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 uint32_t inaddr_nhash; /* hash table size */
331 static uint32_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 uint32_t
inaddr_hashval(uint32_t key)597 inaddr_hashval(uint32_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 struct in_ifaddrhashhead *
inaddr_hashlookup(uint32_t key)612 inaddr_hashlookup(uint32_t key)
613 {
614 return &in_ifaddrhashtbl[inaddr_hashval(key)];
615 }
616
617 __private_extern__ void
ip_proto_dispatch_in(struct mbuf * m,int hlen,u_int8_t proto,ipfilter_t inject_ipfref)618 ip_proto_dispatch_in(struct mbuf *m, int hlen, u_int8_t proto,
619 ipfilter_t inject_ipfref)
620 {
621 struct ipfilter *filter;
622 int seen = (inject_ipfref == NULL);
623 int changed_header = 0;
624 struct ip *ip;
625 void (*pr_input)(struct mbuf *, int len);
626
627 if (!TAILQ_EMPTY(&ipv4_filters)) {
628 ipf_ref();
629 TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
630 if (seen == 0) {
631 if ((struct ipfilter *)inject_ipfref == filter) {
632 seen = 1;
633 }
634 } else if (filter->ipf_filter.ipf_input) {
635 errno_t result;
636
637 if (changed_header == 0) {
638 /*
639 * Perform IP header alignment fixup,
640 * if needed, before passing packet
641 * into filter(s).
642 */
643 IP_HDR_ALIGNMENT_FIXUP(m,
644 m->m_pkthdr.rcvif, ipf_unref());
645
646 /* ipf_unref() already called */
647 if (m == NULL) {
648 return;
649 }
650
651 changed_header = 1;
652 ip = mtod(m, struct ip *);
653 ip->ip_len = htons(ip->ip_len + (uint16_t)hlen);
654 ip->ip_off = htons(ip->ip_off);
655 ip->ip_sum = 0;
656 ip->ip_sum = ip_cksum_hdr_in(m, hlen);
657 }
658 result = filter->ipf_filter.ipf_input(
659 filter->ipf_filter.cookie, (mbuf_t *)&m,
660 hlen, proto);
661 if (result == EJUSTRETURN) {
662 ipf_unref();
663 return;
664 }
665 if (result != 0) {
666 ipf_unref();
667 m_freem(m);
668 return;
669 }
670 }
671 }
672 ipf_unref();
673 }
674
675 /* Perform IP header alignment fixup (post-filters), if needed */
676 IP_HDR_ALIGNMENT_FIXUP(m, m->m_pkthdr.rcvif, return );
677
678 ip = mtod(m, struct ip *);
679
680 if (changed_header) {
681 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
682 ip->ip_off = ntohs(ip->ip_off);
683 }
684
685 /*
686 * If there isn't a specific lock for the protocol
687 * we're about to call, use the generic lock for AF_INET.
688 * otherwise let the protocol deal with its own locking
689 */
690 if ((pr_input = ip_protox[ip->ip_p]->pr_input) == NULL) {
691 m_freem(m);
692 } else if (!(ip_protox[ip->ip_p]->pr_flags & PR_PROTOLOCK)) {
693 lck_mtx_lock(inet_domain_mutex);
694 pr_input(m, hlen);
695 lck_mtx_unlock(inet_domain_mutex);
696 } else {
697 pr_input(m, hlen);
698 }
699 }
700
701 struct pktchain_elm {
702 struct mbuf *pkte_head;
703 struct mbuf *pkte_tail;
704 struct in_addr pkte_saddr;
705 struct in_addr pkte_daddr;
706 uint16_t pkte_npkts;
707 uint16_t pkte_proto;
708 uint32_t pkte_nbytes;
709 };
710
711 typedef struct pktchain_elm pktchain_elm_t;
712
713 /* Store upto PKTTBL_SZ unique flows on the stack */
714 #define PKTTBL_SZ 7
715
716 static struct mbuf *
ip_chain_insert(struct mbuf * packet,pktchain_elm_t * tbl)717 ip_chain_insert(struct mbuf *packet, pktchain_elm_t *tbl)
718 {
719 struct ip* ip;
720 int pkttbl_idx = 0;
721
722 ip = mtod(packet, struct ip*);
723
724 /* reusing the hash function from inaddr_hashval */
725 pkttbl_idx = inaddr_hashval(ntohl(ip->ip_src.s_addr)) % PKTTBL_SZ;
726 if (tbl[pkttbl_idx].pkte_head == NULL) {
727 tbl[pkttbl_idx].pkte_head = packet;
728 tbl[pkttbl_idx].pkte_saddr.s_addr = ip->ip_src.s_addr;
729 tbl[pkttbl_idx].pkte_daddr.s_addr = ip->ip_dst.s_addr;
730 tbl[pkttbl_idx].pkte_proto = ip->ip_p;
731 } else {
732 if ((ip->ip_dst.s_addr == tbl[pkttbl_idx].pkte_daddr.s_addr) &&
733 (ip->ip_src.s_addr == tbl[pkttbl_idx].pkte_saddr.s_addr) &&
734 (ip->ip_p == tbl[pkttbl_idx].pkte_proto)) {
735 } else {
736 return packet;
737 }
738 }
739 if (tbl[pkttbl_idx].pkte_tail != NULL) {
740 mbuf_setnextpkt(tbl[pkttbl_idx].pkte_tail, packet);
741 }
742
743 tbl[pkttbl_idx].pkte_tail = packet;
744 tbl[pkttbl_idx].pkte_npkts += 1;
745 tbl[pkttbl_idx].pkte_nbytes += packet->m_pkthdr.len;
746 return NULL;
747 }
748
749 /* args is a dummy variable here for backward compatibility */
750 static void
ip_input_second_pass_loop_tbl(pktchain_elm_t * tbl,struct ip_fw_in_args * args)751 ip_input_second_pass_loop_tbl(pktchain_elm_t *tbl, struct ip_fw_in_args *args)
752 {
753 int i = 0;
754
755 for (i = 0; i < PKTTBL_SZ; i++) {
756 if (tbl[i].pkte_head != NULL) {
757 struct mbuf *m = tbl[i].pkte_head;
758 ip_input_second_pass(m, m->m_pkthdr.rcvif,
759 tbl[i].pkte_npkts, tbl[i].pkte_nbytes, args);
760
761 if (tbl[i].pkte_npkts > 2) {
762 ipstat.ips_rxc_chainsz_gt2++;
763 }
764 if (tbl[i].pkte_npkts > 4) {
765 ipstat.ips_rxc_chainsz_gt4++;
766 }
767 #if (DEBUG || DEVELOPMENT)
768 if (ip_input_measure) {
769 net_perf_histogram(&net_perf, tbl[i].pkte_npkts);
770 }
771 #endif /* (DEBUG || DEVELOPMENT) */
772 tbl[i].pkte_head = tbl[i].pkte_tail = NULL;
773 tbl[i].pkte_npkts = 0;
774 tbl[i].pkte_nbytes = 0;
775 /* no need to initialize address and protocol in tbl */
776 }
777 }
778 }
779
780 static void
ip_input_cpout_args(struct ip_fw_in_args * args,struct ip_fw_args * args1,boolean_t * done_init)781 ip_input_cpout_args(struct ip_fw_in_args *args, struct ip_fw_args *args1,
782 boolean_t *done_init)
783 {
784 if (*done_init == FALSE) {
785 bzero(args1, sizeof(struct ip_fw_args));
786 *done_init = TRUE;
787 }
788 args1->fwa_pf_rule = args->fwai_pf_rule;
789 }
790
791 static void
ip_input_cpin_args(struct ip_fw_args * args1,struct ip_fw_in_args * args)792 ip_input_cpin_args(struct ip_fw_args *args1, struct ip_fw_in_args *args)
793 {
794 args->fwai_pf_rule = args1->fwa_pf_rule;
795 }
796
797 typedef enum {
798 IPINPUT_DOCHAIN = 0,
799 IPINPUT_DONTCHAIN,
800 IPINPUT_FREED,
801 IPINPUT_DONE
802 } ipinput_chain_ret_t;
803
804 static void
ip_input_update_nstat(struct ifnet * ifp,struct in_addr src_ip,u_int32_t packets,u_int32_t bytes)805 ip_input_update_nstat(struct ifnet *ifp, struct in_addr src_ip,
806 u_int32_t packets, u_int32_t bytes)
807 {
808 if (nstat_collect) {
809 struct rtentry *rt = ifnet_cached_rtlookup_inet(ifp,
810 src_ip);
811 if (rt != NULL) {
812 nstat_route_rx(rt, packets, bytes, 0);
813 rtfree(rt);
814 }
815 }
816 }
817
818 static void
ip_input_dispatch_chain(struct mbuf * m)819 ip_input_dispatch_chain(struct mbuf *m)
820 {
821 struct mbuf *tmp_mbuf = m;
822 struct mbuf *nxt_mbuf = NULL;
823 struct ip *ip = NULL;
824 unsigned int hlen;
825
826 ip = mtod(tmp_mbuf, struct ip *);
827 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
828 while (tmp_mbuf != NULL) {
829 nxt_mbuf = mbuf_nextpkt(tmp_mbuf);
830 mbuf_setnextpkt(tmp_mbuf, NULL);
831 ip_proto_dispatch_in(tmp_mbuf, hlen, ip->ip_p, 0);
832 tmp_mbuf = nxt_mbuf;
833 if (tmp_mbuf) {
834 ip = mtod(tmp_mbuf, struct ip *);
835 /* first mbuf of chain already has adjusted ip_len */
836 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
837 ip->ip_len -= hlen;
838 }
839 }
840 }
841
842 static void
ip_input_setdst_chain(struct mbuf * m,uint16_t ifindex,struct in_ifaddr * ia)843 ip_input_setdst_chain(struct mbuf *m, uint16_t ifindex, struct in_ifaddr *ia)
844 {
845 struct mbuf *tmp_mbuf = m;
846
847 while (tmp_mbuf != NULL) {
848 ip_setdstifaddr_info(tmp_mbuf, ifindex, ia);
849 tmp_mbuf = mbuf_nextpkt(tmp_mbuf);
850 }
851 }
852
853 static void
ip_input_adjust(struct mbuf * m,struct ip * ip,struct ifnet * inifp)854 ip_input_adjust(struct mbuf *m, struct ip *ip, struct ifnet *inifp)
855 {
856 boolean_t adjust = TRUE;
857
858 ASSERT(m_pktlen(m) > ip->ip_len);
859
860 /*
861 * Invalidate hardware checksum info if ip_adj_clear_hwcksum
862 * is set; useful to handle buggy drivers. Note that this
863 * should not be enabled by default, as we may get here due
864 * to link-layer padding.
865 */
866 if (ip_adj_clear_hwcksum &&
867 (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
868 !(inifp->if_flags & IFF_LOOPBACK) &&
869 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
870 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
871 m->m_pkthdr.csum_data = 0;
872 ipstat.ips_adj_hwcsum_clr++;
873 }
874
875 /*
876 * If partial checksum information is available, subtract
877 * out the partial sum of postpended extraneous bytes, and
878 * update the checksum metadata accordingly. By doing it
879 * here, the upper layer transport only needs to adjust any
880 * prepended extraneous bytes (else it will do both.)
881 */
882 if (ip_adj_partial_sum &&
883 (m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
884 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
885 m->m_pkthdr.csum_rx_val = m_adj_sum16(m,
886 m->m_pkthdr.csum_rx_start, m->m_pkthdr.csum_rx_start,
887 (ip->ip_len - m->m_pkthdr.csum_rx_start),
888 m->m_pkthdr.csum_rx_val);
889 } else if ((m->m_pkthdr.csum_flags &
890 (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
891 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
892 /*
893 * If packet has partial checksum info and we decided not
894 * to subtract the partial sum of postpended extraneous
895 * bytes here (not the default case), leave that work to
896 * be handled by the other layers. For now, only TCP, UDP
897 * layers are capable of dealing with this. For all other
898 * protocols (including fragments), trim and ditch the
899 * partial sum as those layers might not implement partial
900 * checksumming (or adjustment) at all.
901 */
902 if ((ip->ip_off & (IP_MF | IP_OFFMASK)) == 0 &&
903 (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_UDP)) {
904 adjust = FALSE;
905 } else {
906 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
907 m->m_pkthdr.csum_data = 0;
908 ipstat.ips_adj_hwcsum_clr++;
909 }
910 }
911
912 if (adjust) {
913 ipstat.ips_adj++;
914 if (m->m_len == m->m_pkthdr.len) {
915 m->m_len = ip->ip_len;
916 m->m_pkthdr.len = ip->ip_len;
917 } else {
918 m_adj(m, ip->ip_len - m->m_pkthdr.len);
919 }
920 }
921 }
922
923 /*
924 * First pass does all essential packet validation and places on a per flow
925 * queue for doing operations that have same outcome for all packets of a flow.
926 */
927 static ipinput_chain_ret_t
ip_input_first_pass(struct mbuf * m,struct ip_fw_in_args * args,struct mbuf ** modm)928 ip_input_first_pass(struct mbuf *m, struct ip_fw_in_args *args, struct mbuf **modm)
929 {
930 struct ip *ip;
931 struct ifnet *inifp;
932 unsigned int hlen;
933 int retval = IPINPUT_DOCHAIN;
934 int len = 0;
935 struct in_addr src_ip;
936 #if DUMMYNET
937 struct m_tag *copy;
938 struct m_tag *p;
939 boolean_t delete = FALSE;
940 struct ip_fw_args args1;
941 boolean_t init = FALSE;
942 #endif /* DUMMYNET */
943 ipfilter_t inject_filter_ref = NULL;
944
945 /* Check if the mbuf is still valid after interface filter processing */
946 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
947 inifp = mbuf_pkthdr_rcvif(m);
948 VERIFY(inifp != NULL);
949
950 /* Perform IP header alignment fixup, if needed */
951 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
952
953 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
954
955 #if DUMMYNET
956 /*
957 * Don't bother searching for tag(s) if there's none.
958 */
959 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
960 goto ipfw_tags_done;
961 }
962
963 /* Grab info from mtags prepended to the chain */
964 p = m_tag_first(m);
965 while (p) {
966 if (p->m_tag_id == KERNEL_MODULE_TAG_ID) {
967 if (p->m_tag_type == KERNEL_TAG_TYPE_DUMMYNET) {
968 struct dn_pkt_tag *dn_tag;
969
970 dn_tag = (struct dn_pkt_tag *)(p + 1);
971 args->fwai_pf_rule = dn_tag->dn_pf_rule;
972 delete = TRUE;
973 }
974
975 if (delete) {
976 copy = p;
977 p = m_tag_next(m, p);
978 m_tag_delete(m, copy);
979 } else {
980 p = m_tag_next(m, p);
981 }
982 } else {
983 p = m_tag_next(m, p);
984 }
985 }
986
987 #if DIAGNOSTIC
988 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
989 panic("ip_input no HDR");
990 }
991 #endif
992
993 if (args->fwai_pf_rule) {
994 /* dummynet already filtered us */
995 ip = mtod(m, struct ip *);
996 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
997 inject_filter_ref = ipf_get_inject_filter(m);
998 if (args->fwai_pf_rule) {
999 goto check_with_pf;
1000 }
1001 }
1002 ipfw_tags_done:
1003 #endif /* DUMMYNET */
1004
1005 /*
1006 * No need to process packet twice if we've already seen it.
1007 */
1008 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1009 inject_filter_ref = ipf_get_inject_filter(m);
1010 }
1011 if (inject_filter_ref != NULL) {
1012 ip = mtod(m, struct ip *);
1013 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1014
1015 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1016 struct ip *, ip, struct ifnet *, inifp,
1017 struct ip *, ip, struct ip6_hdr *, NULL);
1018
1019 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1020 ip->ip_off = ntohs(ip->ip_off);
1021 ip_proto_dispatch_in(m, hlen, ip->ip_p, inject_filter_ref);
1022 return IPINPUT_DONE;
1023 }
1024
1025 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1026 if_ports_used_match_mbuf(inifp, PF_INET, m);
1027 }
1028
1029 if (m->m_pkthdr.len < sizeof(struct ip)) {
1030 OSAddAtomic(1, &ipstat.ips_total);
1031 OSAddAtomic(1, &ipstat.ips_tooshort);
1032 m_freem(m);
1033 return IPINPUT_FREED;
1034 }
1035
1036 if (m->m_len < sizeof(struct ip) &&
1037 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1038 OSAddAtomic(1, &ipstat.ips_total);
1039 OSAddAtomic(1, &ipstat.ips_toosmall);
1040 return IPINPUT_FREED;
1041 }
1042
1043 ip = mtod(m, struct ip *);
1044 *modm = m;
1045
1046 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1047 ip->ip_p, ip->ip_off, ip->ip_len);
1048
1049 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1050 OSAddAtomic(1, &ipstat.ips_total);
1051 OSAddAtomic(1, &ipstat.ips_badvers);
1052 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1053 m_freem(m);
1054 return IPINPUT_FREED;
1055 }
1056
1057 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1058 if (hlen < sizeof(struct ip)) {
1059 OSAddAtomic(1, &ipstat.ips_total);
1060 OSAddAtomic(1, &ipstat.ips_badhlen);
1061 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1062 m_freem(m);
1063 return IPINPUT_FREED;
1064 }
1065
1066 if (hlen > m->m_len) {
1067 if ((m = m_pullup(m, hlen)) == NULL) {
1068 OSAddAtomic(1, &ipstat.ips_total);
1069 OSAddAtomic(1, &ipstat.ips_badhlen);
1070 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1071 return IPINPUT_FREED;
1072 }
1073 ip = mtod(m, struct ip *);
1074 *modm = m;
1075 }
1076
1077 /* 127/8 must not appear on wire - RFC1122 */
1078 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1079 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1080 /*
1081 * Allow for the following exceptions:
1082 *
1083 * 1. If the packet was sent to loopback (i.e. rcvif
1084 * would have been set earlier at output time.)
1085 *
1086 * 2. If the packet was sent out on loopback from a local
1087 * source address which belongs to a non-loopback
1088 * interface (i.e. rcvif may not necessarily be a
1089 * loopback interface, hence the test for PKTF_LOOP.)
1090 * Unlike IPv6, there is no interface scope ID, and
1091 * therefore we don't care so much about PKTF_IFINFO.
1092 */
1093 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1094 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1095 OSAddAtomic(1, &ipstat.ips_total);
1096 OSAddAtomic(1, &ipstat.ips_badaddr);
1097 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1098 m_freem(m);
1099 return IPINPUT_FREED;
1100 }
1101 }
1102
1103 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1104 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1105 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1106 ip_linklocal_stat.iplls_in_total++;
1107 if (ip->ip_ttl != MAXTTL) {
1108 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1109 /* Silently drop link local traffic with bad TTL */
1110 if (!ip_linklocal_in_allowbadttl) {
1111 OSAddAtomic(1, &ipstat.ips_total);
1112 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1113 m_freem(m);
1114 return IPINPUT_FREED;
1115 }
1116 }
1117 }
1118
1119 if (ip_cksum(m, hlen)) {
1120 OSAddAtomic(1, &ipstat.ips_total);
1121 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1122 m_freem(m);
1123 return IPINPUT_FREED;
1124 }
1125
1126 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1127 struct ip *, ip, struct ifnet *, inifp,
1128 struct ip *, ip, struct ip6_hdr *, NULL);
1129
1130 /*
1131 * Convert fields to host representation.
1132 */
1133 #if BYTE_ORDER != BIG_ENDIAN
1134 NTOHS(ip->ip_len);
1135 #endif
1136
1137 if (ip->ip_len < hlen) {
1138 OSAddAtomic(1, &ipstat.ips_total);
1139 OSAddAtomic(1, &ipstat.ips_badlen);
1140 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1141 m_freem(m);
1142 return IPINPUT_FREED;
1143 }
1144
1145 #if BYTE_ORDER != BIG_ENDIAN
1146 NTOHS(ip->ip_off);
1147 #endif
1148
1149 /*
1150 * Check that the amount of data in the buffers
1151 * is as at least much as the IP header would have us expect.
1152 * Trim mbufs if longer than we expect.
1153 * Drop packet if shorter than we expect.
1154 */
1155 if (m->m_pkthdr.len < ip->ip_len) {
1156 OSAddAtomic(1, &ipstat.ips_total);
1157 OSAddAtomic(1, &ipstat.ips_tooshort);
1158 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1159 m_freem(m);
1160 return IPINPUT_FREED;
1161 }
1162
1163 if (m->m_pkthdr.len > ip->ip_len) {
1164 ip_input_adjust(m, ip, inifp);
1165 }
1166
1167 /* for netstat route statistics */
1168 src_ip = ip->ip_src;
1169 len = m->m_pkthdr.len;
1170
1171 #if DUMMYNET
1172 check_with_pf:
1173 #endif /* DUMMYNET */
1174 #if PF
1175 /* Invoke inbound packet filter */
1176 if (PF_IS_ENABLED) {
1177 int error;
1178 ip_input_cpout_args(args, &args1, &init);
1179 ip = mtod(m, struct ip *);
1180 src_ip = ip->ip_src;
1181
1182 #if DUMMYNET
1183 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args1);
1184 #else
1185 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1186 #endif /* DUMMYNET */
1187 if (error != 0 || m == NULL) {
1188 if (m != NULL) {
1189 panic("%s: unexpected packet %p",
1190 __func__, m);
1191 /* NOTREACHED */
1192 }
1193 /* Already freed by callee */
1194 ip_input_update_nstat(inifp, src_ip, 1, len);
1195 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1196 OSAddAtomic(1, &ipstat.ips_total);
1197 return IPINPUT_FREED;
1198 }
1199 ip = mtod(m, struct ip *);
1200 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1201 *modm = m;
1202 ip_input_cpin_args(&args1, args);
1203 }
1204 #endif /* PF */
1205
1206 #if IPSEC
1207 if (ipsec_bypass == 0 && ipsec_gethist(m, NULL)) {
1208 retval = IPINPUT_DONTCHAIN; /* XXX scope for chaining here? */
1209 goto pass;
1210 }
1211 #endif
1212
1213 #if IPSEC
1214 pass:
1215 #endif
1216 /*
1217 * Process options and, if not destined for us,
1218 * ship it on. ip_dooptions returns 1 when an
1219 * error was detected (causing an icmp message
1220 * to be sent and the original packet to be freed).
1221 */
1222 ip_nhops = 0; /* for source routed packets */
1223 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1224 src_ip = ip->ip_src;
1225 ip_input_update_nstat(inifp, src_ip, 1, len);
1226 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1227 OSAddAtomic(1, &ipstat.ips_total);
1228 return IPINPUT_FREED;
1229 }
1230
1231 /*
1232 * Don't chain fragmented packets
1233 */
1234 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1235 return IPINPUT_DONTCHAIN;
1236 }
1237
1238 /* Allow DHCP/BootP responses through */
1239 if ((inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1240 hlen == sizeof(struct ip) && ip->ip_p == IPPROTO_UDP) {
1241 struct udpiphdr *ui;
1242
1243 if (m->m_len < sizeof(struct udpiphdr) &&
1244 (m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1245 OSAddAtomic(1, &udpstat.udps_hdrops);
1246 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1247 OSAddAtomic(1, &ipstat.ips_total);
1248 return IPINPUT_FREED;
1249 }
1250 *modm = m;
1251 ui = mtod(m, struct udpiphdr *);
1252 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1253 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1254 return IPINPUT_DONTCHAIN;
1255 }
1256 }
1257
1258 /* Avoid chaining raw sockets as ipsec checks occur later for them */
1259 if (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR) {
1260 return IPINPUT_DONTCHAIN;
1261 }
1262
1263 return retval;
1264 #if !defined(__i386__) && !defined(__x86_64__)
1265 bad:
1266 m_freem(m);
1267 return IPINPUT_FREED;
1268 #endif
1269 }
1270
1271 /*
1272 * Because the call to m_pullup() may freem the mbuf, the function frees the mbuf packet
1273 * chain before it return IP_CHECK_IF_DROP
1274 */
1275 static ip_check_if_result_t
ip_input_check_interface(struct mbuf ** mp,struct ip * ip,struct ifnet * inifp)1276 ip_input_check_interface(struct mbuf **mp, struct ip *ip, struct ifnet *inifp)
1277 {
1278 struct mbuf *m = *mp;
1279 struct in_ifaddr *ia = NULL;
1280 struct in_ifaddr *best_ia = NULL;
1281 struct ifnet *match_ifp = NULL;
1282 ip_check_if_result_t result = IP_CHECK_IF_NONE;
1283
1284 /*
1285 * Host broadcast and all network broadcast addresses are always a match
1286 */
1287 if (ip->ip_dst.s_addr == (u_int32_t)INADDR_BROADCAST ||
1288 ip->ip_dst.s_addr == INADDR_ANY) {
1289 ip_input_setdst_chain(m, inifp->if_index, NULL);
1290 return IP_CHECK_IF_OURS;
1291 }
1292
1293 /*
1294 * Check for a match in the hash bucket.
1295 */
1296 lck_rw_lock_shared(&in_ifaddr_rwlock);
1297 TAILQ_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
1298 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
1299 best_ia = ia;
1300 match_ifp = best_ia->ia_ifp;
1301
1302 if (ia->ia_ifp == inifp || (inifp->if_flags & IFF_LOOPBACK) ||
1303 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1304 /*
1305 * A locally originated packet or packet from the loopback
1306 * interface is always an exact interface address match
1307 */
1308 match_ifp = inifp;
1309 break;
1310 }
1311 /*
1312 * Continue the loop in case there's a exact match with another
1313 * interface
1314 */
1315 }
1316 }
1317 if (best_ia != NULL) {
1318 if (match_ifp != inifp && ipforwarding == 0 &&
1319 ((ip_checkinterface == IP_CHECKINTERFACE_HYBRID_ES &&
1320 (match_ifp->if_family == IFNET_FAMILY_IPSEC ||
1321 match_ifp->if_family == IFNET_FAMILY_UTUN)) ||
1322 ip_checkinterface == IP_CHECKINTERFACE_STRONG_ES)) {
1323 /*
1324 * Drop when interface address check is strict and forwarding
1325 * is disabled
1326 */
1327 result = IP_CHECK_IF_DROP;
1328 } else {
1329 result = IP_CHECK_IF_OURS;
1330 ip_input_setdst_chain(m, 0, best_ia);
1331 }
1332 }
1333 lck_rw_done(&in_ifaddr_rwlock);
1334
1335 if (result == IP_CHECK_IF_NONE && (inifp->if_flags & IFF_BROADCAST)) {
1336 /*
1337 * Check for broadcast addresses.
1338 *
1339 * Only accept broadcast packets that arrive via the matching
1340 * interface. Reception of forwarded directed broadcasts would be
1341 * handled via ip_forward() and ether_frameout() with the loopback
1342 * into the stack for SIMPLEX interfaces handled by ether_frameout().
1343 */
1344 struct ifaddr *ifa;
1345
1346 ifnet_lock_shared(inifp);
1347 TAILQ_FOREACH(ifa, &inifp->if_addrhead, ifa_link) {
1348 if (ifa->ifa_addr->sa_family != AF_INET) {
1349 continue;
1350 }
1351 ia = ifatoia(ifa);
1352 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == ip->ip_dst.s_addr ||
1353 ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
1354 ip_input_setdst_chain(m, 0, ia);
1355 result = IP_CHECK_IF_OURS;
1356 match_ifp = inifp;
1357 break;
1358 }
1359 }
1360 ifnet_lock_done(inifp);
1361 }
1362
1363 /* Allow DHCP/BootP responses through */
1364 if (result == IP_CHECK_IF_NONE && (inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1365 ip->ip_p == IPPROTO_UDP && (IP_VHL_HL(ip->ip_vhl) << 2) == sizeof(struct ip)) {
1366 struct udpiphdr *ui;
1367
1368 if (m->m_len < sizeof(struct udpiphdr)) {
1369 if ((m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1370 OSAddAtomic(1, &udpstat.udps_hdrops);
1371 *mp = NULL;
1372 return IP_CHECK_IF_DROP;
1373 }
1374 /*
1375 * m_pullup can return a different mbuf
1376 */
1377 *mp = m;
1378 ip = mtod(m, struct ip *);
1379 }
1380 ui = mtod(m, struct udpiphdr *);
1381 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1382 ip_input_setdst_chain(m, inifp->if_index, NULL);
1383 result = IP_CHECK_IF_OURS;
1384 match_ifp = inifp;
1385 }
1386 }
1387
1388 if (result == IP_CHECK_IF_NONE) {
1389 if (ipforwarding == 0) {
1390 result = IP_CHECK_IF_DROP;
1391 } else {
1392 result = IP_CHECK_IF_FORWARD;
1393 ip_input_setdst_chain(m, inifp->if_index, NULL);
1394 }
1395 }
1396
1397 if (result == IP_CHECK_IF_OURS && match_ifp != inifp) {
1398 ipstat.ips_rcv_if_weak_match++;
1399
1400 /* Logging is too noisy when forwarding is enabled */
1401 if (ip_checkinterface_debug != 0 && ipforwarding == 0) {
1402 char src_str[MAX_IPv4_STR_LEN];
1403 char dst_str[MAX_IPv4_STR_LEN];
1404
1405 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1406 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1407 os_log_info(OS_LOG_DEFAULT,
1408 "%s: weak ES interface match to %s for packet from %s to %s proto %u received via %s",
1409 __func__, best_ia->ia_ifp->if_xname, src_str, dst_str, ip->ip_p, inifp->if_xname);
1410 }
1411 } else if (result == IP_CHECK_IF_DROP) {
1412 if (ip_checkinterface_debug > 0) {
1413 char src_str[MAX_IPv4_STR_LEN];
1414 char dst_str[MAX_IPv4_STR_LEN];
1415
1416 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1417 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1418 os_log(OS_LOG_DEFAULT,
1419 "%s: no interface match for packet from %s to %s proto %u received via %s",
1420 __func__, src_str, dst_str, ip->ip_p, inifp->if_xname);
1421 }
1422 struct mbuf *tmp_mbuf = m;
1423 while (tmp_mbuf != NULL) {
1424 ipstat.ips_rcv_if_no_match++;
1425 tmp_mbuf = tmp_mbuf->m_nextpkt;
1426 }
1427 m_freem_list(m);
1428 *mp = NULL;
1429 }
1430
1431 return result;
1432 }
1433
1434 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)1435 ip_input_second_pass(struct mbuf *m, struct ifnet *inifp,
1436 int npkts_in_chain, int bytes_in_chain, struct ip_fw_in_args *args)
1437 {
1438 struct mbuf *tmp_mbuf = NULL;
1439 unsigned int hlen;
1440
1441 #pragma unused (args)
1442
1443 struct ip *ip = mtod(m, struct ip *);
1444 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1445
1446 OSAddAtomic(npkts_in_chain, &ipstat.ips_total);
1447
1448 /*
1449 * Naively assume we can attribute inbound data to the route we would
1450 * use to send to this destination. Asymmetric routing breaks this
1451 * assumption, but it still allows us to account for traffic from
1452 * a remote node in the routing table.
1453 * this has a very significant performance impact so we bypass
1454 * if nstat_collect is disabled. We may also bypass if the
1455 * protocol is tcp in the future because tcp will have a route that
1456 * we can use to attribute the data to. That does mean we would not
1457 * account for forwarded tcp traffic.
1458 */
1459 ip_input_update_nstat(inifp, ip->ip_src, npkts_in_chain,
1460 bytes_in_chain);
1461
1462 /*
1463 * Check our list of addresses, to see if the packet is for us.
1464 * If we don't have any addresses, assume any unicast packet
1465 * we receive might be for us (and let the upper layers deal
1466 * with it).
1467 */
1468 tmp_mbuf = m;
1469 if (TAILQ_EMPTY(&in_ifaddrhead)) {
1470 while (tmp_mbuf != NULL) {
1471 if (!(tmp_mbuf->m_flags & (M_MCAST | M_BCAST))) {
1472 ip_setdstifaddr_info(tmp_mbuf, inifp->if_index,
1473 NULL);
1474 }
1475 tmp_mbuf = mbuf_nextpkt(tmp_mbuf);
1476 }
1477 goto ours;
1478 }
1479
1480 /*
1481 * Enable a consistency check between the destination address
1482 * and the arrival interface for a unicast packet (the RFC 1122
1483 * strong ES model) if IP forwarding is disabled and the packet
1484 * is not locally generated
1485 *
1486 * XXX - Checking also should be disabled if the destination
1487 * address is ipnat'ed to a different interface.
1488 *
1489 * XXX - Checking is incompatible with IP aliases added
1490 * to the loopback interface instead of the interface where
1491 * the packets are received.
1492 */
1493 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1494 ip_check_if_result_t ip_check_if_result = IP_CHECK_IF_NONE;
1495
1496 ip_check_if_result = ip_input_check_interface(&m, ip, inifp);
1497 ASSERT(ip_check_if_result != IP_CHECK_IF_NONE);
1498 if (ip_check_if_result == IP_CHECK_IF_OURS) {
1499 goto ours;
1500 } else if (ip_check_if_result == IP_CHECK_IF_DROP) {
1501 return;
1502 }
1503 } else {
1504 struct in_multi *inm;
1505 /*
1506 * See if we belong to the destination multicast group on the
1507 * arrival interface.
1508 */
1509 in_multihead_lock_shared();
1510 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
1511 in_multihead_lock_done();
1512 if (inm == NULL) {
1513 OSAddAtomic(npkts_in_chain, &ipstat.ips_notmember);
1514 m_freem_list(m);
1515 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1516 return;
1517 }
1518 ip_input_setdst_chain(m, inifp->if_index, NULL);
1519 INM_REMREF(inm);
1520 goto ours;
1521 }
1522
1523 tmp_mbuf = m;
1524 struct mbuf *nxt_mbuf = NULL;
1525 while (tmp_mbuf != NULL) {
1526 nxt_mbuf = mbuf_nextpkt(tmp_mbuf);
1527 /*
1528 * Not for us; forward if possible and desirable.
1529 */
1530 mbuf_setnextpkt(tmp_mbuf, NULL);
1531 if (ipforwarding == 0) {
1532 OSAddAtomic(1, &ipstat.ips_cantforward);
1533 m_freem(tmp_mbuf);
1534 } else {
1535 ip_forward(tmp_mbuf, 0, NULL);
1536 }
1537 tmp_mbuf = nxt_mbuf;
1538 }
1539 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1540 return;
1541 ours:
1542 ip = mtod(m, struct ip *); /* in case it changed */
1543 /*
1544 * If offset is set, must reassemble.
1545 */
1546 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1547 VERIFY(npkts_in_chain == 1);
1548 m = ip_reass(m);
1549 if (m == NULL) {
1550 return;
1551 }
1552 ip = mtod(m, struct ip *);
1553 /* Get the header length of the reassembled packet */
1554 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1555 }
1556
1557 /*
1558 * Further protocols expect the packet length to be w/o the
1559 * IP header.
1560 */
1561 ip->ip_len -= hlen;
1562
1563 #if IPSEC
1564 /*
1565 * enforce IPsec policy checking if we are seeing last header.
1566 * note that we do not visit this with protocols with pcb layer
1567 * code - like udp/tcp/raw ip.
1568 */
1569 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
1570 VERIFY(npkts_in_chain == 1);
1571 if (ipsec4_in_reject(m, NULL)) {
1572 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
1573 goto bad;
1574 }
1575 }
1576 #endif /* IPSEC */
1577
1578 /*
1579 * Switch out to protocol's input routine.
1580 */
1581 OSAddAtomic(npkts_in_chain, &ipstat.ips_delivered);
1582
1583 ip_input_dispatch_chain(m);
1584
1585 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1586 return;
1587 bad:
1588 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1589 m_freem(m);
1590 }
1591
1592 void
ip_input_process_list(struct mbuf * packet_list)1593 ip_input_process_list(struct mbuf *packet_list)
1594 {
1595 pktchain_elm_t pktchain_tbl[PKTTBL_SZ];
1596
1597 struct mbuf *packet = NULL;
1598 struct mbuf *modm = NULL; /* modified mbuf */
1599 int retval = 0;
1600 #if (DEBUG || DEVELOPMENT)
1601 struct timeval start_tv;
1602 #endif /* (DEBUG || DEVELOPMENT) */
1603 int num_pkts = 0;
1604 int chain = 0;
1605 struct ip_fw_in_args args;
1606
1607 if (ip_chaining == 0) {
1608 struct mbuf *m = packet_list;
1609 #if (DEBUG || DEVELOPMENT)
1610 if (ip_input_measure) {
1611 net_perf_start_time(&net_perf, &start_tv);
1612 }
1613 #endif /* (DEBUG || DEVELOPMENT) */
1614
1615 while (m) {
1616 packet_list = mbuf_nextpkt(m);
1617 mbuf_setnextpkt(m, NULL);
1618 ip_input(m);
1619 m = packet_list;
1620 num_pkts++;
1621 }
1622 #if (DEBUG || DEVELOPMENT)
1623 if (ip_input_measure) {
1624 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1625 }
1626 #endif /* (DEBUG || DEVELOPMENT) */
1627 return;
1628 }
1629 #if (DEBUG || DEVELOPMENT)
1630 if (ip_input_measure) {
1631 net_perf_start_time(&net_perf, &start_tv);
1632 }
1633 #endif /* (DEBUG || DEVELOPMENT) */
1634
1635 bzero(&pktchain_tbl, sizeof(pktchain_tbl));
1636 restart_list_process:
1637 chain = 0;
1638 for (packet = packet_list; packet; packet = packet_list) {
1639 m_add_crumb(packet, PKT_CRUMB_IP_INPUT);
1640
1641 packet_list = mbuf_nextpkt(packet);
1642 mbuf_setnextpkt(packet, NULL);
1643
1644 num_pkts++;
1645 modm = NULL;
1646 bzero(&args, sizeof(args));
1647
1648 retval = ip_input_first_pass(packet, &args, &modm);
1649
1650 if (retval == IPINPUT_DOCHAIN) {
1651 if (modm) {
1652 packet = modm;
1653 }
1654 packet = ip_chain_insert(packet, &pktchain_tbl[0]);
1655 if (packet == NULL) {
1656 ipstat.ips_rxc_chained++;
1657 chain++;
1658 if (chain > ip_chainsz) {
1659 break;
1660 }
1661 } else {
1662 ipstat.ips_rxc_collisions++;
1663 break;
1664 }
1665 } else if (retval == IPINPUT_DONTCHAIN) {
1666 /* in order to preserve order, exit from chaining */
1667 if (modm) {
1668 packet = modm;
1669 }
1670 ipstat.ips_rxc_notchain++;
1671 break;
1672 } else {
1673 /* packet was freed or delivered, do nothing. */
1674 }
1675 }
1676
1677 /* do second pass here for pktchain_tbl */
1678 if (chain) {
1679 ip_input_second_pass_loop_tbl(&pktchain_tbl[0], &args);
1680 }
1681
1682 if (packet) {
1683 /*
1684 * equivalent update in chaining case if performed in
1685 * ip_input_second_pass_loop_tbl().
1686 */
1687 #if (DEBUG || DEVELOPMENT)
1688 if (ip_input_measure) {
1689 net_perf_histogram(&net_perf, 1);
1690 }
1691 #endif /* (DEBUG || DEVELOPMENT) */
1692 ip_input_second_pass(packet, packet->m_pkthdr.rcvif,
1693 1, packet->m_pkthdr.len, &args);
1694 }
1695
1696 if (packet_list) {
1697 goto restart_list_process;
1698 }
1699
1700 #if (DEBUG || DEVELOPMENT)
1701 if (ip_input_measure) {
1702 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1703 }
1704 #endif /* (DEBUG || DEVELOPMENT) */
1705 }
1706 /*
1707 * Ip input routine. Checksum and byte swap header. If fragmented
1708 * try to reassemble. Process options. Pass to next level.
1709 */
1710 void
ip_input(struct mbuf * m)1711 ip_input(struct mbuf *m)
1712 {
1713 struct ip *ip;
1714 unsigned int hlen;
1715 u_short sum = 0;
1716 #if DUMMYNET
1717 struct ip_fw_args args;
1718 struct m_tag *tag;
1719 #endif
1720 ipfilter_t inject_filter_ref = NULL;
1721 struct ifnet *inifp;
1722
1723 /* Check if the mbuf is still valid after interface filter processing */
1724 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
1725 inifp = m->m_pkthdr.rcvif;
1726 VERIFY(inifp != NULL);
1727
1728 m_add_crumb(m, PKT_CRUMB_IP_INPUT);
1729
1730 ipstat.ips_rxc_notlist++;
1731
1732 /* Perform IP header alignment fixup, if needed */
1733 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
1734
1735 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
1736
1737 #if DUMMYNET
1738 bzero(&args, sizeof(struct ip_fw_args));
1739
1740 /*
1741 * Don't bother searching for tag(s) if there's none.
1742 */
1743 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
1744 goto ipfw_tags_done;
1745 }
1746
1747 /* Grab info from mtags prepended to the chain */
1748 if ((tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID,
1749 KERNEL_TAG_TYPE_DUMMYNET, NULL)) != NULL) {
1750 struct dn_pkt_tag *dn_tag;
1751
1752 dn_tag = (struct dn_pkt_tag *)(tag + 1);
1753 args.fwa_pf_rule = dn_tag->dn_pf_rule;
1754
1755 m_tag_delete(m, tag);
1756 }
1757
1758 #if DIAGNOSTIC
1759 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
1760 panic("ip_input no HDR");
1761 }
1762 #endif
1763
1764 if (args.fwa_pf_rule) {
1765 /* dummynet already filtered us */
1766 ip = mtod(m, struct ip *);
1767 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1768 inject_filter_ref = ipf_get_inject_filter(m);
1769 if (args.fwa_pf_rule) {
1770 goto check_with_pf;
1771 }
1772 }
1773 ipfw_tags_done:
1774 #endif /* DUMMYNET */
1775
1776 /*
1777 * No need to process packet twice if we've already seen it.
1778 */
1779 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1780 inject_filter_ref = ipf_get_inject_filter(m);
1781 }
1782 if (inject_filter_ref != NULL) {
1783 ip = mtod(m, struct ip *);
1784 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1785
1786 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1787 struct ip *, ip, struct ifnet *, inifp,
1788 struct ip *, ip, struct ip6_hdr *, NULL);
1789
1790 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1791 ip->ip_off = ntohs(ip->ip_off);
1792 ip_proto_dispatch_in(m, hlen, ip->ip_p, inject_filter_ref);
1793 return;
1794 }
1795
1796 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1797 if_ports_used_match_mbuf(inifp, PF_INET, m);
1798 }
1799
1800 OSAddAtomic(1, &ipstat.ips_total);
1801 if (m->m_pkthdr.len < sizeof(struct ip)) {
1802 goto tooshort;
1803 }
1804
1805 if (m->m_len < sizeof(struct ip) &&
1806 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1807 OSAddAtomic(1, &ipstat.ips_toosmall);
1808 return;
1809 }
1810 ip = mtod(m, struct ip *);
1811
1812 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1813 ip->ip_p, ip->ip_off, ip->ip_len);
1814
1815 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1816 OSAddAtomic(1, &ipstat.ips_badvers);
1817 goto bad;
1818 }
1819
1820 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1821 if (hlen < sizeof(struct ip)) { /* minimum header length */
1822 OSAddAtomic(1, &ipstat.ips_badhlen);
1823 goto bad;
1824 }
1825 if (hlen > m->m_len) {
1826 if ((m = m_pullup(m, hlen)) == NULL) {
1827 OSAddAtomic(1, &ipstat.ips_badhlen);
1828 return;
1829 }
1830 ip = mtod(m, struct ip *);
1831 }
1832
1833 /* 127/8 must not appear on wire - RFC1122 */
1834 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1835 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1836 /*
1837 * Allow for the following exceptions:
1838 *
1839 * 1. If the packet was sent to loopback (i.e. rcvif
1840 * would have been set earlier at output time.)
1841 *
1842 * 2. If the packet was sent out on loopback from a local
1843 * source address which belongs to a non-loopback
1844 * interface (i.e. rcvif may not necessarily be a
1845 * loopback interface, hence the test for PKTF_LOOP.)
1846 * Unlike IPv6, there is no interface scope ID, and
1847 * therefore we don't care so much about PKTF_IFINFO.
1848 */
1849 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1850 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1851 OSAddAtomic(1, &ipstat.ips_badaddr);
1852 goto bad;
1853 }
1854 }
1855
1856 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1857 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1858 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1859 ip_linklocal_stat.iplls_in_total++;
1860 if (ip->ip_ttl != MAXTTL) {
1861 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1862 /* Silently drop link local traffic with bad TTL */
1863 if (!ip_linklocal_in_allowbadttl) {
1864 goto bad;
1865 }
1866 }
1867 }
1868
1869 sum = ip_cksum(m, hlen);
1870 if (sum) {
1871 goto bad;
1872 }
1873
1874 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1875 struct ip *, ip, struct ifnet *, inifp,
1876 struct ip *, ip, struct ip6_hdr *, NULL);
1877
1878 /*
1879 * Naively assume we can attribute inbound data to the route we would
1880 * use to send to this destination. Asymmetric routing breaks this
1881 * assumption, but it still allows us to account for traffic from
1882 * a remote node in the routing table.
1883 * this has a very significant performance impact so we bypass
1884 * if nstat_collect is disabled. We may also bypass if the
1885 * protocol is tcp in the future because tcp will have a route that
1886 * we can use to attribute the data to. That does mean we would not
1887 * account for forwarded tcp traffic.
1888 */
1889 if (nstat_collect) {
1890 struct rtentry *rt =
1891 ifnet_cached_rtlookup_inet(inifp, ip->ip_src);
1892 if (rt != NULL) {
1893 nstat_route_rx(rt, 1, m->m_pkthdr.len, 0);
1894 rtfree(rt);
1895 }
1896 }
1897
1898 /*
1899 * Convert fields to host representation.
1900 */
1901 #if BYTE_ORDER != BIG_ENDIAN
1902 NTOHS(ip->ip_len);
1903 #endif
1904
1905 if (ip->ip_len < hlen) {
1906 OSAddAtomic(1, &ipstat.ips_badlen);
1907 goto bad;
1908 }
1909
1910 #if BYTE_ORDER != BIG_ENDIAN
1911 NTOHS(ip->ip_off);
1912 #endif
1913 /*
1914 * Check that the amount of data in the buffers
1915 * is as at least much as the IP header would have us expect.
1916 * Trim mbufs if longer than we expect.
1917 * Drop packet if shorter than we expect.
1918 */
1919 if (m->m_pkthdr.len < ip->ip_len) {
1920 tooshort:
1921 OSAddAtomic(1, &ipstat.ips_tooshort);
1922 goto bad;
1923 }
1924 if (m->m_pkthdr.len > ip->ip_len) {
1925 ip_input_adjust(m, ip, inifp);
1926 }
1927
1928 #if DUMMYNET
1929 check_with_pf:
1930 #endif
1931 #if PF
1932 /* Invoke inbound packet filter */
1933 if (PF_IS_ENABLED) {
1934 int error;
1935 #if DUMMYNET
1936 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args);
1937 #else
1938 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1939 #endif /* DUMMYNET */
1940 if (error != 0 || m == NULL) {
1941 if (m != NULL) {
1942 panic("%s: unexpected packet %p",
1943 __func__, m);
1944 /* NOTREACHED */
1945 }
1946 /* Already freed by callee */
1947 return;
1948 }
1949 ip = mtod(m, struct ip *);
1950 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1951 }
1952 #endif /* PF */
1953
1954 #if IPSEC
1955 if (ipsec_bypass == 0 && ipsec_gethist(m, NULL)) {
1956 goto pass;
1957 }
1958 #endif
1959
1960 pass:
1961 /*
1962 * Process options and, if not destined for us,
1963 * ship it on. ip_dooptions returns 1 when an
1964 * error was detected (causing an icmp message
1965 * to be sent and the original packet to be freed).
1966 */
1967 ip_nhops = 0; /* for source routed packets */
1968 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1969 return;
1970 }
1971
1972 /*
1973 * Check our list of addresses, to see if the packet is for us.
1974 * If we don't have any addresses, assume any unicast packet
1975 * we receive might be for us (and let the upper layers deal
1976 * with it).
1977 */
1978 if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST))) {
1979 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1980 goto ours;
1981 }
1982
1983 /*
1984 * Enable a consistency check between the destination address
1985 * and the arrival interface for a unicast packet (the RFC 1122
1986 * strong ES model) if IP forwarding is disabled and the packet
1987 * is not locally generated and the packet is not subject to
1988 * 'ipfw fwd'.
1989 *
1990 * XXX - Checking also should be disabled if the destination
1991 * address is ipnat'ed to a different interface.
1992 *
1993 * XXX - Checking is incompatible with IP aliases added
1994 * to the loopback interface instead of the interface where
1995 * the packets are received.
1996 */
1997 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1998 ip_check_if_result_t check_if_result = IP_CHECK_IF_NONE;
1999
2000 check_if_result = ip_input_check_interface(&m, ip, inifp);
2001 ASSERT(check_if_result != IP_CHECK_IF_NONE);
2002 if (check_if_result == IP_CHECK_IF_OURS) {
2003 goto ours;
2004 } else if (check_if_result == IP_CHECK_IF_DROP) {
2005 return;
2006 }
2007 } else {
2008 struct in_multi *inm;
2009 /*
2010 * See if we belong to the destination multicast group on the
2011 * arrival interface.
2012 */
2013 in_multihead_lock_shared();
2014 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
2015 in_multihead_lock_done();
2016 if (inm == NULL) {
2017 OSAddAtomic(1, &ipstat.ips_notmember);
2018 m_freem(m);
2019 return;
2020 }
2021 ip_setdstifaddr_info(m, inifp->if_index, NULL);
2022 INM_REMREF(inm);
2023 goto ours;
2024 }
2025
2026 /*
2027 * Not for us; forward if possible and desirable.
2028 */
2029 if (ipforwarding == 0) {
2030 OSAddAtomic(1, &ipstat.ips_cantforward);
2031 m_freem(m);
2032 } else {
2033 ip_forward(m, 0, NULL);
2034 }
2035 return;
2036
2037 ours:
2038 /*
2039 * If offset or IP_MF are set, must reassemble.
2040 */
2041 if (ip->ip_off & ~(IP_DF | IP_RF)) {
2042 m = ip_reass(m);
2043 if (m == NULL) {
2044 return;
2045 }
2046 ip = mtod(m, struct ip *);
2047 /* Get the header length of the reassembled packet */
2048 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2049 }
2050
2051 /*
2052 * Further protocols expect the packet length to be w/o the
2053 * IP header.
2054 */
2055 ip->ip_len -= hlen;
2056
2057
2058 #if IPSEC
2059 /*
2060 * enforce IPsec policy checking if we are seeing last header.
2061 * note that we do not visit this with protocols with pcb layer
2062 * code - like udp/tcp/raw ip.
2063 */
2064 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
2065 if (ipsec4_in_reject(m, NULL)) {
2066 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
2067 goto bad;
2068 }
2069 }
2070 #endif /* IPSEC */
2071
2072 /*
2073 * Switch out to protocol's input routine.
2074 */
2075 OSAddAtomic(1, &ipstat.ips_delivered);
2076
2077 ip_proto_dispatch_in(m, hlen, ip->ip_p, 0);
2078 return;
2079
2080 bad:
2081 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
2082 m_freem(m);
2083 }
2084
2085 static void
ipq_updateparams(void)2086 ipq_updateparams(void)
2087 {
2088 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2089 /*
2090 * -1 for unlimited allocation.
2091 */
2092 if (maxnipq < 0) {
2093 ipq_limit = 0;
2094 }
2095 /*
2096 * Positive number for specific bound.
2097 */
2098 if (maxnipq > 0) {
2099 ipq_limit = maxnipq;
2100 }
2101 /*
2102 * Zero specifies no further fragment queue allocation -- set the
2103 * bound very low, but rely on implementation elsewhere to actually
2104 * prevent allocation and reclaim current queues.
2105 */
2106 if (maxnipq == 0) {
2107 ipq_limit = 1;
2108 }
2109 /*
2110 * Arm the purge timer if not already and if there's work to do
2111 */
2112 frag_sched_timeout();
2113 }
2114
2115 static int
2116 sysctl_maxnipq SYSCTL_HANDLER_ARGS
2117 {
2118 #pragma unused(arg1, arg2)
2119 int error, i;
2120
2121 lck_mtx_lock(&ipqlock);
2122 i = maxnipq;
2123 error = sysctl_handle_int(oidp, &i, 0, req);
2124 if (error || req->newptr == USER_ADDR_NULL) {
2125 goto done;
2126 }
2127 /* impose bounds */
2128 if (i < -1 || i > (nmbclusters / 4)) {
2129 error = EINVAL;
2130 goto done;
2131 }
2132 maxnipq = i;
2133 ipq_updateparams();
2134 done:
2135 lck_mtx_unlock(&ipqlock);
2136 return error;
2137 }
2138
2139 static int
2140 sysctl_maxfragsperpacket SYSCTL_HANDLER_ARGS
2141 {
2142 #pragma unused(arg1, arg2)
2143 int error, i;
2144
2145 lck_mtx_lock(&ipqlock);
2146 i = maxfragsperpacket;
2147 error = sysctl_handle_int(oidp, &i, 0, req);
2148 if (error || req->newptr == USER_ADDR_NULL) {
2149 goto done;
2150 }
2151 maxfragsperpacket = i;
2152 ipq_updateparams(); /* see if we need to arm timer */
2153 done:
2154 lck_mtx_unlock(&ipqlock);
2155 return error;
2156 }
2157
2158 /*
2159 * Take incoming datagram fragment and try to reassemble it into
2160 * whole datagram. If a chain for reassembly of this datagram already
2161 * exists, then it is given as fp; otherwise have to make a chain.
2162 *
2163 * The IP header is *NOT* adjusted out of iplen (but in host byte order).
2164 */
2165 static struct mbuf *
ip_reass(struct mbuf * m)2166 ip_reass(struct mbuf *m)
2167 {
2168 struct ip *ip;
2169 struct mbuf *p, *q, *nq, *t;
2170 struct ipq *fp = NULL;
2171 struct ipqhead *head;
2172 int i, hlen, next;
2173 u_int8_t ecn, ecn0;
2174 uint32_t csum, csum_flags;
2175 uint16_t hash;
2176 struct fq_head dfq;
2177
2178 MBUFQ_INIT(&dfq); /* for deferred frees */
2179
2180 /* If maxnipq or maxfragsperpacket is 0, never accept fragments. */
2181 if (maxnipq == 0 || maxfragsperpacket == 0) {
2182 ipstat.ips_fragments++;
2183 ipstat.ips_fragdropped++;
2184 m_freem(m);
2185 if (nipq > 0) {
2186 lck_mtx_lock(&ipqlock);
2187 frag_sched_timeout(); /* purge stale fragments */
2188 lck_mtx_unlock(&ipqlock);
2189 }
2190 return NULL;
2191 }
2192
2193 ip = mtod(m, struct ip *);
2194 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2195
2196 lck_mtx_lock(&ipqlock);
2197
2198 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
2199 head = &ipq[hash];
2200
2201 /*
2202 * Look for queue of fragments
2203 * of this datagram.
2204 */
2205 TAILQ_FOREACH(fp, head, ipq_list) {
2206 if (ip->ip_id == fp->ipq_id &&
2207 ip->ip_src.s_addr == fp->ipq_src.s_addr &&
2208 ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
2209 ip->ip_p == fp->ipq_p) {
2210 goto found;
2211 }
2212 }
2213
2214 fp = NULL;
2215
2216 /*
2217 * Attempt to trim the number of allocated fragment queues if it
2218 * exceeds the administrative limit.
2219 */
2220 if ((nipq > (unsigned)maxnipq) && (maxnipq > 0)) {
2221 /*
2222 * drop something from the tail of the current queue
2223 * before proceeding further
2224 */
2225 struct ipq *fq = TAILQ_LAST(head, ipqhead);
2226 if (fq == NULL) { /* gak */
2227 for (i = 0; i < IPREASS_NHASH; i++) {
2228 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
2229 if (r) {
2230 ipstat.ips_fragtimeout += r->ipq_nfrags;
2231 frag_freef(&ipq[i], r);
2232 break;
2233 }
2234 }
2235 } else {
2236 ipstat.ips_fragtimeout += fq->ipq_nfrags;
2237 frag_freef(head, fq);
2238 }
2239 }
2240
2241 found:
2242 /*
2243 * Leverage partial checksum offload for IP fragments. Narrow down
2244 * the scope to cover only UDP without IP options, as that is the
2245 * most common case.
2246 *
2247 * Perform 1's complement adjustment of octets that got included/
2248 * excluded in the hardware-calculated checksum value. Ignore cases
2249 * where the value includes the entire IPv4 header span, as the sum
2250 * for those octets would already be 0 by the time we get here; IP
2251 * has already performed its header checksum validation. Also take
2252 * care of any trailing bytes and subtract out their partial sum.
2253 */
2254 if (ip->ip_p == IPPROTO_UDP && hlen == sizeof(struct ip) &&
2255 (m->m_pkthdr.csum_flags &
2256 (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
2257 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
2258 uint32_t start = m->m_pkthdr.csum_rx_start;
2259 int32_t trailer = (m_pktlen(m) - ip->ip_len);
2260 uint32_t swbytes = (uint32_t)trailer;
2261
2262 csum = m->m_pkthdr.csum_rx_val;
2263
2264 ASSERT(trailer >= 0);
2265 if ((start != 0 && start != hlen) || trailer != 0) {
2266 uint32_t datalen = ip->ip_len - hlen;
2267
2268 #if BYTE_ORDER != BIG_ENDIAN
2269 if (start < hlen) {
2270 HTONS(ip->ip_len);
2271 HTONS(ip->ip_off);
2272 }
2273 #endif /* BYTE_ORDER != BIG_ENDIAN */
2274 /* callee folds in sum */
2275 csum = m_adj_sum16(m, start, hlen, datalen, csum);
2276 if (hlen > start) {
2277 swbytes += (hlen - start);
2278 } else {
2279 swbytes += (start - hlen);
2280 }
2281 #if BYTE_ORDER != BIG_ENDIAN
2282 if (start < hlen) {
2283 NTOHS(ip->ip_off);
2284 NTOHS(ip->ip_len);
2285 }
2286 #endif /* BYTE_ORDER != BIG_ENDIAN */
2287 }
2288 csum_flags = m->m_pkthdr.csum_flags;
2289
2290 if (swbytes != 0) {
2291 udp_in_cksum_stats(swbytes);
2292 }
2293 if (trailer != 0) {
2294 m_adj(m, -trailer);
2295 }
2296 } else {
2297 csum = 0;
2298 csum_flags = 0;
2299 }
2300
2301 /* Invalidate checksum */
2302 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
2303
2304 ipstat.ips_fragments++;
2305
2306 /*
2307 * Adjust ip_len to not reflect header,
2308 * convert offset of this to bytes.
2309 */
2310 ip->ip_len -= hlen;
2311 if (ip->ip_off & IP_MF) {
2312 /*
2313 * Make sure that fragments have a data length
2314 * that's a non-zero multiple of 8 bytes.
2315 */
2316 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
2317 OSAddAtomic(1, &ipstat.ips_toosmall);
2318 /*
2319 * Reassembly queue may have been found if previous
2320 * fragments were valid; given that this one is bad,
2321 * we need to drop it. Make sure to set fp to NULL
2322 * if not already, since we don't want to decrement
2323 * ipq_nfrags as it doesn't include this packet.
2324 */
2325 fp = NULL;
2326 goto dropfrag;
2327 }
2328 m->m_flags |= M_FRAG;
2329 } else {
2330 /* Clear the flag in case packet comes from loopback */
2331 m->m_flags &= ~M_FRAG;
2332 }
2333 ip->ip_off <<= 3;
2334
2335 m->m_pkthdr.pkt_hdr = ip;
2336
2337 /* Previous ip_reass() started here. */
2338 /*
2339 * Presence of header sizes in mbufs
2340 * would confuse code below.
2341 */
2342 m->m_data += hlen;
2343 m->m_len -= hlen;
2344
2345 /*
2346 * If first fragment to arrive, create a reassembly queue.
2347 */
2348 if (fp == NULL) {
2349 fp = ipq_alloc(M_DONTWAIT);
2350 if (fp == NULL) {
2351 goto dropfrag;
2352 }
2353 TAILQ_INSERT_HEAD(head, fp, ipq_list);
2354 nipq++;
2355 fp->ipq_nfrags = 1;
2356 fp->ipq_ttl = IPFRAGTTL;
2357 fp->ipq_p = ip->ip_p;
2358 fp->ipq_id = ip->ip_id;
2359 fp->ipq_src = ip->ip_src;
2360 fp->ipq_dst = ip->ip_dst;
2361 fp->ipq_frags = m;
2362 m->m_nextpkt = NULL;
2363 /*
2364 * If the first fragment has valid checksum offload
2365 * info, the rest of fragments are eligible as well.
2366 */
2367 if (csum_flags != 0) {
2368 fp->ipq_csum = csum;
2369 fp->ipq_csum_flags = csum_flags;
2370 }
2371 m = NULL; /* nothing to return */
2372 goto done;
2373 } else {
2374 fp->ipq_nfrags++;
2375 }
2376
2377 #define GETIP(m) ((struct ip *)((m)->m_pkthdr.pkt_hdr))
2378
2379 /*
2380 * Handle ECN by comparing this segment with the first one;
2381 * if CE is set, do not lose CE.
2382 * drop if CE and not-ECT are mixed for the same packet.
2383 */
2384 ecn = ip->ip_tos & IPTOS_ECN_MASK;
2385 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
2386 if (ecn == IPTOS_ECN_CE) {
2387 if (ecn0 == IPTOS_ECN_NOTECT) {
2388 goto dropfrag;
2389 }
2390 if (ecn0 != IPTOS_ECN_CE) {
2391 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
2392 }
2393 }
2394 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
2395 goto dropfrag;
2396 }
2397
2398 /*
2399 * Find a segment which begins after this one does.
2400 */
2401 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2402 if (GETIP(q)->ip_off > ip->ip_off) {
2403 break;
2404 }
2405 }
2406
2407 /*
2408 * If there is a preceding segment, it may provide some of
2409 * our data already. If so, drop the data from the incoming
2410 * segment. If it provides all of our data, drop us, otherwise
2411 * stick new segment in the proper place.
2412 *
2413 * If some of the data is dropped from the preceding
2414 * segment, then it's checksum is invalidated.
2415 */
2416 if (p) {
2417 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
2418 if (i > 0) {
2419 if (i >= ip->ip_len) {
2420 goto dropfrag;
2421 }
2422 m_adj(m, i);
2423 fp->ipq_csum_flags = 0;
2424 ip->ip_off += i;
2425 ip->ip_len -= i;
2426 }
2427 m->m_nextpkt = p->m_nextpkt;
2428 p->m_nextpkt = m;
2429 } else {
2430 m->m_nextpkt = fp->ipq_frags;
2431 fp->ipq_frags = m;
2432 }
2433
2434 /*
2435 * While we overlap succeeding segments trim them or,
2436 * if they are completely covered, dequeue them.
2437 */
2438 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
2439 q = nq) {
2440 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
2441 if (i < GETIP(q)->ip_len) {
2442 GETIP(q)->ip_len -= i;
2443 GETIP(q)->ip_off += i;
2444 m_adj(q, i);
2445 fp->ipq_csum_flags = 0;
2446 break;
2447 }
2448 nq = q->m_nextpkt;
2449 m->m_nextpkt = nq;
2450 ipstat.ips_fragdropped++;
2451 fp->ipq_nfrags--;
2452 /* defer freeing until after lock is dropped */
2453 MBUFQ_ENQUEUE(&dfq, q);
2454 }
2455
2456 /*
2457 * If this fragment contains similar checksum offload info
2458 * as that of the existing ones, accumulate checksum. Otherwise,
2459 * invalidate checksum offload info for the entire datagram.
2460 */
2461 if (csum_flags != 0 && csum_flags == fp->ipq_csum_flags) {
2462 fp->ipq_csum += csum;
2463 } else if (fp->ipq_csum_flags != 0) {
2464 fp->ipq_csum_flags = 0;
2465 }
2466
2467
2468 /*
2469 * Check for complete reassembly and perform frag per packet
2470 * limiting.
2471 *
2472 * Frag limiting is performed here so that the nth frag has
2473 * a chance to complete the packet before we drop the packet.
2474 * As a result, n+1 frags are actually allowed per packet, but
2475 * only n will ever be stored. (n = maxfragsperpacket.)
2476 *
2477 */
2478 next = 0;
2479 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2480 if (GETIP(q)->ip_off != next) {
2481 if (fp->ipq_nfrags > maxfragsperpacket) {
2482 ipstat.ips_fragdropped += fp->ipq_nfrags;
2483 frag_freef(head, fp);
2484 }
2485 m = NULL; /* nothing to return */
2486 goto done;
2487 }
2488 next += GETIP(q)->ip_len;
2489 }
2490 /* Make sure the last packet didn't have the IP_MF flag */
2491 if (p->m_flags & M_FRAG) {
2492 if (fp->ipq_nfrags > maxfragsperpacket) {
2493 ipstat.ips_fragdropped += fp->ipq_nfrags;
2494 frag_freef(head, fp);
2495 }
2496 m = NULL; /* nothing to return */
2497 goto done;
2498 }
2499
2500 /*
2501 * Reassembly is complete. Make sure the packet is a sane size.
2502 */
2503 q = fp->ipq_frags;
2504 ip = GETIP(q);
2505 if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
2506 ipstat.ips_toolong++;
2507 ipstat.ips_fragdropped += fp->ipq_nfrags;
2508 frag_freef(head, fp);
2509 m = NULL; /* nothing to return */
2510 goto done;
2511 }
2512
2513 /*
2514 * Concatenate fragments.
2515 */
2516 m = q;
2517 t = m->m_next;
2518 m->m_next = NULL;
2519 m_cat(m, t);
2520 nq = q->m_nextpkt;
2521 q->m_nextpkt = NULL;
2522 for (q = nq; q != NULL; q = nq) {
2523 nq = q->m_nextpkt;
2524 q->m_nextpkt = NULL;
2525 m_cat(m, q);
2526 }
2527
2528 /*
2529 * Store partial hardware checksum info from the fragment queue;
2530 * the receive start offset is set to 20 bytes (see code at the
2531 * top of this routine.)
2532 */
2533 if (fp->ipq_csum_flags != 0) {
2534 csum = fp->ipq_csum;
2535
2536 ADDCARRY(csum);
2537
2538 m->m_pkthdr.csum_rx_val = (uint16_t)csum;
2539 m->m_pkthdr.csum_rx_start = sizeof(struct ip);
2540 m->m_pkthdr.csum_flags = fp->ipq_csum_flags;
2541 } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
2542 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
2543 /* loopback checksums are always OK */
2544 m->m_pkthdr.csum_data = 0xffff;
2545 m->m_pkthdr.csum_flags =
2546 CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
2547 CSUM_IP_CHECKED | CSUM_IP_VALID;
2548 }
2549
2550 /*
2551 * Create header for new ip packet by modifying header of first
2552 * packet; dequeue and discard fragment reassembly header.
2553 * Make header visible.
2554 */
2555 ip->ip_len = (u_short)((IP_VHL_HL(ip->ip_vhl) << 2) + next);
2556 ip->ip_src = fp->ipq_src;
2557 ip->ip_dst = fp->ipq_dst;
2558
2559 fp->ipq_frags = NULL; /* return to caller as 'm' */
2560 frag_freef(head, fp);
2561 fp = NULL;
2562
2563 m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
2564 m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
2565 /* some debugging cruft by sklower, below, will go away soon */
2566 if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
2567 m_fixhdr(m);
2568 }
2569 ipstat.ips_reassembled++;
2570
2571 /* arm the purge timer if not already and if there's work to do */
2572 frag_sched_timeout();
2573 lck_mtx_unlock(&ipqlock);
2574 /* perform deferred free (if needed) now that lock is dropped */
2575 if (!MBUFQ_EMPTY(&dfq)) {
2576 MBUFQ_DRAIN(&dfq);
2577 }
2578 VERIFY(MBUFQ_EMPTY(&dfq));
2579 return m;
2580
2581 done:
2582 VERIFY(m == NULL);
2583 /* arm the purge timer if not already and if there's work to do */
2584 frag_sched_timeout();
2585 lck_mtx_unlock(&ipqlock);
2586 /* perform deferred free (if needed) */
2587 if (!MBUFQ_EMPTY(&dfq)) {
2588 MBUFQ_DRAIN(&dfq);
2589 }
2590 VERIFY(MBUFQ_EMPTY(&dfq));
2591 return NULL;
2592
2593 dropfrag:
2594 ipstat.ips_fragdropped++;
2595 if (fp != NULL) {
2596 fp->ipq_nfrags--;
2597 }
2598 /* arm the purge timer if not already and if there's work to do */
2599 frag_sched_timeout();
2600 lck_mtx_unlock(&ipqlock);
2601 m_freem(m);
2602 /* perform deferred free (if needed) */
2603 if (!MBUFQ_EMPTY(&dfq)) {
2604 MBUFQ_DRAIN(&dfq);
2605 }
2606 VERIFY(MBUFQ_EMPTY(&dfq));
2607 return NULL;
2608 #undef GETIP
2609 }
2610
2611 /*
2612 * Free a fragment reassembly header and all
2613 * associated datagrams.
2614 */
2615 static void
frag_freef(struct ipqhead * fhp,struct ipq * fp)2616 frag_freef(struct ipqhead *fhp, struct ipq *fp)
2617 {
2618 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2619
2620 fp->ipq_nfrags = 0;
2621 if (fp->ipq_frags != NULL) {
2622 m_freem_list(fp->ipq_frags);
2623 fp->ipq_frags = NULL;
2624 }
2625 TAILQ_REMOVE(fhp, fp, ipq_list);
2626 nipq--;
2627 ipq_free(fp);
2628 }
2629
2630 /*
2631 * IP reassembly timer processing
2632 */
2633 static void
frag_timeout(void * arg)2634 frag_timeout(void *arg)
2635 {
2636 #pragma unused(arg)
2637 struct ipq *fp;
2638 int i;
2639
2640 /*
2641 * Update coarse-grained networking timestamp (in sec.); the idea
2642 * is to piggy-back on the timeout callout to update the counter
2643 * returnable via net_uptime().
2644 */
2645 net_update_uptime();
2646
2647 lck_mtx_lock(&ipqlock);
2648 for (i = 0; i < IPREASS_NHASH; i++) {
2649 for (fp = TAILQ_FIRST(&ipq[i]); fp;) {
2650 struct ipq *fpp;
2651
2652 fpp = fp;
2653 fp = TAILQ_NEXT(fp, ipq_list);
2654 if (--fpp->ipq_ttl == 0) {
2655 ipstat.ips_fragtimeout += fpp->ipq_nfrags;
2656 frag_freef(&ipq[i], fpp);
2657 }
2658 }
2659 }
2660 /*
2661 * If we are over the maximum number of fragments
2662 * (due to the limit being lowered), drain off
2663 * enough to get down to the new limit.
2664 */
2665 if (maxnipq >= 0 && nipq > (unsigned)maxnipq) {
2666 for (i = 0; i < IPREASS_NHASH; i++) {
2667 while (nipq > (unsigned)maxnipq &&
2668 !TAILQ_EMPTY(&ipq[i])) {
2669 ipstat.ips_fragdropped +=
2670 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2671 frag_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
2672 }
2673 }
2674 }
2675 /* re-arm the purge timer if there's work to do */
2676 frag_timeout_run = 0;
2677 frag_sched_timeout();
2678 lck_mtx_unlock(&ipqlock);
2679 }
2680
2681 static void
frag_sched_timeout(void)2682 frag_sched_timeout(void)
2683 {
2684 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2685
2686 if (!frag_timeout_run && nipq > 0) {
2687 frag_timeout_run = 1;
2688 timeout(frag_timeout, NULL, hz);
2689 }
2690 }
2691
2692 /*
2693 * Drain off all datagram fragments.
2694 */
2695 static void
frag_drain(void)2696 frag_drain(void)
2697 {
2698 int i;
2699
2700 lck_mtx_lock(&ipqlock);
2701 for (i = 0; i < IPREASS_NHASH; i++) {
2702 while (!TAILQ_EMPTY(&ipq[i])) {
2703 ipstat.ips_fragdropped +=
2704 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2705 frag_freef(&ipq[i], TAILQ_FIRST(&ipq[i]));
2706 }
2707 }
2708 lck_mtx_unlock(&ipqlock);
2709 }
2710
2711 static struct ipq *
ipq_alloc(int how)2712 ipq_alloc(int how)
2713 {
2714 struct mbuf *t;
2715 struct ipq *fp;
2716
2717 /*
2718 * See comments in ipq_updateparams(). Keep the count separate
2719 * from nipq since the latter represents the elements already
2720 * in the reassembly queues.
2721 */
2722 if (ipq_limit > 0 && ipq_count > ipq_limit) {
2723 return NULL;
2724 }
2725
2726 t = m_get(how, MT_FTABLE);
2727 if (t != NULL) {
2728 atomic_add_32(&ipq_count, 1);
2729 fp = mtod(t, struct ipq *);
2730 bzero(fp, sizeof(*fp));
2731 } else {
2732 fp = NULL;
2733 }
2734 return fp;
2735 }
2736
2737 static void
ipq_free(struct ipq * fp)2738 ipq_free(struct ipq *fp)
2739 {
2740 (void) m_free(dtom(fp));
2741 atomic_add_32(&ipq_count, -1);
2742 }
2743
2744 /*
2745 * Drain callback
2746 */
2747 void
ip_drain(void)2748 ip_drain(void)
2749 {
2750 frag_drain(); /* fragments */
2751 in_rtqdrain(); /* protocol cloned routes */
2752 in_arpdrain(NULL); /* cloned routes: ARP */
2753 }
2754
2755 /*
2756 * Do option processing on a datagram,
2757 * possibly discarding it if bad options are encountered,
2758 * or forwarding it if source-routed.
2759 * The pass argument is used when operating in the IPSTEALTH
2760 * mode to tell what options to process:
2761 * [LS]SRR (pass 0) or the others (pass 1).
2762 * The reason for as many as two passes is that when doing IPSTEALTH,
2763 * non-routing options should be processed only if the packet is for us.
2764 * Returns 1 if packet has been forwarded/freed,
2765 * 0 if the packet should be processed further.
2766 */
2767 static int
ip_dooptions(struct mbuf * m,int pass,struct sockaddr_in * next_hop)2768 ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
2769 {
2770 #pragma unused(pass)
2771 struct ip *ip = mtod(m, struct ip *);
2772 u_char *cp;
2773 struct ip_timestamp *ipt;
2774 struct in_ifaddr *ia;
2775 int opt, optlen, cnt, off, type = ICMP_PARAMPROB, forward = 0;
2776 uint8_t code = 0;
2777 struct in_addr *sin, dst;
2778 u_int32_t ntime;
2779 struct sockaddr_in ipaddr = {
2780 .sin_len = sizeof(ipaddr),
2781 .sin_family = AF_INET,
2782 .sin_port = 0,
2783 .sin_addr = { .s_addr = 0 },
2784 .sin_zero = { 0, }
2785 };
2786
2787 /* Expect 32-bit aligned data pointer on strict-align platforms */
2788 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
2789
2790 dst = ip->ip_dst;
2791 cp = (u_char *)(ip + 1);
2792 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
2793 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2794 opt = cp[IPOPT_OPTVAL];
2795 if (opt == IPOPT_EOL) {
2796 break;
2797 }
2798 if (opt == IPOPT_NOP) {
2799 optlen = 1;
2800 } else {
2801 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
2802 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2803 goto bad;
2804 }
2805 optlen = cp[IPOPT_OLEN];
2806 if (optlen < IPOPT_OLEN + sizeof(*cp) ||
2807 optlen > cnt) {
2808 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2809 goto bad;
2810 }
2811 }
2812 switch (opt) {
2813 default:
2814 break;
2815
2816 /*
2817 * Source routing with record.
2818 * Find interface with current destination address.
2819 * If none on this machine then drop if strictly routed,
2820 * or do nothing if loosely routed.
2821 * Record interface address and bring up next address
2822 * component. If strictly routed make sure next
2823 * address is on directly accessible net.
2824 */
2825 case IPOPT_LSRR:
2826 case IPOPT_SSRR:
2827 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2828 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2829 goto bad;
2830 }
2831 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2832 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2833 goto bad;
2834 }
2835 ipaddr.sin_addr = ip->ip_dst;
2836 ia = (struct in_ifaddr *)ifa_ifwithaddr(SA(&ipaddr));
2837 if (ia == NULL) {
2838 if (opt == IPOPT_SSRR) {
2839 type = ICMP_UNREACH;
2840 code = ICMP_UNREACH_SRCFAIL;
2841 goto bad;
2842 }
2843 if (!ip_dosourceroute) {
2844 goto nosourcerouting;
2845 }
2846 /*
2847 * Loose routing, and not at next destination
2848 * yet; nothing to do except forward.
2849 */
2850 break;
2851 } else {
2852 IFA_REMREF(&ia->ia_ifa);
2853 ia = NULL;
2854 }
2855 off--; /* 0 origin */
2856 if (off > optlen - (int)sizeof(struct in_addr)) {
2857 /*
2858 * End of source route. Should be for us.
2859 */
2860 if (!ip_acceptsourceroute) {
2861 goto nosourcerouting;
2862 }
2863 save_rte(cp, ip->ip_src);
2864 break;
2865 }
2866
2867 if (!ip_dosourceroute) {
2868 if (ipforwarding) {
2869 char buf[MAX_IPv4_STR_LEN];
2870 char buf2[MAX_IPv4_STR_LEN];
2871 /*
2872 * Acting as a router, so generate ICMP
2873 */
2874 nosourcerouting:
2875 log(LOG_WARNING,
2876 "attempted source route from %s "
2877 "to %s\n",
2878 inet_ntop(AF_INET, &ip->ip_src,
2879 buf, sizeof(buf)),
2880 inet_ntop(AF_INET, &ip->ip_dst,
2881 buf2, sizeof(buf2)));
2882 type = ICMP_UNREACH;
2883 code = ICMP_UNREACH_SRCFAIL;
2884 goto bad;
2885 } else {
2886 /*
2887 * Not acting as a router,
2888 * so silently drop.
2889 */
2890 OSAddAtomic(1, &ipstat.ips_cantforward);
2891 m_freem(m);
2892 return 1;
2893 }
2894 }
2895
2896 /*
2897 * locate outgoing interface
2898 */
2899 (void) memcpy(&ipaddr.sin_addr, cp + off,
2900 sizeof(ipaddr.sin_addr));
2901
2902 if (opt == IPOPT_SSRR) {
2903 #define INA struct in_ifaddr *
2904 if ((ia = (INA)ifa_ifwithdstaddr(
2905 SA(&ipaddr))) == NULL) {
2906 ia = (INA)ifa_ifwithnet(SA(&ipaddr));
2907 }
2908 } else {
2909 ia = ip_rtaddr(ipaddr.sin_addr);
2910 }
2911 if (ia == NULL) {
2912 type = ICMP_UNREACH;
2913 code = ICMP_UNREACH_SRCFAIL;
2914 goto bad;
2915 }
2916 ip->ip_dst = ipaddr.sin_addr;
2917 IFA_LOCK(&ia->ia_ifa);
2918 (void) memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
2919 sizeof(struct in_addr));
2920 IFA_UNLOCK(&ia->ia_ifa);
2921 IFA_REMREF(&ia->ia_ifa);
2922 ia = NULL;
2923 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2924 /*
2925 * Let ip_intr's mcast routing check handle mcast pkts
2926 */
2927 forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
2928 break;
2929
2930 case IPOPT_RR:
2931 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2932 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2933 goto bad;
2934 }
2935 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2936 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2937 goto bad;
2938 }
2939 /*
2940 * If no space remains, ignore.
2941 */
2942 off--; /* 0 origin */
2943 if (off > optlen - (int)sizeof(struct in_addr)) {
2944 break;
2945 }
2946 (void) memcpy(&ipaddr.sin_addr, &ip->ip_dst,
2947 sizeof(ipaddr.sin_addr));
2948 /*
2949 * locate outgoing interface; if we're the destination,
2950 * use the incoming interface (should be same).
2951 */
2952 if ((ia = (INA)ifa_ifwithaddr(SA(&ipaddr))) == NULL) {
2953 if ((ia = ip_rtaddr(ipaddr.sin_addr)) == NULL) {
2954 type = ICMP_UNREACH;
2955 code = ICMP_UNREACH_HOST;
2956 goto bad;
2957 }
2958 }
2959 IFA_LOCK(&ia->ia_ifa);
2960 (void) memcpy(cp + off, &(IA_SIN(ia)->sin_addr),
2961 sizeof(struct in_addr));
2962 IFA_UNLOCK(&ia->ia_ifa);
2963 IFA_REMREF(&ia->ia_ifa);
2964 ia = NULL;
2965 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2966 break;
2967
2968 case IPOPT_TS:
2969 code = (uint8_t)(cp - (u_char *)ip);
2970 ipt = (struct ip_timestamp *)(void *)cp;
2971 if (ipt->ipt_len < 4 || ipt->ipt_len > 40) {
2972 code = (uint8_t)((u_char *)&ipt->ipt_len -
2973 (u_char *)ip);
2974 goto bad;
2975 }
2976 if (ipt->ipt_ptr < 5) {
2977 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2978 (u_char *)ip);
2979 goto bad;
2980 }
2981 if (ipt->ipt_ptr >
2982 ipt->ipt_len - (int)sizeof(int32_t)) {
2983 if (++ipt->ipt_oflw == 0) {
2984 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2985 (u_char *)ip);
2986 goto bad;
2987 }
2988 break;
2989 }
2990 sin = (struct in_addr *)(void *)(cp + ipt->ipt_ptr - 1);
2991 switch (ipt->ipt_flg) {
2992 case IPOPT_TS_TSONLY:
2993 break;
2994
2995 case IPOPT_TS_TSANDADDR:
2996 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
2997 sizeof(struct in_addr) > ipt->ipt_len) {
2998 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2999 (u_char *)ip);
3000 goto bad;
3001 }
3002 ipaddr.sin_addr = dst;
3003 ia = (INA)ifaof_ifpforaddr(SA(&ipaddr),
3004 m->m_pkthdr.rcvif);
3005 if (ia == NULL) {
3006 continue;
3007 }
3008 IFA_LOCK(&ia->ia_ifa);
3009 (void) memcpy(sin, &IA_SIN(ia)->sin_addr,
3010 sizeof(struct in_addr));
3011 IFA_UNLOCK(&ia->ia_ifa);
3012 ipt->ipt_ptr += sizeof(struct in_addr);
3013 IFA_REMREF(&ia->ia_ifa);
3014 ia = NULL;
3015 break;
3016
3017 case IPOPT_TS_PRESPEC:
3018 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
3019 sizeof(struct in_addr) > ipt->ipt_len) {
3020 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3021 (u_char *)ip);
3022 goto bad;
3023 }
3024 (void) memcpy(&ipaddr.sin_addr, sin,
3025 sizeof(struct in_addr));
3026 if ((ia = (struct in_ifaddr *)ifa_ifwithaddr(
3027 SA(&ipaddr))) == NULL) {
3028 continue;
3029 }
3030 IFA_REMREF(&ia->ia_ifa);
3031 ia = NULL;
3032 ipt->ipt_ptr += sizeof(struct in_addr);
3033 break;
3034
3035 default:
3036 /* XXX can't take &ipt->ipt_flg */
3037 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3038 (u_char *)ip + 1);
3039 goto bad;
3040 }
3041 ntime = iptime();
3042 (void) memcpy(cp + ipt->ipt_ptr - 1, &ntime,
3043 sizeof(n_time));
3044 ipt->ipt_ptr += sizeof(n_time);
3045 }
3046 }
3047 if (forward && ipforwarding) {
3048 ip_forward(m, 1, next_hop);
3049 return 1;
3050 }
3051 return 0;
3052 bad:
3053 icmp_error(m, type, code, 0, 0);
3054 OSAddAtomic(1, &ipstat.ips_badoptions);
3055 return 1;
3056 }
3057
3058 /*
3059 * Check for the presence of the IP Router Alert option [RFC2113]
3060 * in the header of an IPv4 datagram.
3061 *
3062 * This call is not intended for use from the forwarding path; it is here
3063 * so that protocol domains may check for the presence of the option.
3064 * Given how FreeBSD's IPv4 stack is currently structured, the Router Alert
3065 * option does not have much relevance to the implementation, though this
3066 * may change in future.
3067 * Router alert options SHOULD be passed if running in IPSTEALTH mode and
3068 * we are not the endpoint.
3069 * Length checks on individual options should already have been peformed
3070 * by ip_dooptions() therefore they are folded under DIAGNOSTIC here.
3071 *
3072 * Return zero if not present or options are invalid, non-zero if present.
3073 */
3074 int
ip_checkrouteralert(struct mbuf * m)3075 ip_checkrouteralert(struct mbuf *m)
3076 {
3077 struct ip *ip = mtod(m, struct ip *);
3078 u_char *cp;
3079 int opt, optlen, cnt, found_ra;
3080
3081 found_ra = 0;
3082 cp = (u_char *)(ip + 1);
3083 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3084 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3085 opt = cp[IPOPT_OPTVAL];
3086 if (opt == IPOPT_EOL) {
3087 break;
3088 }
3089 if (opt == IPOPT_NOP) {
3090 optlen = 1;
3091 } else {
3092 #ifdef DIAGNOSTIC
3093 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
3094 break;
3095 }
3096 #endif
3097 optlen = cp[IPOPT_OLEN];
3098 #ifdef DIAGNOSTIC
3099 if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
3100 break;
3101 }
3102 #endif
3103 }
3104 switch (opt) {
3105 case IPOPT_RA:
3106 #ifdef DIAGNOSTIC
3107 if (optlen != IPOPT_OFFSET + sizeof(uint16_t) ||
3108 (*((uint16_t *)(void *)&cp[IPOPT_OFFSET]) != 0)) {
3109 break;
3110 } else
3111 #endif
3112 found_ra = 1;
3113 break;
3114 default:
3115 break;
3116 }
3117 }
3118
3119 return found_ra;
3120 }
3121
3122 /*
3123 * Given address of next destination (final or next hop),
3124 * return internet address info of interface to be used to get there.
3125 */
3126 struct in_ifaddr *
ip_rtaddr(struct in_addr dst)3127 ip_rtaddr(struct in_addr dst)
3128 {
3129 struct sockaddr_in *sin;
3130 struct ifaddr *rt_ifa;
3131 struct route ro;
3132
3133 bzero(&ro, sizeof(ro));
3134 sin = SIN(&ro.ro_dst);
3135 sin->sin_family = AF_INET;
3136 sin->sin_len = sizeof(*sin);
3137 sin->sin_addr = dst;
3138
3139 rtalloc_ign(&ro, RTF_PRCLONING);
3140 if (ro.ro_rt == NULL) {
3141 ROUTE_RELEASE(&ro);
3142 return NULL;
3143 }
3144
3145 RT_LOCK(ro.ro_rt);
3146 if ((rt_ifa = ro.ro_rt->rt_ifa) != NULL) {
3147 IFA_ADDREF(rt_ifa);
3148 }
3149 RT_UNLOCK(ro.ro_rt);
3150 ROUTE_RELEASE(&ro);
3151
3152 return (struct in_ifaddr *)rt_ifa;
3153 }
3154
3155 /*
3156 * Save incoming source route for use in replies,
3157 * to be picked up later by ip_srcroute if the receiver is interested.
3158 */
3159 void
save_rte(u_char * option,struct in_addr dst)3160 save_rte(u_char *option, struct in_addr dst)
3161 {
3162 unsigned olen;
3163
3164 olen = option[IPOPT_OLEN];
3165 #if DIAGNOSTIC
3166 if (ipprintfs) {
3167 printf("save_rte: olen %d\n", olen);
3168 }
3169 #endif
3170 if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst))) {
3171 return;
3172 }
3173 bcopy(option, ip_srcrt.srcopt, olen);
3174 ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
3175 ip_srcrt.dst = dst;
3176 }
3177
3178 /*
3179 * Retrieve incoming source route for use in replies,
3180 * in the same form used by setsockopt.
3181 * The first hop is placed before the options, will be removed later.
3182 */
3183 struct mbuf *
ip_srcroute(void)3184 ip_srcroute(void)
3185 {
3186 struct in_addr *p, *q;
3187 struct mbuf *m;
3188
3189 if (ip_nhops == 0) {
3190 return NULL;
3191 }
3192
3193 m = m_get(M_DONTWAIT, MT_HEADER);
3194 if (m == NULL) {
3195 return NULL;
3196 }
3197
3198 #define OPTSIZ (sizeof (ip_srcrt.nop) + sizeof (ip_srcrt.srcopt))
3199
3200 /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
3201 m->m_len = ip_nhops * sizeof(struct in_addr) +
3202 sizeof(struct in_addr) + OPTSIZ;
3203 #if DIAGNOSTIC
3204 if (ipprintfs) {
3205 printf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
3206 }
3207 #endif
3208
3209 /*
3210 * First save first hop for return route
3211 */
3212 p = &ip_srcrt.route[ip_nhops - 1];
3213 *(mtod(m, struct in_addr *)) = *p--;
3214 #if DIAGNOSTIC
3215 if (ipprintfs) {
3216 printf(" hops %lx",
3217 (u_int32_t)ntohl(mtod(m, struct in_addr *)->s_addr));
3218 }
3219 #endif
3220
3221 /*
3222 * Copy option fields and padding (nop) to mbuf.
3223 */
3224 ip_srcrt.nop = IPOPT_NOP;
3225 ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
3226 (void) memcpy(mtod(m, caddr_t) + sizeof(struct in_addr),
3227 &ip_srcrt.nop, OPTSIZ);
3228 q = (struct in_addr *)(void *)(mtod(m, caddr_t) +
3229 sizeof(struct in_addr) + OPTSIZ);
3230 #undef OPTSIZ
3231 /*
3232 * Record return path as an IP source route,
3233 * reversing the path (pointers are now aligned).
3234 */
3235 while (p >= ip_srcrt.route) {
3236 #if DIAGNOSTIC
3237 if (ipprintfs) {
3238 printf(" %lx", (u_int32_t)ntohl(q->s_addr));
3239 }
3240 #endif
3241 *q++ = *p--;
3242 }
3243 /*
3244 * Last hop goes to final destination.
3245 */
3246 *q = ip_srcrt.dst;
3247 #if DIAGNOSTIC
3248 if (ipprintfs) {
3249 printf(" %lx\n", (u_int32_t)ntohl(q->s_addr));
3250 }
3251 #endif
3252 return m;
3253 }
3254
3255 /*
3256 * Strip out IP options, at higher level protocol in the kernel.
3257 */
3258 void
ip_stripoptions(struct mbuf * m)3259 ip_stripoptions(struct mbuf *m)
3260 {
3261 int i;
3262 struct ip *ip = mtod(m, struct ip *);
3263 caddr_t opts;
3264 int olen;
3265
3266 /* Expect 32-bit aligned data pointer on strict-align platforms */
3267 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
3268
3269 /* use bcopy() since it supports overlapping range */
3270 olen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3271 opts = (caddr_t)(ip + 1);
3272 i = m->m_len - (sizeof(struct ip) + olen);
3273 bcopy(opts + olen, opts, (unsigned)i);
3274 m->m_len -= olen;
3275 if (m->m_flags & M_PKTHDR) {
3276 m->m_pkthdr.len -= olen;
3277 }
3278 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
3279
3280 /*
3281 * We expect ip_{off,len} to be in host order by now, and
3282 * that the original IP header length has been subtracted
3283 * out from ip_len. Temporarily adjust ip_len for checksum
3284 * recalculation, and restore it afterwards.
3285 */
3286 ip->ip_len += sizeof(struct ip);
3287
3288 /* recompute checksum now that IP header is smaller */
3289 #if BYTE_ORDER != BIG_ENDIAN
3290 HTONS(ip->ip_len);
3291 HTONS(ip->ip_off);
3292 #endif /* BYTE_ORDER != BIG_ENDIAN */
3293 ip->ip_sum = in_cksum_hdr(ip);
3294 #if BYTE_ORDER != BIG_ENDIAN
3295 NTOHS(ip->ip_off);
3296 NTOHS(ip->ip_len);
3297 #endif /* BYTE_ORDER != BIG_ENDIAN */
3298
3299 ip->ip_len -= sizeof(struct ip);
3300
3301 /*
3302 * Given that we've just stripped IP options from the header,
3303 * we need to adjust the start offset accordingly if this
3304 * packet had gone thru partial checksum offload.
3305 */
3306 if ((m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
3307 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
3308 if (m->m_pkthdr.csum_rx_start >= (sizeof(struct ip) + olen)) {
3309 /* most common case */
3310 m->m_pkthdr.csum_rx_start -= olen;
3311 } else {
3312 /* compute checksum in software instead */
3313 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
3314 m->m_pkthdr.csum_data = 0;
3315 ipstat.ips_adj_hwcsum_clr++;
3316 }
3317 }
3318 }
3319
3320 u_char inetctlerrmap[PRC_NCMDS] = {
3321 0, 0, 0, 0,
3322 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH,
3323 ENETUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED,
3324 EMSGSIZE, EHOSTUNREACH, 0, 0,
3325 0, 0, EHOSTUNREACH, 0,
3326 ENOPROTOOPT, ECONNREFUSED
3327 };
3328
3329 static int
3330 sysctl_ipforwarding SYSCTL_HANDLER_ARGS
3331 {
3332 #pragma unused(arg1, arg2)
3333 int i, was_ipforwarding = ipforwarding;
3334
3335 i = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req);
3336 if (i != 0 || req->newptr == USER_ADDR_NULL) {
3337 return i;
3338 }
3339
3340 if (was_ipforwarding && !ipforwarding) {
3341 /* clean up IPv4 forwarding cached routes */
3342 ifnet_head_lock_shared();
3343 for (i = 0; i <= if_index; i++) {
3344 struct ifnet *ifp = ifindex2ifnet[i];
3345 if (ifp != NULL) {
3346 lck_mtx_lock(&ifp->if_cached_route_lock);
3347 ROUTE_RELEASE(&ifp->if_fwd_route);
3348 bzero(&ifp->if_fwd_route,
3349 sizeof(ifp->if_fwd_route));
3350 lck_mtx_unlock(&ifp->if_cached_route_lock);
3351 }
3352 }
3353 ifnet_head_done();
3354 }
3355
3356 return 0;
3357 }
3358
3359 /*
3360 * Similar to inp_route_{copyout,copyin} routines except that these copy
3361 * out the cached IPv4 forwarding route from struct ifnet instead of the
3362 * inpcb. See comments for those routines for explanations.
3363 */
3364 static void
ip_fwd_route_copyout(struct ifnet * ifp,struct route * dst)3365 ip_fwd_route_copyout(struct ifnet *ifp, struct route *dst)
3366 {
3367 struct route *src = &ifp->if_fwd_route;
3368
3369 lck_mtx_lock_spin(&ifp->if_cached_route_lock);
3370 lck_mtx_convert_spin(&ifp->if_cached_route_lock);
3371
3372 /* Minor sanity check */
3373 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3374 panic("%s: wrong or corrupted route: %p", __func__, src);
3375 }
3376
3377 route_copyout(dst, src, sizeof(*dst));
3378
3379 lck_mtx_unlock(&ifp->if_cached_route_lock);
3380 }
3381
3382 static void
ip_fwd_route_copyin(struct ifnet * ifp,struct route * src)3383 ip_fwd_route_copyin(struct ifnet *ifp, struct route *src)
3384 {
3385 struct route *dst = &ifp->if_fwd_route;
3386
3387 lck_mtx_lock_spin(&ifp->if_cached_route_lock);
3388 lck_mtx_convert_spin(&ifp->if_cached_route_lock);
3389
3390 /* Minor sanity check */
3391 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3392 panic("%s: wrong or corrupted route: %p", __func__, src);
3393 }
3394
3395 if (ifp->if_fwd_cacheok) {
3396 route_copyin(src, dst, sizeof(*src));
3397 }
3398
3399 lck_mtx_unlock(&ifp->if_cached_route_lock);
3400 }
3401
3402 /*
3403 * Forward a packet. If some error occurs return the sender
3404 * an icmp packet. Note we can't always generate a meaningful
3405 * icmp message because icmp doesn't have a large enough repertoire
3406 * of codes and types.
3407 *
3408 * If not forwarding, just drop the packet. This could be confusing
3409 * if ipforwarding was zero but some routing protocol was advancing
3410 * us as a gateway to somewhere. However, we must let the routing
3411 * protocol deal with that.
3412 *
3413 * The srcrt parameter indicates whether the packet is being forwarded
3414 * via a source route.
3415 */
3416 static void
ip_forward(struct mbuf * m,int srcrt,struct sockaddr_in * next_hop)3417 ip_forward(struct mbuf *m, int srcrt, struct sockaddr_in *next_hop)
3418 {
3419 #pragma unused(next_hop)
3420 struct ip *ip = mtod(m, struct ip *);
3421 struct sockaddr_in *sin;
3422 struct rtentry *rt;
3423 struct route fwd_rt;
3424 int error, type = 0, code = 0;
3425 struct mbuf *mcopy;
3426 n_long dest;
3427 struct in_addr pkt_dst;
3428 u_int32_t nextmtu = 0, len;
3429 struct ip_out_args ipoa;
3430 struct ifnet *rcvifp = m->m_pkthdr.rcvif;
3431
3432 bzero(&ipoa, sizeof(ipoa));
3433 ipoa.ipoa_boundif = IFSCOPE_NONE;
3434 ipoa.ipoa_sotc = SO_TC_UNSPEC;
3435 ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
3436
3437 #if IPSEC
3438 struct secpolicy *sp = NULL;
3439 int ipsecerror;
3440 #endif /* IPSEC */
3441 #if PF
3442 struct pf_mtag *pf_mtag;
3443 #endif /* PF */
3444
3445 dest = 0;
3446 pkt_dst = ip->ip_dst;
3447
3448 #if DIAGNOSTIC
3449 if (ipprintfs) {
3450 printf("forward: src %lx dst %lx ttl %x\n",
3451 (u_int32_t)ip->ip_src.s_addr, (u_int32_t)pkt_dst.s_addr,
3452 ip->ip_ttl);
3453 }
3454 #endif
3455
3456 if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
3457 OSAddAtomic(1, &ipstat.ips_cantforward);
3458 m_freem(m);
3459 return;
3460 }
3461 #if IPSTEALTH
3462 if (!ipstealth) {
3463 #endif /* IPSTEALTH */
3464 if (ip->ip_ttl <= IPTTLDEC) {
3465 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
3466 dest, 0);
3467 return;
3468 }
3469 #if IPSTEALTH
3470 }
3471 #endif /* IPSTEALTH */
3472
3473 #if PF
3474 pf_mtag = pf_find_mtag(m);
3475 if (pf_mtag != NULL && pf_mtag->pftag_rtableid != IFSCOPE_NONE) {
3476 ipoa.ipoa_boundif = pf_mtag->pftag_rtableid;
3477 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
3478 }
3479 #endif /* PF */
3480
3481 ip_fwd_route_copyout(rcvifp, &fwd_rt);
3482
3483 sin = SIN(&fwd_rt.ro_dst);
3484 if (ROUTE_UNUSABLE(&fwd_rt) || pkt_dst.s_addr != sin->sin_addr.s_addr) {
3485 ROUTE_RELEASE(&fwd_rt);
3486
3487 sin->sin_family = AF_INET;
3488 sin->sin_len = sizeof(*sin);
3489 sin->sin_addr = pkt_dst;
3490
3491 rtalloc_scoped_ign(&fwd_rt, RTF_PRCLONING, ipoa.ipoa_boundif);
3492 if (fwd_rt.ro_rt == NULL) {
3493 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
3494 goto done;
3495 }
3496 }
3497 rt = fwd_rt.ro_rt;
3498
3499 /*
3500 * Save the IP header and at most 8 bytes of the payload,
3501 * in case we need to generate an ICMP message to the src.
3502 *
3503 * We don't use m_copy() because it might return a reference
3504 * to a shared cluster. Both this function and ip_output()
3505 * assume exclusive access to the IP header in `m', so any
3506 * data in a cluster may change before we reach icmp_error().
3507 */
3508 MGET(mcopy, M_DONTWAIT, m->m_type);
3509 if (mcopy != NULL && m_dup_pkthdr(mcopy, m, M_DONTWAIT) == 0) {
3510 mcopy->m_len = imin((IP_VHL_HL(ip->ip_vhl) << 2) + 8,
3511 (int)ip->ip_len);
3512 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
3513 }
3514
3515 #if IPSTEALTH
3516 if (!ipstealth) {
3517 #endif /* IPSTEALTH */
3518 ip->ip_ttl -= IPTTLDEC;
3519 #if IPSTEALTH
3520 }
3521 #endif /* IPSTEALTH */
3522
3523 /*
3524 * If forwarding packet using same interface that it came in on,
3525 * perhaps should send a redirect to sender to shortcut a hop.
3526 * Only send redirect if source is sending directly to us,
3527 * and if packet was not source routed (or has any options).
3528 * Also, don't send redirect if forwarding using a default route
3529 * or a route modified by a redirect.
3530 */
3531 RT_LOCK_SPIN(rt);
3532 if (rt->rt_ifp == m->m_pkthdr.rcvif &&
3533 !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
3534 satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
3535 ipsendredirects && !srcrt && rt->rt_ifa != NULL) {
3536 struct in_ifaddr *ia = (struct in_ifaddr *)rt->rt_ifa;
3537 u_int32_t src = ntohl(ip->ip_src.s_addr);
3538
3539 /* Become a regular mutex */
3540 RT_CONVERT_LOCK(rt);
3541 IFA_LOCK_SPIN(&ia->ia_ifa);
3542 if ((src & ia->ia_subnetmask) == ia->ia_subnet) {
3543 if (rt->rt_flags & RTF_GATEWAY) {
3544 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
3545 } else {
3546 dest = pkt_dst.s_addr;
3547 }
3548 /*
3549 * Router requirements says to only send
3550 * host redirects.
3551 */
3552 type = ICMP_REDIRECT;
3553 code = ICMP_REDIRECT_HOST;
3554 #if DIAGNOSTIC
3555 if (ipprintfs) {
3556 printf("redirect (%d) to %lx\n", code,
3557 (u_int32_t)dest);
3558 }
3559 #endif
3560 }
3561 IFA_UNLOCK(&ia->ia_ifa);
3562 }
3563 RT_UNLOCK(rt);
3564
3565
3566 /* Mark this packet as being forwarded from another interface */
3567 m->m_pkthdr.pkt_flags |= PKTF_FORWARDED;
3568 len = m_pktlen(m);
3569
3570 error = ip_output(m, NULL, &fwd_rt, IP_FORWARDING | IP_OUTARGS,
3571 NULL, &ipoa);
3572
3573 /* Refresh rt since the route could have changed while in IP */
3574 rt = fwd_rt.ro_rt;
3575
3576 if (error != 0) {
3577 OSAddAtomic(1, &ipstat.ips_cantforward);
3578 } else {
3579 /*
3580 * Increment stats on the source interface; the ones
3581 * for destination interface has been taken care of
3582 * during output above by virtue of PKTF_FORWARDED.
3583 */
3584 rcvifp->if_fpackets++;
3585 rcvifp->if_fbytes += len;
3586
3587 OSAddAtomic(1, &ipstat.ips_forward);
3588 if (type != 0) {
3589 OSAddAtomic(1, &ipstat.ips_redirectsent);
3590 } else {
3591 if (mcopy != NULL) {
3592 /*
3593 * If we didn't have to go thru ipflow and
3594 * the packet was successfully consumed by
3595 * ip_output, the mcopy is rather a waste;
3596 * this could be further optimized.
3597 */
3598 m_freem(mcopy);
3599 }
3600 goto done;
3601 }
3602 }
3603 if (mcopy == NULL) {
3604 goto done;
3605 }
3606
3607 switch (error) {
3608 case 0: /* forwarded, but need redirect */
3609 /* type, code set above */
3610 break;
3611
3612 case ENETUNREACH: /* shouldn't happen, checked above */
3613 case EHOSTUNREACH:
3614 case ENETDOWN:
3615 case EHOSTDOWN:
3616 default:
3617 type = ICMP_UNREACH;
3618 code = ICMP_UNREACH_HOST;
3619 break;
3620
3621 case EMSGSIZE:
3622 type = ICMP_UNREACH;
3623 code = ICMP_UNREACH_NEEDFRAG;
3624
3625 if (rt == NULL) {
3626 break;
3627 } else {
3628 RT_LOCK_SPIN(rt);
3629 if (rt->rt_ifp != NULL) {
3630 nextmtu = rt->rt_ifp->if_mtu;
3631 }
3632 RT_UNLOCK(rt);
3633 }
3634 #ifdef IPSEC
3635 if (ipsec_bypass) {
3636 break;
3637 }
3638
3639 /*
3640 * If the packet is routed over IPsec tunnel, tell the
3641 * originator the tunnel MTU.
3642 * tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
3643 * XXX quickhack!!!
3644 */
3645 sp = ipsec4_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
3646 IP_FORWARDING, &ipsecerror);
3647
3648 if (sp == NULL) {
3649 break;
3650 }
3651
3652 /*
3653 * find the correct route for outer IPv4
3654 * header, compute tunnel MTU.
3655 */
3656 nextmtu = 0;
3657
3658 if (sp->req != NULL &&
3659 sp->req->saidx.mode == IPSEC_MODE_TUNNEL) {
3660 struct secasindex saidx;
3661 struct secasvar *sav;
3662 struct route *ro;
3663 struct ip *ipm;
3664 size_t ipsechdr;
3665
3666 /* count IPsec header size */
3667 ipsechdr = ipsec_hdrsiz(sp);
3668
3669 ipm = mtod(mcopy, struct ip *);
3670 bcopy(&sp->req->saidx, &saidx, sizeof(saidx));
3671 saidx.mode = sp->req->saidx.mode;
3672 saidx.reqid = sp->req->saidx.reqid;
3673 sin = SIN(&saidx.src);
3674 if (sin->sin_len == 0) {
3675 sin->sin_len = sizeof(*sin);
3676 sin->sin_family = AF_INET;
3677 sin->sin_port = IPSEC_PORT_ANY;
3678 bcopy(&ipm->ip_src, &sin->sin_addr,
3679 sizeof(sin->sin_addr));
3680 }
3681 sin = SIN(&saidx.dst);
3682 if (sin->sin_len == 0) {
3683 sin->sin_len = sizeof(*sin);
3684 sin->sin_family = AF_INET;
3685 sin->sin_port = IPSEC_PORT_ANY;
3686 bcopy(&ipm->ip_dst, &sin->sin_addr,
3687 sizeof(sin->sin_addr));
3688 }
3689 sav = key_allocsa_policy(&saidx);
3690 if (sav != NULL) {
3691 lck_mtx_lock(sadb_mutex);
3692 if (sav->sah != NULL) {
3693 ro = (struct route *)&sav->sah->sa_route;
3694 if (ro->ro_rt != NULL) {
3695 RT_LOCK(ro->ro_rt);
3696 if (ro->ro_rt->rt_ifp != NULL) {
3697 nextmtu = ro->ro_rt->
3698 rt_ifp->if_mtu;
3699 nextmtu -= ipsechdr;
3700 }
3701 RT_UNLOCK(ro->ro_rt);
3702 }
3703 }
3704 key_freesav(sav, KEY_SADB_LOCKED);
3705 lck_mtx_unlock(sadb_mutex);
3706 }
3707 }
3708 key_freesp(sp, KEY_SADB_UNLOCKED);
3709 #endif /* IPSEC */
3710 break;
3711
3712 case ENOBUFS:
3713 /*
3714 * A router should not generate ICMP_SOURCEQUENCH as
3715 * required in RFC1812 Requirements for IP Version 4 Routers.
3716 * Source quench could be a big problem under DoS attacks,
3717 * or if the underlying interface is rate-limited.
3718 * Those who need source quench packets may re-enable them
3719 * via the net.inet.ip.sendsourcequench sysctl.
3720 */
3721 if (ip_sendsourcequench == 0) {
3722 m_freem(mcopy);
3723 goto done;
3724 } else {
3725 type = ICMP_SOURCEQUENCH;
3726 code = 0;
3727 }
3728 break;
3729
3730 case EACCES:
3731 m_freem(mcopy);
3732 goto done;
3733 }
3734
3735 if (type == ICMP_UNREACH && code == ICMP_UNREACH_NEEDFRAG) {
3736 OSAddAtomic(1, &ipstat.ips_cantfrag);
3737 }
3738
3739 icmp_error(mcopy, type, code, dest, nextmtu);
3740 done:
3741 ip_fwd_route_copyin(rcvifp, &fwd_rt);
3742 }
3743
3744 int
ip_savecontrol(struct inpcb * inp,struct mbuf ** mp,struct ip * ip,struct mbuf * m)3745 ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
3746 struct mbuf *m)
3747 {
3748 *mp = NULL;
3749 if (inp->inp_socket->so_options & SO_TIMESTAMP) {
3750 struct timeval tv;
3751
3752 getmicrotime(&tv);
3753 mp = sbcreatecontrol_mbuf((caddr_t)&tv, sizeof(tv),
3754 SCM_TIMESTAMP, SOL_SOCKET, mp);
3755 if (*mp == NULL) {
3756 goto no_mbufs;
3757 }
3758 }
3759 if (inp->inp_socket->so_options & SO_TIMESTAMP_MONOTONIC) {
3760 uint64_t time;
3761
3762 time = mach_absolute_time();
3763 mp = sbcreatecontrol_mbuf((caddr_t)&time, sizeof(time),
3764 SCM_TIMESTAMP_MONOTONIC, SOL_SOCKET, mp);
3765 if (*mp == NULL) {
3766 goto no_mbufs;
3767 }
3768 }
3769 if (inp->inp_socket->so_options & SO_TIMESTAMP_CONTINUOUS) {
3770 uint64_t time;
3771
3772 time = mach_continuous_time();
3773 mp = sbcreatecontrol_mbuf((caddr_t)&time, sizeof(time),
3774 SCM_TIMESTAMP_CONTINUOUS, SOL_SOCKET, mp);
3775 if (*mp == NULL) {
3776 goto no_mbufs;
3777 }
3778 }
3779 if (inp->inp_socket->so_flags & SOF_RECV_TRAFFIC_CLASS) {
3780 int tc = m_get_traffic_class(m);
3781
3782 mp = sbcreatecontrol_mbuf((caddr_t)&tc, sizeof(tc),
3783 SO_TRAFFIC_CLASS, SOL_SOCKET, mp);
3784 if (*mp == NULL) {
3785 goto no_mbufs;
3786 }
3787 }
3788 if ((inp->inp_socket->so_flags & SOF_RECV_WAKE_PKT) &&
3789 (m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
3790 int flag = 1;
3791
3792 mp = sbcreatecontrol_mbuf((caddr_t)&flag, sizeof(flag),
3793 SO_RECV_WAKE_PKT, SOL_SOCKET, mp);
3794 if (*mp == NULL) {
3795 goto no_mbufs;
3796 }
3797 }
3798
3799 if (inp->inp_flags & INP_RECVDSTADDR || SOFLOW_ENABLED(inp->inp_socket)) {
3800 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_dst,
3801 sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP, mp);
3802 if (*mp == NULL) {
3803 goto no_mbufs;
3804 }
3805 }
3806 #ifdef notyet
3807 /*
3808 * XXX
3809 * Moving these out of udp_input() made them even more broken
3810 * than they already were.
3811 */
3812 /* options were tossed already */
3813 if (inp->inp_flags & INP_RECVOPTS) {
3814 mp = sbcreatecontrol_mbuf((caddr_t)opts_deleted_above,
3815 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP, mp);
3816 if (*mp == NULL) {
3817 goto no_mbufs;
3818 }
3819 }
3820 /* ip_srcroute doesn't do what we want here, need to fix */
3821 if (inp->inp_flags & INP_RECVRETOPTS) {
3822 mp = sbcreatecontrol_mbuf((caddr_t)ip_srcroute(),
3823 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP, mp);
3824 if (*mp == NULL) {
3825 goto no_mbufs;
3826 }
3827 }
3828 #endif /* notyet */
3829 if (inp->inp_flags & INP_RECVIF) {
3830 struct ifnet *ifp;
3831 uint8_t sdlbuf[SOCK_MAXADDRLEN + 1];
3832 struct sockaddr_dl *sdl2 = SDL(&sdlbuf);
3833
3834 /*
3835 * Make sure to accomodate the largest possible
3836 * size of SA(if_lladdr)->sa_len.
3837 */
3838 _CASSERT(sizeof(sdlbuf) == (SOCK_MAXADDRLEN + 1));
3839
3840 ifnet_head_lock_shared();
3841 if ((ifp = m->m_pkthdr.rcvif) != NULL &&
3842 ifp->if_index && (ifp->if_index <= if_index)) {
3843 struct ifaddr *ifa = ifnet_addrs[ifp->if_index - 1];
3844 struct sockaddr_dl *sdp;
3845
3846 if (!ifa || !ifa->ifa_addr) {
3847 goto makedummy;
3848 }
3849
3850 IFA_LOCK_SPIN(ifa);
3851 sdp = SDL(ifa->ifa_addr);
3852 /*
3853 * Change our mind and don't try copy.
3854 */
3855 if (sdp->sdl_family != AF_LINK) {
3856 IFA_UNLOCK(ifa);
3857 goto makedummy;
3858 }
3859 /* the above _CASSERT ensures sdl_len fits in sdlbuf */
3860 bcopy(sdp, sdl2, sdp->sdl_len);
3861 IFA_UNLOCK(ifa);
3862 } else {
3863 makedummy:
3864 sdl2->sdl_len =
3865 offsetof(struct sockaddr_dl, sdl_data[0]);
3866 sdl2->sdl_family = AF_LINK;
3867 sdl2->sdl_index = 0;
3868 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
3869 }
3870 ifnet_head_done();
3871 mp = sbcreatecontrol_mbuf((caddr_t)sdl2, sdl2->sdl_len,
3872 IP_RECVIF, IPPROTO_IP, mp);
3873 if (*mp == NULL) {
3874 goto no_mbufs;
3875 }
3876 }
3877 if (inp->inp_flags & INP_RECVTTL) {
3878 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_ttl,
3879 sizeof(ip->ip_ttl), IP_RECVTTL, IPPROTO_IP, mp);
3880 if (*mp == NULL) {
3881 goto no_mbufs;
3882 }
3883 }
3884 if (inp->inp_flags & INP_PKTINFO) {
3885 struct in_pktinfo pi;
3886
3887 bzero(&pi, sizeof(struct in_pktinfo));
3888 bcopy(&ip->ip_dst, &pi.ipi_addr, sizeof(struct in_addr));
3889 pi.ipi_ifindex = (m != NULL && m->m_pkthdr.rcvif != NULL) ?
3890 m->m_pkthdr.rcvif->if_index : 0;
3891
3892 mp = sbcreatecontrol_mbuf((caddr_t)&pi,
3893 sizeof(struct in_pktinfo), IP_RECVPKTINFO, IPPROTO_IP, mp);
3894 if (*mp == NULL) {
3895 goto no_mbufs;
3896 }
3897 }
3898 if (inp->inp_flags & INP_RECVTOS) {
3899 mp = sbcreatecontrol_mbuf((caddr_t)&ip->ip_tos,
3900 sizeof(u_char), IP_RECVTOS, IPPROTO_IP, mp);
3901 if (*mp == NULL) {
3902 goto no_mbufs;
3903 }
3904 }
3905 return 0;
3906
3907 no_mbufs:
3908 ipstat.ips_pktdropcntrl++;
3909 return ENOBUFS;
3910 }
3911
3912 static inline u_short
ip_cksum(struct mbuf * m,int hlen)3913 ip_cksum(struct mbuf *m, int hlen)
3914 {
3915 u_short sum;
3916
3917 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
3918 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
3919 } else if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
3920 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
3921 /*
3922 * The packet arrived on an interface which isn't capable
3923 * of performing IP header checksum; compute it now.
3924 */
3925 sum = ip_cksum_hdr_in(m, hlen);
3926 } else {
3927 sum = 0;
3928 m->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
3929 CSUM_IP_CHECKED | CSUM_IP_VALID);
3930 m->m_pkthdr.csum_data = 0xffff;
3931 }
3932
3933 if (sum != 0) {
3934 OSAddAtomic(1, &ipstat.ips_badsum);
3935 }
3936
3937 return sum;
3938 }
3939
3940 static int
3941 ip_getstat SYSCTL_HANDLER_ARGS
3942 {
3943 #pragma unused(oidp, arg1, arg2)
3944 if (req->oldptr == USER_ADDR_NULL) {
3945 req->oldlen = (size_t)sizeof(struct ipstat);
3946 }
3947
3948 return SYSCTL_OUT(req, &ipstat, MIN(sizeof(ipstat), req->oldlen));
3949 }
3950
3951 void
ip_setsrcifaddr_info(struct mbuf * m,uint16_t src_idx,struct in_ifaddr * ia)3952 ip_setsrcifaddr_info(struct mbuf *m, uint16_t src_idx, struct in_ifaddr *ia)
3953 {
3954 VERIFY(m->m_flags & M_PKTHDR);
3955
3956 /*
3957 * If the source ifaddr is specified, pick up the information
3958 * from there; otherwise just grab the passed-in ifindex as the
3959 * caller may not have the ifaddr available.
3960 */
3961 if (ia != NULL) {
3962 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3963 m->m_pkthdr.src_ifindex = ia->ia_ifp->if_index;
3964 } else {
3965 m->m_pkthdr.src_ifindex = src_idx;
3966 if (src_idx != 0) {
3967 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3968 }
3969 }
3970 }
3971
3972 void
ip_setdstifaddr_info(struct mbuf * m,uint16_t dst_idx,struct in_ifaddr * ia)3973 ip_setdstifaddr_info(struct mbuf *m, uint16_t dst_idx, struct in_ifaddr *ia)
3974 {
3975 VERIFY(m->m_flags & M_PKTHDR);
3976
3977 /*
3978 * If the destination ifaddr is specified, pick up the information
3979 * from there; otherwise just grab the passed-in ifindex as the
3980 * caller may not have the ifaddr available.
3981 */
3982 if (ia != NULL) {
3983 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3984 m->m_pkthdr.dst_ifindex = ia->ia_ifp->if_index;
3985 } else {
3986 m->m_pkthdr.dst_ifindex = dst_idx;
3987 if (dst_idx != 0) {
3988 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3989 }
3990 }
3991 }
3992
3993 int
ip_getsrcifaddr_info(struct mbuf * m,uint32_t * src_idx,uint32_t * iaf)3994 ip_getsrcifaddr_info(struct mbuf *m, uint32_t *src_idx, uint32_t *iaf)
3995 {
3996 VERIFY(m->m_flags & M_PKTHDR);
3997
3998 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
3999 return -1;
4000 }
4001
4002 if (src_idx != NULL) {
4003 *src_idx = m->m_pkthdr.src_ifindex;
4004 }
4005
4006 if (iaf != NULL) {
4007 *iaf = 0;
4008 }
4009
4010 return 0;
4011 }
4012
4013 int
ip_getdstifaddr_info(struct mbuf * m,uint32_t * dst_idx,uint32_t * iaf)4014 ip_getdstifaddr_info(struct mbuf *m, uint32_t *dst_idx, uint32_t *iaf)
4015 {
4016 VERIFY(m->m_flags & M_PKTHDR);
4017
4018 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
4019 return -1;
4020 }
4021
4022 if (dst_idx != NULL) {
4023 *dst_idx = m->m_pkthdr.dst_ifindex;
4024 }
4025
4026 if (iaf != NULL) {
4027 *iaf = 0;
4028 }
4029
4030 return 0;
4031 }
4032
4033 /*
4034 * Protocol input handler for IPPROTO_GRE.
4035 */
4036 void
gre_input(struct mbuf * m,int off)4037 gre_input(struct mbuf *m, int off)
4038 {
4039 gre_input_func_t fn = gre_input_func;
4040
4041 /*
4042 * If there is a registered GRE input handler, pass mbuf to it.
4043 */
4044 if (fn != NULL) {
4045 lck_mtx_unlock(inet_domain_mutex);
4046 m = fn(m, off, (mtod(m, struct ip *))->ip_p);
4047 lck_mtx_lock(inet_domain_mutex);
4048 }
4049
4050 /*
4051 * If no matching tunnel that is up is found, we inject
4052 * the mbuf to raw ip socket to see if anyone picks it up.
4053 */
4054 if (m != NULL) {
4055 rip_input(m, off);
4056 }
4057 }
4058
4059 /*
4060 * Private KPI for PPP/PPTP.
4061 */
4062 int
ip_gre_register_input(gre_input_func_t fn)4063 ip_gre_register_input(gre_input_func_t fn)
4064 {
4065 lck_mtx_lock(inet_domain_mutex);
4066 gre_input_func = fn;
4067 lck_mtx_unlock(inet_domain_mutex);
4068
4069 return 0;
4070 }
4071
4072 #if (DEBUG || DEVELOPMENT)
4073 static int
4074 sysctl_reset_ip_input_stats SYSCTL_HANDLER_ARGS
4075 {
4076 #pragma unused(arg1, arg2)
4077 int error, i;
4078
4079 i = ip_input_measure;
4080 error = sysctl_handle_int(oidp, &i, 0, req);
4081 if (error || req->newptr == USER_ADDR_NULL) {
4082 goto done;
4083 }
4084 /* impose bounds */
4085 if (i < 0 || i > 1) {
4086 error = EINVAL;
4087 goto done;
4088 }
4089 if (ip_input_measure != i && i == 1) {
4090 net_perf_initialize(&net_perf, ip_input_measure_bins);
4091 }
4092 ip_input_measure = i;
4093 done:
4094 return error;
4095 }
4096
4097 static int
4098 sysctl_ip_input_measure_bins SYSCTL_HANDLER_ARGS
4099 {
4100 #pragma unused(arg1, arg2)
4101 int error;
4102 uint64_t i;
4103
4104 i = ip_input_measure_bins;
4105 error = sysctl_handle_quad(oidp, &i, 0, req);
4106 if (error || req->newptr == USER_ADDR_NULL) {
4107 goto done;
4108 }
4109 /* validate data */
4110 if (!net_perf_validate_bins(i)) {
4111 error = EINVAL;
4112 goto done;
4113 }
4114 ip_input_measure_bins = i;
4115 done:
4116 return error;
4117 }
4118
4119 static int
4120 sysctl_ip_input_getperf SYSCTL_HANDLER_ARGS
4121 {
4122 #pragma unused(oidp, arg1, arg2)
4123 if (req->oldptr == USER_ADDR_NULL) {
4124 req->oldlen = (size_t)sizeof(struct ipstat);
4125 }
4126
4127 return SYSCTL_OUT(req, &net_perf, MIN(sizeof(net_perf), req->oldlen));
4128 }
4129 #endif /* (DEBUG || DEVELOPMENT) */
4130
4131 static int
4132 sysctl_ip_checkinterface SYSCTL_HANDLER_ARGS
4133 {
4134 #pragma unused(arg1, arg2)
4135 int error, i;
4136
4137 i = ip_checkinterface;
4138 error = sysctl_handle_int(oidp, &i, 0, req);
4139 if (error != 0 || req->newptr == USER_ADDR_NULL) {
4140 return error;
4141 }
4142
4143 switch (i) {
4144 case IP_CHECKINTERFACE_WEAK_ES:
4145 case IP_CHECKINTERFACE_HYBRID_ES:
4146 case IP_CHECKINTERFACE_STRONG_ES:
4147 if (ip_checkinterface != i) {
4148 ip_checkinterface = i;
4149 os_log(OS_LOG_DEFAULT, "%s: ip_checkinterface is now %d\n",
4150 __func__, ip_checkinterface);
4151 }
4152 break;
4153 default:
4154 error = EINVAL;
4155 break;
4156 }
4157 return error;
4158 }
4159