val);
883 } else if ((m->m_pkthdr.csum_flags &
884 (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
885 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
886 /*
887 * If packet has partial checksum info and we decided not
888 * to subtract the partial sum of postpended extraneous
889 * bytes here (not the default case), leave that work to
890 * be handled by the other layers. For now, only TCP, UDP
891 * layers are capable of dealing with this. For all other
892 * protocols (including fragments), trim and ditch the
893 * partial sum as those layers might not implement partial
894 * checksumming (or adjustment) at all.
895 */
896 if ((ip->ip_off & (IP_MF | IP_OFFMASK)) == 0 &&
897 (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_UDP)) {
898 adjust = FALSE;
899 } else {
900 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
901 m->m_pkthdr.csum_data = 0;
902 ipstat.ips_adj_hwcsum_clr++;
903 }
904 }
905
906 if (adjust) {
907 ipstat.ips_adj++;
908 if (m->m_len == m->m_pkthdr.len) {
909 m->m_len = ip->ip_len;
910 m->m_pkthdr.len = ip->ip_len;
911 } else {
912 m_adj(m, ip->ip_len - m->m_pkthdr.len);
913 }
914 }
915 }
916
917 /*
918 * First pass does all essential packet validation and places on a per flow
919 * queue for doing operations that have same outcome for all packets of a flow.
920 */
921 static ipinput_chain_ret_t
ip_input_first_pass(struct mbuf * m,struct ip_fw_in_args * args,struct mbuf ** modm)922 ip_input_first_pass(struct mbuf *m, struct ip_fw_in_args *args, struct mbuf **modm)
923 {
924 struct ip *ip;
925 struct ifnet *inifp;
926 unsigned int hlen;
927 int retval = IPINPUT_DOCHAIN;
928 int len = 0;
929 struct in_addr src_ip;
930 #if DUMMYNET
931 struct m_tag *copy;
932 struct m_tag *p;
933 boolean_t delete = FALSE;
934 struct ip_fw_args args1;
935 boolean_t init = FALSE;
936 #endif /* DUMMYNET */
937 ipfilter_t inject_filter_ref = NULL;
938
939 /* Check if the mbuf is still valid after interface filter processing */
940 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
941 inifp = mbuf_pkthdr_rcvif(m);
942 VERIFY(inifp != NULL);
943
944 /* Perform IP header alignment fixup, if needed */
945 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
946
947 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
948
949 #if DUMMYNET
950 /*
951 * Don't bother searching for tag(s) if there's none.
952 */
953 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
954 goto ipfw_tags_done;
955 }
956
957 /* Grab info from mtags prepended to the chain */
958 p = m_tag_first(m);
959 while (p) {
960 if (p->m_tag_id == KERNEL_MODULE_TAG_ID) {
961 if (p->m_tag_type == KERNEL_TAG_TYPE_DUMMYNET) {
962 struct dn_pkt_tag *dn_tag;
963
964 dn_tag = (struct dn_pkt_tag *)(p + 1);
965 args->fwai_pf_rule = dn_tag->dn_pf_rule;
966 delete = TRUE;
967 }
968
969 if (delete) {
970 copy = p;
971 p = m_tag_next(m, p);
972 m_tag_delete(m, copy);
973 } else {
974 p = m_tag_next(m, p);
975 }
976 } else {
977 p = m_tag_next(m, p);
978 }
979 }
980
981 #if DIAGNOSTIC
982 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
983 panic("ip_input no HDR");
984 }
985 #endif
986
987 if (args->fwai_pf_rule) {
988 /* dummynet already filtered us */
989 ip = mtod(m, struct ip *);
990 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
991 inject_filter_ref = ipf_get_inject_filter(m);
992 if (args->fwai_pf_rule) {
993 goto check_with_pf;
994 }
995 }
996 ipfw_tags_done:
997 #endif /* DUMMYNET */
998
999 /*
1000 * No need to process packet twice if we've already seen it.
1001 */
1002 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1003 inject_filter_ref = ipf_get_inject_filter(m);
1004 }
1005 if (inject_filter_ref != NULL) {
1006 ip = mtod(m, struct ip *);
1007 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1008
1009 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1010 struct ip *, ip, struct ifnet *, inifp,
1011 struct ip *, ip, struct ip6_hdr *, NULL);
1012
1013 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1014 ip->ip_off = ntohs(ip->ip_off);
1015 ip_proto_dispatch_in(m, hlen, ip->ip_p, inject_filter_ref);
1016 return IPINPUT_DONE;
1017 }
1018
1019 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1020 if_ports_used_match_mbuf(inifp, PF_INET, m);
1021 }
1022
1023 if (m->m_pkthdr.len < sizeof(struct ip)) {
1024 OSAddAtomic(1, &ipstat.ips_total);
1025 OSAddAtomic(1, &ipstat.ips_tooshort);
1026 m_freem(m);
1027 return IPINPUT_FREED;
1028 }
1029
1030 if (m->m_len < sizeof(struct ip) &&
1031 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1032 OSAddAtomic(1, &ipstat.ips_total);
1033 OSAddAtomic(1, &ipstat.ips_toosmall);
1034 return IPINPUT_FREED;
1035 }
1036
1037 ip = mtod(m, struct ip *);
1038 *modm = m;
1039
1040 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1041 ip->ip_p, ip->ip_off, ip->ip_len);
1042
1043 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1044 OSAddAtomic(1, &ipstat.ips_total);
1045 OSAddAtomic(1, &ipstat.ips_badvers);
1046 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1047 m_freem(m);
1048 return IPINPUT_FREED;
1049 }
1050
1051 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1052 if (hlen < sizeof(struct ip)) {
1053 OSAddAtomic(1, &ipstat.ips_total);
1054 OSAddAtomic(1, &ipstat.ips_badhlen);
1055 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1056 m_freem(m);
1057 return IPINPUT_FREED;
1058 }
1059
1060 if (hlen > m->m_len) {
1061 if ((m = m_pullup(m, hlen)) == NULL) {
1062 OSAddAtomic(1, &ipstat.ips_total);
1063 OSAddAtomic(1, &ipstat.ips_badhlen);
1064 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1065 return IPINPUT_FREED;
1066 }
1067 ip = mtod(m, struct ip *);
1068 *modm = m;
1069 }
1070
1071 /* 127/8 must not appear on wire - RFC1122 */
1072 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1073 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1074 /*
1075 * Allow for the following exceptions:
1076 *
1077 * 1. If the packet was sent to loopback (i.e. rcvif
1078 * would have been set earlier at output time.)
1079 *
1080 * 2. If the packet was sent out on loopback from a local
1081 * source address which belongs to a non-loopback
1082 * interface (i.e. rcvif may not necessarily be a
1083 * loopback interface, hence the test for PKTF_LOOP.)
1084 * Unlike IPv6, there is no interface scope ID, and
1085 * therefore we don't care so much about PKTF_IFINFO.
1086 */
1087 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1088 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1089 OSAddAtomic(1, &ipstat.ips_total);
1090 OSAddAtomic(1, &ipstat.ips_badaddr);
1091 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1092 m_freem(m);
1093 return IPINPUT_FREED;
1094 }
1095 }
1096
1097 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1098 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1099 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1100 ip_linklocal_stat.iplls_in_total++;
1101 if (ip->ip_ttl != MAXTTL) {
1102 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1103 /* Silently drop link local traffic with bad TTL */
1104 if (!ip_linklocal_in_allowbadttl) {
1105 OSAddAtomic(1, &ipstat.ips_total);
1106 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1107 m_freem(m);
1108 return IPINPUT_FREED;
1109 }
1110 }
1111 }
1112
1113 if (ip_cksum(m, hlen)) {
1114 OSAddAtomic(1, &ipstat.ips_total);
1115 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1116 m_freem(m);
1117 return IPINPUT_FREED;
1118 }
1119
1120 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1121 struct ip *, ip, struct ifnet *, inifp,
1122 struct ip *, ip, struct ip6_hdr *, NULL);
1123
1124 /*
1125 * Convert fields to host representation.
1126 */
1127 #if BYTE_ORDER != BIG_ENDIAN
1128 NTOHS(ip->ip_len);
1129 #endif
1130
1131 if (ip->ip_len < hlen) {
1132 OSAddAtomic(1, &ipstat.ips_total);
1133 OSAddAtomic(1, &ipstat.ips_badlen);
1134 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1135 m_freem(m);
1136 return IPINPUT_FREED;
1137 }
1138
1139 #if BYTE_ORDER != BIG_ENDIAN
1140 NTOHS(ip->ip_off);
1141 #endif
1142
1143 /*
1144 * Check that the amount of data in the buffers
1145 * is as at least much as the IP header would have us expect.
1146 * Trim mbufs if longer than we expect.
1147 * Drop packet if shorter than we expect.
1148 */
1149 if (m->m_pkthdr.len < ip->ip_len) {
1150 OSAddAtomic(1, &ipstat.ips_total);
1151 OSAddAtomic(1, &ipstat.ips_tooshort);
1152 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1153 m_freem(m);
1154 return IPINPUT_FREED;
1155 }
1156
1157 if (m->m_pkthdr.len > ip->ip_len) {
1158 ip_input_adjust(m, ip, inifp);
1159 }
1160
1161 /* for netstat route statistics */
1162 src_ip = ip->ip_src;
1163 len = m->m_pkthdr.len;
1164
1165 #if DUMMYNET
1166 check_with_pf:
1167 #endif /* DUMMYNET */
1168 #if PF
1169 /* Invoke inbound packet filter */
1170 if (PF_IS_ENABLED) {
1171 int error;
1172 ip_input_cpout_args(args, &args1, &init);
1173 ip = mtod(m, struct ip *);
1174 src_ip = ip->ip_src;
1175
1176 #if DUMMYNET
1177 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args1);
1178 #else
1179 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1180 #endif /* DUMMYNET */
1181 if (error != 0 || m == NULL) {
1182 if (m != NULL) {
1183 panic("%s: unexpected packet %p",
1184 __func__, m);
1185 /* NOTREACHED */
1186 }
1187 /* Already freed by callee */
1188 ip_input_update_nstat(inifp, src_ip, 1, len);
1189 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1190 OSAddAtomic(1, &ipstat.ips_total);
1191 return IPINPUT_FREED;
1192 }
1193 ip = mtod(m, struct ip *);
1194 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1195 *modm = m;
1196 ip_input_cpin_args(&args1, args);
1197 }
1198 #endif /* PF */
1199
1200 #if IPSEC
1201 if (ipsec_bypass == 0 && ipsec_gethist(m, NULL)) {
1202 retval = IPINPUT_DONTCHAIN; /* XXX scope for chaining here? */
1203 goto pass;
1204 }
1205 #endif
1206
1207 #if IPSEC
1208 pass:
1209 #endif
1210 /*
1211 * Process options and, if not destined for us,
1212 * ship it on. ip_dooptions returns 1 when an
1213 * error was detected (causing an icmp message
1214 * to be sent and the original packet to be freed).
1215 */
1216 ip_nhops = 0; /* for source routed packets */
1217 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1218 src_ip = ip->ip_src;
1219 ip_input_update_nstat(inifp, src_ip, 1, len);
1220 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1221 OSAddAtomic(1, &ipstat.ips_total);
1222 return IPINPUT_FREED;
1223 }
1224
1225 /*
1226 * Don't chain fragmented packets
1227 */
1228 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1229 return IPINPUT_DONTCHAIN;
1230 }
1231
1232 /* Allow DHCP/BootP responses through */
1233 if ((inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1234 hlen == sizeof(struct ip) && ip->ip_p == IPPROTO_UDP) {
1235 struct udpiphdr *ui;
1236
1237 if (m->m_len < sizeof(struct udpiphdr) &&
1238 (m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1239 OSAddAtomic(1, &udpstat.udps_hdrops);
1240 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1241 OSAddAtomic(1, &ipstat.ips_total);
1242 return IPINPUT_FREED;
1243 }
1244 *modm = m;
1245 ui = mtod(m, struct udpiphdr *);
1246 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1247 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1248 return IPINPUT_DONTCHAIN;
1249 }
1250 }
1251
1252 /* Avoid chaining raw sockets as ipsec checks occur later for them */
1253 if (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR) {
1254 return IPINPUT_DONTCHAIN;
1255 }
1256
1257 return retval;
1258 #if !defined(__i386__) && !defined(__x86_64__)
1259 bad:
1260 m_freem(m);
1261 return IPINPUT_FREED;
1262 #endif
1263 }
1264
1265 /*
1266 * Because the call to m_pullup() may freem the mbuf, the function frees the mbuf packet
1267 * chain before it return IP_CHECK_IF_DROP
1268 */
1269 static ip_check_if_result_t
ip_input_check_interface(struct mbuf ** mp,struct ip * ip,struct ifnet * inifp)1270 ip_input_check_interface(struct mbuf **mp, struct ip *ip, struct ifnet *inifp)
1271 {
1272 struct mbuf *m = *mp;
1273 struct in_ifaddr *ia = NULL;
1274 struct in_ifaddr *best_ia = NULL;
1275 struct ifnet *match_ifp = NULL;
1276 ip_check_if_result_t result = IP_CHECK_IF_NONE;
1277
1278 /*
1279 * Host broadcast and all network broadcast addresses are always a match
1280 */
1281 if (ip->ip_dst.s_addr == (u_int32_t)INADDR_BROADCAST ||
1282 ip->ip_dst.s_addr == INADDR_ANY) {
1283 ip_input_setdst_chain(m, inifp->if_index, NULL);
1284 return IP_CHECK_IF_OURS;
1285 }
1286
1287 /*
1288 * Check for a match in the hash bucket.
1289 */
1290 lck_rw_lock_shared(&in_ifaddr_rwlock);
1291 TAILQ_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
1292 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
1293 best_ia = ia;
1294 match_ifp = best_ia->ia_ifp;
1295
1296 if (ia->ia_ifp == inifp || (inifp->if_flags & IFF_LOOPBACK) ||
1297 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1298 /*
1299 * A locally originated packet or packet from the loopback
1300 * interface is always an exact interface address match
1301 */
1302 match_ifp = inifp;
1303 break;
1304 }
1305 /*
1306 * Continue the loop in case there's a exact match with another
1307 * interface
1308 */
1309 }
1310 }
1311 if (best_ia != NULL) {
1312 if (match_ifp != inifp && ipforwarding == 0 &&
1313 ((ip_checkinterface == IP_CHECKINTERFACE_HYBRID_ES &&
1314 (match_ifp->if_family == IFNET_FAMILY_IPSEC ||
1315 match_ifp->if_family == IFNET_FAMILY_UTUN)) ||
1316 ip_checkinterface == IP_CHECKINTERFACE_STRONG_ES)) {
1317 /*
1318 * Drop when interface address check is strict and forwarding
1319 * is disabled
1320 */
1321 result = IP_CHECK_IF_DROP;
1322 } else {
1323 result = IP_CHECK_IF_OURS;
1324 ip_input_setdst_chain(m, 0, best_ia);
1325 }
1326 }
1327 lck_rw_done(&in_ifaddr_rwlock);
1328
1329 if (result == IP_CHECK_IF_NONE && (inifp->if_flags & IFF_BROADCAST)) {
1330 /*
1331 * Check for broadcast addresses.
1332 *
1333 * Only accept broadcast packets that arrive via the matching
1334 * interface. Reception of forwarded directed broadcasts would be
1335 * handled via ip_forward() and ether_frameout() with the loopback
1336 * into the stack for SIMPLEX interfaces handled by ether_frameout().
1337 */
1338 struct ifaddr *ifa;
1339
1340 ifnet_lock_shared(inifp);
1341 TAILQ_FOREACH(ifa, &inifp->if_addrhead, ifa_link) {
1342 if (ifa->ifa_addr->sa_family != AF_INET) {
1343 continue;
1344 }
1345 ia = ifatoia(ifa);
1346 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == ip->ip_dst.s_addr ||
1347 ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
1348 ip_input_setdst_chain(m, 0, ia);
1349 result = IP_CHECK_IF_OURS;
1350 match_ifp = inifp;
1351 break;
1352 }
1353 }
1354 ifnet_lock_done(inifp);
1355 }
1356
1357 /* Allow DHCP/BootP responses through */
1358 if (result == IP_CHECK_IF_NONE && (inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1359 ip->ip_p == IPPROTO_UDP && (IP_VHL_HL(ip->ip_vhl) << 2) == sizeof(struct ip)) {
1360 struct udpiphdr *ui;
1361
1362 if (m->m_len < sizeof(struct udpiphdr)) {
1363 if ((m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1364 OSAddAtomic(1, &udpstat.udps_hdrops);
1365 *mp = NULL;
1366 return IP_CHECK_IF_DROP;
1367 }
1368 /*
1369 * m_pullup can return a different mbuf
1370 */
1371 *mp = m;
1372 ip = mtod(m, struct ip *);
1373 }
1374 ui = mtod(m, struct udpiphdr *);
1375 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1376 ip_input_setdst_chain(m, inifp->if_index, NULL);
1377 result = IP_CHECK_IF_OURS;
1378 match_ifp = inifp;
1379 }
1380 }
1381
1382 if (result == IP_CHECK_IF_NONE) {
1383 if (ipforwarding == 0) {
1384 result = IP_CHECK_IF_DROP;
1385 } else {
1386 result = IP_CHECK_IF_FORWARD;
1387 ip_input_setdst_chain(m, inifp->if_index, NULL);
1388 }
1389 }
1390
1391 if (result == IP_CHECK_IF_OURS && match_ifp != inifp) {
1392 ipstat.ips_rcv_if_weak_match++;
1393
1394 /* Logging is too noisy when forwarding is enabled */
1395 if (ip_checkinterface_debug != 0 && ipforwarding == 0) {
1396 char src_str[MAX_IPv4_STR_LEN];
1397 char dst_str[MAX_IPv4_STR_LEN];
1398
1399 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1400 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1401 os_log_info(OS_LOG_DEFAULT,
1402 "%s: weak ES interface match to %s for packet from %s to %s proto %u received via %s",
1403 __func__, best_ia->ia_ifp->if_xname, src_str, dst_str, ip->ip_p, inifp->if_xname);
1404 }
1405 } else if (result == IP_CHECK_IF_DROP) {
1406 if (ip_checkinterface_debug > 0) {
1407 char src_str[MAX_IPv4_STR_LEN];
1408 char dst_str[MAX_IPv4_STR_LEN];
1409
1410 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1411 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1412 os_log(OS_LOG_DEFAULT,
1413 "%s: no interface match for packet from %s to %s proto %u received via %s",
1414 __func__, src_str, dst_str, ip->ip_p, inifp->if_xname);
1415 }
1416 struct mbuf *tmp_mbuf = m;
1417 while (tmp_mbuf != NULL) {
1418 ipstat.ips_rcv_if_no_match++;
1419 tmp_mbuf = tmp_mbuf->m_nextpkt;
1420 }
1421 m_freem_list(m);
1422 *mp = NULL;
1423 }
1424
1425 return result;
1426 }
1427
1428 static void
ip_input_second_pass(struct mbuf * m,struct ifnet * inifp,int npkts_in_chain,int bytes_in_chain,struct ip_fw_in_args * args)1429 ip_input_second_pass(struct mbuf *m, struct ifnet *inifp,
1430 int npkts_in_chain, int bytes_in_chain, struct ip_fw_in_args *args)
1431 {
1432 struct mbuf *tmp_mbuf = NULL;
1433 unsigned int hlen;
1434
1435 #pragma unused (args)
1436
1437 struct ip *ip = mtod(m, struct ip *);
1438 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1439
1440 OSAddAtomic(npkts_in_chain, &ipstat.ips_total);
1441
1442 /*
1443 * Naively assume we can attribute inbound data to the route we would
1444 * use to send to this destination. Asymmetric routing breaks this
1445 * assumption, but it still allows us to account for traffic from
1446 * a remote node in the routing table.
1447 * this has a very significant performance impact so we bypass
1448 * if nstat_collect is disabled. We may also bypass if the
1449 * protocol is tcp in the future because tcp will have a route that
1450 * we can use to attribute the data to. That does mean we would not
1451 * account for forwarded tcp traffic.
1452 */
1453 ip_input_update_nstat(inifp, ip->ip_src, npkts_in_chain,
1454 bytes_in_chain);
1455
1456 /*
1457 * Check our list of addresses, to see if the packet is for us.
1458 * If we don't have any addresses, assume any unicast packet
1459 * we receive might be for us (and let the upper layers deal
1460 * with it).
1461 */
1462 tmp_mbuf = m;
1463 if (TAILQ_EMPTY(&in_ifaddrhead)) {
1464 while (tmp_mbuf != NULL) {
1465 if (!(tmp_mbuf->m_flags & (M_MCAST | M_BCAST))) {
1466 ip_setdstifaddr_info(tmp_mbuf, inifp->if_index,
1467 NULL);
1468 }
1469 tmp_mbuf = mbuf_nextpkt(tmp_mbuf);
1470 }
1471 goto ours;
1472 }
1473
1474 /*
1475 * Enable a consistency check between the destination address
1476 * and the arrival interface for a unicast packet (the RFC 1122
1477 * strong ES model) if IP forwarding is disabled and the packet
1478 * is not locally generated
1479 *
1480 * XXX - Checking also should be disabled if the destination
1481 * address is ipnat'ed to a different interface.
1482 *
1483 * XXX - Checking is incompatible with IP aliases added
1484 * to the loopback interface instead of the interface where
1485 * the packets are received.
1486 */
1487 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1488 ip_check_if_result_t ip_check_if_result = IP_CHECK_IF_NONE;
1489
1490 ip_check_if_result = ip_input_check_interface(&m, ip, inifp);
1491 ASSERT(ip_check_if_result != IP_CHECK_IF_NONE);
1492 if (ip_check_if_result == IP_CHECK_IF_OURS) {
1493 goto ours;
1494 } else if (ip_check_if_result == IP_CHECK_IF_DROP) {
1495 return;
1496 }
1497 } else {
1498 struct in_multi *inm;
1499 /*
1500 * See if we belong to the destination multicast group on the
1501 * arrival interface.
1502 */
1503 in_multihead_lock_shared();
1504 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
1505 in_multihead_lock_done();
1506 if (inm == NULL) {
1507 OSAddAtomic(npkts_in_chain, &ipstat.ips_notmember);
1508 m_freem_list(m);
1509 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1510 return;
1511 }
1512 ip_input_setdst_chain(m, inifp->if_index, NULL);
1513 INM_REMREF(inm);
1514 goto ours;
1515 }
1516
1517 tmp_mbuf = m;
1518 struct mbuf *nxt_mbuf = NULL;
1519 while (tmp_mbuf != NULL) {
1520 nxt_mbuf = mbuf_nextpkt(tmp_mbuf);
1521 /*
1522 * Not for us; forward if possible and desirable.
1523 */
1524 mbuf_setnextpkt(tmp_mbuf, NULL);
1525 if (ipforwarding == 0) {
1526 OSAddAtomic(1, &ipstat.ips_cantforward);
1527 m_freem(tmp_mbuf);
1528 } else {
1529 ip_forward(tmp_mbuf, 0, NULL);
1530 }
1531 tmp_mbuf = nxt_mbuf;
1532 }
1533 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1534 return;
1535 ours:
1536 ip = mtod(m, struct ip *); /* in case it changed */
1537 /*
1538 * If offset is set, must reassemble.
1539 */
1540 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1541 VERIFY(npkts_in_chain == 1);
1542 m = ip_reass(m);
1543 if (m == NULL) {
1544 return;
1545 }
1546 ip = mtod(m, struct ip *);
1547 /* Get the header length of the reassembled packet */
1548 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1549 }
1550
1551 /*
1552 * Further protocols expect the packet length to be w/o the
1553 * IP header.
1554 */
1555 ip->ip_len -= hlen;
1556
1557 #if IPSEC
1558 /*
1559 * enforce IPsec policy checking if we are seeing last header.
1560 * note that we do not visit this with protocols with pcb layer
1561 * code - like udp/tcp/raw ip.
1562 */
1563 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
1564 VERIFY(npkts_in_chain == 1);
1565 if (ipsec4_in_reject(m, NULL)) {
1566 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
1567 goto bad;
1568 }
1569 }
1570 #endif /* IPSEC */
1571
1572 /*
1573 * Switch out to protocol's input routine.
1574 */
1575 OSAddAtomic(npkts_in_chain, &ipstat.ips_delivered);
1576
1577 ip_input_dispatch_chain(m);
1578
1579 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1580 return;
1581 bad:
1582 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1583 m_freem(m);
1584 }
1585
1586 void
ip_input_process_list(struct mbuf * packet_list)1587 ip_input_process_list(struct mbuf *packet_list)
1588 {
1589 pktchain_elm_t pktchain_tbl[PKTTBL_SZ];
1590
1591 struct mbuf *packet = NULL;
1592 struct mbuf *modm = NULL; /* modified mbuf */
1593 int retval = 0;
1594 #if (DEBUG || DEVELOPMENT)
1595 struct timeval start_tv;
1596 #endif /* (DEBUG || DEVELOPMENT) */
1597 int num_pkts = 0;
1598 int chain = 0;
1599 struct ip_fw_in_args args;
1600
1601 if (ip_chaining == 0) {
1602 struct mbuf *m = packet_list;
1603 #if (DEBUG || DEVELOPMENT)
1604 if (ip_input_measure) {
1605 net_perf_start_time(&net_perf, &start_tv);
1606 }
1607 #endif /* (DEBUG || DEVELOPMENT) */
1608
1609 while (m) {
1610 packet_list = mbuf_nextpkt(m);
1611 mbuf_setnextpkt(m, NULL);
1612 ip_input(m);
1613 m = packet_list;
1614 num_pkts++;
1615 }
1616 #if (DEBUG || DEVELOPMENT)
1617 if (ip_input_measure) {
1618 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1619 }
1620 #endif /* (DEBUG || DEVELOPMENT) */
1621 return;
1622 }
1623 #if (DEBUG || DEVELOPMENT)
1624 if (ip_input_measure) {
1625 net_perf_start_time(&net_perf, &start_tv);
1626 }
1627 #endif /* (DEBUG || DEVELOPMENT) */
1628
1629 bzero(&pktchain_tbl, sizeof(pktchain_tbl));
1630 restart_list_process:
1631 chain = 0;
1632 for (packet = packet_list; packet; packet = packet_list) {
1633 m_add_crumb(packet, PKT_CRUMB_IP_INPUT);
1634
1635 packet_list = mbuf_nextpkt(packet);
1636 mbuf_setnextpkt(packet, NULL);
1637
1638 num_pkts++;
1639 modm = NULL;
1640 bzero(&args, sizeof(args));
1641
1642 retval = ip_input_first_pass(packet, &args, &modm);
1643
1644 if (retval == IPINPUT_DOCHAIN) {
1645 if (modm) {
1646 packet = modm;
1647 }
1648 packet = ip_chain_insert(packet, &pktchain_tbl[0]);
1649 if (packet == NULL) {
1650 ipstat.ips_rxc_chained++;
1651 chain++;
1652 if (chain > ip_chainsz) {
1653 break;
1654 }
1655 } else {
1656 ipstat.ips_rxc_collisions++;
1657 break;
1658 }
1659 } else if (retval == IPINPUT_DONTCHAIN) {
1660 /* in order to preserve order, exit from chaining */
1661 if (modm) {
1662 packet = modm;
1663 }
1664 ipstat.ips_rxc_notchain++;
1665 break;
1666 } else {
1667 /* packet was freed or delivered, do nothing. */
1668 }
1669 }
1670
1671 /* do second pass here for pktchain_tbl */
1672 if (chain) {
1673 ip_input_second_pass_loop_tbl(&pktchain_tbl[0], &args);
1674 }
1675
1676 if (packet) {
1677 /*
1678 * equivalent update in chaining case if performed in
1679 * ip_input_second_pass_loop_tbl().
1680 */
1681 #if (DEBUG || DEVELOPMENT)
1682 if (ip_input_measure) {
1683 net_perf_histogram(&net_perf, 1);
1684 }
1685 #endif /* (DEBUG || DEVELOPMENT) */
1686 ip_input_second_pass(packet, packet->m_pkthdr.rcvif,
1687 1, packet->m_pkthdr.len, &args);
1688 }
1689
1690 if (packet_list) {
1691 goto restart_list_process;
1692 }
1693
1694 #if (DEBUG || DEVELOPMENT)
1695 if (ip_input_measure) {
1696 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1697 }
1698 #endif /* (DEBUG || DEVELOPMENT) */
1699 }
1700 /*
1701 * Ip input routine. Checksum and byte swap header. If fragmented
1702 * try to reassemble. Process options. Pass to next level.
1703 */
1704 void
ip_input(struct mbuf * m)1705 ip_input(struct mbuf *m)
1706 {
1707 struct ip *ip;
1708 unsigned int hlen;
1709 u_short sum = 0;
1710 #if DUMMYNET
1711 struct ip_fw_args args;
1712 struct m_tag *tag;
1713 #endif
1714 ipfilter_t inject_filter_ref = NULL;
1715 struct ifnet *inifp;
1716
1717 /* Check if the mbuf is still valid after interface filter processing */
1718 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
1719 inifp = m->m_pkthdr.rcvif;
1720 VERIFY(inifp != NULL);
1721
1722 m_add_crumb(m, PKT_CRUMB_IP_INPUT);
1723
1724 ipstat.ips_rxc_notlist++;
1725
1726 /* Perform IP header alignment fixup, if needed */
1727 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
1728
1729 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
1730
1731 #if DUMMYNET
1732 bzero(&args, sizeof(struct ip_fw_args));
1733
1734 /*
1735 * Don't bother searching for tag(s) if there's none.
1736 */
1737 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
1738 goto ipfw_tags_done;
1739 }
1740
1741 /* Grab info from mtags prepended to the chain */
1742 if ((tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID,
1743 KERNEL_TAG_TYPE_DUMMYNET, NULL)) != NULL) {
1744 struct dn_pkt_tag *dn_tag;
1745
1746 dn_tag = (struct dn_pkt_tag *)(tag + 1);
1747 args.fwa_pf_rule = dn_tag->dn_pf_rule;
1748
1749 m_tag_delete(m, tag);
1750 }
1751
1752 #if DIAGNOSTIC
1753 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
1754 panic("ip_input no HDR");
1755 }
1756 #endif
1757
1758 if (args.fwa_pf_rule) {
1759 /* dummynet already filtered us */
1760 ip = mtod(m, struct ip *);
1761 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1762 inject_filter_ref = ipf_get_inject_filter(m);
1763 if (args.fwa_pf_rule) {
1764 goto check_with_pf;
1765 }
1766 }
1767 ipfw_tags_done:
1768 #endif /* DUMMYNET */
1769
1770 /*
1771 * No need to process packet twice if we've already seen it.
1772 */
1773 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1774 inject_filter_ref = ipf_get_inject_filter(m);
1775 }
1776 if (inject_filter_ref != NULL) {
1777 ip = mtod(m, struct ip *);
1778 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1779
1780 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1781 struct ip *, ip, struct ifnet *, inifp,
1782 struct ip *, ip, struct ip6_hdr *, NULL);
1783
1784 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1785 ip->ip_off = ntohs(ip->ip_off);
1786 ip_proto_dispatch_in(m, hlen, ip->ip_p, inject_filter_ref);
1787 return;
1788 }
1789
1790 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1791 if_ports_used_match_mbuf(inifp, PF_INET, m);
1792 }
1793
1794 OSAddAtomic(1, &ipstat.ips_total);
1795 if (m->m_pkthdr.len < sizeof(struct ip)) {
1796 goto tooshort;
1797 }
1798
1799 if (m->m_len < sizeof(struct ip) &&
1800 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1801 OSAddAtomic(1, &ipstat.ips_toosmall);
1802 return;
1803 }
1804 ip = mtod(m, struct ip *);
1805
1806 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1807 ip->ip_p, ip->ip_off, ip->ip_len);
1808
1809 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1810 OSAddAtomic(1, &ipstat.ips_badvers);
1811 goto bad;
1812 }
1813
1814 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1815 if (hlen < sizeof(struct ip)) { /* minimum header length */
1816 OSAddAtomic(1, &ipstat.ips_badhlen);
1817 goto bad;
1818 }
1819 if (hlen > m->m_len) {
1820 if ((m = m_pullup(m, hlen)) == NULL) {
1821 OSAddAtomic(1, &ipstat.ips_badhlen);
1822 return;
1823 }
1824 ip = mtod(m, struct ip *);
1825 }
1826
1827 /* 127/8 must not appear on wire - RFC1122 */
1828 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1829 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1830 /*
1831 * Allow for the following exceptions:
1832 *
1833 * 1. If the packet was sent to loopback (i.e. rcvif
1834 * would have been set earlier at output time.)
1835 *
1836 * 2. If the packet was sent out on loopback from a local
1837 * source address which belongs to a non-loopback
1838 * interface (i.e. rcvif may not necessarily be a
1839 * loopback interface, hence the test for PKTF_LOOP.)
1840 * Unlike IPv6, there is no interface scope ID, and
1841 * therefore we don't care so much about PKTF_IFINFO.
1842 */
1843 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1844 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1845 OSAddAtomic(1, &ipstat.ips_badaddr);
1846 goto bad;
1847 }
1848 }
1849
1850 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1851 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1852 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1853 ip_linklocal_stat.iplls_in_total++;
1854 if (ip->ip_ttl != MAXTTL) {
1855 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1856 /* Silently drop link local traffic with bad TTL */
1857 if (!ip_linklocal_in_allowbadttl) {
1858 goto bad;
1859 }
1860 }
1861 }
1862
1863 sum = ip_cksum(m, hlen);
1864 if (sum) {
1865 goto bad;
1866 }
1867
1868 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1869 struct ip *, ip, struct ifnet *, inifp,
1870 struct ip *, ip, struct ip6_hdr *, NULL);
1871
1872 /*
1873 * Naively assume we can attribute inbound data to the route we would
1874 * use to send to this destination. Asymmetric routing breaks this
1875 * assumption, but it still allows us to account for traffic from
1876 * a remote node in the routing table.
1877 * this has a very significant performance impact so we bypass
1878 * if nstat_collect is disabled. We may also bypass if the
1879 * protocol is tcp in the future because tcp will have a route that
1880 * we can use to attribute the data to. That does mean we would not
1881 * account for forwarded tcp traffic.
1882 */
1883 if (nstat_collect) {
1884 struct rtentry *rt =
1885 ifnet_cached_rtlookup_inet(inifp, ip->ip_src);
1886 if (rt != NULL) {
1887 nstat_route_rx(rt, 1, m->m_pkthdr.len, 0);
1888 rtfree(rt);
1889 }
1890 }
1891
1892 /*
1893 * Convert fields to host representation.
1894 */
1895 #if BYTE_ORDER != BIG_ENDIAN
1896 NTOHS(ip->ip_len);
1897 #endif
1898
1899 if (ip->ip_len < hlen) {
1900 OSAddAtomic(1, &ipstat.ips_badlen);
1901 goto bad;
1902 }
1903
1904 #if BYTE_ORDER != BIG_ENDIAN
1905 NTOHS(ip->ip_off);
1906 #endif
1907 /*
1908 * Check that the amount of data in the buffers
1909 * is as at least much as the IP header would have us expect.
1910 * Trim mbufs if longer than we expect.
1911 * Drop packet if shorter than we expect.
1912 */
1913 if (m->m_pkthdr.len < ip->ip_len) {
1914 tooshort:
1915 OSAddAtomic(1, &ipstat.ips_tooshort);
1916 goto bad;
1917 }
1918 if (m->m_pkthdr.len > ip->ip_len) {
1919 ip_input_adjust(m, ip, inifp);
1920 }
1921
1922 #if DUMMYNET
1923 check_with_pf:
1924 #endif
1925 #if PF
1926 /* Invoke inbound packet filter */
1927 if (PF_IS_ENABLED) {
1928 int error;
1929 #if DUMMYNET
1930 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args);
1931 #else
1932 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1933 #endif /* DUMMYNET */
1934 if (error != 0 || m == NULL) {
1935 if (m != NULL) {
1936 panic("%s: unexpected packet %p",
1937 __func__, m);
1938 /* NOTREACHED */
1939 }
1940 /* Already freed by callee */
1941 return;
1942 }
1943 ip = mtod(m, struct ip *);
1944 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1945 }
1946 #endif /* PF */
1947
1948 #if IPSEC
1949 if (ipsec_bypass == 0 && ipsec_gethist(m, NULL)) {
1950 goto pass;
1951 }
1952 #endif
1953
1954 pass:
1955 /*
1956 * Process options and, if not destined for us,
1957 * ship it on. ip_dooptions returns 1 when an
1958 * error was detected (causing an icmp message
1959 * to be sent and the original packet to be freed).
1960 */
1961 ip_nhops = 0; /* for source routed packets */
1962 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1963 return;
1964 }
1965
1966 /*
1967 * Check our list of addresses, to see if the packet is for us.
1968 * If we don't have any addresses, assume any unicast packet
1969 * we receive might be for us (and let the upper layers deal
1970 * with it).
1971 */
1972 if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST))) {
1973 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1974 goto ours;
1975 }
1976
1977 /*
1978 * Enable a consistency check between the destination address
1979 * and the arrival interface for a unicast packet (the RFC 1122
1980 * strong ES model) if IP forwarding is disabled and the packet
1981 * is not locally generated and the packet is not subject to
1982 * 'ipfw fwd'.
1983 *
1984 * XXX - Checking also should be disabled if the destination
1985 * address is ipnat'ed to a different interface.
1986 *
1987 * XXX - Checking is incompatible with IP aliases added
1988 * to the loopback interface instead of the interface where
1989 * the packets are received.
1990 */
1991 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1992 ip_check_if_result_t check_if_result = IP_CHECK_IF_NONE;
1993
1994 check_if_result = ip_input_check_interface(&m, ip, inifp);
1995 ASSERT(check_if_result != IP_CHECK_IF_NONE);
1996 if (check_if_result == IP_CHECK_IF_OURS) {
1997 goto ours;
1998 } else if (check_if_result == IP_CHECK_IF_DROP) {
1999 return;
2000 }
2001 } else {
2002 struct in_multi *inm;
2003 /*
2004 * See if we belong to the destination multicast group on the
2005 * arrival interface.
2006 */
2007 in_multihead_lock_shared();
2008 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
2009 in_multihead_lock_done();
2010 if (inm == NULL) {
2011 OSAddAtomic(1, &ipstat.ips_notmember);
2012 m_freem(m);
2013 return;
2014 }
2015 ip_setdstifaddr_info(m, inifp->if_index, NULL);
2016 INM_REMREF(inm);
2017 goto ours;
2018 }
2019
2020 /*
2021 * Not for us; forward if possible and desirable.
2022 */
2023 if (ipforwarding == 0) {
2024 OSAddAtomic(1, &ipstat.ips_cantforward);
2025 m_freem(m);
2026 } else {
2027 ip_forward(m, 0, NULL);
2028 }
2029 return;
2030
2031 ours:
2032 /*
2033 * If offset or IP_MF are set, must reassemble.
2034 */
2035 if (ip->ip_off & ~(IP_DF | IP_RF)) {
2036 m = ip_reass(m);
2037 if (m == NULL) {
2038 return;
2039 }
2040 ip = mtod(m, struct ip *);
2041 /* Get the header length of the reassembled packet */
2042 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2043 }
2044
2045 /*
2046 * Further protocols expect the packet length to be w/o the
2047 * IP header.
2048 */
2049 ip->ip_len -= hlen;
2050
2051
2052 #if IPSEC
2053 /*
2054 * enforce IPsec policy checking if we are seeing last header.
2055 * note that we do not visit this with protocols with pcb layer
2056 * code - like udp/tcp/raw ip.
2057 */
2058 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
2059 if (ipsec4_in_reject(m, NULL)) {
2060 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
2061 goto bad;
2062 }
2063 }
2064 #endif /* IPSEC */
2065
2066 /*
2067 * Switch out to protocol's input routine.
2068 */
2069 OSAddAtomic(1, &ipstat.ips_delivered);
2070
2071 ip_proto_dispatch_in(m, hlen, ip->ip_p, 0);
2072 return;
2073
2074 bad:
2075 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
2076 m_freem(m);
2077 }
2078
2079 static void
ipq_updateparams(void)2080 ipq_updateparams(void)
2081 {
2082 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2083 /*
2084 * -1 for unlimited allocation.
2085 */
2086 if (maxnipq < 0) {
2087 ipq_limit = 0;
2088 }
2089 /*
2090 * Positive number for specific bound.
2091 */
2092 if (maxnipq > 0) {
2093 ipq_limit = maxnipq;
2094 }
2095 /*
2096 * Zero specifies no further fragment queue allocation -- set the
2097 * bound very low, but rely on implementation elsewhere to actually
2098 * prevent allocation and reclaim current queues.
2099 */
2100 if (maxnipq == 0) {
2101 ipq_limit = 1;
2102 }
2103 /*
2104 * Arm the purge timer if not already and if there's work to do
2105 */
2106 frag_sched_timeout();
2107 }
2108
2109 static int
2110 sysctl_maxnipq SYSCTL_HANDLER_ARGS
2111 {
2112 #pragma unused(arg1, arg2)
2113 int error, i;
2114
2115 lck_mtx_lock(&ipqlock);
2116 i = maxnipq;
2117 error = sysctl_handle_int(oidp, &i, 0, req);
2118 if (error || req->newptr == USER_ADDR_NULL) {
2119 goto done;
2120 }
2121 /* impose bounds */
2122 if (i < -1 || i > (nmbclusters / 4)) {
2123 error = EINVAL;
2124 goto done;
2125 }
2126 maxnipq = i;
2127 ipq_updateparams();
2128 done:
2129 lck_mtx_unlock(&ipqlock);
2130 return error;
2131 }
2132
2133 static int
2134 sysctl_maxfragsperpacket SYSCTL_HANDLER_ARGS
2135 {
2136 #pragma unused(arg1, arg2)
2137 int error, i;
2138
2139 lck_mtx_lock(&ipqlock);
2140 i = maxfragsperpacket;
2141 error = sysctl_handle_int(oidp, &i, 0, req);
2142 if (error || req->newptr == USER_ADDR_NULL) {
2143 goto done;
2144 }
2145 maxfragsperpacket = i;
2146 ipq_updateparams(); /* see if we need to arm timer */
2147 done:
2148 lck_mtx_unlock(&ipqlock);
2149 return error;
2150 }
2151
2152 /*
2153 * Take incoming datagram fragment and try to reassemble it into
2154 * whole datagram. If a chain for reassembly of this datagram already
2155 * exists, then it is given as fp; otherwise have to make a chain.
2156 *
2157 * The IP header is *NOT* adjusted out of iplen (but in host byte order).
2158 */
2159 static struct mbuf *
ip_reass(struct mbuf * m)2160 ip_reass(struct mbuf *m)
2161 {
2162 struct ip *ip;
2163 struct mbuf *p, *q, *nq, *t;
2164 struct ipq *fp = NULL;
2165 struct ipqhead *head;
2166 int i, hlen, next;
2167 u_int8_t ecn, ecn0;
2168 uint32_t csum, csum_flags;
2169 uint16_t hash;
2170 struct fq_head dfq;
2171
2172 MBUFQ_INIT(&dfq); /* for deferred frees */
2173
2174 /* If maxnipq or maxfragsperpacket is 0, never accept fragments. */
2175 if (maxnipq == 0 || maxfragsperpacket == 0) {
2176 ipstat.ips_fragments++;
2177 ipstat.ips_fragdropped++;
2178 m_freem(m);
2179 if (nipq > 0) {
2180 lck_mtx_lock(&ipqlock);
2181 frag_sched_timeout(); /* purge stale fragments */
2182 lck_mtx_unlock(&ipqlock);
2183 }
2184 return NULL;
2185 }
2186
2187 ip = mtod(m, struct ip *);
2188 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2189
2190 lck_mtx_lock(&ipqlock);
2191
2192 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
2193 head = &ipq[hash];
2194
2195 /*
2196 * Look for queue of fragments
2197 * of this datagram.
2198 */
2199 TAILQ_FOREACH(fp, head, ipq_list) {
2200 if (ip->ip_id == fp->ipq_id &&
2201 ip->ip_src.s_addr == fp->ipq_src.s_addr &&
2202 ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
2203 ip->ip_p == fp->ipq_p) {
2204 goto found;
2205 }
2206 }
2207
2208 fp = NULL;
2209
2210 /*
2211 * Attempt to trim the number of allocated fragment queues if it
2212 * exceeds the administrative limit.
2213 */
2214 if ((nipq > (unsigned)maxnipq) && (maxnipq > 0)) {
2215 /*
2216 * drop something from the tail of the current queue
2217 * before proceeding further
2218 */
2219 struct ipq *fq = TAILQ_LAST(head, ipqhead);
2220 if (fq == NULL) { /* gak */
2221 for (i = 0; i < IPREASS_NHASH; i++) {
2222 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
2223 if (r) {
2224 ipstat.ips_fragtimeout += r->ipq_nfrags;
2225 frag_freef(&ipq[i], r);
2226 break;
2227 }
2228 }
2229 } else {
2230 ipstat.ips_fragtimeout += fq->ipq_nfrags;
2231 frag_freef(head, fq);
2232 }
2233 }
2234
2235 found:
2236 /*
2237 * Leverage partial checksum offload for IP fragments. Narrow down
2238 * the scope to cover only UDP without IP options, as that is the
2239 * most common case.
2240 *
2241 * Perform 1's complement adjustment of octets that got included/
2242 * excluded in the hardware-calculated checksum value. Ignore cases
2243 * where the value includes the entire IPv4 header span, as the sum
2244 * for those octets would already be 0 by the time we get here; IP
2245 * has already performed its header checksum validation. Also take
2246 * care of any trailing bytes and subtract out their partial sum.
2247 */
2248 if (ip->ip_p == IPPROTO_UDP && hlen == sizeof(struct ip) &&
2249 (m->m_pkthdr.csum_flags &
2250 (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
2251 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
2252 uint32_t start = m->m_pkthdr.csum_rx_start;
2253 int32_t trailer = (m_pktlen(m) - ip->ip_len);
2254 uint32_t swbytes = (uint32_t)trailer;
2255
2256 csum = m->m_pkthdr.csum_rx_val;
2257
2258 ASSERT(trailer >= 0);
2259 if ((start != 0 && start != hlen) || trailer != 0) {
2260 uint32_t datalen = ip->ip_len - hlen;
2261
2262 #if BYTE_ORDER != BIG_ENDIAN
2263 if (start < hlen) {
2264 HTONS(ip->ip_len);
2265 HTONS(ip->ip_off);
2266 }
2267 #endif /* BYTE_ORDER != BIG_ENDIAN */
2268 /* callee folds in sum */
2269 csum = m_adj_sum16(m, start, hlen, datalen, csum);
2270 if (hlen > start) {
2271 swbytes += (hlen - start);
2272 } else {
2273 swbytes += (start - hlen);
2274 }
2275 #if BYTE_ORDER != BIG_ENDIAN
2276 if (start < hlen) {
2277 NTOHS(ip->ip_off);
2278 NTOHS(ip->ip_len);
2279 }
2280 #endif /* BYTE_ORDER != BIG_ENDIAN */
2281 }
2282 csum_flags = m->m_pkthdr.csum_flags;
2283
2284 if (swbytes != 0) {
2285 udp_in_cksum_stats(swbytes);
2286 }
2287 if (trailer != 0) {
2288 m_adj(m, -trailer);
2289 }
2290 } else {
2291 csum = 0;
2292 csum_flags = 0;
2293 }
2294
2295 /* Invalidate checksum */
2296 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
2297
2298 ipstat.ips_fragments++;
2299
2300 /*
2301 * Adjust ip_len to not reflect header,
2302 * convert offset of this to bytes.
2303 */
2304 ip->ip_len -= hlen;
2305 if (ip->ip_off & IP_MF) {
2306 /*
2307 * Make sure that fragments have a data length
2308 * that's a non-zero multiple of 8 bytes.
2309 */
2310 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
2311 OSAddAtomic(1, &ipstat.ips_toosmall);
2312 /*
2313 * Reassembly queue may have been found if previous
2314 * fragments were valid; given that this one is bad,
2315 * we need to drop it. Make sure to set fp to NULL
2316 * if not already, since we don't want to decrement
2317 * ipq_nfrags as it doesn't include this packet.
2318 */
2319 fp = NULL;
2320 goto dropfrag;
2321 }
2322 m->m_flags |= M_FRAG;
2323 } else {
2324 /* Clear the flag in case packet comes from loopback */
2325 m->m_flags &= ~M_FRAG;
2326 }
2327 ip->ip_off <<= 3;
2328
2329 m->m_pkthdr.pkt_hdr = ip;
2330
2331 /* Previous ip_reass() started here. */
2332 /*
2333 * Presence of header sizes in mbufs
2334 * would confuse code below.
2335 */
2336 m->m_data += hlen;
2337 m->m_len -= hlen;
2338
2339 /*
2340 * If first fragment to arrive, create a reassembly queue.
2341 */
2342 if (fp == NULL) {
2343 fp = ipq_alloc(M_DONTWAIT);
2344 if (fp == NULL) {
2345 goto dropfrag;
2346 }
2347 TAILQ_INSERT_HEAD(head, fp, ipq_list);
2348 nipq++;
2349 fp->ipq_nfrags = 1;
2350 fp->ipq_ttl = IPFRAGTTL;
2351 fp->ipq_p = ip->ip_p;
2352 fp->ipq_id = ip->ip_id;
2353 fp->ipq_src = ip->ip_src;
2354 fp->ipq_dst = ip->ip_dst;
2355 fp->ipq_frags = m;
2356 m->m_nextpkt = NULL;
2357 /*
2358 * If the first fragment has valid checksum offload
2359 * info, the rest of fragments are eligible as well.
2360 */
2361 if (csum_flags != 0) {
2362 fp->ipq_csum = csum;
2363 fp->ipq_csum_flags = csum_flags;
2364 }
2365 m = NULL; /* nothing to return */
2366 goto done;
2367 } else {
2368 fp->ipq_nfrags++;
2369 }
2370
2371 #define GETIP(m) ((struct ip *)((m)->m_pkthdr.pkt_hdr))
2372
2373 /*
2374 * Handle ECN by comparing this segment with the first one;
2375 * if CE is set, do not lose CE.
2376 * drop if CE and not-ECT are mixed for the same packet.
2377 */
2378 ecn = ip->ip_tos & IPTOS_ECN_MASK;
2379 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
2380 if (ecn == IPTOS_ECN_CE) {
2381 if (ecn0 == IPTOS_ECN_NOTECT) {
2382 goto dropfrag;
2383 }
2384 if (ecn0 != IPTOS_ECN_CE) {
2385 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
2386 }
2387 }
2388 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
2389 goto dropfrag;
2390 }
2391
2392 /*
2393 * Find a segment which begins after this one does.
2394 */
2395 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2396 if (GETIP(q)->ip_off > ip->ip_off) {
2397 break;
2398 }
2399 }
2400
2401 /*
2402 * If there is a preceding segment, it may provide some of
2403 * our data already. If so, drop the data from the incoming
2404 * segment. If it provides all of our data, drop us, otherwise
2405 * stick new segment in the proper place.
2406 *
2407 * If some of the data is dropped from the preceding
2408 * segment, then it's checksum is invalidated.
2409 */
2410 if (p) {
2411 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
2412 if (i > 0) {
2413 if (i >= ip->ip_len) {
2414 goto dropfrag;
2415 }
2416 m_adj(m, i);
2417 fp->ipq_csum_flags = 0;
2418 ip->ip_off += i;
2419 ip->ip_len -= i;
2420 }
2421 m->m_nextpkt = p->m_nextpkt;
2422 p->m_nextpkt = m;
2423 } else {
2424 m->m_nextpkt = fp->ipq_frags;
2425 fp->ipq_frags = m;
2426 }
2427
2428 /*
2429 * While we overlap succeeding segments trim them or,
2430 * if they are completely covered, dequeue them.
2431 */
2432 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
2433 q = nq) {
2434 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
2435 if (i < GETIP(q)->ip_len) {
2436 GETIP(q)->ip_len -= i;
2437 GETIP(q)->ip_off += i;
2438 m_adj(q, i);
2439 fp->ipq_csum_flags = 0;
2440 break;
2441 }
2442 nq = q->m_nextpkt;
2443 m->m_nextpkt = nq;
2444 ipstat.ips_fragdropped++;
2445 fp->ipq_nfrags--;
2446 /* defer freeing until after lock is dropped */
2447 MBUFQ_ENQUEUE(&dfq, q);
2448 }
2449
2450 /*
2451 * If this fragment contains similar checksum offload info
2452 * as that of the existing ones, accumulate checksum. Otherwise,
2453 * invalidate checksum offload info for the entire datagram.
2454 */
2455 if (csum_flags != 0 && csum_flags == fp->ipq_csum_flags) {
2456 fp->ipq_csum += csum;
2457 } else if (fp->ipq_csum_flags != 0) {
2458 fp->ipq_csum_flags = 0;
2459 }
2460
2461
2462 /*
2463 * Check for complete reassembly and perform frag per packet
2464 * limiting.
2465 *
2466 * Frag limiting is performed here so that the nth frag has
2467 * a chance to complete the packet before we drop the packet.
2468 * As a result, n+1 frags are actually allowed per packet, but
2469 * only n will ever be stored. (n = maxfragsperpacket.)
2470 *
2471 */
2472 next = 0;
2473 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2474 if (GETIP(q)->ip_off != next) {
2475 if (fp->ipq_nfrags > maxfragsperpacket) {
2476 ipstat.ips_fragdropped += fp->ipq_nfrags;
2477 frag_freef(head, fp);
2478 }
2479 m = NULL; /* nothing to return */
2480 goto done;
2481 }
2482 next += GETIP(q)->ip_len;
2483 }
2484 /* Make sure the last packet didn't have the IP_MF flag */
2485 if (p->m_flags & M_FRAG) {
2486 if (fp->ipq_nfrags > maxfragsperpacket) {
2487 ipstat.ips_fragdropped += fp->ipq_nfrags;
2488 frag_freef(head, fp);
2489 }
2490 m = NULL; /* nothing to return */
2491 goto done;
2492 }
2493
2494 /*
2495 * Reassembly is complete. Make sure the packet is a sane size.
2496 */
2497 q = fp->ipq_frags;
2498 ip = GETIP(q);
2499 if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
2500 ipstat.ips_toolong++;
2501 ipstat.ips_fragdropped += fp->ipq_nfrags;
2502 frag_freef(head, fp);
2503 m = NULL; /* nothing to return */
2504 goto done;
2505 }
2506
2507 /*
2508 * Concatenate fragments.
2509 */
2510 m = q;
2511 t = m->m_next;
2512 m->m_next = NULL;
2513 m_cat(m, t);
2514 nq = q->m_nextpkt;
2515 q->m_nextpkt = NULL;
2516 for (q = nq; q != NULL; q = nq) {
2517 nq = q->m_nextpkt;
2518 q->m_nextpkt = NULL;
2519 m_cat(m, q);
2520 }
2521
2522 /*
2523 * Store partial hardware checksum info from the fragment queue;
2524 * the receive start offset is set to 20 bytes (see code at the
2525 * top of this routine.)
2526 */
2527 if (fp->ipq_csum_flags != 0) {
2528 csum = fp->ipq_csum;
2529
2530 ADDCARRY(csum);
2531
2532 m->m_pkthdr.csum_rx_val = (uint16_t)csum;
2533 m->m_pkthdr.csum_rx_start = sizeof(struct ip);
2534 m->m_pkthdr.csum_flags = fp->ipq_csum_flags;
2535 } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
2536 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
2537 /* loopback checksums are always OK */
2538 m->m_pkthdr.csum_data = 0xffff;
2539 m->m_pkthdr.csum_flags =
2540 CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
2541 CSUM_IP_CHECKED | CSUM_IP_VALID;
2542 }
2543
2544 /*
2545 * Create header for new ip packet by modifying header of first
2546 * packet; dequeue and discard fragment reassembly header.
2547 * Make header visible.
2548 */
2549 ip->ip_len = (u_short)((IP_VHL_HL(ip->ip_vhl) << 2) + next);
2550 ip->ip_src = fp->ipq_src;
2551 ip->ip_dst = fp->ipq_dst;
2552
2553 fp->ipq_frags = NULL; /* return to caller as 'm' */
2554 frag_freef(head, fp);
2555 fp = NULL;
2556
2557 m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
2558 m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
2559 /* some debugging cruft by sklower, below, will go away soon */
2560 if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
2561 m_fixhdr(m);
2562 }
2563 ipstat.ips_reassembled++;
2564
2565 /* arm the purge timer if not already and if there's work to do */
2566 frag_sched_timeout();
2567 lck_mtx_unlock(&ipqlock);
2568 /* perform deferred free (if needed) now that lock is dropped */
2569 if (!MBUFQ_EMPTY(&dfq)) {
2570 MBUFQ_DRAIN(&dfq);
2571 }
2572 VERIFY(MBUFQ_EMPTY(&dfq));
2573 return m;
2574
2575 done:
2576 VERIFY(m == NULL);
2577 /* arm the purge timer if not already and if there's work to do */
2578 frag_sched_timeout();
2579 lck_mtx_unlock(&ipqlock);
2580 /* perform deferred free (if needed) */
2581 if (!MBUFQ_EMPTY(&dfq)) {
2582 MBUFQ_DRAIN(&dfq);
2583 }
2584 VERIFY(MBUFQ_EMPTY(&dfq));
2585 return NULL;
2586
2587 dropfrag:
2588 ipstat.ips_fragdropped++;
2589 if (fp != NULL) {
2590 fp->ipq_nfrags--;
2591 }
2592 /* arm the purge timer if not already and if there's work to do */
2593 frag_sched_timeout();
2594 lck_mtx_unlock(&ipqlock);
2595 m_freem(m);
2596 /* perform deferred free (if needed) */
2597 if (!MBUFQ_EMPTY(&dfq)) {
2598 MBUFQ_DRAIN(&dfq);
2599 }
2600 VERIFY(MBUFQ_EMPTY(&dfq));
2601 return NULL;
2602 #undef GETIP
2603 }
2604
2605 /*
2606 * Free a fragment reassembly header and all
2607 * associated datagrams.
2608 */
2609 static void
frag_freef(struct ipqhead * fhp,struct ipq * fp)2610 frag_freef(struct ipqhead *fhp, struct ipq *fp)
2611 {
2612 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2613
2614 fp->ipq_nfrags = 0;
2615 if (fp->ipq_frags != NULL) {
2616 m_freem_list(fp->ipq_frags);
2617 fp->ipq_frags = NULL;
2618 }
2619 TAILQ_REMOVE(fhp, fp, ipq_list);
2620 nipq--;
2621 ipq_free(fp);
2622 }
2623
2624 /*
2625 * IP reassembly timer processing
2626 */
2627 static void
frag_timeout(void * arg)2628 frag_timeout(void *arg)
2629 {
2630 #pragma unused(arg)
2631 struct ipq *fp;
2632 int i;
2633
2634 /*
2635 * Update coarse-grained networking timestamp (in sec.); the idea
2636 * is to piggy-back on the timeout callout to update the counter
2637 * returnable via net_uptime().
2638 */
2639 net_update_uptime();
2640
2641 lck_mtx_lock(&ipqlock);
2642 for (i = 0; i < IPREASS_NHASH; i++) {
2643 for (fp = TAILQ_FIRST(&ipq[i]); fp;) {
2644 struct ipq *fpp;
2645
2646 fpp = fp;
2647 fp = TAILQ_NEXT(fp, ipq_list);
2648 if (--fpp->ipq_ttl == 0) {
2649 ipstat.ips_fragtimeout += fpp->ipq_nfrags;
2650 frag_freef(&ipq[i], fpp);
2651 }
2652 }
2653 }
2654 /*
2655 * If we are over the maximum number of fragments
2656 * (due to the limit being lowered), drain off
2657 * enough to get down to the new limit.
2658 */
2659 if (maxnipq >= 0 && nipq > (unsigned)maxnipq) {
2660 for (i = 0; i < IPREASS_NHASH; i++) {
2661 while (nipq > (unsigned)maxnipq &&
2662 !TAILQ_EMPTY(&ipq[i])) {
2663 ipstat.ips_fragdropped +=
2664 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2665 frag_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
2666 }
2667 }
2668 }
2669 /* re-arm the purge timer if there's work to do */
2670 frag_timeout_run = 0;
2671 frag_sched_timeout();
2672 lck_mtx_unlock(&ipqlock);
2673 }
2674
2675 static void
frag_sched_timeout(void)2676 frag_sched_timeout(void)
2677 {
2678 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2679
2680 if (!frag_timeout_run && nipq > 0) {
2681 frag_timeout_run = 1;
2682 timeout(frag_timeout, NULL, hz);
2683 }
2684 }
2685
2686 /*
2687 * Drain off all datagram fragments.
2688 */
2689 static void
frag_drain(void)2690 frag_drain(void)
2691 {
2692 int i;
2693
2694 lck_mtx_lock(&ipqlock);
2695 for (i = 0; i < IPREASS_NHASH; i++) {
2696 while (!TAILQ_EMPTY(&ipq[i])) {
2697 ipstat.ips_fragdropped +=
2698 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2699 frag_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
2700 }
2701 }
2702 lck_mtx_unlock(&ipqlock);
2703 }
2704
2705 static struct ipq *
ipq_alloc(int how)2706 ipq_alloc(int how)
2707 {
2708 struct mbuf *t;
2709 struct ipq *fp;
2710
2711 /*
2712 * See comments in ipq_updateparams(). Keep the count separate
2713 * from nipq since the latter represents the elements already
2714 * in the reassembly queues.
2715 */
2716 if (ipq_limit > 0 && ipq_count > ipq_limit) {
2717 return NULL;
2718 }
2719
2720 t = m_get(how, MT_FTABLE);
2721 if (t != NULL) {
2722 atomic_add_32(&ipq_count, 1);
2723 fp = mtod(t, struct ipq *);
2724 bzero(fp, sizeof(*fp));
2725 } else {
2726 fp = NULL;
2727 }
2728 return fp;
2729 }
2730
2731 static void
ipq_free(struct ipq * fp)2732 ipq_free(struct ipq *fp)
2733 {
2734 (void) m_free(dtom(fp));
2735 atomic_add_32(&ipq_count, -1);
2736 }
2737
2738 /*
2739 * Drain callback
2740 */
2741 void
ip_drain(void)2742 ip_drain(void)
2743 {
2744 frag_drain(); /* fragments */
2745 in_rtqdrain(); /* protocol cloned routes */
2746 in_arpdrain(NULL); /* cloned routes: ARP */
2747 }
2748
2749 /*
2750 * Do option processing on a datagram,
2751 * possibly discarding it if bad options are encountered,
2752 * or forwarding it if source-routed.
2753 * The pass argument is used when operating in the IPSTEALTH
2754 * mode to tell what options to process:
2755 * [LS]SRR (pass 0) or the others (pass 1).
2756 * The reason for as many as two passes is that when doing IPSTEALTH,
2757 * non-routing options should be processed only if the packet is for us.
2758 * Returns 1 if packet has been forwarded/freed,
2759 * 0 if the packet should be processed further.
2760 */
2761 static int
ip_dooptions(struct mbuf * m,int pass,struct sockaddr_in * next_hop)2762 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
2763 {
2764 #pragma unused(pass)
2765 struct ip *ip = mtod(m, struct ip *);
2766 u_char *cp;
2767 struct ip_timestamp *ipt;
2768 struct in_ifaddr *ia;
2769 int opt, optlen, cnt, off, type = ICMP_PARAMPROB, forward = 0;
2770 uint8_t code = 0;
2771 struct in_addr *sin, dst;
2772 u_int32_t ntime;
2773 struct sockaddr_in ipaddr = {
2774 .sin_len = sizeof(ipaddr),
2775 .sin_family = AF_INET,
2776 .sin_port = 0,
2777 .sin_addr = { .s_addr = 0 },
2778 .sin_zero = { 0, }
2779 };
2780
2781 /* Expect 32-bit aligned data pointer on strict-align platforms */
2782 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
2783
2784 dst = ip->ip_dst;
2785 cp = (u_char *)(ip + 1);
2786 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
2787 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2788 opt = cp[IPOPT_OPTVAL];
2789 if (opt == IPOPT_EOL) {
2790 break;
2791 }
2792 if (opt == IPOPT_NOP) {
2793 optlen = 1;
2794 } else {
2795 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
2796 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2797 goto bad;
2798 }
2799 optlen = cp[IPOPT_OLEN];
2800 if (optlen < IPOPT_OLEN + sizeof(*cp) ||
2801 optlen > cnt) {
2802 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2803 goto bad;
2804 }
2805 }
2806 switch (opt) {
2807 default:
2808 break;
2809
2810 /*
2811 * Source routing with record.
2812 * Find interface with current destination address.
2813 * If none on this machine then drop if strictly routed,
2814 * or do nothing if loosely routed.
2815 * Record interface address and bring up next address
2816 * component. If strictly routed make sure next
2817 * address is on directly accessible net.
2818 */
2819 case IPOPT_LSRR:
2820 case IPOPT_SSRR:
2821 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2822 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2823 goto bad;
2824 }
2825 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2826 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2827 goto bad;
2828 }
2829 ipaddr.sin_addr = ip->ip_dst;
2830 ia = (struct in_ifaddr *)ifa_ifwithaddr(SA(&ipaddr));
2831 if (ia == NULL) {
2832 if (opt == IPOPT_SSRR) {
2833 type = ICMP_UNREACH;
2834 code = ICMP_UNREACH_SRCFAIL;
2835 goto bad;
2836 }
2837 if (!ip_dosourceroute) {
2838 goto nosourcerouting;
2839 }
2840 /*
2841 * Loose routing, and not at next destination
2842 * yet; nothing to do except forward.
2843 */
2844 break;
2845 } else {
2846 IFA_REMREF(&ia->ia_ifa);
2847 ia = NULL;
2848 }
2849 off--; /* 0 origin */
2850 if (off > optlen - (int)sizeof(struct in_addr)) {
2851 /*
2852 * End of source route. Should be for us.
2853 */
2854 if (!ip_acceptsourceroute) {
2855 goto nosourcerouting;
2856 }
2857 save_rte(cp, ip->ip_src);
2858 break;
2859 }
2860
2861 if (!ip_dosourceroute) {
2862 if (ipforwarding) {
2863 char buf[MAX_IPv4_STR_LEN];
2864 char buf2[MAX_IPv4_STR_LEN];
2865 /*
2866 * Acting as a router, so generate ICMP
2867 */
2868 nosourcerouting:
2869 log(LOG_WARNING,
2870 "attempted source route from %s "
2871 "to %s\n",
2872 inet_ntop(AF_INET, &ip->ip_src,
2873 buf, sizeof(buf)),
2874 inet_ntop(AF_INET, &ip->ip_dst,
2875 buf2, sizeof(buf2)));
2876 type = ICMP_UNREACH;
2877 code = ICMP_UNREACH_SRCFAIL;
2878 goto bad;
2879 } else {
2880 /*
2881 * Not acting as a router,
2882 * so silently drop.
2883 */
2884 OSAddAtomic(1, &ipstat.ips_cantforward);
2885 m_freem(m);
2886 return 1;
2887 }
2888 }
2889
2890 /*
2891 * locate outgoing interface
2892 */
2893 (void) memcpy(&ipaddr.sin_addr, cp + off,
2894 sizeof(ipaddr.sin_addr));
2895
2896 if (opt == IPOPT_SSRR) {
2897 #define INA struct in_ifaddr *
2898 if ((ia = (INA)ifa_ifwithdstaddr(
2899 SA(&ipaddr))) == NULL) {
2900 ia = (INA)ifa_ifwithnet(SA(&ipaddr));
2901 }
2902 } else {
2903 ia = ip_rtaddr(ipaddr.sin_addr);
2904 }
2905 if (ia == NULL) {
2906 type = ICMP_UNREACH;
2907 code = ICMP_UNREACH_SRCFAIL;
2908 goto bad;
2909 }
2910 ip->ip_dst = ipaddr.sin_addr;
2911 IFA_LOCK(&ia->ia_ifa);
2912 (void) memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
2913 sizeof(struct in_addr));
2914 IFA_UNLOCK(&ia->ia_ifa);
2915 IFA_REMREF(&ia->ia_ifa);
2916 ia = NULL;
2917 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2918 /*
2919 * Let ip_intr's mcast routing check handle mcast pkts
2920 */
2921 forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
2922 break;
2923
2924 case IPOPT_RR:
2925 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2926 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2927 goto bad;
2928 }
2929 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2930 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2931 goto bad;
2932 }
2933 /*
2934 * If no space remains, ignore.
2935 */
2936 off--; /* 0 origin */
2937 if (off > optlen - (int)sizeof(struct in_addr)) {
2938 break;
2939 }
2940 (void) memcpy(&ipaddr.sin_addr, &ip->ip_dst,
2941 sizeof(ipaddr.sin_addr));
2942 /*
2943 * locate outgoing interface; if we're the destination,
2944 * use the incoming interface (should be same).
2945 */
2946 if ((ia = (INA)ifa_ifwithaddr(SA(&ipaddr))) == NULL) {
2947 if ((ia = ip_rtaddr(ipaddr.sin_addr)) == NULL) {
2948 type = ICMP_UNREACH;
2949 code = ICMP_UNREACH_HOST;
2950 goto bad;
2951 }
2952 }
2953 IFA_LOCK(&ia->ia_ifa);
2954 (void) memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
2955 sizeof(struct in_addr));
2956 IFA_UNLOCK(&ia->ia_ifa);
2957 IFA_REMREF(&ia->ia_ifa);
2958 ia = NULL;
2959 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2960 break;
2961
2962 case IPOPT_TS:
2963 code = (uint8_t)(cp - (u_char *)ip);
2964 ipt = (struct ip_timestamp *)(void *)cp;
2965 if (ipt->ipt_len < 4 || ipt->ipt_len > 40) {
2966 code = (uint8_t)((u_char *)&ipt->ipt_len -
2967 (u_char *)ip);
2968 goto bad;
2969 }
2970 if (ipt->ipt_ptr < 5) {
2971 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2972 (u_char *)ip);
2973 goto bad;
2974 }
2975 if (ipt->ipt_ptr >
2976 ipt->ipt_len - (int)sizeof(int32_t)) {
2977 if (++ipt->ipt_oflw == 0) {
2978 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2979 (u_char *)ip);
2980 goto bad;
2981 }
2982 break;
2983 }
2984 sin = (struct in_addr *)(void *)(cp + ipt->ipt_ptr - 1);
2985 switch (ipt->ipt_flg) {
2986 case IPOPT_TS_TSONLY:
2987 break;
2988
2989 case IPOPT_TS_TSANDADDR:
2990 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
2991 sizeof(struct in_addr) > ipt->ipt_len) {
2992 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2993 (u_char *)ip);
2994 goto bad;
2995 }
2996 ipaddr.sin_addr = dst;
2997 ia = (INA)ifaof_ifpforaddr(SA(&ipaddr),
2998 m->m_pkthdr.rcvif);
2999 if (ia == NULL) {
3000 continue;
3001 }
3002 IFA_LOCK(&ia->ia_ifa);
3003 (void) memcpy(sin, &IA_SIN(ia)->sin_addr,
3004 sizeof(struct in_addr));
3005 IFA_UNLOCK(&ia->ia_ifa);
3006 ipt->ipt_ptr += sizeof(struct in_addr);
3007 IFA_REMREF(&ia->ia_ifa);
3008 ia = NULL;
3009 break;
3010
3011 case IPOPT_TS_PRESPEC:
3012 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
3013 sizeof(struct in_addr) > ipt->ipt_len) {
3014 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3015 (u_char *)ip);
3016 goto bad;
3017 }
3018 (void) memcpy(&ipaddr.sin_addr, sin,
3019 sizeof(struct in_addr));
3020 if ((ia = (struct in_ifaddr *)ifa_ifwithaddr(
3021 SA(&ipaddr))) == NULL) {
3022 continue;
3023 }
3024 IFA_REMREF(&ia->ia_ifa);
3025 ia = NULL;
3026 ipt->ipt_ptr += sizeof(struct in_addr);
3027 break;
3028
3029 default:
3030 /* XXX can't take &ipt->ipt_flg */
3031 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3032 (u_char *)ip + 1);
3033 goto bad;
3034 }
3035 ntime = iptime();
3036 (void) memcpy(cp + ipt->ipt_ptr - 1, &ntime,
3037 sizeof(n_time));
3038 ipt->ipt_ptr += sizeof(n_time);
3039 }
3040 }
3041 if (forward && ipforwarding) {
3042 ip_forward(m, 1, next_hop);
3043 return 1;
3044 }
3045 return 0;
3046 bad:
3047 icmp_error(m, type, code, 0, 0);
3048 OSAddAtomic(1, &ipstat.ips_badoptions);
3049 return 1;
3050 }
3051
3052 /*
3053 * Check for the presence of the IP Router Alert option [RFC2113]
3054 * in the header of an IPv4 datagram.
3055 *
3056 * This call is not intended for use from the forwarding path; it is here
3057 * so that protocol domains may check for the presence of the option.
3058 * Given how FreeBSD's IPv4 stack is currently structured, the Router Alert
3059 * option does not have much relevance to the implementation, though this
3060 * may change in future.
3061 * Router alert options SHOULD be passed if running in IPSTEALTH mode and
3062 * we are not the endpoint.
3063 * Length checks on individual options should already have been peformed
3064 * by ip_dooptions() therefore they are folded under DIAGNOSTIC here.
3065 *
3066 * Return zero if not present or options are invalid, non-zero if present.
3067 */
3068 int
ip_checkrouteralert(struct mbuf * m)3069 ip_checkrouteralert(struct mbuf *m)
3070 {
3071 struct ip *ip = mtod(m, struct ip *);
3072 u_char *cp;
3073 int opt, optlen, cnt, found_ra;
3074
3075 found_ra = 0;
3076 cp = (u_char *)(ip + 1);
3077 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3078 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3079 opt = cp[IPOPT_OPTVAL];
3080 if (opt == IPOPT_EOL) {
3081 break;
3082 }
3083 if (opt == IPOPT_NOP) {
3084 optlen = 1;
3085 } else {
3086 #ifdef DIAGNOSTIC
3087 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
3088 break;
3089 }
3090 #endif
3091 optlen = cp[IPOPT_OLEN];
3092 #ifdef DIAGNOSTIC
3093 if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
3094 break;
3095 }
3096 #endif
3097 }
3098 switch (opt) {
3099 case IPOPT_RA:
3100 #ifdef DIAGNOSTIC
3101 if (optlen != IPOPT_OFFSET + sizeof(uint16_t) ||
3102 (*((uint16_t *)(void *)&cp[IPOPT_OFFSET]) != 0)) {
3103 break;
3104 } else
3105 #endif
3106 found_ra = 1;
3107 break;
3108 default:
3109 break;
3110 }
3111 }
3112
3113 return found_ra;
3114 }
3115
3116 /*
3117 * Given address of next destination (final or next hop),
3118 * return internet address info of interface to be used to get there.
3119 */
3120 struct in_ifaddr *
ip_rtaddr(struct in_addr dst)3121 ip_rtaddr(struct in_addr dst)
3122 {
3123 struct sockaddr_in *sin;
3124 struct ifaddr *rt_ifa;
3125 struct route ro;
3126
3127 bzero(&ro, sizeof(ro));
3128 sin = SIN(&ro.ro_dst);
3129 sin->sin_family = AF_INET;
3130 sin->sin_len = sizeof(*sin);
3131 sin->sin_addr = dst;
3132
3133 rtalloc_ign(&ro, RTF_PRCLONING);
3134 if (ro.ro_rt == NULL) {
3135 ROUTE_RELEASE(&ro);
3136 return NULL;
3137 }
3138
3139 RT_LOCK(ro.ro_rt);
3140 if ((rt_ifa = ro.ro_rt->rt_ifa) != NULL) {
3141 IFA_ADDREF(rt_ifa);
3142 }
3143 RT_UNLOCK(ro.ro_rt);
3144 ROUTE_RELEASE(&ro);
3145
3146 return (struct in_ifaddr *)rt_ifa;
3147 }
3148
3149 /*
3150 * Save incoming source route for use in replies,
3151 * to be picked up later by ip_srcroute if the receiver is interested.
3152 */
3153 void
save_rte(u_char * option,struct in_addr dst)3154 save_rte(u_char *option, struct in_addr dst)
3155 {
3156 unsigned olen;
3157
3158 olen = option[IPOPT_OLEN];
3159 #if DIAGNOSTIC
3160 if (ipprintfs) {
3161 printf("save_rte: olen %d\n", olen);
3162 }
3163 #endif
3164 if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst))) {
3165 return;
3166 }
3167 bcopy(option, ip_srcrt.srcopt, olen);
3168 ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
3169 ip_srcrt.dst = dst;
3170 }
3171
3172 /*
3173 * Retrieve incoming source route for use in replies,
3174 * in the same form used by setsockopt.
3175 * The first hop is placed before the options, will be removed later.
3176 */
3177 struct mbuf *
ip_srcroute(void)3178 ip_srcroute(void)
3179 {
3180 struct in_addr *p, *q;
3181 struct mbuf *m;
3182
3183 if (ip_nhops == 0) {
3184 return NULL;
3185 }
3186
3187 m = m_get(M_DONTWAIT, MT_HEADER);
3188 if (m == NULL) {
3189 return NULL;
3190 }
3191
3192 #define OPTSIZ (sizeof (ip_srcrt.nop) + sizeof (ip_srcrt.srcopt))
3193
3194 /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
3195 m->m_len = ip_nhops * sizeof(struct in_addr) +
3196 sizeof(struct in_addr) + OPTSIZ;
3197 #if DIAGNOSTIC
3198 if (ipprintfs) {
3199 printf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
3200 }
3201 #endif
3202
3203 /*
3204 * First save first hop for return route
3205 */
3206 p = &ip_srcrt.route[ip_nhops - 1];
3207 *(mtod(m, struct in_addr *)) = *p--;
3208 #if DIAGNOSTIC
3209 if (ipprintfs) {
3210 printf(" hops %lx",
3211 (u_int32_t)ntohl(mtod(m, struct in_addr *)->s_addr));
3212 }
3213 #endif
3214
3215 /*
3216 * Copy option fields and padding (nop) to mbuf.
3217 */
3218 ip_srcrt.nop = IPOPT_NOP;
3219 ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
3220 (void) memcpy(mtod(m, caddr_t) + sizeof(struct in_addr),
3221 &ip_srcrt.nop, OPTSIZ);
3222 q = (struct in_addr *)(void *)(mtod(m, caddr_t) +
3223 sizeof(struct in_addr) + OPTSIZ);
3224 #undef OPTSIZ
3225 /*
3226 * Record return path as an IP source route,
3227 * reversing the path (pointers are now aligned).
3228 */
3229 while (p >= ip_srcrt.route) {
3230 #if DIAGNOSTIC
3231 if (ipprintfs) {
3232 printf(" %lx", (u_int32_t)ntohl(q->s_addr));
3233 }
3234 #endif
3235 *q++ = *p--;
3236 }
3237 /*
3238 * Last hop goes to final destination.
3239 */
3240 *q = ip_srcrt.dst;
3241 #if DIAGNOSTIC
3242 if (ipprintfs) {
3243 printf(" %lx\n", (u_int32_t)ntohl(q->s_addr));
3244 }
3245 #endif
3246 return m;
3247 }
3248
3249 /*
3250 * Strip out IP options, at higher level protocol in the kernel.
3251 */
3252 void
ip_stripoptions(struct mbuf * m)3253 ip_stripoptions(struct mbuf *m)
3254 {
3255 int i;
3256 struct ip *ip = mtod(m, struct ip *);
3257 caddr_t opts;
3258 int olen;
3259
3260 /* Expect 32-bit aligned data pointer on strict-align platforms */
3261 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
3262
3263 /* use bcopy() since it supports overlapping range */
3264 olen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3265 opts = (caddr_t)(ip + 1);
3266 i = m->m_len - (sizeof(struct ip) + olen);
3267 bcopy(opts + olen, opts, (unsigned)i);
3268 m->m_len -= olen;
3269 if (m->m_flags & M_PKTHDR) {
3270 m->m_pkthdr.len -= olen;
3271 }
3272 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
3273
3274 /*
3275 * We expect ip_{off,len} to be in host order by now, and
3276 * that the original IP header length has been subtracted
3277 * out from ip_len. Temporarily adjust ip_len for checksum
3278 * recalculation, and restore it afterwards.
3279 */
3280 ip->ip_len += sizeof(struct ip);
3281
3282 /* recompute checksum now that IP header is smaller */
3283 #if BYTE_ORDER != BIG_ENDIAN
3284 HTONS(ip->ip_len);
3285 HTONS(ip->ip_off);
3286 #endif /* BYTE_ORDER != BIG_ENDIAN */
3287 ip->ip_sum = in_cksum_hdr(ip);
3288 #if BYTE_ORDER != BIG_ENDIAN
3289 NTOHS(ip->ip_off);
3290 NTOHS(ip->ip_len);
3291 #endif /* BYTE_ORDER != BIG_ENDIAN */
3292
3293 ip->ip_len -= sizeof(struct ip);
3294
3295 /*
3296 * Given that we've just stripped IP options from the header,
3297 * we need to adjust the start offset accordingly if this
3298 * packet had gone thru partial checksum offload.
3299 */
3300 if ((m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
3301 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
3302 if (m->m_pkthdr.csum_rx_start >= (sizeof(struct ip) + olen)) {
3303 /* most common case */
3304 m->m_pkthdr.csum_rx_start -= olen;
3305 } else {
3306 /* compute checksum in software instead */
3307 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
3308 m->m_pkthdr.csum_data = 0;
3309 ipstat.ips_adj_hwcsum_clr++;
3310 }
3311 }
3312 }
3313
3314 u_char inetctlerrmap[PRC_NCMDS] = {
3315 0, 0, 0, 0,
3316 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH,
3317 ENETUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED,
3318 EMSGSIZE, EHOSTUNREACH, 0, 0,
3319 0, 0, EHOSTUNREACH, 0,
3320 ENOPROTOOPT, ECONNREFUSED
3321 };
3322
3323 static int
3324 sysctl_ipforwarding SYSCTL_HANDLER_ARGS
3325 {
3326 #pragma unused(arg1, arg2)
3327 int i, was_ipforwarding = ipforwarding;
3328
3329 i = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
3330 if (i != 0 || req->newptr == USER_ADDR_NULL) {
3331 return i;
3332 }
3333
3334 if (was_ipforwarding && !ipforwarding) {
3335 /* clean up IPv4 forwarding cached routes */
3336 ifnet_head_lock_shared();
3337 for (i = 0; i <= if_index; i++) {
3338 struct ifnet *ifp = ifindex2ifnet[i];
3339 if (ifp != NULL) {
3340 lck_mtx_lock(&ifp->if_cached_route_lock);
3341 ROUTE_RELEASE(&ifp->if_fwd_route);
3342 bzero(&ifp->if_fwd_route,
3343 sizeof(ifp->if_fwd_route));
3344 lck_mtx_unlock(&ifp->if_cached_route_lock);
3345 }
3346 }
3347 ifnet_head_done();
3348 }
3349
3350 return 0;
3351 }
3352
3353 /*
3354 * Similar to inp_route_{copyout,copyin} routines except that these copy
3355 * out the cached IPv4 forwarding route from struct ifnet instead of the
3356 * inpcb. See comments for those routines for explanations.
3357 */
3358 static void
ip_fwd_route_copyout(struct ifnet * ifp,struct route * dst)3359 ip_fwd_route_copyout(struct ifnet *ifp, struct route *dst)
3360 {
3361 struct route *src = &ifp->if_fwd_route;
3362
3363 lck_mtx_lock_spin(&ifp->if_cached_route_lock);
3364 lck_mtx_convert_spin(&ifp->if_cached_route_lock);
3365
3366 /* Minor sanity check */
3367 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3368 panic("%s: wrong or corrupted route: %p", __func__, src);
3369 }
3370
3371 route_copyout(dst, src, sizeof(*dst));
3372
3373 lck_mtx_unlock(&ifp->if_cached_route_lock);
3374 }
3375
3376 static void
ip_fwd_route_copyin(struct ifnet * ifp,struct route * src)3377 ip_fwd_route_copyin(struct ifnet *ifp, struct route *src)
3378 {
3379 struct route *dst = &ifp->if_fwd_route;
3380
3381 lck_mtx_lock_spin(&ifp->if_cached_route_lock);
3382 lck_mtx_convert_spin(&ifp->if_cached_route_lock);
3383
3384 /* Minor sanity check */
3385 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3386 panic("%s: wrong or corrupted route: %p", __func__, src);
3387 }
3388
3389 if (ifp->if_fwd_cacheok) {
3390 route_copyin(src, dst, sizeof(*src));
3391 }
3392
3393 lck_mtx_unlock(&ifp->if_cached_route_lock);
3394 }
3395
3396 /*
3397 * Forward a packet. If some error occurs return the sender
3398 * an icmp packet. Note we can't always generate a meaningful
3399 * icmp message because icmp doesn't have a large enough repertoire
3400 * of codes and types.
3401 *
3402 * If not forwarding, just drop the packet. This could be confusing
3403 * if ipforwarding was zero but some routing protocol was advancing
3404 * us as a gateway to somewhere. However, we must let the routing
3405 * protocol deal with that.
3406 *
3407 * The srcrt parameter indicates whether the packet is being forwarded
3408 * via a source route.
3409 */
3410 static void
ip_forward(struct mbuf * m,int srcrt,struct sockaddr_in * next_hop)3411 ip_forward(struct mbuf *m, int srcrt, struct sockaddr_in *next_hop)
3412 {
3413 #pragma unused(next_hop)
3414 struct ip *ip = mtod(m, struct ip *);
3415 struct sockaddr_in *sin;
3416 struct rtentry *rt;
3417 struct route fwd_rt;
3418 int error, type = 0, code = 0;
3419 struct mbuf *mcopy;
3420 n_long dest;
3421 struct in_addr pkt_dst;
3422 u_int32_t nextmtu = 0, len;
3423 struct ip_out_args ipoa;
3424 struct ifnet *rcvifp = m->m_pkthdr.rcvif;
3425
3426 bzero(&ipoa, sizeof(ipoa));
3427 ipoa.ipoa_boundif = IFSCOPE_NONE;
3428 ipoa.ipoa_sotc = SO_TC_UNSPEC;
3429 ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
3430
3431 #if IPSEC
3432 struct secpolicy *sp = NULL;
3433 int ipsecerror;
3434 #endif /* IPSEC */
3435 #if PF
3436 struct pf_mtag *pf_mtag;
3437 #endif /* PF */
3438
3439 dest = 0;
3440 pkt_dst = ip->ip_dst;
3441
3442 #if DIAGNOSTIC
3443 if (ipprintfs) {
3444 printf("forward: src %lx dst %lx ttl %x\n",
3445 (u_int32_t)ip->ip_src.s_addr, (u_int32_t)pkt_dst.s_addr,
3446 ip->ip_ttl);
3447 }
3448 #endif
3449
3450 if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
3451 OSAddAtomic(1, &ipstat.ips_cantforward);
3452 m_freem(m);
3453 return;
3454 }
3455 #if IPSTEALTH
3456 if (!ipstealth) {
3457 #endif /* IPSTEALTH */
3458 if (ip->ip_ttl <= IPTTLDEC) {
3459 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
3460 dest, 0);
3461 return;
3462 }
3463 #if IPSTEALTH
3464 }
3465 #endif /* IPSTEALTH */
3466
3467 #if PF
3468 pf_mtag = pf_find_mtag(m);
3469 if (pf_mtag != NULL && pf_mtag->pftag_rtableid != IFSCOPE_NONE) {
3470 ipoa.ipoa_boundif = pf_mtag->pftag_rtableid;
3471 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
3472 }
3473 #endif /* PF */
3474
3475 ip_fwd_route_copyout(rcvifp, &fwd_rt);
3476
3477 sin = SIN(&fwd_rt.ro_dst);
3478 if (ROUTE_UNUSABLE(&fwd_rt) || pkt_dst.s_addr != sin->sin_addr.s_addr) {
3479 ROUTE_RELEASE(&fwd_rt);
3480
3481 sin->sin_family = AF_INET;
3482 sin->sin_len = sizeof(*sin);
3483 sin->sin_addr = pkt_dst;
3484
3485 rtalloc_scoped_ign(&fwd_rt, RTF_PRCLONING, ipoa.ipoa_boundif);
3486 if (fwd_rt.ro_rt == NULL) {
3487 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
3488 goto done;
3489 }
3490 }
3491 rt = fwd_rt.ro_rt;
3492
3493 /*
3494 * Save the IP header and at most 8 bytes of the payload,
3495 * in case we need to generate an ICMP message to the src.
3496 *
3497 * We don't use m_copy() because it might return a reference
3498 * to a shared cluster. Both this function and ip_output()
3499 * assume exclusive access to the IP header in `m', so any
3500 * data in a cluster may change before we reach icmp_error().
3501 */
3502 MGET(mcopy, M_DONTWAIT, m->m_type);
3503 if (mcopy != NULL && m_dup_pkthdr(mcopy, m, M_DONTWAIT) == 0) {
3504 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
3505 (int)ip->ip_len);
3506 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
3507 }
3508
3509 #if IPSTEALTH
3510 if (!ipstealth) {
3511 #endif /* IPSTEALTH */
3512 ip->ip_ttl -= IPTTLDEC;
3513 #if IPSTEALTH
3514 }
3515 #endif /* IPSTEALTH */
3516
3517 /*
3518 * If forwarding packet using same interface that it came in on,
3519 * perhaps should send a redirect to sender to shortcut a hop.
3520 * Only send redirect if source is sending directly to us,
3521 * and if packet was not source routed (or has any options).
3522 * Also, don't send redirect if forwarding using a default route
3523 * or a route modified by a redirect.
3524 */
3525 RT_LOCK_SPIN(rt);
3526 if (rt->rt_ifp == m->m_pkthdr.rcvif &&
3527 !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
3528 satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
3529 ipsendredirects && !srcrt && rt->rt_ifa != NULL) {
3530 struct in_ifaddr *ia = (struct in_ifaddr *)rt->rt_ifa;
3531 u_int32_t src = ntohl(ip->ip_src.s_addr);
3532
3533 /* Become a regular mutex */
3534 RT_CONVERT_LOCK(rt);
3535 IFA_LOCK_SPIN(&ia->ia_ifa);
3536 if ((src & ia->ia_subnetmask) == ia->ia_subnet) {
3537 if (rt->rt_flags & RTF_GATEWAY) {
3538 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
3539 } else {
3540 dest = pkt_dst.s_addr;
3541 }
3542 /*
3543 * Router requirements says to only send
3544 * host redirects.
3545 */
3546 type = ICMP_REDIRECT;
3547 code = ICMP_REDIRECT_HOST;
3548 #if DIAGNOSTIC
3549 if (ipprintfs) {
3550 printf("redirect (%d) to %lx\n", code,
3551 (u_int32_t)dest);
3552 }
3553 #endif
3554 }
3555 IFA_UNLOCK(&ia->ia_ifa);
3556 }
3557 RT_UNLOCK(rt);
3558
3559
3560 /* Mark this packet as being forwarded from another interface */
3561 m->m_pkthdr.pkt_flags |= PKTF_FORWARDED;
3562 len = m_pktlen(m);
3563
3564 error = ip_output(m, NULL, &fwd_rt, IP_FORWARDING | IP_OUTARGS,
3565 NULL, &ipoa);
3566
3567 /* Refresh rt since the route could have changed while in IP */
3568 rt = fwd_rt.ro_rt;
3569
3570 if (error != 0) {
3571 OSAddAtomic(1, &ipstat.ips_cantforward);
3572 } else {
3573 /*
3574 * Increment stats on the source interface; the ones
3575 * for destination interface has been taken care of
3576 * during output above by virtue of PKTF_FORWARDED.
3577 */
3578 rcvifp->if_fpackets++;
3579 rcvifp->if_fbytes += len;
3580
3581 OSAddAtomic(1, &ipstat.ips_forward);
3582 if (type != 0) {
3583 OSAddAtomic(1, &ipstat.ips_redirectsent);
3584 } else {
3585 if (mcopy != NULL) {
3586 /*
3587 * If we didn't have to go thru ipflow and
3588 * the packet was successfully consumed by
3589 * ip_output, the mcopy is rather a waste;
3590 * this could be further optimized.
3591 */
3592 m_freem(mcopy);
3593 }
3594 goto done;
3595 }
3596 }
3597 if (mcopy == NULL) {
3598 goto done;
3599 }
3600
3601 switch (error) {
3602 case 0: /* forwarded, but need redirect */
3603 /* type, code set above */
3604 break;
3605
3606 case ENETUNREACH: /* shouldn't happen, checked above */
3607 case EHOSTUNREACH:
3608 case ENETDOWN:
3609 case EHOSTDOWN:
3610 default:
3611 type = ICMP_UNREACH;
3612 code = ICMP_UNREACH_HOST;
3613 break;
3614
3615 case EMSGSIZE:
3616 type = ICMP_UNREACH;
3617 code = ICMP_UNREACH_NEEDFRAG;
3618
3619 if (rt == NULL) {
3620 break;
3621 } else {
3622 RT_LOCK_SPIN(rt);
3623 if (rt->rt_ifp != NULL) {
3624 nextmtu = rt->rt_ifp->if_mtu;
3625 }
3626 RT_UNLOCK(rt);
3627 }
3628 #ifdef IPSEC
3629 if (ipsec_bypass) {
3630 break;
3631 }
3632
3633 /*
3634 * If the packet is routed over IPsec tunnel, tell the
3635 * originator the tunnel MTU.
3636 * tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
3637 * XXX quickhack!!!
3638 */
3639 sp = ipsec4_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
3640 IP_FORWARDING, &ipsecerror);
3641
3642 if (sp == NULL) {
3643 break;
3644 }
3645
3646 /*
3647 * find the correct route for outer IPv4
3648 * header, compute tunnel MTU.
3649 */
3650 nextmtu = 0;
3651
3652 if (sp->req != NULL &&
3653 sp->req->saidx.mode == IPSEC_MODE_TUNNEL) {
3654 struct secasindex saidx;
3655 struct secasvar *sav;
3656 struct route *ro;
3657 struct ip *ipm;
3658 size_t ipsechdr;
3659
3660 /* count IPsec header size */
3661 ipsechdr = ipsec_hdrsiz(sp);
3662
3663 ipm = mtod(mcopy, struct ip *);
3664 bcopy(&sp->req->saidx, &saidx, sizeof(saidx));
3665 saidx.mode = sp->req->saidx.mode;
3666 saidx.reqid = sp->req->saidx.reqid;
3667 sin = SIN(&saidx.src);
3668 if (sin->sin_len == 0) {
3669 sin->sin_len = sizeof(*sin);
3670 sin->sin_family = AF_INET;
3671 sin->sin_port = IPSEC_PORT_ANY;
3672 bcopy(&ipm->ip_src, &sin->sin_addr,
3673 sizeof(sin->sin_addr));
3674 }
3675 sin = SIN(&saidx.dst);
3676 if (sin->sin_len == 0) {
3677 sin->sin_len = sizeof(*sin);
3678 sin->sin_family = AF_INET;
3679 sin->sin_port = IPSEC_PORT_ANY;
3680 bcopy(&ipm->ip_dst, &sin->sin_addr,
3681 sizeof(sin->sin_addr));
3682 }
3683 sav = key_allocsa_policy(&saidx);
3684 if (sav != NULL) {
3685 lck_mtx_lock(sadb_mutex);
3686 if (sav->sah != NULL) {
3687 ro = (struct route *)&sav->sah->sa_route;
3688 if (ro->ro_rt != NULL) {
3689 RT_LOCK(ro->ro_rt);
3690 if (ro->ro_rt->rt_ifp != NULL) {
3691 nextmtu = ro->ro_rt->
3692 rt_ifp->if_mtu;
3693 nextmtu -= ipsechdr;
3694 }
3695 RT_UNLOCK(ro->ro_rt);
3696 }
3697 }
3698 key_freesav(sav, KEY_SADB_LOCKED);
3699 lck_mtx_unlock(sadb_mutex);
3700 }
3701 }
3702 key_freesp(sp, KEY_SADB_UNLOCKED);
3703 #endif /* IPSEC */
3704 break;
3705
3706 case ENOBUFS:
3707 /*
3708 * A router should not generate ICMP_SOURCEQUENCH as
3709 * required in RFC1812 Requirements for IP Version 4 Routers.
3710 * Source quench could be a big problem under DoS attacks,
3711 * or if the underlying interface is rate-limited.
3712 * Those who need source quench packets may re-enable them
3713 * via the net.inet.ip.sendsourcequench sysctl.
3714 */
3715 if (ip_sendsourcequench == 0) {
3716 m_freem(mcopy);
3717 goto done;
3718 } else {
3719 type = ICMP_SOURCEQUENCH;
3720 code = 0;
3721 }
3722 break;
3723
3724 case EACCES:
3725 m_freem(mcopy);
3726 goto done;
3727 }
3728
3729 if (type == ICMP_UNREACH && code == ICMP_UNREACH_NEEDFRAG) {
3730 OSAddAtomic(1, &ipstat.ips_cantfrag);
3731 }
3732
3733 icmp_error(mcopy, type, code, dest, nextmtu);
3734 done:
3735 ip_fwd_route_copyin(rcvifp, &fwd_rt);
3736 }
3737
3738 int
ip_savecontrol(struct inpcb * inp,struct mbuf ** mp,struct ip * ip,struct mbuf * m)3739 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
3740 struct mbuf *m)
3741 {
3742 *mp = NULL;
3743 if (inp->inp_socket->so_options & SO_TIMESTAMP) {
3744 struct timeval tv;
3745
3746 getmicrotime(&tv);
3747 mp = sbcreatecontrol_mbuf((caddr_t)&tv, sizeof(tv),
3748 SCM_TIMESTAMP, SOL_SOCKET, mp);
3749 if (*mp == NULL) {
3750 goto no_mbufs;
3751 }
3752 }
3753 if (inp->inp_socket->so_options & SO_TIMESTAMP_MONOTONIC) {
3754 uint64_t time;
3755
3756 time = mach_absolute_time();
3757 mp = sbcreatecontrol_mbuf((caddr_t)&time, sizeof(time),
3758 SCM_TIMESTAMP_MONOTONIC, SOL_SOCKET, mp);
3759 if (*mp == NULL) {
3760 goto no_mbufs;
3761 }
3762 }
3763 if (inp->inp_socket->so_options & SO_TIMESTAMP_CONTINUOUS) {
3764 uint64_t time;
3765
3766 time = mach_continuous_time();
3767 mp = sbcreatecontrol_mbuf((caddr_t)&time, sizeof(time),
3768 SCM_TIMESTAMP_CONTINUOUS, SOL_SOCKET, mp);
3769 if (*mp == NULL) {
3770 goto no_mbufs;
3771 }
3772 }
3773 if (inp->inp_socket->so_flags & SOF_RECV_TRAFFIC_CLASS) {
3774 int tc = m_get_traffic_class(m);
3775
3776 mp = sbcreatecontrol_mbuf((caddr_t)&tc, sizeof(tc),
3777 SO_TRAFFIC_CLASS, SOL_SOCKET, mp);
3778 if (*mp == NULL) {
3779 goto no_mbufs;
3780 }
3781 }
3782 if ((inp->inp_socket->so_flags & SOF_RECV_WAKE_PKT) &&
3783 (m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
3784 int flag = 1;
3785
3786 mp = sbcreatecontrol_mbuf((caddr_t)&flag, sizeof(flag),
3787 SO_RECV_WAKE_PKT, SOL_SOCKET, mp);
3788 if (*mp == NULL) {
3789 goto no_mbufs;
3790 }
3791 }
3792
3793 if (inp->inp_flags & INP_RECVDSTADDR || SOFLOW_ENABLED(inp->inp_socket)) {
3794 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_dst,
3795 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP, mp);
3796 if (*mp == NULL) {
3797 goto no_mbufs;
3798 }
3799 }
3800 #ifdef notyet
3801 /*
3802 * XXX
3803 * Moving these out of udp_input() made them even more broken
3804 * than they already were.
3805 */
3806 /* options were tossed already */
3807 if (inp->inp_flags & INP_RECVOPTS) {
3808 mp = sbcreatecontrol_mbuf((caddr_t)opts_deleted_above,
3809 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP, mp);
3810 if (*mp == NULL) {
3811 goto no_mbufs;
3812 }
3813 }
3814 /* ip_srcroute doesn't do what we want here, need to fix */
3815 if (inp->inp_flags & INP_RECVRETOPTS) {
3816 mp = sbcreatecontrol_mbuf((caddr_t)ip_srcroute(),
3817 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP, mp);
3818 if (*mp == NULL) {
3819 goto no_mbufs;
3820 }
3821 }
3822 #endif /* notyet */
3823 if (inp->inp_flags & INP_RECVIF) {
3824 struct ifnet *ifp;
3825 uint8_t sdlbuf[SOCK_MAXADDRLEN + 1];
3826 struct sockaddr_dl *sdl2 = SDL(&sdlbuf);
3827
3828 /*
3829 * Make sure to accomodate the largest possible
3830 * size of SA(if_lladdr)->sa_len.
3831 */
3832 _CASSERT(sizeof(sdlbuf) == (SOCK_MAXADDRLEN + 1));
3833
3834 ifnet_head_lock_shared();
3835 if ((ifp = m->m_pkthdr.rcvif) != NULL &&
3836 ifp->if_index && (ifp->if_index <= if_index)) {
3837 struct ifaddr *ifa = ifnet_addrs[ifp->if_index - 1];
3838 struct sockaddr_dl *sdp;
3839
3840 if (!ifa || !ifa->ifa_addr) {
3841 goto makedummy;
3842 }
3843
3844 IFA_LOCK_SPIN(ifa);
3845 sdp = SDL(ifa->ifa_addr);
3846 /*
3847 * Change our mind and don't try copy.
3848 */
3849 if (sdp->sdl_family != AF_LINK) {
3850 IFA_UNLOCK(ifa);
3851 goto makedummy;
3852 }
3853 /* the above _CASSERT ensures sdl_len fits in sdlbuf */
3854 bcopy(sdp, sdl2, sdp->sdl_len);
3855 IFA_UNLOCK(ifa);
3856 } else {
3857 makedummy:
3858 sdl2->sdl_len =
3859 offsetof(struct sockaddr_dl, sdl_data[0]);
3860 sdl2->sdl_family = AF_LINK;
3861 sdl2->sdl_index = 0;
3862 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
3863 }
3864 ifnet_head_done();
3865 mp = sbcreatecontrol_mbuf((caddr_t)sdl2, sdl2->sdl_len,
3866 IP_RECVIF, IPPROTO_IP, mp);
3867 if (*mp == NULL) {
3868 goto no_mbufs;
3869 }
3870 }
3871 if (inp->inp_flags & INP_RECVTTL) {
3872 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_ttl,
3873 sizeof(ip->ip_ttl), IP_RECVTTL, IPPROTO_IP, mp);
3874 if (*mp == NULL) {
3875 goto no_mbufs;
3876 }
3877 }
3878 if (inp->inp_flags & INP_PKTINFO) {
3879 struct in_pktinfo pi;
3880
3881 bzero(&pi, sizeof(struct in_pktinfo));
3882 bcopy(&ip->ip_dst, &pi.ipi_addr, sizeof(struct in_addr));
3883 pi.ipi_ifindex = (m != NULL && m->m_pkthdr.rcvif != NULL) ?
3884 m->m_pkthdr.rcvif->if_index : 0;
3885
3886 mp = sbcreatecontrol_mbuf((caddr_t)&pi,
3887 sizeof(struct in_pktinfo), IP_RECVPKTINFO, IPPROTO_IP, mp);
3888 if (*mp == NULL) {
3889 goto no_mbufs;
3890 }
3891 }
3892 if (inp->inp_flags & INP_RECVTOS) {
3893 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_tos,
3894 sizeof(u_char), IP_RECVTOS, IPPROTO_IP, mp);
3895 if (*mp == NULL) {
3896 goto no_mbufs;
3897 }
3898 }
3899 return 0;
3900
3901 no_mbufs:
3902 ipstat.ips_pktdropcntrl++;
3903 return ENOBUFS;
3904 }
3905
3906 static inline u_short
ip_cksum(struct mbuf * m,int hlen)3907 ip_cksum(struct mbuf *m, int hlen)
3908 {
3909 u_short sum;
3910
3911 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
3912 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
3913 } else if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
3914 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
3915 /*
3916 * The packet arrived on an interface which isn't capable
3917 * of performing IP header checksum; compute it now.
3918 */
3919 sum = ip_cksum_hdr_in(m, hlen);
3920 } else {
3921 sum = 0;
3922 m->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
3923 CSUM_IP_CHECKED | CSUM_IP_VALID);
3924 m->m_pkthdr.csum_data = 0xffff;
3925 }
3926
3927 if (sum != 0) {
3928 OSAddAtomic(1, &ipstat.ips_badsum);
3929 }
3930
3931 return sum;
3932 }
3933
3934 static int
3935 ip_getstat SYSCTL_HANDLER_ARGS
3936 {
3937 #pragma unused(oidp, arg1, arg2)
3938 if (req->oldptr == USER_ADDR_NULL) {
3939 req->oldlen = (size_t)sizeof(struct ipstat);
3940 }
3941
3942 return SYSCTL_OUT(req, &ipstat, MIN(sizeof(ipstat), req->oldlen));
3943 }
3944
3945 void
ip_setsrcifaddr_info(struct mbuf * m,uint16_t src_idx,struct in_ifaddr * ia)3946 ip_setsrcifaddr_info(struct mbuf *m, uint16_t src_idx, struct in_ifaddr *ia)
3947 {
3948 VERIFY(m->m_flags & M_PKTHDR);
3949
3950 /*
3951 * If the source ifaddr is specified, pick up the information
3952 * from there; otherwise just grab the passed-in ifindex as the
3953 * caller may not have the ifaddr available.
3954 */
3955 if (ia != NULL) {
3956 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3957 m->m_pkthdr.src_ifindex = ia->ia_ifp->if_index;
3958 } else {
3959 m->m_pkthdr.src_ifindex = src_idx;
3960 if (src_idx != 0) {
3961 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3962 }
3963 }
3964 }
3965
3966 void
ip_setdstifaddr_info(struct mbuf * m,uint16_t dst_idx,struct in_ifaddr * ia)3967 ip_setdstifaddr_info(struct mbuf *m, uint16_t dst_idx, struct in_ifaddr *ia)
3968 {
3969 VERIFY(m->m_flags & M_PKTHDR);
3970
3971 /*
3972 * If the destination ifaddr is specified, pick up the information
3973 * from there; otherwise just grab the passed-in ifindex as the
3974 * caller may not have the ifaddr available.
3975 */
3976 if (ia != NULL) {
3977 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3978 m->m_pkthdr.dst_ifindex = ia->ia_ifp->if_index;
3979 } else {
3980 m->m_pkthdr.dst_ifindex = dst_idx;
3981 if (dst_idx != 0) {
3982 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3983 }
3984 }
3985 }
3986
3987 int
ip_getsrcifaddr_info(struct mbuf * m,uint32_t * src_idx,uint32_t * iaf)3988 ip_getsrcifaddr_info(struct mbuf *m, uint32_t *src_idx, uint32_t *iaf)
3989 {
3990 VERIFY(m->m_flags & M_PKTHDR);
3991
3992 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
3993 return -1;
3994 }
3995
3996 if (src_idx != NULL) {
3997 *src_idx = m->m_pkthdr.src_ifindex;
3998 }
3999
4000 if (iaf != NULL) {
4001 *iaf = 0;
4002 }
4003
4004 return 0;
4005 }
4006
4007 int
ip_getdstifaddr_info(struct mbuf * m,uint32_t * dst_idx,uint32_t * iaf)4008 ip_getdstifaddr_info(struct mbuf *m, uint32_t *dst_idx, uint32_t *iaf)
4009 {
4010 VERIFY(m->m_flags & M_PKTHDR);
4011
4012 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
4013 return -1;
4014 }
4015
4016 if (dst_idx != NULL) {
4017 *dst_idx = m->m_pkthdr.dst_ifindex;
4018 }
4019
4020 if (iaf != NULL) {
4021 *iaf = 0;
4022 }
4023
4024 return 0;
4025 }
4026
4027 /*
4028 * Protocol input handler for IPPROTO_GRE.
4029 */
4030 void
gre_input(struct mbuf * m,int off)4031 gre_input(struct mbuf *m, int off)
4032 {
4033 gre_input_func_t fn = gre_input_func;
4034
4035 /*
4036 * If there is a registered GRE input handler, pass mbuf to it.
4037 */
4038 if (fn != NULL) {
4039 lck_mtx_unlock(inet_domain_mutex);
4040 m = fn(m, off, (mtod(m, struct ip *))->ip_p);
4041 lck_mtx_lock(inet_domain_mutex);
4042 }
4043
4044 /*
4045 * If no matching tunnel that is up is found, we inject
4046 * the mbuf to raw ip socket to see if anyone picks it up.
4047 */
4048 if (m != NULL) {
4049 rip_input(m, off);
4050 }
4051 }
4052
4053 /*
4054 * Private KPI for PPP/PPTP.
4055 */
4056 int
ip_gre_register_input(gre_input_func_t fn)4057 ip_gre_register_input(gre_input_func_t fn)
4058 {
4059 lck_mtx_lock(inet_domain_mutex);
4060 gre_input_func = fn;
4061 lck_mtx_unlock(inet_domain_mutex);
4062
4063 return 0;
4064 }
4065
4066 #if (DEBUG || DEVELOPMENT)
4067 static int
4068 sysctl_reset_ip_input_stats SYSCTL_HANDLER_ARGS
4069 {
4070 #pragma unused(arg1, arg2)
4071 int error, i;
4072
4073 i = ip_input_measure;
4074 error = sysctl_handle_int(oidp, &i, 0, req);
4075 if (error || req->newptr == USER_ADDR_NULL) {
4076 goto done;
4077 }
4078 /* impose bounds */
4079 if (i < 0 || i > 1) {
4080 error = EINVAL;
4081 goto done;
4082 }
4083 if (ip_input_measure != i && i == 1) {
4084 net_perf_initialize(&net_perf, ip_input_measure_bins);
4085 }
4086 ip_input_measure = i;
4087 done:
4088 return error;
4089 }
4090
4091 static int
4092 sysctl_ip_input_measure_bins SYSCTL_HANDLER_ARGS
4093 {
4094 #pragma unused(arg1, arg2)
4095 int error;
4096 uint64_t i;
4097
4098 i = ip_input_measure_bins;
4099 error = sysctl_handle_quad(oidp, &i, 0, req);
4100 if (error || req->newptr == USER_ADDR_NULL) {
4101 goto done;
4102 }
4103 /* validate data */
4104 if (!net_perf_validate_bins(i)) {
4105 error = EINVAL;
4106 goto done;
4107 }
4108 ip_input_measure_bins = i;
4109 done:
4110 return error;
4111 }
4112
4113 static int
4114 sysctl_ip_input_getperf SYSCTL_HANDLER_ARGS
4115 {
4116 #pragma unused(oidp, arg1, arg2)
4117 if (req->oldptr == USER_ADDR_NULL) {
4118 req->oldlen = (size_t)sizeof(struct ipstat);
4119 }
4120
4121 return SYSCTL_OUT(req, &net_perf, MIN(sizeof(net_perf), req->oldlen));
4122 }
4123 #endif /* (DEBUG || DEVELOPMENT) */
4124
4125 static int
4126 sysctl_ip_checkinterface SYSCTL_HANDLER_ARGS
4127 {
4128 #pragma unused(arg1, arg2)
4129 int error, i;
4130
4131 i = ip_checkinterface;
4132 error = sysctl_handle_int(oidp, &i, 0, req);
4133 if (error != 0 || req->newptr == USER_ADDR_NULL) {
4134 return error;
4135 }
4136
4137 switch (i) {
4138 case IP_CHECKINTERFACE_WEAK_ES:
4139 case IP_CHECKINTERFACE_HYBRID_ES:
4140 case IP_CHECKINTERFACE_STRONG_ES:
4141 if (ip_checkinterface != i) {
4142 ip_checkinterface = i;
4143 os_log(OS_LOG_DEFAULT, "%s: ip_checkinterface is now %d\n",
4144 __func__, ip_checkinterface);
4145 }
4146 break;
4147 default:
4148 error = EINVAL;
4149 break;
4150 }
4151 return error;
4152 }
4153