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 <net/sockaddr_utils.h>
135
136#include <os/log.h>
137
138#define DBG_LAYER_BEG NETDBG_CODE(DBG_NETIP, 0)
139#define DBG_LAYER_END NETDBG_CODE(DBG_NETIP, 2)
140#define DBG_FNC_IP_INPUT NETDBG_CODE(DBG_NETIP, (2 << 8))
141
142#if IPSEC
143extern int ipsec_bypass;
144#endif /* IPSEC */
145
146MBUFQ_HEAD(fq_head);
147
148static int frag_timeout_run; /* frag timer is scheduled to run */
149static void frag_timeout(void *);
150static void frag_sched_timeout(void);
151
152static struct ipq *ipq_alloc(void);
153static void ipq_free(struct ipq *);
154static void ipq_updateparams(void);
155static void ip_input_second_pass(struct mbuf *, struct ifnet *,
156 int, int, struct ip_fw_in_args *);
157
158static LCK_GRP_DECLARE(ipqlock_grp, "ipqlock");
159static LCK_MTX_DECLARE(ipqlock, &ipqlock_grp);
160
161
162/* Packet reassembly stuff */
163#define IPREASS_NHASH_LOG2 6
164#define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2)
165#define IPREASS_HMASK (IPREASS_NHASH - 1)
166#define IPREASS_HASH(x, y) \
167 (((((x) & 0xF) | ((((x) >> 8) & 0xF) << 4)) ^ (y)) & IPREASS_HMASK)
168
169/* IP fragment reassembly queues (protected by ipqlock) */
170static TAILQ_HEAD(ipqhead, ipq) ipq[IPREASS_NHASH]; /* ip reassembly queues */
171static int maxnipq; /* max packets in reass queues */
172static u_int32_t maxfragsperpacket; /* max frags/packet in reass queues */
173static u_int32_t nipq; /* # of packets in reass queues */
174static u_int32_t ipq_limit; /* ipq allocation limit */
175static u_int32_t ipq_count; /* current # of allocated ipq's */
176
177static int sysctl_ipforwarding SYSCTL_HANDLER_ARGS;
178static int sysctl_maxnipq SYSCTL_HANDLER_ARGS;
179static int sysctl_maxfragsperpacket SYSCTL_HANDLER_ARGS;
180
181#if (DEBUG || DEVELOPMENT)
182static int sysctl_reset_ip_input_stats SYSCTL_HANDLER_ARGS;
183static int sysctl_ip_input_measure_bins SYSCTL_HANDLER_ARGS;
184static int sysctl_ip_input_getperf SYSCTL_HANDLER_ARGS;
185#endif /* (DEBUG || DEVELOPMENT) */
186
187int ipforwarding = 0;
188SYSCTL_PROC(_net_inet_ip, IPCTL_FORWARDING, forwarding,
189 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &ipforwarding, 0,
190 sysctl_ipforwarding, "I", "Enable IP forwarding between interfaces");
191
192static int ipsendredirects = 1; /* XXX */
193SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect,
194 CTLFLAG_RW | CTLFLAG_LOCKED, &ipsendredirects, 0,
195 "Enable sending IP redirects");
196
197int ip_defttl = IPDEFTTL;
198SYSCTL_INT(_net_inet_ip, IPCTL_DEFTTL, ttl, CTLFLAG_RW | CTLFLAG_LOCKED,
199 &ip_defttl, 0, "Maximum TTL on IP packets");
200
201static int ip_dosourceroute = 0;
202SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute,
203 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_dosourceroute, 0,
204 "Enable forwarding source routed IP packets");
205
206static int ip_acceptsourceroute = 0;
207SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute,
208 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_acceptsourceroute, 0,
209 "Enable accepting source routed IP packets");
210
211static int ip_sendsourcequench = 0;
212SYSCTL_INT(_net_inet_ip, OID_AUTO, sendsourcequench,
213 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_sendsourcequench, 0,
214 "Enable the transmission of source quench packets");
215
216SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets,
217 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &maxnipq, 0, sysctl_maxnipq,
218 "I", "Maximum number of IPv4 fragment reassembly queue entries");
219
220SYSCTL_UINT(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_RD | CTLFLAG_LOCKED,
221 &nipq, 0, "Current number of IPv4 fragment reassembly queue entries");
222
223SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragsperpacket,
224 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &maxfragsperpacket, 0,
225 sysctl_maxfragsperpacket, "I",
226 "Maximum number of IPv4 fragments allowed per packet");
227
228static uint32_t ip_adj_clear_hwcksum = 0;
229SYSCTL_UINT(_net_inet_ip, OID_AUTO, adj_clear_hwcksum,
230 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_adj_clear_hwcksum, 0,
231 "Invalidate hwcksum info when adjusting length");
232
233static uint32_t ip_adj_partial_sum = 1;
234SYSCTL_UINT(_net_inet_ip, OID_AUTO, adj_partial_sum,
235 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_adj_partial_sum, 0,
236 "Perform partial sum adjustment of trailing bytes at IP layer");
237
238/*
239 * ip_checkinterface controls the receive side of the models for multihoming
240 * that are discussed in RFC 1122.
241 *
242 * ip_checkinterface values are:
243 * IP_CHECKINTERFACE_WEAK_ES:
244 * This corresponds to the Weak End-System model where incoming packets from
245 * any interface are accepted provided the destination address of the incoming packet
246 * is assigned to some interface.
247 *
248 * IP_CHECKINTERFACE_HYBRID_ES:
249 * The Hybrid End-System model use the Strong End-System for tunnel interfaces
250 * (ipsec and utun) and the weak End-System model for other interfaces families.
251 * This prevents a rogue middle box to probe for signs of TCP connections
252 * that use the tunnel interface.
253 *
254 * IP_CHECKINTERFACE_STRONG_ES:
255 * The Strong model model requires the packet arrived on an interface that
256 * is assigned the destination address of the packet.
257 *
258 * Since the routing table and transmit implementation do not implement the Strong ES model,
259 * setting this to a value different from IP_CHECKINTERFACE_WEAK_ES may lead to unexpected results.
260 *
261 * When forwarding is enabled, the system reverts to the Weak ES model as a router
262 * is expected by design to receive packets from several interfaces to the same address.
263 *
264 * XXX - ip_checkinterface currently must be set to IP_CHECKINTERFACE_WEAK_ES if you use ipnat
265 * to translate the destination address to another local interface.
266 *
267 * XXX - ip_checkinterface must be set to IP_CHECKINTERFACE_WEAK_ES if you add IP aliases
268 * to the loopback interface instead of the interface where the
269 * packets for those addresses are received.
270 */
271#define IP_CHECKINTERFACE_WEAK_ES 0
272#define IP_CHECKINTERFACE_HYBRID_ES 1
273#define IP_CHECKINTERFACE_STRONG_ES 2
274
275static int ip_checkinterface = IP_CHECKINTERFACE_HYBRID_ES;
276
277static int sysctl_ip_checkinterface SYSCTL_HANDLER_ARGS;
278SYSCTL_PROC(_net_inet_ip, OID_AUTO, check_interface,
279 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
280 0, 0, sysctl_ip_checkinterface, "I", "Verify packet arrives on correct interface");
281
282#if (DEBUG || DEVELOPMENT)
283#define IP_CHECK_IF_DEBUG 1
284#else
285#define IP_CHECK_IF_DEBUG 0
286#endif /* (DEBUG || DEVELOPMENT) */
287static int ip_checkinterface_debug = IP_CHECK_IF_DEBUG;
288SYSCTL_INT(_net_inet_ip, OID_AUTO, checkinterface_debug, CTLFLAG_RW | CTLFLAG_LOCKED,
289 &ip_checkinterface_debug, IP_CHECK_IF_DEBUG, "");
290
291static int ip_chaining = 1;
292SYSCTL_INT(_net_inet_ip, OID_AUTO, rx_chaining, CTLFLAG_RW | CTLFLAG_LOCKED,
293 &ip_chaining, 1, "Do receive side ip address based chaining");
294
295static int ip_chainsz = 6;
296SYSCTL_INT(_net_inet_ip, OID_AUTO, rx_chainsz, CTLFLAG_RW | CTLFLAG_LOCKED,
297 &ip_chainsz, 1, "IP receive side max chaining");
298
299#if (DEBUG || DEVELOPMENT)
300static int ip_input_measure = 0;
301SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf,
302 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
303 &ip_input_measure, 0, sysctl_reset_ip_input_stats, "I", "Do time measurement");
304
305static uint64_t ip_input_measure_bins = 0;
306SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf_bins,
307 CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &ip_input_measure_bins, 0,
308 sysctl_ip_input_measure_bins, "I",
309 "bins for chaining performance data histogram");
310
311static net_perf_t net_perf;
312SYSCTL_PROC(_net_inet_ip, OID_AUTO, input_perf_data,
313 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED,
314 0, 0, sysctl_ip_input_getperf, "S,net_perf",
315 "IP input performance data (struct net_perf, net/net_perf.h)");
316#endif /* (DEBUG || DEVELOPMENT) */
317
318#if DIAGNOSTIC
319static int ipprintfs = 0;
320#endif
321
322struct protosw *ip_protox[IPPROTO_MAX];
323
324static LCK_GRP_DECLARE(in_ifaddr_rwlock_grp, "in_ifaddr_rwlock");
325LCK_RW_DECLARE(in_ifaddr_rwlock, &in_ifaddr_rwlock_grp);
326
327/* Protected by in_ifaddr_rwlock */
328struct in_ifaddrhead in_ifaddrhead; /* first inet address */
329struct in_ifaddrhashhead *in_ifaddrhashtbl; /* inet addr hash table */
330
331#define INADDR_NHASH 61
332static uint32_t inaddr_nhash; /* hash table size */
333static uint32_t inaddr_hashp; /* next largest prime */
334
335static int ip_getstat SYSCTL_HANDLER_ARGS;
336struct ipstat ipstat;
337SYSCTL_PROC(_net_inet_ip, IPCTL_STATS, stats,
338 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED,
339 0, 0, ip_getstat, "S,ipstat",
340 "IP statistics (struct ipstat, netinet/ip_var.h)");
341
342#if IPCTL_DEFMTU
343SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW | CTLFLAG_LOCKED,
344 &ip_mtu, 0, "Default MTU");
345#endif /* IPCTL_DEFMTU */
346
347#if IPSTEALTH
348static int ipstealth = 0;
349SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_RW | CTLFLAG_LOCKED,
350 &ipstealth, 0, "");
351#endif /* IPSTEALTH */
352
353#if DUMMYNET
354ip_dn_io_t *ip_dn_io_ptr;
355#endif /* DUMMYNET */
356
357SYSCTL_NODE(_net_inet_ip, OID_AUTO, linklocal,
358 CTLFLAG_RW | CTLFLAG_LOCKED, 0, "link local");
359
360struct ip_linklocal_stat ip_linklocal_stat;
361SYSCTL_STRUCT(_net_inet_ip_linklocal, OID_AUTO, stat,
362 CTLFLAG_RD | CTLFLAG_LOCKED, &ip_linklocal_stat, ip_linklocal_stat,
363 "Number of link local packets with TTL less than 255");
364
365SYSCTL_NODE(_net_inet_ip_linklocal, OID_AUTO, in,
366 CTLFLAG_RW | CTLFLAG_LOCKED, 0, "link local input");
367
368int ip_linklocal_in_allowbadttl = 1;
369SYSCTL_INT(_net_inet_ip_linklocal_in, OID_AUTO, allowbadttl,
370 CTLFLAG_RW | CTLFLAG_LOCKED, &ip_linklocal_in_allowbadttl, 0,
371 "Allow incoming link local packets with TTL less than 255");
372
373
374/*
375 * We need to save the IP options in case a protocol wants to respond
376 * to an incoming packet over the same route if the packet got here
377 * using IP source routing. This allows connection establishment and
378 * maintenance when the remote end is on a network that is not known
379 * to us.
380 */
381static int ip_nhops = 0;
382static struct ip_srcrt {
383 struct in_addr dst; /* final destination */
384 char nop; /* one NOP to align */
385 char srcopt[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN and OFFSET */
386 struct in_addr route[MAX_IPOPTLEN / sizeof(struct in_addr)];
387} ip_srcrt;
388
389static void in_ifaddrhashtbl_init(void);
390static void save_rte(u_char *, struct in_addr);
391static int ip_dooptions(struct mbuf *, int, struct sockaddr_in *);
392static void ip_forward(struct mbuf *, int, struct sockaddr_in *);
393static void frag_freef(struct ipqhead *, struct ipq *);
394static struct mbuf *ip_reass(struct mbuf *);
395static void ip_fwd_route_copyout(struct ifnet *, struct route *);
396static void ip_fwd_route_copyin(struct ifnet *, struct route *);
397static inline u_short ip_cksum(struct mbuf *, int);
398
399/*
400 * On platforms which require strict alignment (currently for anything but
401 * i386 or x86_64 or arm64), check if the IP header pointer is 32-bit aligned; if not,
402 * copy the contents of the mbuf chain into a new chain, and free the original
403 * one. Create some head room in the first mbuf of the new chain, in case
404 * it's needed later on.
405 */
406#if defined(__i386__) || defined(__x86_64__) || defined(__arm64__)
407#define IP_HDR_ALIGNMENT_FIXUP(_m, _ifp, _action) do { } while (0)
408#else /* !__i386__ && !__x86_64__ && !__arm64__ */
409#define IP_HDR_ALIGNMENT_FIXUP(_m, _ifp, _action) do { \
410 if (!IP_HDR_ALIGNED_P(mtod(_m, caddr_t))) { \
411 struct mbuf *_n; \
412 struct ifnet *__ifp = (_ifp); \
413 os_atomic_inc(&(__ifp)->if_alignerrs, relaxed); \
414 if (((_m)->m_flags & M_PKTHDR) && \
415 (_m)->m_pkthdr.pkt_hdr != NULL) \
416 (_m)->m_pkthdr.pkt_hdr = NULL; \
417 _n = m_defrag_offset(_m, max_linkhdr, M_NOWAIT); \
418 if (_n == NULL) { \
419 os_atomic_inc(&ipstat.ips_toosmall, relaxed); \
420 m_freem(_m); \
421 (_m) = NULL; \
422 _action; \
423 } else { \
424 VERIFY(_n != (_m)); \
425 (_m) = _n; \
426 } \
427 } \
428} while (0)
429#endif /* !__i386__ && !__x86_64__ && !__arm64__ */
430
431
432typedef enum ip_check_if_result {
433 IP_CHECK_IF_NONE = 0,
434 IP_CHECK_IF_OURS = 1,
435 IP_CHECK_IF_DROP = 2,
436 IP_CHECK_IF_FORWARD = 3
437} ip_check_if_result_t;
438
439static ip_check_if_result_t ip_input_check_interface(struct mbuf **, struct ip *, struct ifnet *);
440
441/*
442 * GRE input handler function, settable via ip_gre_register_input() for PPTP.
443 */
444static gre_input_func_t gre_input_func;
445
446static void
447ip_init_delayed(void)
448{
449 struct ifreq ifr;
450 int error;
451 struct sockaddr_in *sin;
452
453 bzero(s: &ifr, n: sizeof(ifr));
454 strlcpy(dst: ifr.ifr_name, src: "lo0", n: sizeof(ifr.ifr_name));
455 sin = SIN(&ifr.ifr_addr);
456 sin->sin_len = sizeof(struct sockaddr_in);
457 sin->sin_family = AF_INET;
458 sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
459 error = in_control(NULL, SIOCSIFADDR, (caddr_t)&ifr, lo_ifp, kernproc);
460 if (error) {
461 printf("%s: failed to initialise lo0's address, error=%d\n",
462 __func__, error);
463 }
464}
465
466/*
467 * IP initialization: fill in IP protocol switch table.
468 * All protocols not implemented in kernel go to raw IP protocol handler.
469 */
470void
471ip_init(struct protosw *pp, struct domain *dp)
472{
473 static int ip_initialized = 0;
474 struct protosw *pr;
475 struct timeval tv;
476 int i;
477
478 domain_proto_mtx_lock_assert_held();
479 VERIFY((pp->pr_flags & (PR_INITIALIZED | PR_ATTACHED)) == PR_ATTACHED);
480
481 /*
482 * Some ioctls (e.g. SIOCAIFADDR) use ifaliasreq struct, which is
483 * interchangeable with in_aliasreq; they must have the same size.
484 */
485 _CASSERT(sizeof(struct ifaliasreq) == sizeof(struct in_aliasreq));
486
487 if (ip_initialized) {
488 return;
489 }
490 ip_initialized = 1;
491
492 TAILQ_INIT(&in_ifaddrhead);
493 in_ifaddrhashtbl_init();
494
495 ip_moptions_init();
496
497 pr = pffindproto_locked(PF_INET, IPPROTO_RAW, SOCK_RAW);
498 if (pr == NULL) {
499 panic("%s: Unable to find [PF_INET,IPPROTO_RAW,SOCK_RAW]",
500 __func__);
501 /* NOTREACHED */
502 }
503
504 /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */
505 for (i = 0; i < IPPROTO_MAX; i++) {
506 ip_protox[i] = pr;
507 }
508 /*
509 * Cycle through IP protocols and put them into the appropriate place
510 * in ip_protox[], skipping protocols IPPROTO_{IP,RAW}.
511 */
512 VERIFY(dp == inetdomain && dp->dom_family == PF_INET);
513 TAILQ_FOREACH(pr, &dp->dom_protosw, pr_entry) {
514 VERIFY(pr->pr_domain == dp);
515 if (pr->pr_protocol != 0 && pr->pr_protocol != IPPROTO_RAW) {
516 /* Be careful to only index valid IP protocols. */
517 if (pr->pr_protocol < IPPROTO_MAX) {
518 ip_protox[pr->pr_protocol] = pr;
519 }
520 }
521 }
522
523 lck_mtx_lock(lck: &ipqlock);
524 /* Initialize IP reassembly queue. */
525 for (i = 0; i < IPREASS_NHASH; i++) {
526 TAILQ_INIT(&ipq[i]);
527 }
528
529 maxnipq = nmbclusters / 32;
530 maxfragsperpacket = 128; /* enough for 64k in 512 byte fragments */
531 ipq_updateparams();
532 lck_mtx_unlock(lck: &ipqlock);
533
534 getmicrotime(&tv);
535 ip_id = (u_short)(RandomULong() ^ tv.tv_usec);
536
537 PE_parse_boot_argn(arg_string: "ip_checkinterface", arg_ptr: &i, max_arg: sizeof(i));
538 switch (i) {
539 case IP_CHECKINTERFACE_WEAK_ES:
540 case IP_CHECKINTERFACE_HYBRID_ES:
541 case IP_CHECKINTERFACE_STRONG_ES:
542 ip_checkinterface = i;
543 break;
544 default:
545 break;
546 }
547
548 arp_init();
549 net_init_add(init_func: ip_init_delayed);
550}
551
552/*
553 * Initialize IPv4 source address hash table.
554 */
555static void
556in_ifaddrhashtbl_init(void)
557{
558 int i, k, p;
559
560 if (in_ifaddrhashtbl != NULL) {
561 return;
562 }
563
564 PE_parse_boot_argn(arg_string: "inaddr_nhash", arg_ptr: &inaddr_nhash,
565 max_arg: sizeof(inaddr_nhash));
566 if (inaddr_nhash == 0) {
567 inaddr_nhash = INADDR_NHASH;
568 }
569
570 in_ifaddrhashtbl = zalloc_permanent(
571 inaddr_nhash * sizeof(*in_ifaddrhashtbl),
572 ZALIGN_PTR);
573
574 /*
575 * Generate the next largest prime greater than inaddr_nhash.
576 */
577 k = (inaddr_nhash % 2 == 0) ? inaddr_nhash + 1 : inaddr_nhash + 2;
578 for (;;) {
579 p = 1;
580 for (i = 3; i * i <= k; i += 2) {
581 if (k % i == 0) {
582 p = 0;
583 }
584 }
585 if (p == 1) {
586 break;
587 }
588 k += 2;
589 }
590 inaddr_hashp = k;
591}
592
593uint32_t
594inaddr_hashval(uint32_t key)
595{
596 /*
597 * The hash index is the computed prime times the key modulo
598 * the hash size, as documented in "Introduction to Algorithms"
599 * (Cormen, Leiserson, Rivest).
600 */
601 if (inaddr_nhash > 1) {
602 return (key * inaddr_hashp) % inaddr_nhash;
603 } else {
604 return 0;
605 }
606}
607
608struct in_ifaddrhashhead *
609inaddr_hashlookup(uint32_t key)
610{
611 return &in_ifaddrhashtbl[inaddr_hashval(key)];
612}
613
614__private_extern__ void
615ip_proto_dispatch_in(struct mbuf *m, int hlen, u_int8_t proto,
616 ipfilter_t inject_ipfref)
617{
618 struct ipfilter *filter;
619 int seen = (inject_ipfref == NULL);
620 int changed_header = 0;
621 struct ip *ip;
622 void (*pr_input)(struct mbuf *, int len);
623
624 if (!TAILQ_EMPTY(&ipv4_filters)) {
625 ipf_ref();
626 TAILQ_FOREACH(filter, &ipv4_filters, ipf_link) {
627 if (seen == 0) {
628 if ((struct ipfilter *)inject_ipfref == filter) {
629 seen = 1;
630 }
631 } else if (filter->ipf_filter.ipf_input) {
632 errno_t result;
633
634 if (changed_header == 0) {
635 /*
636 * Perform IP header alignment fixup,
637 * if needed, before passing packet
638 * into filter(s).
639 */
640 IP_HDR_ALIGNMENT_FIXUP(m,
641 m->m_pkthdr.rcvif, ipf_unref());
642
643 /* ipf_unref() already called */
644 if (m == NULL) {
645 return;
646 }
647
648 changed_header = 1;
649 ip = mtod(m, struct ip *);
650 ip->ip_len = htons(ip->ip_len + (uint16_t)hlen);
651 ip->ip_off = htons(ip->ip_off);
652 ip->ip_sum = 0;
653 ip->ip_sum = ip_cksum_hdr_in(m, hlen);
654 }
655 result = filter->ipf_filter.ipf_input(
656 filter->ipf_filter.cookie, (mbuf_t *)&m,
657 hlen, proto);
658 if (result == EJUSTRETURN) {
659 ipf_unref();
660 return;
661 }
662 if (result != 0) {
663 ipf_unref();
664 m_freem(m);
665 return;
666 }
667 }
668 }
669 ipf_unref();
670 }
671
672 /* Perform IP header alignment fixup (post-filters), if needed */
673 IP_HDR_ALIGNMENT_FIXUP(m, m->m_pkthdr.rcvif, return );
674
675 ip = mtod(m, struct ip *);
676
677 if (changed_header) {
678 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
679 ip->ip_off = ntohs(ip->ip_off);
680 }
681
682 /*
683 * If there isn't a specific lock for the protocol
684 * we're about to call, use the generic lock for AF_INET.
685 * otherwise let the protocol deal with its own locking
686 */
687 if ((pr_input = ip_protox[ip->ip_p]->pr_input) == NULL) {
688 m_freem(m);
689 } else if (!(ip_protox[ip->ip_p]->pr_flags & PR_PROTOLOCK)) {
690 lck_mtx_lock(lck: inet_domain_mutex);
691 pr_input(m, hlen);
692 lck_mtx_unlock(lck: inet_domain_mutex);
693 } else {
694 pr_input(m, hlen);
695 }
696}
697
698struct pktchain_elm {
699 struct mbuf *pkte_head;
700 struct mbuf *pkte_tail;
701 struct in_addr pkte_saddr;
702 struct in_addr pkte_daddr;
703 uint16_t pkte_npkts;
704 uint16_t pkte_proto;
705 uint32_t pkte_nbytes;
706};
707
708typedef struct pktchain_elm pktchain_elm_t;
709
710/* Store upto PKTTBL_SZ unique flows on the stack */
711#define PKTTBL_SZ 7
712
713static struct mbuf *
714ip_chain_insert(struct mbuf *packet, pktchain_elm_t *tbl)
715{
716 struct ip* ip;
717 int pkttbl_idx = 0;
718
719 ip = mtod(packet, struct ip*);
720
721 /* reusing the hash function from inaddr_hashval */
722 pkttbl_idx = inaddr_hashval(ntohl(ip->ip_src.s_addr)) % PKTTBL_SZ;
723 if (tbl[pkttbl_idx].pkte_head == NULL) {
724 tbl[pkttbl_idx].pkte_head = packet;
725 tbl[pkttbl_idx].pkte_saddr.s_addr = ip->ip_src.s_addr;
726 tbl[pkttbl_idx].pkte_daddr.s_addr = ip->ip_dst.s_addr;
727 tbl[pkttbl_idx].pkte_proto = ip->ip_p;
728 } else {
729 if ((ip->ip_dst.s_addr == tbl[pkttbl_idx].pkte_daddr.s_addr) &&
730 (ip->ip_src.s_addr == tbl[pkttbl_idx].pkte_saddr.s_addr) &&
731 (ip->ip_p == tbl[pkttbl_idx].pkte_proto)) {
732 } else {
733 return packet;
734 }
735 }
736 if (tbl[pkttbl_idx].pkte_tail != NULL) {
737 mbuf_setnextpkt(mbuf: tbl[pkttbl_idx].pkte_tail, nextpkt: packet);
738 }
739
740 tbl[pkttbl_idx].pkte_tail = packet;
741 tbl[pkttbl_idx].pkte_npkts += 1;
742 tbl[pkttbl_idx].pkte_nbytes += packet->m_pkthdr.len;
743 return NULL;
744}
745
746/* args is a dummy variable here for backward compatibility */
747static void
748ip_input_second_pass_loop_tbl(pktchain_elm_t *tbl, struct ip_fw_in_args *args)
749{
750 int i = 0;
751
752 for (i = 0; i < PKTTBL_SZ; i++) {
753 if (tbl[i].pkte_head != NULL) {
754 struct mbuf *m = tbl[i].pkte_head;
755 ip_input_second_pass(m, m->m_pkthdr.rcvif,
756 tbl[i].pkte_npkts, tbl[i].pkte_nbytes, args);
757
758 if (tbl[i].pkte_npkts > 2) {
759 ipstat.ips_rxc_chainsz_gt2++;
760 }
761 if (tbl[i].pkte_npkts > 4) {
762 ipstat.ips_rxc_chainsz_gt4++;
763 }
764#if (DEBUG || DEVELOPMENT)
765 if (ip_input_measure) {
766 net_perf_histogram(&net_perf, tbl[i].pkte_npkts);
767 }
768#endif /* (DEBUG || DEVELOPMENT) */
769 tbl[i].pkte_head = tbl[i].pkte_tail = NULL;
770 tbl[i].pkte_npkts = 0;
771 tbl[i].pkte_nbytes = 0;
772 /* no need to initialize address and protocol in tbl */
773 }
774 }
775}
776
777static void
778ip_input_cpout_args(struct ip_fw_in_args *args, struct ip_fw_args *args1,
779 boolean_t *done_init)
780{
781 if (*done_init == FALSE) {
782 bzero(s: args1, n: sizeof(struct ip_fw_args));
783 *done_init = TRUE;
784 }
785 args1->fwa_pf_rule = args->fwai_pf_rule;
786}
787
788static void
789ip_input_cpin_args(struct ip_fw_args *args1, struct ip_fw_in_args *args)
790{
791 args->fwai_pf_rule = args1->fwa_pf_rule;
792}
793
794typedef enum {
795 IPINPUT_DOCHAIN = 0,
796 IPINPUT_DONTCHAIN,
797 IPINPUT_FREED,
798 IPINPUT_DONE
799} ipinput_chain_ret_t;
800
801static void
802ip_input_update_nstat(struct ifnet *ifp, struct in_addr src_ip,
803 u_int32_t packets, u_int32_t bytes)
804{
805 if (nstat_collect) {
806 struct rtentry *rt = ifnet_cached_rtlookup_inet(ifp,
807 src_ip);
808 if (rt != NULL) {
809 nstat_route_rx(rte: rt, packets, bytes, flags: 0);
810 rtfree(rt);
811 }
812 }
813}
814
815static void
816ip_input_dispatch_chain(struct mbuf *m)
817{
818 struct mbuf *tmp_mbuf = m;
819 struct mbuf *nxt_mbuf = NULL;
820 struct ip *ip = NULL;
821 unsigned int hlen;
822
823 ip = mtod(tmp_mbuf, struct ip *);
824 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
825 while (tmp_mbuf != NULL) {
826 nxt_mbuf = mbuf_nextpkt(mbuf: tmp_mbuf);
827 mbuf_setnextpkt(mbuf: tmp_mbuf, NULL);
828 ip_proto_dispatch_in(m: tmp_mbuf, hlen, proto: ip->ip_p, inject_ipfref: 0);
829 tmp_mbuf = nxt_mbuf;
830 if (tmp_mbuf) {
831 ip = mtod(tmp_mbuf, struct ip *);
832 /* first mbuf of chain already has adjusted ip_len */
833 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
834 ip->ip_len -= hlen;
835 }
836 }
837}
838
839static void
840ip_input_setdst_chain(struct mbuf *m, uint16_t ifindex, struct in_ifaddr *ia)
841{
842 struct mbuf *tmp_mbuf = m;
843
844 while (tmp_mbuf != NULL) {
845 ip_setdstifaddr_info(tmp_mbuf, ifindex, ia);
846 tmp_mbuf = mbuf_nextpkt(mbuf: tmp_mbuf);
847 }
848}
849
850static void
851ip_input_adjust(struct mbuf *m, struct ip *ip, struct ifnet *inifp)
852{
853 boolean_t adjust = TRUE;
854
855 ASSERT(m_pktlen(m) > ip->ip_len);
856
857 /*
858 * Invalidate hardware checksum info if ip_adj_clear_hwcksum
859 * is set; useful to handle buggy drivers. Note that this
860 * should not be enabled by default, as we may get here due
861 * to link-layer padding.
862 */
863 if (ip_adj_clear_hwcksum &&
864 (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) &&
865 !(inifp->if_flags & IFF_LOOPBACK) &&
866 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
867 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
868 m->m_pkthdr.csum_data = 0;
869 ipstat.ips_adj_hwcsum_clr++;
870 }
871
872 /*
873 * If partial checksum information is available, subtract
874 * out the partial sum of postpended extraneous bytes, and
875 * update the checksum metadata accordingly. By doing it
876 * here, the upper layer transport only needs to adjust any
877 * prepended extraneous bytes (else it will do both.)
878 */
879 if (ip_adj_partial_sum &&
880 (m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
881 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
882 m->m_pkthdr.csum_rx_val = m_adj_sum16(m,
883 m->m_pkthdr.csum_rx_start, m->m_pkthdr.csum_rx_start,
884 (ip->ip_len - m->m_pkthdr.csum_rx_start),
885 m->m_pkthdr.csum_rx_val);
886 } else if ((m->m_pkthdr.csum_flags &
887 (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
888 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
889 /*
890 * If packet has partial checksum info and we decided not
891 * to subtract the partial sum of postpended extraneous
892 * bytes here (not the default case), leave that work to
893 * be handled by the other layers. For now, only TCP, UDP
894 * layers are capable of dealing with this. For all other
895 * protocols (including fragments), trim and ditch the
896 * partial sum as those layers might not implement partial
897 * checksumming (or adjustment) at all.
898 */
899 if ((ip->ip_off & (IP_MF | IP_OFFMASK)) == 0 &&
900 (ip->ip_p == IPPROTO_TCP || ip->ip_p == IPPROTO_UDP)) {
901 adjust = FALSE;
902 } else {
903 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
904 m->m_pkthdr.csum_data = 0;
905 ipstat.ips_adj_hwcsum_clr++;
906 }
907 }
908
909 if (adjust) {
910 ipstat.ips_adj++;
911 if (m->m_len == m->m_pkthdr.len) {
912 m->m_len = ip->ip_len;
913 m->m_pkthdr.len = ip->ip_len;
914 } else {
915 m_adj(m, ip->ip_len - m->m_pkthdr.len);
916 }
917 }
918}
919
920/*
921 * First pass does all essential packet validation and places on a per flow
922 * queue for doing operations that have same outcome for all packets of a flow.
923 */
924static ipinput_chain_ret_t
925ip_input_first_pass(struct mbuf *m, struct ip_fw_in_args *args, struct mbuf **modm)
926{
927 struct ip *ip;
928 struct ifnet *inifp;
929 unsigned int hlen;
930 int retval = IPINPUT_DOCHAIN;
931 int len = 0;
932 struct in_addr src_ip;
933#if DUMMYNET
934 struct m_tag *copy;
935 struct m_tag *p;
936 boolean_t delete = FALSE;
937 struct ip_fw_args args1;
938 boolean_t init = FALSE;
939#endif /* DUMMYNET */
940 ipfilter_t inject_filter_ref = NULL;
941
942 /* Check if the mbuf is still valid after interface filter processing */
943 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
944 inifp = mbuf_pkthdr_rcvif(mbuf: m);
945 VERIFY(inifp != NULL);
946
947 /* Perform IP header alignment fixup, if needed */
948 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
949
950 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
951
952#if DUMMYNET
953 /*
954 * Don't bother searching for tag(s) if there's none.
955 */
956 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
957 goto ipfw_tags_done;
958 }
959
960 /* Grab info from mtags prepended to the chain */
961 p = m_tag_first(m);
962 while (p) {
963 if (p->m_tag_id == KERNEL_MODULE_TAG_ID) {
964 if (p->m_tag_type == KERNEL_TAG_TYPE_DUMMYNET) {
965 struct dn_pkt_tag *dn_tag;
966
967 dn_tag = (struct dn_pkt_tag *)(p->m_tag_data);
968 args->fwai_pf_rule = dn_tag->dn_pf_rule;
969 delete = TRUE;
970 }
971
972 if (delete) {
973 copy = p;
974 p = m_tag_next(m, p);
975 m_tag_delete(m, copy);
976 } else {
977 p = m_tag_next(m, p);
978 }
979 } else {
980 p = m_tag_next(m, p);
981 }
982 }
983
984#if DIAGNOSTIC
985 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
986 panic("ip_input no HDR");
987 }
988#endif
989
990 if (args->fwai_pf_rule) {
991 /* dummynet already filtered us */
992 ip = mtod(m, struct ip *);
993 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
994 inject_filter_ref = ipf_get_inject_filter(m);
995 if (args->fwai_pf_rule) {
996 goto check_with_pf;
997 }
998 }
999ipfw_tags_done:
1000#endif /* DUMMYNET */
1001
1002 /*
1003 * No need to process packet twice if we've already seen it.
1004 */
1005 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1006 inject_filter_ref = ipf_get_inject_filter(m);
1007 }
1008 if (inject_filter_ref != NULL) {
1009 ip = mtod(m, struct ip *);
1010 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1011
1012 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1013 struct ip *, ip, struct ifnet *, inifp,
1014 struct ip *, ip, struct ip6_hdr *, NULL);
1015
1016 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1017 ip->ip_off = ntohs(ip->ip_off);
1018 ip_proto_dispatch_in(m, hlen, proto: ip->ip_p, inject_ipfref: inject_filter_ref);
1019 return IPINPUT_DONE;
1020 }
1021
1022 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1023 if_ports_used_match_mbuf(ifp: inifp, PF_INET, m);
1024 }
1025
1026 if (m->m_pkthdr.len < sizeof(struct ip)) {
1027 OSAddAtomic(1, &ipstat.ips_total);
1028 OSAddAtomic(1, &ipstat.ips_tooshort);
1029 m_freem(m);
1030 return IPINPUT_FREED;
1031 }
1032
1033 if (m->m_len < sizeof(struct ip) &&
1034 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1035 OSAddAtomic(1, &ipstat.ips_total);
1036 OSAddAtomic(1, &ipstat.ips_toosmall);
1037 return IPINPUT_FREED;
1038 }
1039
1040 ip = mtod(m, struct ip *);
1041 *modm = m;
1042
1043 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1044 ip->ip_p, ip->ip_off, ip->ip_len);
1045
1046 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1047 OSAddAtomic(1, &ipstat.ips_total);
1048 OSAddAtomic(1, &ipstat.ips_badvers);
1049 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1050 m_freem(m);
1051 return IPINPUT_FREED;
1052 }
1053
1054 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1055 if (hlen < sizeof(struct ip)) {
1056 OSAddAtomic(1, &ipstat.ips_total);
1057 OSAddAtomic(1, &ipstat.ips_badhlen);
1058 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1059 m_freem(m);
1060 return IPINPUT_FREED;
1061 }
1062
1063 if (hlen > m->m_len) {
1064 if ((m = m_pullup(m, hlen)) == NULL) {
1065 OSAddAtomic(1, &ipstat.ips_total);
1066 OSAddAtomic(1, &ipstat.ips_badhlen);
1067 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1068 return IPINPUT_FREED;
1069 }
1070 ip = mtod(m, struct ip *);
1071 *modm = m;
1072 }
1073
1074 if ((ip->ip_tos & IPTOS_ECN_MASK) == IPTOS_ECN_ECT1) {
1075 m->m_pkthdr.pkt_ext_flags |= PKTF_EXT_L4S;
1076 }
1077
1078 /* 127/8 must not appear on wire - RFC1122 */
1079 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1080 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1081 /*
1082 * Allow for the following exceptions:
1083 *
1084 * 1. If the packet was sent to loopback (i.e. rcvif
1085 * would have been set earlier at output time.)
1086 *
1087 * 2. If the packet was sent out on loopback from a local
1088 * source address which belongs to a non-loopback
1089 * interface (i.e. rcvif may not necessarily be a
1090 * loopback interface, hence the test for PKTF_LOOP.)
1091 * Unlike IPv6, there is no interface scope ID, and
1092 * therefore we don't care so much about PKTF_IFINFO.
1093 */
1094 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1095 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1096 OSAddAtomic(1, &ipstat.ips_total);
1097 OSAddAtomic(1, &ipstat.ips_badaddr);
1098 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1099 m_freem(m);
1100 return IPINPUT_FREED;
1101 }
1102 }
1103
1104 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1105 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1106 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1107 ip_linklocal_stat.iplls_in_total++;
1108 if (ip->ip_ttl != MAXTTL) {
1109 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1110 /* Silently drop link local traffic with bad TTL */
1111 if (!ip_linklocal_in_allowbadttl) {
1112 OSAddAtomic(1, &ipstat.ips_total);
1113 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1114 m_freem(m);
1115 return IPINPUT_FREED;
1116 }
1117 }
1118 }
1119
1120 if (ip_cksum(m, hlen)) {
1121 OSAddAtomic(1, &ipstat.ips_total);
1122 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1123 m_freem(m);
1124 return IPINPUT_FREED;
1125 }
1126
1127 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1128 struct ip *, ip, struct ifnet *, inifp,
1129 struct ip *, ip, struct ip6_hdr *, NULL);
1130
1131 /*
1132 * Convert fields to host representation.
1133 */
1134#if BYTE_ORDER != BIG_ENDIAN
1135 NTOHS(ip->ip_len);
1136#endif
1137
1138 if (ip->ip_len < hlen) {
1139 OSAddAtomic(1, &ipstat.ips_total);
1140 OSAddAtomic(1, &ipstat.ips_badlen);
1141 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1142 m_freem(m);
1143 return IPINPUT_FREED;
1144 }
1145
1146#if BYTE_ORDER != BIG_ENDIAN
1147 NTOHS(ip->ip_off);
1148#endif
1149
1150 /*
1151 * Check that the amount of data in the buffers
1152 * is as at least much as the IP header would have us expect.
1153 * Trim mbufs if longer than we expect.
1154 * Drop packet if shorter than we expect.
1155 */
1156 if (m->m_pkthdr.len < ip->ip_len) {
1157 OSAddAtomic(1, &ipstat.ips_total);
1158 OSAddAtomic(1, &ipstat.ips_tooshort);
1159 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1160 m_freem(m);
1161 return IPINPUT_FREED;
1162 }
1163
1164 if (m->m_pkthdr.len > ip->ip_len) {
1165 ip_input_adjust(m, ip, inifp);
1166 }
1167
1168 /* for netstat route statistics */
1169 src_ip = ip->ip_src;
1170 len = m->m_pkthdr.len;
1171
1172#if DUMMYNET
1173check_with_pf:
1174#endif /* DUMMYNET */
1175#if PF
1176 /* Invoke inbound packet filter */
1177 if (PF_IS_ENABLED) {
1178 int error;
1179 ip_input_cpout_args(args, args1: &args1, done_init: &init);
1180 ip = mtod(m, struct ip *);
1181 src_ip = ip->ip_src;
1182
1183#if DUMMYNET
1184 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args1);
1185#else
1186 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1187#endif /* DUMMYNET */
1188 if (error != 0 || m == NULL) {
1189 if (m != NULL) {
1190 panic("%s: unexpected packet %p",
1191 __func__, m);
1192 /* NOTREACHED */
1193 }
1194 /* Already freed by callee */
1195 ip_input_update_nstat(ifp: inifp, src_ip, packets: 1, bytes: len);
1196 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1197 OSAddAtomic(1, &ipstat.ips_total);
1198 return IPINPUT_FREED;
1199 }
1200 ip = mtod(m, struct ip *);
1201 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1202 *modm = m;
1203 ip_input_cpin_args(args1: &args1, args);
1204 }
1205#endif /* PF */
1206
1207#if IPSEC
1208 if (ipsec_bypass == 0 && ipsec_get_history_count(m)) {
1209 retval = IPINPUT_DONTCHAIN; /* XXX scope for chaining here? */
1210 goto pass;
1211 }
1212#endif
1213
1214#if IPSEC
1215pass:
1216#endif
1217 /*
1218 * Process options and, if not destined for us,
1219 * ship it on. ip_dooptions returns 1 when an
1220 * error was detected (causing an icmp message
1221 * to be sent and the original packet to be freed).
1222 */
1223 ip_nhops = 0; /* for source routed packets */
1224 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1225 src_ip = ip->ip_src;
1226 ip_input_update_nstat(ifp: inifp, src_ip, packets: 1, bytes: len);
1227 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1228 OSAddAtomic(1, &ipstat.ips_total);
1229 return IPINPUT_FREED;
1230 }
1231
1232 /*
1233 * Don't chain fragmented packets
1234 */
1235 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1236 return IPINPUT_DONTCHAIN;
1237 }
1238
1239 /* Allow DHCP/BootP responses through */
1240 if ((inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1241 hlen == sizeof(struct ip) && ip->ip_p == IPPROTO_UDP) {
1242 struct udpiphdr *ui;
1243
1244 if (m->m_len < sizeof(struct udpiphdr) &&
1245 (m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1246 OSAddAtomic(1, &udpstat.udps_hdrops);
1247 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1248 OSAddAtomic(1, &ipstat.ips_total);
1249 return IPINPUT_FREED;
1250 }
1251 *modm = m;
1252 ui = mtod(m, struct udpiphdr *);
1253 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1254 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1255 return IPINPUT_DONTCHAIN;
1256 }
1257 }
1258
1259 /* Avoid chaining raw sockets as ipsec checks occur later for them */
1260 if (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR) {
1261 return IPINPUT_DONTCHAIN;
1262 }
1263
1264 return retval;
1265#if !defined(__i386__) && !defined(__x86_64__) && !defined(__arm64__)
1266bad:
1267 m_freem(m);
1268 return IPINPUT_FREED;
1269#endif
1270}
1271
1272/*
1273 * Because the call to m_pullup() may freem the mbuf, the function frees the mbuf packet
1274 * chain before it return IP_CHECK_IF_DROP
1275 */
1276static ip_check_if_result_t
1277ip_input_check_interface(struct mbuf **mp, struct ip *ip, struct ifnet *inifp)
1278{
1279 struct mbuf *m = *mp;
1280 struct in_ifaddr *ia = NULL;
1281 struct in_ifaddr *best_ia = NULL;
1282 struct ifnet *match_ifp = NULL;
1283 ip_check_if_result_t result = IP_CHECK_IF_NONE;
1284
1285 /*
1286 * Host broadcast and all network broadcast addresses are always a match
1287 */
1288 if (ip->ip_dst.s_addr == (u_int32_t)INADDR_BROADCAST ||
1289 ip->ip_dst.s_addr == INADDR_ANY) {
1290 ip_input_setdst_chain(m, ifindex: inifp->if_index, NULL);
1291 return IP_CHECK_IF_OURS;
1292 }
1293
1294 /*
1295 * Check for a match in the hash bucket.
1296 */
1297 lck_rw_lock_shared(lck: &in_ifaddr_rwlock);
1298 TAILQ_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
1299 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
1300 best_ia = ia;
1301 match_ifp = best_ia->ia_ifp;
1302
1303 if (ia->ia_ifp == inifp || (inifp->if_flags & IFF_LOOPBACK) ||
1304 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1305 /*
1306 * A locally originated packet or packet from the loopback
1307 * interface is always an exact interface address match
1308 */
1309 match_ifp = inifp;
1310 break;
1311 }
1312 /*
1313 * Continue the loop in case there's a exact match with another
1314 * interface
1315 */
1316 }
1317 }
1318 if (best_ia != NULL) {
1319 if (match_ifp != inifp && ipforwarding == 0 &&
1320 ((ip_checkinterface == IP_CHECKINTERFACE_HYBRID_ES &&
1321 (match_ifp->if_family == IFNET_FAMILY_IPSEC ||
1322 match_ifp->if_family == IFNET_FAMILY_UTUN)) ||
1323 ip_checkinterface == IP_CHECKINTERFACE_STRONG_ES)) {
1324 /*
1325 * Drop when interface address check is strict and forwarding
1326 * is disabled
1327 */
1328 result = IP_CHECK_IF_DROP;
1329 } else {
1330 result = IP_CHECK_IF_OURS;
1331 ip_input_setdst_chain(m, ifindex: 0, ia: best_ia);
1332 }
1333 }
1334 lck_rw_done(lck: &in_ifaddr_rwlock);
1335
1336 if (result == IP_CHECK_IF_NONE && (inifp->if_flags & IFF_BROADCAST)) {
1337 /*
1338 * Check for broadcast addresses.
1339 *
1340 * Only accept broadcast packets that arrive via the matching
1341 * interface. Reception of forwarded directed broadcasts would be
1342 * handled via ip_forward() and ether_frameout() with the loopback
1343 * into the stack for SIMPLEX interfaces handled by ether_frameout().
1344 */
1345 struct ifaddr *ifa;
1346
1347 ifnet_lock_shared(ifp: inifp);
1348 TAILQ_FOREACH(ifa, &inifp->if_addrhead, ifa_link) {
1349 if (ifa->ifa_addr->sa_family != AF_INET) {
1350 continue;
1351 }
1352 ia = ifatoia(ifa);
1353 if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == ip->ip_dst.s_addr ||
1354 ia->ia_netbroadcast.s_addr == ip->ip_dst.s_addr) {
1355 ip_input_setdst_chain(m, ifindex: 0, ia);
1356 result = IP_CHECK_IF_OURS;
1357 match_ifp = inifp;
1358 break;
1359 }
1360 }
1361 ifnet_lock_done(ifp: inifp);
1362 }
1363
1364 /* Allow DHCP/BootP responses through */
1365 if (result == IP_CHECK_IF_NONE && (inifp->if_eflags & IFEF_AUTOCONFIGURING) &&
1366 ip->ip_p == IPPROTO_UDP && (IP_VHL_HL(ip->ip_vhl) << 2) == sizeof(struct ip)) {
1367 struct udpiphdr *ui;
1368
1369 if (m->m_len < sizeof(struct udpiphdr)) {
1370 if ((m = m_pullup(m, sizeof(struct udpiphdr))) == NULL) {
1371 OSAddAtomic(1, &udpstat.udps_hdrops);
1372 *mp = NULL;
1373 return IP_CHECK_IF_DROP;
1374 }
1375 /*
1376 * m_pullup can return a different mbuf
1377 */
1378 *mp = m;
1379 ip = mtod(m, struct ip *);
1380 }
1381 ui = mtod(m, struct udpiphdr *);
1382 if (ntohs(ui->ui_dport) == IPPORT_BOOTPC) {
1383 ip_input_setdst_chain(m, ifindex: inifp->if_index, NULL);
1384 result = IP_CHECK_IF_OURS;
1385 match_ifp = inifp;
1386 }
1387 }
1388
1389 if (result == IP_CHECK_IF_NONE) {
1390 if (ipforwarding == 0) {
1391 result = IP_CHECK_IF_DROP;
1392 } else {
1393 result = IP_CHECK_IF_FORWARD;
1394 ip_input_setdst_chain(m, ifindex: inifp->if_index, NULL);
1395 }
1396 }
1397
1398 if (result == IP_CHECK_IF_OURS && match_ifp != inifp) {
1399 ipstat.ips_rcv_if_weak_match++;
1400
1401 /* Logging is too noisy when forwarding is enabled */
1402 if (ip_checkinterface_debug != 0 && ipforwarding == 0) {
1403 char src_str[MAX_IPv4_STR_LEN];
1404 char dst_str[MAX_IPv4_STR_LEN];
1405
1406 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1407 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1408 os_log_info(OS_LOG_DEFAULT,
1409 "%s: weak ES interface match to %s for packet from %s to %s proto %u received via %s",
1410 __func__, best_ia->ia_ifp->if_xname, src_str, dst_str, ip->ip_p, inifp->if_xname);
1411 }
1412 } else if (result == IP_CHECK_IF_DROP) {
1413 if (ip_checkinterface_debug > 0) {
1414 char src_str[MAX_IPv4_STR_LEN];
1415 char dst_str[MAX_IPv4_STR_LEN];
1416
1417 inet_ntop(AF_INET, &ip->ip_src, src_str, sizeof(src_str));
1418 inet_ntop(AF_INET, &ip->ip_dst, dst_str, sizeof(dst_str));
1419 os_log(OS_LOG_DEFAULT,
1420 "%s: no interface match for packet from %s to %s proto %u received via %s",
1421 __func__, src_str, dst_str, ip->ip_p, inifp->if_xname);
1422 }
1423 struct mbuf *tmp_mbuf = m;
1424 while (tmp_mbuf != NULL) {
1425 ipstat.ips_rcv_if_no_match++;
1426 tmp_mbuf = tmp_mbuf->m_nextpkt;
1427 }
1428 m_freem_list(m);
1429 *mp = NULL;
1430 }
1431
1432 return result;
1433}
1434
1435static void
1436ip_input_second_pass(struct mbuf *m, struct ifnet *inifp,
1437 int npkts_in_chain, int bytes_in_chain, struct ip_fw_in_args *args)
1438{
1439 struct mbuf *tmp_mbuf = NULL;
1440 unsigned int hlen;
1441
1442#pragma unused (args)
1443
1444 struct ip *ip = mtod(m, struct ip *);
1445 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1446
1447 OSAddAtomic(npkts_in_chain, &ipstat.ips_total);
1448
1449 /*
1450 * Naively assume we can attribute inbound data to the route we would
1451 * use to send to this destination. Asymmetric routing breaks this
1452 * assumption, but it still allows us to account for traffic from
1453 * a remote node in the routing table.
1454 * this has a very significant performance impact so we bypass
1455 * if nstat_collect is disabled. We may also bypass if the
1456 * protocol is tcp in the future because tcp will have a route that
1457 * we can use to attribute the data to. That does mean we would not
1458 * account for forwarded tcp traffic.
1459 */
1460 ip_input_update_nstat(ifp: inifp, src_ip: ip->ip_src, packets: npkts_in_chain,
1461 bytes: bytes_in_chain);
1462
1463 /*
1464 * Check our list of addresses, to see if the packet is for us.
1465 * If we don't have any addresses, assume any unicast packet
1466 * we receive might be for us (and let the upper layers deal
1467 * with it).
1468 */
1469 tmp_mbuf = m;
1470 if (TAILQ_EMPTY(&in_ifaddrhead)) {
1471 while (tmp_mbuf != NULL) {
1472 if (!(tmp_mbuf->m_flags & (M_MCAST | M_BCAST))) {
1473 ip_setdstifaddr_info(tmp_mbuf, inifp->if_index,
1474 NULL);
1475 }
1476 tmp_mbuf = mbuf_nextpkt(mbuf: tmp_mbuf);
1477 }
1478 goto ours;
1479 }
1480
1481 /*
1482 * Enable a consistency check between the destination address
1483 * and the arrival interface for a unicast packet (the RFC 1122
1484 * strong ES model) if IP forwarding is disabled and the packet
1485 * is not locally generated
1486 *
1487 * XXX - Checking also should be disabled if the destination
1488 * address is ipnat'ed to a different interface.
1489 *
1490 * XXX - Checking is incompatible with IP aliases added
1491 * to the loopback interface instead of the interface where
1492 * the packets are received.
1493 */
1494 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1495 ip_check_if_result_t ip_check_if_result = IP_CHECK_IF_NONE;
1496
1497 ip_check_if_result = ip_input_check_interface(mp: &m, ip, inifp);
1498 ASSERT(ip_check_if_result != IP_CHECK_IF_NONE);
1499 if (ip_check_if_result == IP_CHECK_IF_OURS) {
1500 goto ours;
1501 } else if (ip_check_if_result == IP_CHECK_IF_DROP) {
1502 return;
1503 }
1504 } else {
1505 struct in_multi *inm;
1506 /*
1507 * See if we belong to the destination multicast group on the
1508 * arrival interface.
1509 */
1510 in_multihead_lock_shared();
1511 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
1512 in_multihead_lock_done();
1513 if (inm == NULL) {
1514 OSAddAtomic(npkts_in_chain, &ipstat.ips_notmember);
1515 m_freem_list(m);
1516 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1517 return;
1518 }
1519 ip_input_setdst_chain(m, ifindex: inifp->if_index, NULL);
1520 INM_REMREF(inm);
1521 goto ours;
1522 }
1523
1524 tmp_mbuf = m;
1525 struct mbuf *nxt_mbuf = NULL;
1526 while (tmp_mbuf != NULL) {
1527 nxt_mbuf = mbuf_nextpkt(mbuf: tmp_mbuf);
1528 /*
1529 * Not for us; forward if possible and desirable.
1530 */
1531 mbuf_setnextpkt(mbuf: tmp_mbuf, NULL);
1532 if (ipforwarding == 0) {
1533 OSAddAtomic(1, &ipstat.ips_cantforward);
1534 m_freem(tmp_mbuf);
1535 } else {
1536 ip_forward(tmp_mbuf, 0, NULL);
1537 }
1538 tmp_mbuf = nxt_mbuf;
1539 }
1540 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1541 return;
1542ours:
1543 ip = mtod(m, struct ip *); /* in case it changed */
1544 /*
1545 * If offset is set, must reassemble.
1546 */
1547 if (ip->ip_off & ~(IP_DF | IP_RF)) {
1548 VERIFY(npkts_in_chain == 1);
1549 m = ip_reass(m);
1550 if (m == NULL) {
1551 return;
1552 }
1553 ip = mtod(m, struct ip *);
1554 /* Get the header length of the reassembled packet */
1555 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1556 }
1557
1558 /*
1559 * Further protocols expect the packet length to be w/o the
1560 * IP header.
1561 */
1562 ip->ip_len -= hlen;
1563
1564#if IPSEC
1565 /*
1566 * enforce IPsec policy checking if we are seeing last header.
1567 * note that we do not visit this with protocols with pcb layer
1568 * code - like udp/tcp/raw ip.
1569 */
1570 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
1571 VERIFY(npkts_in_chain == 1);
1572 if (ipsec4_in_reject(m, NULL)) {
1573 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
1574 goto bad;
1575 }
1576 }
1577#endif /* IPSEC */
1578
1579 /*
1580 * Switch out to protocol's input routine.
1581 */
1582 OSAddAtomic(npkts_in_chain, &ipstat.ips_delivered);
1583
1584 ip_input_dispatch_chain(m);
1585
1586 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1587 return;
1588bad:
1589 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
1590 m_freem(m);
1591}
1592
1593void
1594ip_input_process_list(struct mbuf *packet_list)
1595{
1596 pktchain_elm_t pktchain_tbl[PKTTBL_SZ];
1597
1598 struct mbuf *packet = NULL;
1599 struct mbuf *modm = NULL; /* modified mbuf */
1600 int retval = 0;
1601#if (DEBUG || DEVELOPMENT)
1602 struct timeval start_tv;
1603#endif /* (DEBUG || DEVELOPMENT) */
1604 int num_pkts = 0;
1605 int chain = 0;
1606 struct ip_fw_in_args args;
1607
1608 if (ip_chaining == 0) {
1609 struct mbuf *m = packet_list;
1610#if (DEBUG || DEVELOPMENT)
1611 if (ip_input_measure) {
1612 net_perf_start_time(&net_perf, &start_tv);
1613 }
1614#endif /* (DEBUG || DEVELOPMENT) */
1615
1616 while (m) {
1617 packet_list = mbuf_nextpkt(mbuf: m);
1618 mbuf_setnextpkt(mbuf: m, NULL);
1619 ip_input(m);
1620 m = packet_list;
1621 num_pkts++;
1622 }
1623#if (DEBUG || DEVELOPMENT)
1624 if (ip_input_measure) {
1625 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1626 }
1627#endif /* (DEBUG || DEVELOPMENT) */
1628 return;
1629 }
1630#if (DEBUG || DEVELOPMENT)
1631 if (ip_input_measure) {
1632 net_perf_start_time(&net_perf, &start_tv);
1633 }
1634#endif /* (DEBUG || DEVELOPMENT) */
1635
1636 bzero(s: &pktchain_tbl, n: sizeof(pktchain_tbl));
1637restart_list_process:
1638 chain = 0;
1639 for (packet = packet_list; packet; packet = packet_list) {
1640 m_add_crumb(packet, PKT_CRUMB_IP_INPUT);
1641
1642 packet_list = mbuf_nextpkt(mbuf: packet);
1643 mbuf_setnextpkt(mbuf: packet, NULL);
1644
1645 num_pkts++;
1646 modm = NULL;
1647 bzero(s: &args, n: sizeof(args));
1648
1649 retval = ip_input_first_pass(m: packet, args: &args, modm: &modm);
1650
1651 if (retval == IPINPUT_DOCHAIN) {
1652 if (modm) {
1653 packet = modm;
1654 }
1655 packet = ip_chain_insert(packet, tbl: &pktchain_tbl[0]);
1656 if (packet == NULL) {
1657 ipstat.ips_rxc_chained++;
1658 chain++;
1659 if (chain > ip_chainsz) {
1660 break;
1661 }
1662 } else {
1663 ipstat.ips_rxc_collisions++;
1664 break;
1665 }
1666 } else if (retval == IPINPUT_DONTCHAIN) {
1667 /* in order to preserve order, exit from chaining */
1668 if (modm) {
1669 packet = modm;
1670 }
1671 ipstat.ips_rxc_notchain++;
1672 break;
1673 } else {
1674 /* packet was freed or delivered, do nothing. */
1675 }
1676 }
1677
1678 /* do second pass here for pktchain_tbl */
1679 if (chain) {
1680 ip_input_second_pass_loop_tbl(tbl: &pktchain_tbl[0], args: &args);
1681 }
1682
1683 if (packet) {
1684 /*
1685 * equivalent update in chaining case if performed in
1686 * ip_input_second_pass_loop_tbl().
1687 */
1688#if (DEBUG || DEVELOPMENT)
1689 if (ip_input_measure) {
1690 net_perf_histogram(&net_perf, 1);
1691 }
1692#endif /* (DEBUG || DEVELOPMENT) */
1693 ip_input_second_pass(m: packet, inifp: packet->m_pkthdr.rcvif,
1694 npkts_in_chain: 1, bytes_in_chain: packet->m_pkthdr.len, args: &args);
1695 }
1696
1697 if (packet_list) {
1698 goto restart_list_process;
1699 }
1700
1701#if (DEBUG || DEVELOPMENT)
1702 if (ip_input_measure) {
1703 net_perf_measure_time(&net_perf, &start_tv, num_pkts);
1704 }
1705#endif /* (DEBUG || DEVELOPMENT) */
1706}
1707/*
1708 * Ip input routine. Checksum and byte swap header. If fragmented
1709 * try to reassemble. Process options. Pass to next level.
1710 */
1711void
1712ip_input(struct mbuf *m)
1713{
1714 struct ip *ip;
1715 unsigned int hlen;
1716 u_short sum = 0;
1717#if DUMMYNET
1718 struct ip_fw_args args;
1719 struct m_tag *tag;
1720#endif
1721 ipfilter_t inject_filter_ref = NULL;
1722 struct ifnet *inifp;
1723
1724 /* Check if the mbuf is still valid after interface filter processing */
1725 MBUF_INPUT_CHECK(m, m->m_pkthdr.rcvif);
1726 inifp = m->m_pkthdr.rcvif;
1727 VERIFY(inifp != NULL);
1728
1729 m_add_crumb(m, PKT_CRUMB_IP_INPUT);
1730
1731 ipstat.ips_rxc_notlist++;
1732
1733 /* Perform IP header alignment fixup, if needed */
1734 IP_HDR_ALIGNMENT_FIXUP(m, inifp, goto bad);
1735
1736 m->m_pkthdr.pkt_flags &= ~PKTF_FORWARDED;
1737
1738#if DUMMYNET
1739 bzero(s: &args, n: sizeof(struct ip_fw_args));
1740
1741 /*
1742 * Don't bother searching for tag(s) if there's none.
1743 */
1744 if (SLIST_EMPTY(&m->m_pkthdr.tags)) {
1745 goto ipfw_tags_done;
1746 }
1747
1748 /* Grab info from mtags prepended to the chain */
1749 if ((tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID,
1750 KERNEL_TAG_TYPE_DUMMYNET)) != NULL) {
1751 struct dn_pkt_tag *dn_tag;
1752
1753 dn_tag = (struct dn_pkt_tag *)(tag->m_tag_data);
1754 args.fwa_pf_rule = dn_tag->dn_pf_rule;
1755
1756 m_tag_delete(m, tag);
1757 }
1758
1759#if DIAGNOSTIC
1760 if (m == NULL || !(m->m_flags & M_PKTHDR)) {
1761 panic("ip_input no HDR");
1762 }
1763#endif
1764
1765 if (args.fwa_pf_rule) {
1766 /* dummynet already filtered us */
1767 ip = mtod(m, struct ip *);
1768 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1769 inject_filter_ref = ipf_get_inject_filter(m);
1770 if (args.fwa_pf_rule) {
1771 goto check_with_pf;
1772 }
1773 }
1774ipfw_tags_done:
1775#endif /* DUMMYNET */
1776
1777 /*
1778 * No need to process packet twice if we've already seen it.
1779 */
1780 if (!SLIST_EMPTY(&m->m_pkthdr.tags)) {
1781 inject_filter_ref = ipf_get_inject_filter(m);
1782 }
1783 if (inject_filter_ref != NULL) {
1784 ip = mtod(m, struct ip *);
1785 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1786
1787 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1788 struct ip *, ip, struct ifnet *, inifp,
1789 struct ip *, ip, struct ip6_hdr *, NULL);
1790
1791 ip->ip_len = ntohs(ip->ip_len) - (u_short)hlen;
1792 ip->ip_off = ntohs(ip->ip_off);
1793 ip_proto_dispatch_in(m, hlen, proto: ip->ip_p, inject_ipfref: inject_filter_ref);
1794 return;
1795 }
1796
1797 if (__improbable(m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
1798 if_ports_used_match_mbuf(ifp: inifp, PF_INET, m);
1799 }
1800
1801 OSAddAtomic(1, &ipstat.ips_total);
1802 if (m->m_pkthdr.len < sizeof(struct ip)) {
1803 goto tooshort;
1804 }
1805
1806 if (m->m_len < sizeof(struct ip) &&
1807 (m = m_pullup(m, sizeof(struct ip))) == NULL) {
1808 OSAddAtomic(1, &ipstat.ips_toosmall);
1809 return;
1810 }
1811 ip = mtod(m, struct ip *);
1812
1813 KERNEL_DEBUG(DBG_LAYER_BEG, ip->ip_dst.s_addr, ip->ip_src.s_addr,
1814 ip->ip_p, ip->ip_off, ip->ip_len);
1815
1816 if (IP_VHL_V(ip->ip_vhl) != IPVERSION) {
1817 OSAddAtomic(1, &ipstat.ips_badvers);
1818 goto bad;
1819 }
1820
1821 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1822 if (hlen < sizeof(struct ip)) { /* minimum header length */
1823 OSAddAtomic(1, &ipstat.ips_badhlen);
1824 goto bad;
1825 }
1826 if (hlen > m->m_len) {
1827 if ((m = m_pullup(m, hlen)) == NULL) {
1828 OSAddAtomic(1, &ipstat.ips_badhlen);
1829 return;
1830 }
1831 ip = mtod(m, struct ip *);
1832 }
1833
1834 /* 127/8 must not appear on wire - RFC1122 */
1835 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1836 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
1837 /*
1838 * Allow for the following exceptions:
1839 *
1840 * 1. If the packet was sent to loopback (i.e. rcvif
1841 * would have been set earlier at output time.)
1842 *
1843 * 2. If the packet was sent out on loopback from a local
1844 * source address which belongs to a non-loopback
1845 * interface (i.e. rcvif may not necessarily be a
1846 * loopback interface, hence the test for PKTF_LOOP.)
1847 * Unlike IPv6, there is no interface scope ID, and
1848 * therefore we don't care so much about PKTF_IFINFO.
1849 */
1850 if (!(inifp->if_flags & IFF_LOOPBACK) &&
1851 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
1852 OSAddAtomic(1, &ipstat.ips_badaddr);
1853 goto bad;
1854 }
1855 }
1856
1857 /* IPv4 Link-Local Addresses as defined in RFC3927 */
1858 if ((IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) ||
1859 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)))) {
1860 ip_linklocal_stat.iplls_in_total++;
1861 if (ip->ip_ttl != MAXTTL) {
1862 OSAddAtomic(1, &ip_linklocal_stat.iplls_in_badttl);
1863 /* Silently drop link local traffic with bad TTL */
1864 if (!ip_linklocal_in_allowbadttl) {
1865 goto bad;
1866 }
1867 }
1868 }
1869
1870 sum = ip_cksum(m, hlen);
1871 if (sum) {
1872 goto bad;
1873 }
1874
1875 DTRACE_IP6(receive, struct mbuf *, m, struct inpcb *, NULL,
1876 struct ip *, ip, struct ifnet *, inifp,
1877 struct ip *, ip, struct ip6_hdr *, NULL);
1878
1879 /*
1880 * Naively assume we can attribute inbound data to the route we would
1881 * use to send to this destination. Asymmetric routing breaks this
1882 * assumption, but it still allows us to account for traffic from
1883 * a remote node in the routing table.
1884 * this has a very significant performance impact so we bypass
1885 * if nstat_collect is disabled. We may also bypass if the
1886 * protocol is tcp in the future because tcp will have a route that
1887 * we can use to attribute the data to. That does mean we would not
1888 * account for forwarded tcp traffic.
1889 */
1890 if (nstat_collect) {
1891 struct rtentry *rt =
1892 ifnet_cached_rtlookup_inet(inifp, ip->ip_src);
1893 if (rt != NULL) {
1894 nstat_route_rx(rte: rt, packets: 1, bytes: m->m_pkthdr.len, flags: 0);
1895 rtfree(rt);
1896 }
1897 }
1898
1899 /*
1900 * Convert fields to host representation.
1901 */
1902#if BYTE_ORDER != BIG_ENDIAN
1903 NTOHS(ip->ip_len);
1904#endif
1905
1906 if (ip->ip_len < hlen) {
1907 OSAddAtomic(1, &ipstat.ips_badlen);
1908 goto bad;
1909 }
1910
1911#if BYTE_ORDER != BIG_ENDIAN
1912 NTOHS(ip->ip_off);
1913#endif
1914 /*
1915 * Check that the amount of data in the buffers
1916 * is as at least much as the IP header would have us expect.
1917 * Trim mbufs if longer than we expect.
1918 * Drop packet if shorter than we expect.
1919 */
1920 if (m->m_pkthdr.len < ip->ip_len) {
1921tooshort:
1922 OSAddAtomic(1, &ipstat.ips_tooshort);
1923 goto bad;
1924 }
1925 if (m->m_pkthdr.len > ip->ip_len) {
1926 ip_input_adjust(m, ip, inifp);
1927 }
1928
1929#if DUMMYNET
1930check_with_pf:
1931#endif
1932#if PF
1933 /* Invoke inbound packet filter */
1934 if (PF_IS_ENABLED) {
1935 int error;
1936#if DUMMYNET
1937 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, &args);
1938#else
1939 error = pf_af_hook(inifp, NULL, &m, AF_INET, TRUE, NULL);
1940#endif /* DUMMYNET */
1941 if (error != 0 || m == NULL) {
1942 if (m != NULL) {
1943 panic("%s: unexpected packet %p",
1944 __func__, m);
1945 /* NOTREACHED */
1946 }
1947 /* Already freed by callee */
1948 return;
1949 }
1950 ip = mtod(m, struct ip *);
1951 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1952 }
1953#endif /* PF */
1954
1955#if IPSEC
1956 if (ipsec_bypass == 0 && ipsec_get_history_count(m)) {
1957 goto pass;
1958 }
1959#endif
1960
1961pass:
1962 /*
1963 * Process options and, if not destined for us,
1964 * ship it on. ip_dooptions returns 1 when an
1965 * error was detected (causing an icmp message
1966 * to be sent and the original packet to be freed).
1967 */
1968 ip_nhops = 0; /* for source routed packets */
1969 if (hlen > sizeof(struct ip) && ip_dooptions(m, 0, NULL)) {
1970 return;
1971 }
1972
1973 /*
1974 * Check our list of addresses, to see if the packet is for us.
1975 * If we don't have any addresses, assume any unicast packet
1976 * we receive might be for us (and let the upper layers deal
1977 * with it).
1978 */
1979 if (TAILQ_EMPTY(&in_ifaddrhead) && !(m->m_flags & (M_MCAST | M_BCAST))) {
1980 ip_setdstifaddr_info(m, inifp->if_index, NULL);
1981 goto ours;
1982 }
1983
1984 /*
1985 * Enable a consistency check between the destination address
1986 * and the arrival interface for a unicast packet (the RFC 1122
1987 * strong ES model) if IP forwarding is disabled and the packet
1988 * is not locally generated and the packet is not subject to
1989 * 'ipfw fwd'.
1990 *
1991 * XXX - Checking also should be disabled if the destination
1992 * address is ipnat'ed to a different interface.
1993 *
1994 * XXX - Checking is incompatible with IP aliases added
1995 * to the loopback interface instead of the interface where
1996 * the packets are received.
1997 */
1998 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
1999 ip_check_if_result_t check_if_result = IP_CHECK_IF_NONE;
2000
2001 check_if_result = ip_input_check_interface(mp: &m, ip, inifp);
2002 ASSERT(check_if_result != IP_CHECK_IF_NONE);
2003 if (check_if_result == IP_CHECK_IF_OURS) {
2004 goto ours;
2005 } else if (check_if_result == IP_CHECK_IF_DROP) {
2006 return;
2007 }
2008 } else {
2009 struct in_multi *inm;
2010 /*
2011 * See if we belong to the destination multicast group on the
2012 * arrival interface.
2013 */
2014 in_multihead_lock_shared();
2015 IN_LOOKUP_MULTI(&ip->ip_dst, inifp, inm);
2016 in_multihead_lock_done();
2017 if (inm == NULL) {
2018 OSAddAtomic(1, &ipstat.ips_notmember);
2019 m_freem(m);
2020 return;
2021 }
2022 ip_setdstifaddr_info(m, inifp->if_index, NULL);
2023 INM_REMREF(inm);
2024 goto ours;
2025 }
2026
2027 /*
2028 * Not for us; forward if possible and desirable.
2029 */
2030 if (ipforwarding == 0) {
2031 OSAddAtomic(1, &ipstat.ips_cantforward);
2032 m_freem(m);
2033 } else {
2034 ip_forward(m, 0, NULL);
2035 }
2036 return;
2037
2038ours:
2039 /*
2040 * If offset or IP_MF are set, must reassemble.
2041 */
2042 if (ip->ip_off & ~(IP_DF | IP_RF)) {
2043 m = ip_reass(m);
2044 if (m == NULL) {
2045 return;
2046 }
2047 ip = mtod(m, struct ip *);
2048 /* Get the header length of the reassembled packet */
2049 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2050 }
2051
2052 /*
2053 * Further protocols expect the packet length to be w/o the
2054 * IP header.
2055 */
2056 ip->ip_len -= hlen;
2057
2058
2059#if IPSEC
2060 /*
2061 * enforce IPsec policy checking if we are seeing last header.
2062 * note that we do not visit this with protocols with pcb layer
2063 * code - like udp/tcp/raw ip.
2064 */
2065 if (ipsec_bypass == 0 && (ip_protox[ip->ip_p]->pr_flags & PR_LASTHDR)) {
2066 if (ipsec4_in_reject(m, NULL)) {
2067 IPSEC_STAT_INCREMENT(ipsecstat.in_polvio);
2068 goto bad;
2069 }
2070 }
2071#endif /* IPSEC */
2072
2073 /*
2074 * Switch out to protocol's input routine.
2075 */
2076 OSAddAtomic(1, &ipstat.ips_delivered);
2077
2078 ip_proto_dispatch_in(m, hlen, proto: ip->ip_p, inject_ipfref: 0);
2079 return;
2080
2081bad:
2082 KERNEL_DEBUG(DBG_LAYER_END, 0, 0, 0, 0, 0);
2083 m_freem(m);
2084}
2085
2086static void
2087ipq_updateparams(void)
2088{
2089 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2090 /*
2091 * -1 for unlimited allocation.
2092 */
2093 if (maxnipq < 0) {
2094 ipq_limit = 0;
2095 }
2096 /*
2097 * Positive number for specific bound.
2098 */
2099 if (maxnipq > 0) {
2100 ipq_limit = maxnipq;
2101 }
2102 /*
2103 * Zero specifies no further fragment queue allocation -- set the
2104 * bound very low, but rely on implementation elsewhere to actually
2105 * prevent allocation and reclaim current queues.
2106 */
2107 if (maxnipq == 0) {
2108 ipq_limit = 1;
2109 }
2110 /*
2111 * Arm the purge timer if not already and if there's work to do
2112 */
2113 frag_sched_timeout();
2114}
2115
2116static int
2117sysctl_maxnipq SYSCTL_HANDLER_ARGS
2118{
2119#pragma unused(arg1, arg2)
2120 int error, i;
2121
2122 lck_mtx_lock(lck: &ipqlock);
2123 i = maxnipq;
2124 error = sysctl_handle_int(oidp, arg1: &i, arg2: 0, req);
2125 if (error || req->newptr == USER_ADDR_NULL) {
2126 goto done;
2127 }
2128 /* impose bounds */
2129 if (i < -1 || i > (nmbclusters / 4)) {
2130 error = EINVAL;
2131 goto done;
2132 }
2133 maxnipq = i;
2134 ipq_updateparams();
2135done:
2136 lck_mtx_unlock(lck: &ipqlock);
2137 return error;
2138}
2139
2140static int
2141sysctl_maxfragsperpacket SYSCTL_HANDLER_ARGS
2142{
2143#pragma unused(arg1, arg2)
2144 int error, i;
2145
2146 lck_mtx_lock(lck: &ipqlock);
2147 i = maxfragsperpacket;
2148 error = sysctl_handle_int(oidp, arg1: &i, arg2: 0, req);
2149 if (error || req->newptr == USER_ADDR_NULL) {
2150 goto done;
2151 }
2152 maxfragsperpacket = i;
2153 ipq_updateparams(); /* see if we need to arm timer */
2154done:
2155 lck_mtx_unlock(lck: &ipqlock);
2156 return error;
2157}
2158
2159/*
2160 * Take incoming datagram fragment and try to reassemble it into
2161 * whole datagram. If a chain for reassembly of this datagram already
2162 * exists, then it is given as fp; otherwise have to make a chain.
2163 *
2164 * The IP header is *NOT* adjusted out of iplen (but in host byte order).
2165 */
2166static struct mbuf *
2167ip_reass(struct mbuf *m)
2168{
2169 struct ip *ip;
2170 struct mbuf *p, *q, *nq, *t;
2171 struct ipq *fp = NULL;
2172 struct ipqhead *head;
2173 int i, hlen, next;
2174 u_int8_t ecn, ecn0;
2175 uint32_t csum, csum_flags;
2176 uint16_t hash;
2177 struct fq_head dfq;
2178
2179 MBUFQ_INIT(&dfq); /* for deferred frees */
2180
2181 /* If maxnipq or maxfragsperpacket is 0, never accept fragments. */
2182 if (maxnipq == 0 || maxfragsperpacket == 0) {
2183 ipstat.ips_fragments++;
2184 ipstat.ips_fragdropped++;
2185 m_freem(m);
2186 if (nipq > 0) {
2187 lck_mtx_lock(lck: &ipqlock);
2188 frag_sched_timeout(); /* purge stale fragments */
2189 lck_mtx_unlock(lck: &ipqlock);
2190 }
2191 return NULL;
2192 }
2193
2194 ip = mtod(m, struct ip *);
2195 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
2196
2197 lck_mtx_lock(lck: &ipqlock);
2198
2199 hash = IPREASS_HASH(ip->ip_src.s_addr, ip->ip_id);
2200 head = &ipq[hash];
2201
2202 /*
2203 * Look for queue of fragments
2204 * of this datagram.
2205 */
2206 TAILQ_FOREACH(fp, head, ipq_list) {
2207 if (ip->ip_id == fp->ipq_id &&
2208 ip->ip_src.s_addr == fp->ipq_src.s_addr &&
2209 ip->ip_dst.s_addr == fp->ipq_dst.s_addr &&
2210 ip->ip_p == fp->ipq_p) {
2211 goto found;
2212 }
2213 }
2214
2215 fp = NULL;
2216
2217 /*
2218 * Attempt to trim the number of allocated fragment queues if it
2219 * exceeds the administrative limit.
2220 */
2221 if ((nipq > (unsigned)maxnipq) && (maxnipq > 0)) {
2222 /*
2223 * drop something from the tail of the current queue
2224 * before proceeding further
2225 */
2226 struct ipq *fq = TAILQ_LAST(head, ipqhead);
2227 if (fq == NULL) { /* gak */
2228 for (i = 0; i < IPREASS_NHASH; i++) {
2229 struct ipq *r = TAILQ_LAST(&ipq[i], ipqhead);
2230 if (r) {
2231 ipstat.ips_fragtimeout += r->ipq_nfrags;
2232 frag_freef(&ipq[i], r);
2233 break;
2234 }
2235 }
2236 } else {
2237 ipstat.ips_fragtimeout += fq->ipq_nfrags;
2238 frag_freef(head, fq);
2239 }
2240 }
2241
2242found:
2243 /*
2244 * Leverage partial checksum offload for IP fragments. Narrow down
2245 * the scope to cover only UDP without IP options, as that is the
2246 * most common case.
2247 *
2248 * Perform 1's complement adjustment of octets that got included/
2249 * excluded in the hardware-calculated checksum value. Ignore cases
2250 * where the value includes the entire IPv4 header span, as the sum
2251 * for those octets would already be 0 by the time we get here; IP
2252 * has already performed its header checksum validation. Also take
2253 * care of any trailing bytes and subtract out their partial sum.
2254 */
2255 if (ip->ip_p == IPPROTO_UDP && hlen == sizeof(struct ip) &&
2256 (m->m_pkthdr.csum_flags &
2257 (CSUM_DATA_VALID | CSUM_PARTIAL | CSUM_PSEUDO_HDR)) ==
2258 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
2259 uint32_t start = m->m_pkthdr.csum_rx_start;
2260 int32_t trailer = (m_pktlen(m) - ip->ip_len);
2261 uint32_t swbytes = (uint32_t)trailer;
2262
2263 csum = m->m_pkthdr.csum_rx_val;
2264
2265 ASSERT(trailer >= 0);
2266 if ((start != 0 && start != hlen) || trailer != 0) {
2267 uint32_t datalen = ip->ip_len - hlen;
2268
2269#if BYTE_ORDER != BIG_ENDIAN
2270 if (start < hlen) {
2271 HTONS(ip->ip_len);
2272 HTONS(ip->ip_off);
2273 }
2274#endif /* BYTE_ORDER != BIG_ENDIAN */
2275 /* callee folds in sum */
2276 csum = m_adj_sum16(m, start, hlen, datalen, csum);
2277 if (hlen > start) {
2278 swbytes += (hlen - start);
2279 } else {
2280 swbytes += (start - hlen);
2281 }
2282#if BYTE_ORDER != BIG_ENDIAN
2283 if (start < hlen) {
2284 NTOHS(ip->ip_off);
2285 NTOHS(ip->ip_len);
2286 }
2287#endif /* BYTE_ORDER != BIG_ENDIAN */
2288 }
2289 csum_flags = m->m_pkthdr.csum_flags;
2290
2291 if (swbytes != 0) {
2292 udp_in_cksum_stats(swbytes);
2293 }
2294 if (trailer != 0) {
2295 m_adj(m, -trailer);
2296 }
2297 } else {
2298 csum = 0;
2299 csum_flags = 0;
2300 }
2301
2302 /* Invalidate checksum */
2303 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
2304
2305 ipstat.ips_fragments++;
2306
2307 /*
2308 * Adjust ip_len to not reflect header,
2309 * convert offset of this to bytes.
2310 */
2311 ip->ip_len -= hlen;
2312 if (ip->ip_off & IP_MF) {
2313 /*
2314 * Make sure that fragments have a data length
2315 * that's a non-zero multiple of 8 bytes.
2316 */
2317 if (ip->ip_len == 0 || (ip->ip_len & 0x7) != 0) {
2318 OSAddAtomic(1, &ipstat.ips_toosmall);
2319 /*
2320 * Reassembly queue may have been found if previous
2321 * fragments were valid; given that this one is bad,
2322 * we need to drop it. Make sure to set fp to NULL
2323 * if not already, since we don't want to decrement
2324 * ipq_nfrags as it doesn't include this packet.
2325 */
2326 fp = NULL;
2327 goto dropfrag;
2328 }
2329 m->m_flags |= M_FRAG;
2330 } else {
2331 /* Clear the flag in case packet comes from loopback */
2332 m->m_flags &= ~M_FRAG;
2333 }
2334 ip->ip_off = (u_short)(ip->ip_off << 3);
2335
2336 m->m_pkthdr.pkt_hdr = ip;
2337
2338 /* Previous ip_reass() started here. */
2339 /*
2340 * Presence of header sizes in mbufs
2341 * would confuse code below.
2342 */
2343 m->m_data += hlen;
2344 m->m_len -= hlen;
2345
2346 /*
2347 * If first fragment to arrive, create a reassembly queue.
2348 */
2349 if (fp == NULL) {
2350 fp = ipq_alloc();
2351 if (fp == NULL) {
2352 goto dropfrag;
2353 }
2354 TAILQ_INSERT_HEAD(head, fp, ipq_list);
2355 nipq++;
2356 fp->ipq_nfrags = 1;
2357 fp->ipq_ttl = IPFRAGTTL;
2358 fp->ipq_p = ip->ip_p;
2359 fp->ipq_id = ip->ip_id;
2360 fp->ipq_src = ip->ip_src;
2361 fp->ipq_dst = ip->ip_dst;
2362 fp->ipq_frags = m;
2363 m->m_nextpkt = NULL;
2364 /*
2365 * If the first fragment has valid checksum offload
2366 * info, the rest of fragments are eligible as well.
2367 */
2368 if (csum_flags != 0) {
2369 fp->ipq_csum = csum;
2370 fp->ipq_csum_flags = csum_flags;
2371 }
2372 m = NULL; /* nothing to return */
2373 goto done;
2374 } else {
2375 fp->ipq_nfrags++;
2376 }
2377
2378#define GETIP(m) ((struct ip *)((m)->m_pkthdr.pkt_hdr))
2379
2380 /*
2381 * Handle ECN by comparing this segment with the first one;
2382 * if CE is set, do not lose CE.
2383 * drop if CE and not-ECT are mixed for the same packet.
2384 */
2385 ecn = ip->ip_tos & IPTOS_ECN_MASK;
2386 ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK;
2387 if (ecn == IPTOS_ECN_CE) {
2388 if (ecn0 == IPTOS_ECN_NOTECT) {
2389 goto dropfrag;
2390 }
2391 if (ecn0 != IPTOS_ECN_CE) {
2392 GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE;
2393 }
2394 }
2395 if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) {
2396 goto dropfrag;
2397 }
2398
2399 /*
2400 * Find a segment which begins after this one does.
2401 */
2402 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2403 if (GETIP(q)->ip_off > ip->ip_off) {
2404 break;
2405 }
2406 }
2407
2408 /*
2409 * If there is a preceding segment, it may provide some of
2410 * our data already. If so, drop the data from the incoming
2411 * segment. If it provides all of our data, drop us, otherwise
2412 * stick new segment in the proper place.
2413 *
2414 * If some of the data is dropped from the preceding
2415 * segment, then it's checksum is invalidated.
2416 */
2417 if (p) {
2418 i = GETIP(p)->ip_off + GETIP(p)->ip_len - ip->ip_off;
2419 if (i > 0) {
2420 if (i >= ip->ip_len) {
2421 goto dropfrag;
2422 }
2423 m_adj(m, i);
2424 fp->ipq_csum_flags = 0;
2425 ip->ip_off += i;
2426 ip->ip_len -= i;
2427 }
2428 m->m_nextpkt = p->m_nextpkt;
2429 p->m_nextpkt = m;
2430 } else {
2431 m->m_nextpkt = fp->ipq_frags;
2432 fp->ipq_frags = m;
2433 }
2434
2435 /*
2436 * While we overlap succeeding segments trim them or,
2437 * if they are completely covered, dequeue them.
2438 */
2439 for (; q != NULL && ip->ip_off + ip->ip_len > GETIP(q)->ip_off;
2440 q = nq) {
2441 i = (ip->ip_off + ip->ip_len) - GETIP(q)->ip_off;
2442 if (i < GETIP(q)->ip_len) {
2443 GETIP(q)->ip_len -= i;
2444 GETIP(q)->ip_off += i;
2445 m_adj(q, i);
2446 fp->ipq_csum_flags = 0;
2447 break;
2448 }
2449 nq = q->m_nextpkt;
2450 m->m_nextpkt = nq;
2451 ipstat.ips_fragdropped++;
2452 fp->ipq_nfrags--;
2453 /* defer freeing until after lock is dropped */
2454 MBUFQ_ENQUEUE(&dfq, q);
2455 }
2456
2457 /*
2458 * If this fragment contains similar checksum offload info
2459 * as that of the existing ones, accumulate checksum. Otherwise,
2460 * invalidate checksum offload info for the entire datagram.
2461 */
2462 if (csum_flags != 0 && csum_flags == fp->ipq_csum_flags) {
2463 fp->ipq_csum += csum;
2464 } else if (fp->ipq_csum_flags != 0) {
2465 fp->ipq_csum_flags = 0;
2466 }
2467
2468
2469 /*
2470 * Check for complete reassembly and perform frag per packet
2471 * limiting.
2472 *
2473 * Frag limiting is performed here so that the nth frag has
2474 * a chance to complete the packet before we drop the packet.
2475 * As a result, n+1 frags are actually allowed per packet, but
2476 * only n will ever be stored. (n = maxfragsperpacket.)
2477 *
2478 */
2479 next = 0;
2480 for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) {
2481 if (GETIP(q)->ip_off != next) {
2482 if (fp->ipq_nfrags > maxfragsperpacket) {
2483 ipstat.ips_fragdropped += fp->ipq_nfrags;
2484 frag_freef(head, fp);
2485 }
2486 m = NULL; /* nothing to return */
2487 goto done;
2488 }
2489 next += GETIP(q)->ip_len;
2490 }
2491 /* Make sure the last packet didn't have the IP_MF flag */
2492 if (p->m_flags & M_FRAG) {
2493 if (fp->ipq_nfrags > maxfragsperpacket) {
2494 ipstat.ips_fragdropped += fp->ipq_nfrags;
2495 frag_freef(head, fp);
2496 }
2497 m = NULL; /* nothing to return */
2498 goto done;
2499 }
2500
2501 /*
2502 * Reassembly is complete. Make sure the packet is a sane size.
2503 */
2504 q = fp->ipq_frags;
2505 ip = GETIP(q);
2506 if (next + (IP_VHL_HL(ip->ip_vhl) << 2) > IP_MAXPACKET) {
2507 ipstat.ips_toolong++;
2508 ipstat.ips_fragdropped += fp->ipq_nfrags;
2509 frag_freef(head, fp);
2510 m = NULL; /* nothing to return */
2511 goto done;
2512 }
2513
2514 /*
2515 * Concatenate fragments.
2516 */
2517 m = q;
2518 t = m->m_next;
2519 m->m_next = NULL;
2520 m_cat(m, t);
2521 nq = q->m_nextpkt;
2522 q->m_nextpkt = NULL;
2523 for (q = nq; q != NULL; q = nq) {
2524 nq = q->m_nextpkt;
2525 q->m_nextpkt = NULL;
2526 m_cat(m, q);
2527 }
2528
2529 /*
2530 * Store partial hardware checksum info from the fragment queue;
2531 * the receive start offset is set to 20 bytes (see code at the
2532 * top of this routine.)
2533 */
2534 if (fp->ipq_csum_flags != 0) {
2535 csum = fp->ipq_csum;
2536
2537 ADDCARRY(csum);
2538
2539 m->m_pkthdr.csum_rx_val = (uint16_t)csum;
2540 m->m_pkthdr.csum_rx_start = sizeof(struct ip);
2541 m->m_pkthdr.csum_flags = fp->ipq_csum_flags;
2542 } else if ((m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) ||
2543 (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
2544 /* loopback checksums are always OK */
2545 m->m_pkthdr.csum_data = 0xffff;
2546 m->m_pkthdr.csum_flags =
2547 CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
2548 CSUM_IP_CHECKED | CSUM_IP_VALID;
2549 }
2550
2551 /*
2552 * Create header for new ip packet by modifying header of first
2553 * packet; dequeue and discard fragment reassembly header.
2554 * Make header visible.
2555 */
2556 ip->ip_len = (u_short)((IP_VHL_HL(ip->ip_vhl) << 2) + next);
2557 ip->ip_src = fp->ipq_src;
2558 ip->ip_dst = fp->ipq_dst;
2559
2560 fp->ipq_frags = NULL; /* return to caller as 'm' */
2561 frag_freef(head, fp);
2562 fp = NULL;
2563
2564 m->m_len += (IP_VHL_HL(ip->ip_vhl) << 2);
2565 m->m_data -= (IP_VHL_HL(ip->ip_vhl) << 2);
2566 /* some debugging cruft by sklower, below, will go away soon */
2567 if (m->m_flags & M_PKTHDR) { /* XXX this should be done elsewhere */
2568 m_fixhdr(m);
2569 }
2570 ipstat.ips_reassembled++;
2571
2572 /* arm the purge timer if not already and if there's work to do */
2573 frag_sched_timeout();
2574 lck_mtx_unlock(lck: &ipqlock);
2575 /* perform deferred free (if needed) now that lock is dropped */
2576 if (!MBUFQ_EMPTY(&dfq)) {
2577 MBUFQ_DRAIN(&dfq);
2578 }
2579 VERIFY(MBUFQ_EMPTY(&dfq));
2580 return m;
2581
2582done:
2583 VERIFY(m == NULL);
2584 /* arm the purge timer if not already and if there's work to do */
2585 frag_sched_timeout();
2586 lck_mtx_unlock(lck: &ipqlock);
2587 /* perform deferred free (if needed) */
2588 if (!MBUFQ_EMPTY(&dfq)) {
2589 MBUFQ_DRAIN(&dfq);
2590 }
2591 VERIFY(MBUFQ_EMPTY(&dfq));
2592 return NULL;
2593
2594dropfrag:
2595 ipstat.ips_fragdropped++;
2596 if (fp != NULL) {
2597 fp->ipq_nfrags--;
2598 }
2599 /* arm the purge timer if not already and if there's work to do */
2600 frag_sched_timeout();
2601 lck_mtx_unlock(lck: &ipqlock);
2602 m_freem(m);
2603 /* perform deferred free (if needed) */
2604 if (!MBUFQ_EMPTY(&dfq)) {
2605 MBUFQ_DRAIN(&dfq);
2606 }
2607 VERIFY(MBUFQ_EMPTY(&dfq));
2608 return NULL;
2609#undef GETIP
2610}
2611
2612/*
2613 * Free a fragment reassembly header and all
2614 * associated datagrams.
2615 */
2616static void
2617frag_freef(struct ipqhead *fhp, struct ipq *fp)
2618{
2619 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2620
2621 fp->ipq_nfrags = 0;
2622 if (fp->ipq_frags != NULL) {
2623 m_freem_list(fp->ipq_frags);
2624 fp->ipq_frags = NULL;
2625 }
2626 TAILQ_REMOVE(fhp, fp, ipq_list);
2627 nipq--;
2628 ipq_free(fp);
2629}
2630
2631/*
2632 * IP reassembly timer processing
2633 */
2634static void
2635frag_timeout(void *arg)
2636{
2637#pragma unused(arg)
2638 struct ipq *fp;
2639 int i;
2640
2641 /*
2642 * Update coarse-grained networking timestamp (in sec.); the idea
2643 * is to piggy-back on the timeout callout to update the counter
2644 * returnable via net_uptime().
2645 */
2646 net_update_uptime();
2647
2648 lck_mtx_lock(lck: &ipqlock);
2649 for (i = 0; i < IPREASS_NHASH; i++) {
2650 for (fp = TAILQ_FIRST(&ipq[i]); fp;) {
2651 struct ipq *fpp;
2652
2653 fpp = fp;
2654 fp = TAILQ_NEXT(fp, ipq_list);
2655 if (--fpp->ipq_ttl == 0) {
2656 ipstat.ips_fragtimeout += fpp->ipq_nfrags;
2657 frag_freef(fhp: &ipq[i], fp: fpp);
2658 }
2659 }
2660 }
2661 /*
2662 * If we are over the maximum number of fragments
2663 * (due to the limit being lowered), drain off
2664 * enough to get down to the new limit.
2665 */
2666 if (maxnipq >= 0 && nipq > (unsigned)maxnipq) {
2667 for (i = 0; i < IPREASS_NHASH; i++) {
2668 while (nipq > (unsigned)maxnipq &&
2669 !TAILQ_EMPTY(&ipq[i])) {
2670 ipstat.ips_fragdropped +=
2671 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2672 frag_freef(fhp: &ipq[i], TAILQ_FIRST(&ipq[i]));
2673 }
2674 }
2675 }
2676 /* re-arm the purge timer if there's work to do */
2677 frag_timeout_run = 0;
2678 frag_sched_timeout();
2679 lck_mtx_unlock(lck: &ipqlock);
2680}
2681
2682static void
2683frag_sched_timeout(void)
2684{
2685 LCK_MTX_ASSERT(&ipqlock, LCK_MTX_ASSERT_OWNED);
2686
2687 if (!frag_timeout_run && nipq > 0) {
2688 frag_timeout_run = 1;
2689 timeout(frag_timeout, NULL, ticks: hz);
2690 }
2691}
2692
2693/*
2694 * Drain off all datagram fragments.
2695 */
2696static void
2697frag_drain(void)
2698{
2699 int i;
2700
2701 lck_mtx_lock(lck: &ipqlock);
2702 for (i = 0; i < IPREASS_NHASH; i++) {
2703 while (!TAILQ_EMPTY(&ipq[i])) {
2704 ipstat.ips_fragdropped +=
2705 TAILQ_FIRST(&ipq[i])->ipq_nfrags;
2706 frag_freef(fhp: &ipq[i], TAILQ_FIRST(&ipq[i]));
2707 }
2708 }
2709 lck_mtx_unlock(lck: &ipqlock);
2710}
2711
2712static struct ipq *
2713ipq_alloc(void)
2714{
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 fp = kalloc_type(struct ipq, Z_NOWAIT | Z_ZERO);
2727 if (fp != NULL) {
2728 os_atomic_inc(&ipq_count, relaxed);
2729 }
2730 return fp;
2731}
2732
2733static void
2734ipq_free(struct ipq *fp)
2735{
2736 kfree_type(struct ipq, fp);
2737 os_atomic_dec(&ipq_count, relaxed);
2738}
2739
2740/*
2741 * Drain callback
2742 */
2743void
2744ip_drain(void)
2745{
2746 frag_drain(); /* fragments */
2747 in_rtqdrain(); /* protocol cloned routes */
2748 in_arpdrain(NULL); /* cloned routes: ARP */
2749}
2750
2751/*
2752 * Do option processing on a datagram,
2753 * possibly discarding it if bad options are encountered,
2754 * or forwarding it if source-routed.
2755 * The pass argument is used when operating in the IPSTEALTH
2756 * mode to tell what options to process:
2757 * [LS]SRR (pass 0) or the others (pass 1).
2758 * The reason for as many as two passes is that when doing IPSTEALTH,
2759 * non-routing options should be processed only if the packet is for us.
2760 * Returns 1 if packet has been forwarded/freed,
2761 * 0 if the packet should be processed further.
2762 */
2763static int
2764ip_dooptions(struct mbuf *m, int pass, struct sockaddr_in *next_hop)
2765{
2766#pragma unused(pass)
2767 struct ip *ip = mtod(m, struct ip *);
2768 u_char *cp;
2769 struct ip_timestamp *ipt;
2770 struct in_ifaddr *ia;
2771 int opt, optlen, cnt, off, type = ICMP_PARAMPROB, forward = 0;
2772 uint8_t code = 0;
2773 struct in_addr *sin, dst;
2774 u_int32_t ntime;
2775 struct sockaddr_in ipaddr = {
2776 .sin_len = sizeof(ipaddr),
2777 .sin_family = AF_INET,
2778 .sin_port = 0,
2779 .sin_addr = { .s_addr = 0 },
2780 .sin_zero = { 0, }
2781 };
2782
2783 /* Expect 32-bit aligned data pointer on strict-align platforms */
2784 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
2785
2786 dst = ip->ip_dst;
2787 cp = (u_char *)(ip + 1);
2788 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
2789 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2790 opt = cp[IPOPT_OPTVAL];
2791 if (opt == IPOPT_EOL) {
2792 break;
2793 }
2794 if (opt == IPOPT_NOP) {
2795 optlen = 1;
2796 } else {
2797 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
2798 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2799 goto bad;
2800 }
2801 optlen = cp[IPOPT_OLEN];
2802 if (optlen < IPOPT_OLEN + sizeof(*cp) ||
2803 optlen > cnt) {
2804 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2805 goto bad;
2806 }
2807 }
2808 switch (opt) {
2809 default:
2810 break;
2811
2812 /*
2813 * Source routing with record.
2814 * Find interface with current destination address.
2815 * If none on this machine then drop if strictly routed,
2816 * or do nothing if loosely routed.
2817 * Record interface address and bring up next address
2818 * component. If strictly routed make sure next
2819 * address is on directly accessible net.
2820 */
2821 case IPOPT_LSRR:
2822 case IPOPT_SSRR:
2823 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2824 code = (uint8_t)(&cp[IPOPT_OLEN] - (u_char *)ip);
2825 goto bad;
2826 }
2827 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2828 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2829 goto bad;
2830 }
2831 ipaddr.sin_addr = ip->ip_dst;
2832 ia = (struct in_ifaddr *)ifa_ifwithaddr(SA(&ipaddr));
2833 if (ia == NULL) {
2834 if (opt == IPOPT_SSRR) {
2835 type = ICMP_UNREACH;
2836 code = ICMP_UNREACH_SRCFAIL;
2837 goto bad;
2838 }
2839 if (!ip_dosourceroute) {
2840 goto nosourcerouting;
2841 }
2842 /*
2843 * Loose routing, and not at next destination
2844 * yet; nothing to do except forward.
2845 */
2846 break;
2847 } else {
2848 ifa_remref(ifa: &ia->ia_ifa);
2849 ia = NULL;
2850 }
2851 off--; /* 0 origin */
2852 if (off > optlen - (int)sizeof(struct in_addr)) {
2853 /*
2854 * End of source route. Should be for us.
2855 */
2856 if (!ip_acceptsourceroute) {
2857 goto nosourcerouting;
2858 }
2859 save_rte(cp, ip->ip_src);
2860 break;
2861 }
2862
2863 if (!ip_dosourceroute) {
2864 if (ipforwarding) {
2865 char buf[MAX_IPv4_STR_LEN];
2866 char buf2[MAX_IPv4_STR_LEN];
2867 /*
2868 * Acting as a router, so generate ICMP
2869 */
2870nosourcerouting:
2871 log(LOG_WARNING,
2872 "attempted source route from %s "
2873 "to %s\n",
2874 inet_ntop(AF_INET, &ip->ip_src,
2875 buf, sizeof(buf)),
2876 inet_ntop(AF_INET, &ip->ip_dst,
2877 buf2, sizeof(buf2)));
2878 type = ICMP_UNREACH;
2879 code = ICMP_UNREACH_SRCFAIL;
2880 goto bad;
2881 } else {
2882 /*
2883 * Not acting as a router,
2884 * so silently drop.
2885 */
2886 OSAddAtomic(1, &ipstat.ips_cantforward);
2887 m_freem(m);
2888 return 1;
2889 }
2890 }
2891
2892 /*
2893 * locate outgoing interface
2894 */
2895 (void) memcpy(dst: &ipaddr.sin_addr, src: cp + off,
2896 n: sizeof(ipaddr.sin_addr));
2897
2898 if (opt == IPOPT_SSRR) {
2899#define INA struct in_ifaddr *
2900 if ((ia = (INA)ifa_ifwithdstaddr(
2901 SA(&ipaddr))) == NULL) {
2902 ia = (INA)ifa_ifwithnet(SA(&ipaddr));
2903 }
2904 } else {
2905 ia = ip_rtaddr(ipaddr.sin_addr);
2906 }
2907 if (ia == NULL) {
2908 type = ICMP_UNREACH;
2909 code = ICMP_UNREACH_SRCFAIL;
2910 goto bad;
2911 }
2912 ip->ip_dst = ipaddr.sin_addr;
2913 IFA_LOCK(&ia->ia_ifa);
2914 (void) memcpy(dst: cp + off, src: &(IA_SIN(ia)->sin_addr),
2915 n: sizeof(struct in_addr));
2916 IFA_UNLOCK(&ia->ia_ifa);
2917 ifa_remref(ifa: &ia->ia_ifa);
2918 ia = NULL;
2919 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2920 /*
2921 * Let ip_intr's mcast routing check handle mcast pkts
2922 */
2923 forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr));
2924 break;
2925
2926 case IPOPT_RR:
2927 if (optlen < IPOPT_OFFSET + sizeof(*cp)) {
2928 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2929 goto bad;
2930 }
2931 if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) {
2932 code = (uint8_t)(&cp[IPOPT_OFFSET] - (u_char *)ip);
2933 goto bad;
2934 }
2935 /*
2936 * If no space remains, ignore.
2937 */
2938 off--; /* 0 origin */
2939 if (off > optlen - (int)sizeof(struct in_addr)) {
2940 break;
2941 }
2942 (void) memcpy(dst: &ipaddr.sin_addr, src: &ip->ip_dst,
2943 n: sizeof(ipaddr.sin_addr));
2944 /*
2945 * locate outgoing interface; if we're the destination,
2946 * use the incoming interface (should be same).
2947 */
2948 if ((ia = (INA)ifa_ifwithaddr(SA(&ipaddr))) == NULL) {
2949 if ((ia = ip_rtaddr(ipaddr.sin_addr)) == NULL) {
2950 type = ICMP_UNREACH;
2951 code = ICMP_UNREACH_HOST;
2952 goto bad;
2953 }
2954 }
2955 IFA_LOCK(&ia->ia_ifa);
2956 (void) memcpy(dst: cp + off, src: &(IA_SIN(ia)->sin_addr),
2957 n: sizeof(struct in_addr));
2958 IFA_UNLOCK(&ia->ia_ifa);
2959 ifa_remref(ifa: &ia->ia_ifa);
2960 ia = NULL;
2961 cp[IPOPT_OFFSET] += sizeof(struct in_addr);
2962 break;
2963
2964 case IPOPT_TS:
2965 code = (uint8_t)(cp - (u_char *)ip);
2966 ipt = (struct ip_timestamp *)(void *)cp;
2967 if (ipt->ipt_len < 4 || ipt->ipt_len > 40) {
2968 code = (uint8_t)((u_char *)&ipt->ipt_len -
2969 (u_char *)ip);
2970 goto bad;
2971 }
2972 if (ipt->ipt_ptr < 5) {
2973 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2974 (u_char *)ip);
2975 goto bad;
2976 }
2977 if (ipt->ipt_ptr >
2978 ipt->ipt_len - (int)sizeof(int32_t)) {
2979 if (++ipt->ipt_oflw == 0) {
2980 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2981 (u_char *)ip);
2982 goto bad;
2983 }
2984 break;
2985 }
2986 sin = (struct in_addr *)(void *)(cp + ipt->ipt_ptr - 1);
2987 switch (ipt->ipt_flg) {
2988 case IPOPT_TS_TSONLY:
2989 break;
2990
2991 case IPOPT_TS_TSANDADDR:
2992 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
2993 sizeof(struct in_addr) > ipt->ipt_len) {
2994 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
2995 (u_char *)ip);
2996 goto bad;
2997 }
2998 ipaddr.sin_addr = dst;
2999 ia = (INA)ifaof_ifpforaddr(SA(&ipaddr),
3000 m->m_pkthdr.rcvif);
3001 if (ia == NULL) {
3002 continue;
3003 }
3004 IFA_LOCK(&ia->ia_ifa);
3005 (void) memcpy(dst: sin, src: &IA_SIN(ia)->sin_addr,
3006 n: sizeof(struct in_addr));
3007 IFA_UNLOCK(&ia->ia_ifa);
3008 ipt->ipt_ptr += sizeof(struct in_addr);
3009 ifa_remref(ifa: &ia->ia_ifa);
3010 ia = NULL;
3011 break;
3012
3013 case IPOPT_TS_PRESPEC:
3014 if (ipt->ipt_ptr - 1 + sizeof(n_time) +
3015 sizeof(struct in_addr) > ipt->ipt_len) {
3016 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3017 (u_char *)ip);
3018 goto bad;
3019 }
3020 (void) memcpy(dst: &ipaddr.sin_addr, src: sin,
3021 n: sizeof(struct in_addr));
3022 if ((ia = (struct in_ifaddr *)ifa_ifwithaddr(
3023 SA(&ipaddr))) == NULL) {
3024 continue;
3025 }
3026 ifa_remref(ifa: &ia->ia_ifa);
3027 ia = NULL;
3028 ipt->ipt_ptr += sizeof(struct in_addr);
3029 break;
3030
3031 default:
3032 /* XXX can't take &ipt->ipt_flg */
3033 code = (uint8_t)((u_char *)&ipt->ipt_ptr -
3034 (u_char *)ip + 1);
3035 goto bad;
3036 }
3037 ntime = iptime();
3038 (void) memcpy(dst: cp + ipt->ipt_ptr - 1, src: &ntime,
3039 n: sizeof(n_time));
3040 ipt->ipt_ptr += sizeof(n_time);
3041 }
3042 }
3043 if (forward && ipforwarding) {
3044 ip_forward(m, 1, next_hop);
3045 return 1;
3046 }
3047 return 0;
3048bad:
3049 icmp_error(m, type, code, 0, 0);
3050 OSAddAtomic(1, &ipstat.ips_badoptions);
3051 return 1;
3052}
3053
3054/*
3055 * Check for the presence of the IP Router Alert option [RFC2113]
3056 * in the header of an IPv4 datagram.
3057 *
3058 * This call is not intended for use from the forwarding path; it is here
3059 * so that protocol domains may check for the presence of the option.
3060 * Given how FreeBSD's IPv4 stack is currently structured, the Router Alert
3061 * option does not have much relevance to the implementation, though this
3062 * may change in future.
3063 * Router alert options SHOULD be passed if running in IPSTEALTH mode and
3064 * we are not the endpoint.
3065 * Length checks on individual options should already have been peformed
3066 * by ip_dooptions() therefore they are folded under DIAGNOSTIC here.
3067 *
3068 * Return zero if not present or options are invalid, non-zero if present.
3069 */
3070int
3071ip_checkrouteralert(struct mbuf *m)
3072{
3073 struct ip *ip = mtod(m, struct ip *);
3074 u_char *cp;
3075 int opt, optlen, cnt, found_ra;
3076
3077 found_ra = 0;
3078 cp = (u_char *)(ip + 1);
3079 cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3080 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3081 opt = cp[IPOPT_OPTVAL];
3082 if (opt == IPOPT_EOL) {
3083 break;
3084 }
3085 if (opt == IPOPT_NOP) {
3086 optlen = 1;
3087 } else {
3088#ifdef DIAGNOSTIC
3089 if (cnt < IPOPT_OLEN + sizeof(*cp)) {
3090 break;
3091 }
3092#endif
3093 optlen = cp[IPOPT_OLEN];
3094#ifdef DIAGNOSTIC
3095 if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) {
3096 break;
3097 }
3098#endif
3099 }
3100 switch (opt) {
3101 case IPOPT_RA:
3102#ifdef DIAGNOSTIC
3103 if (optlen != IPOPT_OFFSET + sizeof(uint16_t) ||
3104 (*((uint16_t *)(void *)&cp[IPOPT_OFFSET]) != 0)) {
3105 break;
3106 } else
3107#endif
3108 found_ra = 1;
3109 break;
3110 default:
3111 break;
3112 }
3113 }
3114
3115 return found_ra;
3116}
3117
3118/*
3119 * Given address of next destination (final or next hop),
3120 * return internet address info of interface to be used to get there.
3121 */
3122struct in_ifaddr *
3123ip_rtaddr(struct in_addr dst)
3124{
3125 struct sockaddr_in *sin;
3126 struct ifaddr *rt_ifa;
3127 struct route ro;
3128
3129 bzero(s: &ro, n: sizeof(ro));
3130 sin = SIN(&ro.ro_dst);
3131 sin->sin_family = AF_INET;
3132 sin->sin_len = sizeof(*sin);
3133 sin->sin_addr = dst;
3134
3135 rtalloc_ign(&ro, RTF_PRCLONING);
3136 if (ro.ro_rt == NULL) {
3137 ROUTE_RELEASE(&ro);
3138 return NULL;
3139 }
3140
3141 RT_LOCK(ro.ro_rt);
3142 if ((rt_ifa = ro.ro_rt->rt_ifa) != NULL) {
3143 ifa_addref(ifa: rt_ifa);
3144 }
3145 RT_UNLOCK(ro.ro_rt);
3146 ROUTE_RELEASE(&ro);
3147
3148 return (struct in_ifaddr *)rt_ifa;
3149}
3150
3151/*
3152 * Save incoming source route for use in replies,
3153 * to be picked up later by ip_srcroute if the receiver is interested.
3154 */
3155void
3156save_rte(u_char *option, struct in_addr dst)
3157{
3158 unsigned olen;
3159
3160 olen = option[IPOPT_OLEN];
3161#if DIAGNOSTIC
3162 if (ipprintfs) {
3163 printf("save_rte: olen %d\n", olen);
3164 }
3165#endif
3166 if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst))) {
3167 return;
3168 }
3169 bcopy(src: option, dst: ip_srcrt.srcopt, n: olen);
3170 ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr);
3171 ip_srcrt.dst = dst;
3172}
3173
3174/*
3175 * Retrieve incoming source route for use in replies,
3176 * in the same form used by setsockopt.
3177 * The first hop is placed before the options, will be removed later.
3178 */
3179struct mbuf *
3180ip_srcroute(void)
3181{
3182 struct in_addr *p, *q;
3183 struct mbuf *m;
3184
3185 if (ip_nhops == 0) {
3186 return NULL;
3187 }
3188
3189 m = m_get(M_DONTWAIT, MT_HEADER);
3190 if (m == NULL) {
3191 return NULL;
3192 }
3193
3194#define OPTSIZ (sizeof (ip_srcrt.nop) + sizeof (ip_srcrt.srcopt))
3195
3196 /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */
3197 m->m_len = ip_nhops * sizeof(struct in_addr) +
3198 sizeof(struct in_addr) + OPTSIZ;
3199#if DIAGNOSTIC
3200 if (ipprintfs) {
3201 printf("ip_srcroute: nhops %d mlen %d", ip_nhops, m->m_len);
3202 }
3203#endif
3204
3205 /*
3206 * First save first hop for return route
3207 */
3208 p = &ip_srcrt.route[ip_nhops - 1];
3209 *(mtod(m, struct in_addr *)) = *p--;
3210#if DIAGNOSTIC
3211 if (ipprintfs) {
3212 printf(" hops %lx",
3213 (u_int32_t)ntohl(mtod(m, struct in_addr *)->s_addr));
3214 }
3215#endif
3216
3217 /*
3218 * Copy option fields and padding (nop) to mbuf.
3219 */
3220 ip_srcrt.nop = IPOPT_NOP;
3221 ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF;
3222 (void) __nochk_memcpy(mtod(m, caddr_t) + sizeof(struct in_addr),
3223 src: &ip_srcrt.nop, OPTSIZ);
3224 q = (struct in_addr *)(void *)(mtod(m, caddr_t) +
3225 sizeof(struct in_addr) + OPTSIZ);
3226#undef OPTSIZ
3227 /*
3228 * Record return path as an IP source route,
3229 * reversing the path (pointers are now aligned).
3230 */
3231 while (p >= ip_srcrt.route) {
3232#if DIAGNOSTIC
3233 if (ipprintfs) {
3234 printf(" %lx", (u_int32_t)ntohl(q->s_addr));
3235 }
3236#endif
3237 *q++ = *p--;
3238 }
3239 /*
3240 * Last hop goes to final destination.
3241 */
3242 *q = ip_srcrt.dst;
3243#if DIAGNOSTIC
3244 if (ipprintfs) {
3245 printf(" %lx\n", (u_int32_t)ntohl(q->s_addr));
3246 }
3247#endif
3248 return m;
3249}
3250
3251/*
3252 * Strip out IP options, at higher level protocol in the kernel.
3253 */
3254void
3255ip_stripoptions(struct mbuf *m)
3256{
3257 int i;
3258 struct ip *ip = mtod(m, struct ip *);
3259 caddr_t opts;
3260 int olen;
3261
3262 /* Expect 32-bit aligned data pointer on strict-align platforms */
3263 MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m);
3264
3265 /* use bcopy() since it supports overlapping range */
3266 olen = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
3267 opts = (caddr_t)(ip + 1);
3268 i = m->m_len - (sizeof(struct ip) + olen);
3269 bcopy(src: opts + olen, dst: opts, n: (unsigned)i);
3270 m->m_len -= olen;
3271 if (m->m_flags & M_PKTHDR) {
3272 m->m_pkthdr.len -= olen;
3273 }
3274 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, sizeof(struct ip) >> 2);
3275
3276 /*
3277 * We expect ip_{off,len} to be in host order by now, and
3278 * that the original IP header length has been subtracted
3279 * out from ip_len. Temporarily adjust ip_len for checksum
3280 * recalculation, and restore it afterwards.
3281 */
3282 ip->ip_len += sizeof(struct ip);
3283
3284 /* recompute checksum now that IP header is smaller */
3285#if BYTE_ORDER != BIG_ENDIAN
3286 HTONS(ip->ip_len);
3287 HTONS(ip->ip_off);
3288#endif /* BYTE_ORDER != BIG_ENDIAN */
3289 ip->ip_sum = in_cksum_hdr(ip);
3290#if BYTE_ORDER != BIG_ENDIAN
3291 NTOHS(ip->ip_off);
3292 NTOHS(ip->ip_len);
3293#endif /* BYTE_ORDER != BIG_ENDIAN */
3294
3295 ip->ip_len -= sizeof(struct ip);
3296
3297 /*
3298 * Given that we've just stripped IP options from the header,
3299 * we need to adjust the start offset accordingly if this
3300 * packet had gone thru partial checksum offload.
3301 */
3302 if ((m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PARTIAL)) ==
3303 (CSUM_DATA_VALID | CSUM_PARTIAL)) {
3304 if (m->m_pkthdr.csum_rx_start >= (sizeof(struct ip) + olen)) {
3305 /* most common case */
3306 m->m_pkthdr.csum_rx_start -= olen;
3307 } else {
3308 /* compute checksum in software instead */
3309 m->m_pkthdr.csum_flags &= ~CSUM_DATA_VALID;
3310 m->m_pkthdr.csum_data = 0;
3311 ipstat.ips_adj_hwcsum_clr++;
3312 }
3313 }
3314}
3315
3316u_char inetctlerrmap[PRC_NCMDS] = {
3317 0, 0, 0, 0,
3318 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH,
3319 ENETUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED,
3320 EMSGSIZE, EHOSTUNREACH, 0, 0,
3321 0, 0, EHOSTUNREACH, 0,
3322 ENOPROTOOPT, ECONNREFUSED
3323};
3324
3325static int
3326sysctl_ipforwarding SYSCTL_HANDLER_ARGS
3327{
3328#pragma unused(arg1, arg2)
3329 int i, was_ipforwarding = ipforwarding;
3330
3331 i = sysctl_handle_int(oidp, arg1: oidp->oid_arg1, arg2: oidp->oid_arg2, req);
3332 if (i != 0 || req->newptr == USER_ADDR_NULL) {
3333 return i;
3334 }
3335
3336 if (was_ipforwarding && !ipforwarding) {
3337 /* clean up IPv4 forwarding cached routes */
3338 ifnet_head_lock_shared();
3339 for (i = 0; i <= if_index; i++) {
3340 struct ifnet *ifp = ifindex2ifnet[i];
3341 if (ifp != NULL) {
3342 lck_mtx_lock(lck: &ifp->if_cached_route_lock);
3343 ROUTE_RELEASE(&ifp->if_fwd_route);
3344 bzero(s: &ifp->if_fwd_route,
3345 n: sizeof(ifp->if_fwd_route));
3346 lck_mtx_unlock(lck: &ifp->if_cached_route_lock);
3347 }
3348 }
3349 ifnet_head_done();
3350 }
3351
3352 return 0;
3353}
3354
3355/*
3356 * Similar to inp_route_{copyout,copyin} routines except that these copy
3357 * out the cached IPv4 forwarding route from struct ifnet instead of the
3358 * inpcb. See comments for those routines for explanations.
3359 */
3360static void
3361ip_fwd_route_copyout(struct ifnet *ifp, struct route *dst)
3362{
3363 struct route *src = &ifp->if_fwd_route;
3364
3365 lck_mtx_lock_spin(lck: &ifp->if_cached_route_lock);
3366 lck_mtx_convert_spin(lck: &ifp->if_cached_route_lock);
3367
3368 /* Minor sanity check */
3369 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3370 panic("%s: wrong or corrupted route: %p", __func__, src);
3371 }
3372
3373 route_copyout(dst, src, sizeof(*dst));
3374
3375 lck_mtx_unlock(lck: &ifp->if_cached_route_lock);
3376}
3377
3378static void
3379ip_fwd_route_copyin(struct ifnet *ifp, struct route *src)
3380{
3381 struct route *dst = &ifp->if_fwd_route;
3382
3383 lck_mtx_lock_spin(lck: &ifp->if_cached_route_lock);
3384 lck_mtx_convert_spin(lck: &ifp->if_cached_route_lock);
3385
3386 /* Minor sanity check */
3387 if (src->ro_rt != NULL && rt_key(src->ro_rt)->sa_family != AF_INET) {
3388 panic("%s: wrong or corrupted route: %p", __func__, src);
3389 }
3390
3391 if (ifp->if_fwd_cacheok) {
3392 route_copyin(src, dst, sizeof(*src));
3393 }
3394
3395 lck_mtx_unlock(lck: &ifp->if_cached_route_lock);
3396}
3397
3398/*
3399 * Forward a packet. If some error occurs return the sender
3400 * an icmp packet. Note we can't always generate a meaningful
3401 * icmp message because icmp doesn't have a large enough repertoire
3402 * of codes and types.
3403 *
3404 * If not forwarding, just drop the packet. This could be confusing
3405 * if ipforwarding was zero but some routing protocol was advancing
3406 * us as a gateway to somewhere. However, we must let the routing
3407 * protocol deal with that.
3408 *
3409 * The srcrt parameter indicates whether the packet is being forwarded
3410 * via a source route.
3411 */
3412static void
3413ip_forward(struct mbuf *m, int srcrt, struct sockaddr_in *next_hop)
3414{
3415#pragma unused(next_hop)
3416 struct ip *ip = mtod(m, struct ip *);
3417 struct sockaddr_in *sin;
3418 struct rtentry *rt;
3419 struct route fwd_rt;
3420 int error, type = 0, code = 0;
3421 struct mbuf *mcopy;
3422 n_long dest;
3423 struct in_addr pkt_dst;
3424 u_int32_t nextmtu = 0, len;
3425 struct ip_out_args ipoa;
3426 struct ifnet *rcvifp = m->m_pkthdr.rcvif;
3427
3428 bzero(s: &ipoa, n: sizeof(ipoa));
3429 ipoa.ipoa_boundif = IFSCOPE_NONE;
3430 ipoa.ipoa_sotc = SO_TC_UNSPEC;
3431 ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
3432
3433#if IPSEC
3434 struct secpolicy *sp = NULL;
3435 int ipsecerror;
3436#endif /* IPSEC */
3437#if PF
3438 struct pf_mtag *pf_mtag;
3439#endif /* PF */
3440
3441 dest = 0;
3442 pkt_dst = ip->ip_dst;
3443
3444#if DIAGNOSTIC
3445 if (ipprintfs) {
3446 printf("forward: src %lx dst %lx ttl %x\n",
3447 (u_int32_t)ip->ip_src.s_addr, (u_int32_t)pkt_dst.s_addr,
3448 ip->ip_ttl);
3449 }
3450#endif
3451
3452 if (m->m_flags & (M_BCAST | M_MCAST) || !in_canforward(pkt_dst)) {
3453 OSAddAtomic(1, &ipstat.ips_cantforward);
3454 m_freem(m);
3455 return;
3456 }
3457#if IPSTEALTH
3458 if (!ipstealth) {
3459#endif /* IPSTEALTH */
3460 if (ip->ip_ttl <= IPTTLDEC) {
3461 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS,
3462 dest, 0);
3463 return;
3464 }
3465#if IPSTEALTH
3466}
3467#endif /* IPSTEALTH */
3468
3469#if PF
3470 pf_mtag = pf_find_mtag(m);
3471 if (pf_mtag != NULL && pf_mtag->pftag_rtableid != IFSCOPE_NONE) {
3472 ipoa.ipoa_boundif = pf_mtag->pftag_rtableid;
3473 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
3474 }
3475#endif /* PF */
3476
3477 ip_fwd_route_copyout(ifp: rcvifp, dst: &fwd_rt);
3478
3479 sin = SIN(&fwd_rt.ro_dst);
3480 if (ROUTE_UNUSABLE(&fwd_rt) || pkt_dst.s_addr != sin->sin_addr.s_addr) {
3481 ROUTE_RELEASE(&fwd_rt);
3482
3483 sin->sin_family = AF_INET;
3484 sin->sin_len = sizeof(*sin);
3485 sin->sin_addr = pkt_dst;
3486
3487 rtalloc_scoped_ign(&fwd_rt, RTF_PRCLONING, ipoa.ipoa_boundif);
3488 if (fwd_rt.ro_rt == NULL) {
3489 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, dest, 0);
3490 goto done;
3491 }
3492 }
3493 rt = fwd_rt.ro_rt;
3494
3495 /*
3496 * Save the IP header and at most 8 bytes of the payload,
3497 * in case we need to generate an ICMP message to the src.
3498 *
3499 * We don't use m_copy() because it might return a reference
3500 * to a shared cluster. Both this function and ip_output()
3501 * assume exclusive access to the IP header in `m', so any
3502 * data in a cluster may change before we reach icmp_error().
3503 */
3504 MGET(mcopy, M_DONTWAIT, m->m_type);
3505 if (mcopy != NULL && m_dup_pkthdr(mcopy, m, M_DONTWAIT) == 0) {
3506 mcopy->m_len = imin(a: (IP_VHL_HL(ip->ip_vhl) << 2) + 8,
3507 b: (int)ip->ip_len);
3508 m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t));
3509 }
3510
3511#if IPSTEALTH
3512 if (!ipstealth) {
3513#endif /* IPSTEALTH */
3514 ip->ip_ttl -= IPTTLDEC;
3515#if IPSTEALTH
3516}
3517#endif /* IPSTEALTH */
3518
3519 /*
3520 * If forwarding packet using same interface that it came in on,
3521 * perhaps should send a redirect to sender to shortcut a hop.
3522 * Only send redirect if source is sending directly to us,
3523 * and if packet was not source routed (or has any options).
3524 * Also, don't send redirect if forwarding using a default route
3525 * or a route modified by a redirect.
3526 */
3527 RT_LOCK_SPIN(rt);
3528 if (rt->rt_ifp == m->m_pkthdr.rcvif &&
3529 !(rt->rt_flags & (RTF_DYNAMIC | RTF_MODIFIED)) &&
3530 satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY &&
3531 ipsendredirects && !srcrt && rt->rt_ifa != NULL) {
3532 struct in_ifaddr *ia = (struct in_ifaddr *)rt->rt_ifa;
3533 u_int32_t src = ntohl(ip->ip_src.s_addr);
3534
3535 /* Become a regular mutex */
3536 RT_CONVERT_LOCK(rt);
3537 IFA_LOCK_SPIN(&ia->ia_ifa);
3538 if ((src & ia->ia_subnetmask) == ia->ia_subnet) {
3539 if (rt->rt_flags & RTF_GATEWAY) {
3540 dest = satosin(rt->rt_gateway)->sin_addr.s_addr;
3541 } else {
3542 dest = pkt_dst.s_addr;
3543 }
3544 /*
3545 * Router requirements says to only send
3546 * host redirects.
3547 */
3548 type = ICMP_REDIRECT;
3549 code = ICMP_REDIRECT_HOST;
3550#if DIAGNOSTIC
3551 if (ipprintfs) {
3552 printf("redirect (%d) to %lx\n", code,
3553 (u_int32_t)dest);
3554 }
3555#endif
3556 }
3557 IFA_UNLOCK(&ia->ia_ifa);
3558 }
3559 RT_UNLOCK(rt);
3560
3561
3562 /* Mark this packet as being forwarded from another interface */
3563 m->m_pkthdr.pkt_flags |= PKTF_FORWARDED;
3564 len = m_pktlen(m);
3565
3566 error = ip_output(m, NULL, &fwd_rt, IP_FORWARDING | IP_OUTARGS,
3567 NULL, &ipoa);
3568
3569 /* Refresh rt since the route could have changed while in IP */
3570 rt = fwd_rt.ro_rt;
3571
3572 if (error != 0) {
3573 OSAddAtomic(1, &ipstat.ips_cantforward);
3574 } else {
3575 /*
3576 * Increment stats on the source interface; the ones
3577 * for destination interface has been taken care of
3578 * during output above by virtue of PKTF_FORWARDED.
3579 */
3580 rcvifp->if_fpackets++;
3581 rcvifp->if_fbytes += len;
3582
3583 OSAddAtomic(1, &ipstat.ips_forward);
3584 if (type != 0) {
3585 OSAddAtomic(1, &ipstat.ips_redirectsent);
3586 } else {
3587 if (mcopy != NULL) {
3588 /*
3589 * If we didn't have to go thru ipflow and
3590 * the packet was successfully consumed by
3591 * ip_output, the mcopy is rather a waste;
3592 * this could be further optimized.
3593 */
3594 m_freem(mcopy);
3595 }
3596 goto done;
3597 }
3598 }
3599 if (mcopy == NULL) {
3600 goto done;
3601 }
3602
3603 switch (error) {
3604 case 0: /* forwarded, but need redirect */
3605 /* type, code set above */
3606 break;
3607
3608 case ENETUNREACH: /* shouldn't happen, checked above */
3609 case EHOSTUNREACH:
3610 case ENETDOWN:
3611 case EHOSTDOWN:
3612 default:
3613 type = ICMP_UNREACH;
3614 code = ICMP_UNREACH_HOST;
3615 break;
3616
3617 case EMSGSIZE:
3618 type = ICMP_UNREACH;
3619 code = ICMP_UNREACH_NEEDFRAG;
3620
3621 if (rt == NULL) {
3622 break;
3623 } else {
3624 RT_LOCK_SPIN(rt);
3625 if (rt->rt_ifp != NULL) {
3626 nextmtu = rt->rt_ifp->if_mtu;
3627 }
3628 RT_UNLOCK(rt);
3629 }
3630#ifdef IPSEC
3631 if (ipsec_bypass) {
3632 break;
3633 }
3634
3635 /*
3636 * If the packet is routed over IPsec tunnel, tell the
3637 * originator the tunnel MTU.
3638 * tunnel MTU = if MTU - sizeof(IP) - ESP/AH hdrsiz
3639 * XXX quickhack!!!
3640 */
3641 sp = ipsec4_getpolicybyaddr(mcopy, IPSEC_DIR_OUTBOUND,
3642 IP_FORWARDING, &ipsecerror);
3643
3644 if (sp == NULL) {
3645 break;
3646 }
3647
3648 /*
3649 * find the correct route for outer IPv4
3650 * header, compute tunnel MTU.
3651 */
3652 nextmtu = 0;
3653
3654 if (sp->req != NULL &&
3655 sp->req->saidx.mode == IPSEC_MODE_TUNNEL) {
3656 struct secasindex saidx;
3657 struct secasvar *sav;
3658 struct route *ro;
3659 struct ip *ipm;
3660 size_t ipsechdr;
3661
3662 /* count IPsec header size */
3663 ipsechdr = ipsec_hdrsiz(sp);
3664
3665 ipm = mtod(mcopy, struct ip *);
3666 bcopy(src: &sp->req->saidx, dst: &saidx, n: sizeof(saidx));
3667 saidx.mode = sp->req->saidx.mode;
3668 saidx.reqid = sp->req->saidx.reqid;
3669 sin = SIN(&saidx.src);
3670 if (sin->sin_len == 0) {
3671 sin->sin_len = sizeof(*sin);
3672 sin->sin_family = AF_INET;
3673 sin->sin_port = IPSEC_PORT_ANY;
3674 bcopy(src: &ipm->ip_src, dst: &sin->sin_addr,
3675 n: sizeof(sin->sin_addr));
3676 }
3677 sin = SIN(&saidx.dst);
3678 if (sin->sin_len == 0) {
3679 sin->sin_len = sizeof(*sin);
3680 sin->sin_family = AF_INET;
3681 sin->sin_port = IPSEC_PORT_ANY;
3682 bcopy(src: &ipm->ip_dst, dst: &sin->sin_addr,
3683 n: sizeof(sin->sin_addr));
3684 }
3685 sav = key_allocsa_policy(&saidx);
3686 if (sav != NULL) {
3687 lck_mtx_lock(sadb_mutex);
3688 if (sav->sah != NULL) {
3689 ro = (struct route *)&sav->sah->sa_route;
3690 if (ro->ro_rt != NULL) {
3691 RT_LOCK(ro->ro_rt);
3692 if (ro->ro_rt->rt_ifp != NULL) {
3693 nextmtu = ro->ro_rt->
3694 rt_ifp->if_mtu;
3695 nextmtu -= ipsechdr;
3696 }
3697 RT_UNLOCK(ro->ro_rt);
3698 }
3699 }
3700 key_freesav(sav, KEY_SADB_LOCKED);
3701 lck_mtx_unlock(sadb_mutex);
3702 }
3703 }
3704 key_freesp(sp, KEY_SADB_UNLOCKED);
3705#endif /* IPSEC */
3706 break;
3707
3708 case ENOBUFS:
3709 /*
3710 * A router should not generate ICMP_SOURCEQUENCH as
3711 * required in RFC1812 Requirements for IP Version 4 Routers.
3712 * Source quench could be a big problem under DoS attacks,
3713 * or if the underlying interface is rate-limited.
3714 * Those who need source quench packets may re-enable them
3715 * via the net.inet.ip.sendsourcequench sysctl.
3716 */
3717 if (ip_sendsourcequench == 0) {
3718 m_freem(mcopy);
3719 goto done;
3720 } else {
3721 type = ICMP_SOURCEQUENCH;
3722 code = 0;
3723 }
3724 break;
3725
3726 case EACCES:
3727 m_freem(mcopy);
3728 goto done;
3729 }
3730
3731 if (type == ICMP_UNREACH && code == ICMP_UNREACH_NEEDFRAG) {
3732 OSAddAtomic(1, &ipstat.ips_cantfrag);
3733 }
3734
3735 icmp_error(mcopy, type, code, dest, nextmtu);
3736done:
3737 ip_fwd_route_copyin(ifp: rcvifp, src: &fwd_rt);
3738}
3739
3740int
3741ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
3742 struct mbuf *m)
3743{
3744 *mp = NULL;
3745 if (inp->inp_socket->so_options & SO_TIMESTAMP) {
3746 struct timeval tv;
3747
3748 getmicrotime(&tv);
3749 mp = sbcreatecontrol_mbuf(p: (caddr_t)&tv, size: sizeof(tv),
3750 SCM_TIMESTAMP, SOL_SOCKET, m: mp);
3751 if (*mp == NULL) {
3752 goto no_mbufs;
3753 }
3754 }
3755 if (inp->inp_socket->so_options & SO_TIMESTAMP_MONOTONIC) {
3756 uint64_t time;
3757
3758 time = mach_absolute_time();
3759 mp = sbcreatecontrol_mbuf(p: (caddr_t)&time, size: sizeof(time),
3760 SCM_TIMESTAMP_MONOTONIC, SOL_SOCKET, m: mp);
3761 if (*mp == NULL) {
3762 goto no_mbufs;
3763 }
3764 }
3765 if (inp->inp_socket->so_options & SO_TIMESTAMP_CONTINUOUS) {
3766 uint64_t time;
3767
3768 time = mach_continuous_time();
3769 mp = sbcreatecontrol_mbuf(p: (caddr_t)&time, size: sizeof(time),
3770 SCM_TIMESTAMP_CONTINUOUS, SOL_SOCKET, m: mp);
3771 if (*mp == NULL) {
3772 goto no_mbufs;
3773 }
3774 }
3775 if (inp->inp_socket->so_flags & SOF_RECV_TRAFFIC_CLASS) {
3776 int tc = m_get_traffic_class(m);
3777
3778 mp = sbcreatecontrol_mbuf(p: (caddr_t)&tc, size: sizeof(tc),
3779 SO_TRAFFIC_CLASS, SOL_SOCKET, m: mp);
3780 if (*mp == NULL) {
3781 goto no_mbufs;
3782 }
3783 }
3784 if ((inp->inp_socket->so_flags & SOF_RECV_WAKE_PKT) &&
3785 (m->m_pkthdr.pkt_flags & PKTF_WAKE_PKT)) {
3786 int flag = 1;
3787
3788 mp = sbcreatecontrol_mbuf(p: (caddr_t)&flag, size: sizeof(flag),
3789 SO_RECV_WAKE_PKT, SOL_SOCKET, m: mp);
3790 if (*mp == NULL) {
3791 goto no_mbufs;
3792 }
3793 }
3794
3795 if (inp->inp_flags & INP_RECVDSTADDR || SOFLOW_ENABLED(inp->inp_socket)) {
3796 mp = sbcreatecontrol_mbuf(p: (caddr_t)&ip->ip_dst,
3797 size: sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP, m: mp);
3798 if (*mp == NULL) {
3799 goto no_mbufs;
3800 }
3801 }
3802#ifdef notyet
3803 /*
3804 * XXX
3805 * Moving these out of udp_input() made them even more broken
3806 * than they already were.
3807 */
3808 /* options were tossed already */
3809 if (inp->inp_flags & INP_RECVOPTS) {
3810 mp = sbcreatecontrol_mbuf((caddr_t)opts_deleted_above,
3811 sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP, mp);
3812 if (*mp == NULL) {
3813 goto no_mbufs;
3814 }
3815 }
3816 /* ip_srcroute doesn't do what we want here, need to fix */
3817 if (inp->inp_flags & INP_RECVRETOPTS) {
3818 mp = sbcreatecontrol_mbuf((caddr_t)ip_srcroute(),
3819 sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP, mp);
3820 if (*mp == NULL) {
3821 goto no_mbufs;
3822 }
3823 }
3824#endif /* notyet */
3825 if (inp->inp_flags & INP_RECVIF) {
3826 struct ifnet *ifp;
3827 uint8_t sdlbuf[SOCK_MAXADDRLEN + 1];
3828 struct sockaddr_dl *sdl2 = SDL(sdlbuf);
3829
3830 /*
3831 * Make sure to accomodate the largest possible
3832 * size of SA(if_lladdr)->sa_len.
3833 */
3834 _CASSERT(sizeof(sdlbuf) == (SOCK_MAXADDRLEN + 1));
3835
3836 ifnet_head_lock_shared();
3837 if ((ifp = m->m_pkthdr.rcvif) != NULL &&
3838 ifp->if_index && IF_INDEX_IN_RANGE(ifp->if_index)) {
3839 struct ifaddr *ifa = ifnet_addrs[ifp->if_index - 1];
3840 struct sockaddr_dl *sdp;
3841
3842 if (!ifa || !ifa->ifa_addr) {
3843 goto makedummy;
3844 }
3845
3846 IFA_LOCK_SPIN(ifa);
3847 sdp = SDL(ifa->ifa_addr);
3848 /*
3849 * Change our mind and don't try copy.
3850 */
3851 if (sdp->sdl_family != AF_LINK) {
3852 IFA_UNLOCK(ifa);
3853 goto makedummy;
3854 }
3855 /* the above _CASSERT ensures sdl_len fits in sdlbuf */
3856 SOCKADDR_COPY(sdp, sdl2, sdp->sdl_len);
3857 IFA_UNLOCK(ifa);
3858 } else {
3859makedummy:
3860 sdl2->sdl_len =
3861 offsetof(struct sockaddr_dl, sdl_data[0]);
3862 sdl2->sdl_family = AF_LINK;
3863 sdl2->sdl_index = 0;
3864 sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0;
3865 }
3866 ifnet_head_done();
3867 mp = sbcreatecontrol_mbuf(p: (caddr_t)sdl2, size: sdl2->sdl_len,
3868 IP_RECVIF, IPPROTO_IP, m: mp);
3869 if (*mp == NULL) {
3870 goto no_mbufs;
3871 }
3872 }
3873 if (inp->inp_flags & INP_RECVTTL) {
3874 mp = sbcreatecontrol_mbuf(p: (caddr_t)&ip->ip_ttl,
3875 size: sizeof(ip->ip_ttl), IP_RECVTTL, IPPROTO_IP, m: mp);
3876 if (*mp == NULL) {
3877 goto no_mbufs;
3878 }
3879 }
3880 if (inp->inp_flags & INP_PKTINFO) {
3881 struct in_pktinfo pi;
3882
3883 bzero(s: &pi, n: sizeof(struct in_pktinfo));
3884 bcopy(src: &ip->ip_dst, dst: &pi.ipi_addr, n: sizeof(struct in_addr));
3885 pi.ipi_ifindex = (m != NULL && m->m_pkthdr.rcvif != NULL) ?
3886 m->m_pkthdr.rcvif->if_index : 0;
3887
3888 mp = sbcreatecontrol_mbuf(p: (caddr_t)&pi,
3889 size: sizeof(struct in_pktinfo), IP_RECVPKTINFO, IPPROTO_IP, m: mp);
3890 if (*mp == NULL) {
3891 goto no_mbufs;
3892 }
3893 }
3894 if (inp->inp_flags & INP_RECVTOS) {
3895 mp = sbcreatecontrol_mbuf(p: (caddr_t)&ip->ip_tos,
3896 size: sizeof(u_char), IP_RECVTOS, IPPROTO_IP, m: mp);
3897 if (*mp == NULL) {
3898 goto no_mbufs;
3899 }
3900 }
3901 return 0;
3902
3903no_mbufs:
3904 ipstat.ips_pktdropcntrl++;
3905 return ENOBUFS;
3906}
3907
3908static inline u_short
3909ip_cksum(struct mbuf *m, int hlen)
3910{
3911 u_short sum;
3912
3913 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) {
3914 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID);
3915 } else if (!(m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) &&
3916 !(m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
3917 /*
3918 * The packet arrived on an interface which isn't capable
3919 * of performing IP header checksum; compute it now.
3920 */
3921 sum = ip_cksum_hdr_in(m, hlen);
3922 } else {
3923 sum = 0;
3924 m->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR |
3925 CSUM_IP_CHECKED | CSUM_IP_VALID);
3926 m->m_pkthdr.csum_data = 0xffff;
3927 }
3928
3929 if (sum != 0) {
3930 OSAddAtomic(1, &ipstat.ips_badsum);
3931 }
3932
3933 return sum;
3934}
3935
3936static int
3937ip_getstat SYSCTL_HANDLER_ARGS
3938{
3939#pragma unused(oidp, arg1, arg2)
3940 if (req->oldptr == USER_ADDR_NULL) {
3941 req->oldlen = (size_t)sizeof(struct ipstat);
3942 }
3943
3944 return SYSCTL_OUT(req, &ipstat, MIN(sizeof(ipstat), req->oldlen));
3945}
3946
3947void
3948ip_setsrcifaddr_info(struct mbuf *m, uint16_t src_idx, struct in_ifaddr *ia)
3949{
3950 VERIFY(m->m_flags & M_PKTHDR);
3951
3952 /*
3953 * If the source ifaddr is specified, pick up the information
3954 * from there; otherwise just grab the passed-in ifindex as the
3955 * caller may not have the ifaddr available.
3956 */
3957 if (ia != NULL) {
3958 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3959 m->m_pkthdr.src_ifindex = ia->ia_ifp->if_index;
3960 } else {
3961 m->m_pkthdr.src_ifindex = src_idx;
3962 if (src_idx != 0) {
3963 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3964 }
3965 }
3966}
3967
3968void
3969ip_setdstifaddr_info(struct mbuf *m, uint16_t dst_idx, struct in_ifaddr *ia)
3970{
3971 VERIFY(m->m_flags & M_PKTHDR);
3972
3973 /*
3974 * If the destination ifaddr is specified, pick up the information
3975 * from there; otherwise just grab the passed-in ifindex as the
3976 * caller may not have the ifaddr available.
3977 */
3978 if (ia != NULL) {
3979 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3980 m->m_pkthdr.dst_ifindex = ia->ia_ifp->if_index;
3981 } else {
3982 m->m_pkthdr.dst_ifindex = dst_idx;
3983 if (dst_idx != 0) {
3984 m->m_pkthdr.pkt_flags |= PKTF_IFAINFO;
3985 }
3986 }
3987}
3988
3989int
3990ip_getsrcifaddr_info(struct mbuf *m, uint32_t *src_idx, uint32_t *iaf)
3991{
3992 VERIFY(m->m_flags & M_PKTHDR);
3993
3994 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
3995 return -1;
3996 }
3997
3998 if (src_idx != NULL) {
3999 *src_idx = m->m_pkthdr.src_ifindex;
4000 }
4001
4002 if (iaf != NULL) {
4003 *iaf = 0;
4004 }
4005
4006 return 0;
4007}
4008
4009int
4010ip_getdstifaddr_info(struct mbuf *m, uint32_t *dst_idx, uint32_t *iaf)
4011{
4012 VERIFY(m->m_flags & M_PKTHDR);
4013
4014 if (!(m->m_pkthdr.pkt_flags & PKTF_IFAINFO)) {
4015 return -1;
4016 }
4017
4018 if (dst_idx != NULL) {
4019 *dst_idx = m->m_pkthdr.dst_ifindex;
4020 }
4021
4022 if (iaf != NULL) {
4023 *iaf = 0;
4024 }
4025
4026 return 0;
4027}
4028
4029/*
4030 * Protocol input handler for IPPROTO_GRE.
4031 */
4032void
4033gre_input(struct mbuf *m, int off)
4034{
4035 gre_input_func_t fn = gre_input_func;
4036
4037 /*
4038 * If there is a registered GRE input handler, pass mbuf to it.
4039 */
4040 if (fn != NULL) {
4041 lck_mtx_unlock(lck: inet_domain_mutex);
4042 m = fn(m, off, (mtod(m, struct ip *))->ip_p);
4043 lck_mtx_lock(lck: inet_domain_mutex);
4044 }
4045
4046 /*
4047 * If no matching tunnel that is up is found, we inject
4048 * the mbuf to raw ip socket to see if anyone picks it up.
4049 */
4050 if (m != NULL) {
4051 rip_input(m, off);
4052 }
4053}
4054
4055/*
4056 * Private KPI for PPP/PPTP.
4057 */
4058int
4059ip_gre_register_input(gre_input_func_t fn)
4060{
4061 lck_mtx_lock(lck: inet_domain_mutex);
4062 gre_input_func = fn;
4063 lck_mtx_unlock(lck: inet_domain_mutex);
4064
4065 return 0;
4066}
4067
4068#if (DEBUG || DEVELOPMENT)
4069static int
4070sysctl_reset_ip_input_stats SYSCTL_HANDLER_ARGS
4071{
4072#pragma unused(arg1, arg2)
4073 int error, i;
4074
4075 i = ip_input_measure;
4076 error = sysctl_handle_int(oidp, &i, 0, req);
4077 if (error || req->newptr == USER_ADDR_NULL) {
4078 goto done;
4079 }
4080 /* impose bounds */
4081 if (i < 0 || i > 1) {
4082 error = EINVAL;
4083 goto done;
4084 }
4085 if (ip_input_measure != i && i == 1) {
4086 net_perf_initialize(&net_perf, ip_input_measure_bins);
4087 }
4088 ip_input_measure = i;
4089done:
4090 return error;
4091}
4092
4093static int
4094sysctl_ip_input_measure_bins SYSCTL_HANDLER_ARGS
4095{
4096#pragma unused(arg1, arg2)
4097 int error;
4098 uint64_t i;
4099
4100 i = ip_input_measure_bins;
4101 error = sysctl_handle_quad(oidp, &i, 0, req);
4102 if (error || req->newptr == USER_ADDR_NULL) {
4103 goto done;
4104 }
4105 /* validate data */
4106 if (!net_perf_validate_bins(i)) {
4107 error = EINVAL;
4108 goto done;
4109 }
4110 ip_input_measure_bins = i;
4111done:
4112 return error;
4113}
4114
4115static int
4116sysctl_ip_input_getperf SYSCTL_HANDLER_ARGS
4117{
4118#pragma unused(oidp, arg1, arg2)
4119 if (req->oldptr == USER_ADDR_NULL) {
4120 req->oldlen = (size_t)sizeof(struct ipstat);
4121 }
4122
4123 return SYSCTL_OUT(req, &net_perf, MIN(sizeof(net_perf), req->oldlen));
4124}
4125#endif /* (DEBUG || DEVELOPMENT) */
4126
4127static int
4128sysctl_ip_checkinterface SYSCTL_HANDLER_ARGS
4129{
4130#pragma unused(arg1, arg2)
4131 int error, i;
4132
4133 i = ip_checkinterface;
4134 error = sysctl_handle_int(oidp, arg1: &i, arg2: 0, req);
4135 if (error != 0 || req->newptr == USER_ADDR_NULL) {
4136 return error;
4137 }
4138
4139 switch (i) {
4140 case IP_CHECKINTERFACE_WEAK_ES:
4141 case IP_CHECKINTERFACE_HYBRID_ES:
4142 case IP_CHECKINTERFACE_STRONG_ES:
4143 if (ip_checkinterface != i) {
4144 ip_checkinterface = i;
4145 os_log(OS_LOG_DEFAULT, "%s: ip_checkinterface is now %d\n",
4146 __func__, ip_checkinterface);
4147 }
4148 break;
4149 default:
4150 error = EINVAL;
4151 break;
4152 }
4153 return error;
4154}
4155