1 /*-
2 * Copyright (c) 1982, 1986, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2010-2011 Juniper Networks, Inc.
5 * All rights reserved.
6 *
7 * Portions of this software were developed by Robert N. M. Watson under
8 * contract to Juniper Networks, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)in_pcb.h 8.1 (Berkeley) 6/10/93
35 * $FreeBSD: stable/10/sys/netinet/in_pcb.h 309108 2016-11-24 14:48:46Z jch $
36 */
37
38 #ifndef _NETINET_IN_PCB_H_
39 #define _NETINET_IN_PCB_H_
40
41 #include <sys/queue.h>
42 #include <sys/_lock.h>
43 #include <sys/_mutex.h>
44 #include <sys/_rwlock.h>
45
46 #ifdef _KERNEL
47 #include <sys/lock.h>
48 #include <sys/rwlock.h>
49 #include <net/vnet.h>
50 #include <vm/uma.h>
51 #endif
52
53 #define in6pcb inpcb /* for KAME src sync over BSD*'s */
54 #define in6p_sp inp_sp /* for KAME src sync over BSD*'s */
55 struct inpcbpolicy;
56
57 /*
58 * struct inpcb is the common protocol control block structure used in most
59 * IP transport protocols.
60 *
61 * Pointers to local and foreign host table entries, local and foreign socket
62 * numbers, and pointers up (to a socket structure) and down (to a
63 * protocol-specific control block) are stored here.
64 */
65 LIST_HEAD(inpcbhead, inpcb);
66 LIST_HEAD(inpcbporthead, inpcbport);
67 typedef u_quad_t inp_gen_t;
68
69 /*
70 * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet.
71 * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing
72 * the following structure.
73 */
74 struct in_addr_4in6 {
75 u_int32_t ia46_pad32[3];
76 struct in_addr ia46_addr4;
77 };
78
79 /*
80 * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has
81 * some extra padding to accomplish this.
82 */
83 struct in_endpoints {
84 u_int16_t ie_fport; /* foreign port */
85 u_int16_t ie_lport; /* local port */
86 /* protocol dependent part, local and foreign addr */
87 union {
88 /* foreign host table entry */
89 struct in_addr_4in6 ie46_foreign;
90 struct in6_addr ie6_foreign;
91 } ie_dependfaddr;
92 union {
93 /* local host table entry */
94 struct in_addr_4in6 ie46_local;
95 struct in6_addr ie6_local;
96 } ie_dependladdr;
97 };
98 #define ie_faddr ie_dependfaddr.ie46_foreign.ia46_addr4
99 #define ie_laddr ie_dependladdr.ie46_local.ia46_addr4
100 #define ie6_faddr ie_dependfaddr.ie6_foreign
101 #define ie6_laddr ie_dependladdr.ie6_local
102
103 /*
104 * XXX The defines for inc_* are hacks and should be changed to direct
105 * references.
106 */
107 struct in_conninfo {
108 u_int8_t inc_flags;
109 u_int8_t inc_len;
110 u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */
111 /* protocol dependent part */
112 struct in_endpoints inc_ie;
113 };
114
115 /*
116 * Flags for inc_flags.
117 */
118 #define INC_ISIPV6 0x01
119
120 #define inc_isipv6 inc_flags /* temp compatability */
121 #define inc_fport inc_ie.ie_fport
122 #define inc_lport inc_ie.ie_lport
123 #define inc_faddr inc_ie.ie_faddr
124 #define inc_laddr inc_ie.ie_laddr
125 #define inc6_faddr inc_ie.ie6_faddr
126 #define inc6_laddr inc_ie.ie6_laddr
127
128 struct icmp6_filter;
129
130 /*-
131 * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and
132 * IPv6 sockets. In the case of TCP and UDP, further per-connection state is
133 * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb
134 * are static after creation or protected by a per-inpcb rwlock, inp_lock. A
135 * few fields are protected by multiple locks as indicated in the locking notes
136 * below. For these fields, all of the listed locks must be write-locked for
137 * any modifications. However, these fields can be safely read while any one of
138 * the listed locks are read-locked. This model can permit greater concurrency
139 * for read operations. For example, connections can be looked up while only
140 * holding a read lock on the global pcblist lock. This is important for
141 * performance when attempting to find the connection for a packet given its IP
142 * and port tuple.
143 *
144 * One noteworthy exception is that the global pcbinfo lock follows a different
145 * set of rules in relation to the inp_list field. Rather than being
146 * write-locked for modifications and read-locked for list iterations, it must
147 * be read-locked during modifications and write-locked during list iterations.
148 * This ensures that the relatively rare global list iterations safely walk a
149 * stable snapshot of connections while allowing more common list modifications
150 * to safely grab the pcblist lock just while adding or removing a connection
151 * from the global list.
152 *
153 * Key:
154 * (c) - Constant after initialization
155 * (g) - Protected by the pcbgroup lock
156 * (i) - Protected by the inpcb lock
157 * (p) - Protected by the pcbinfo lock for the inpcb
158 * (l) - Protected by the pcblist lock for the inpcb
159 * (h) - Protected by the pcbhash lock for the inpcb
160 * (s) - Protected by another subsystem's locks
161 * (x) - Undefined locking
162 *
163 * A few other notes:
164 *
165 * When a read lock is held, stability of the field is guaranteed; to write
166 * to a field, a write lock must generally be held.
167 *
168 * netinet/netinet6-layer code should not assume that the inp_socket pointer
169 * is safe to dereference without inp_lock being held, even for protocols
170 * other than TCP (where the inpcb persists during TIMEWAIT even after the
171 * socket has been freed), or there may be close(2)-related races.
172 *
173 * The inp_vflag field is overloaded, and would otherwise ideally be (c).
174 *
175 * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock
176 * read-lock usage during modification, this model can be applied to other
177 * protocols (especially SCTP).
178 */
179 struct inpcb {
180 LIST_ENTRY(inpcb) inp_hash; /* (h/i) hash list */
181 LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */
182 LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */
183 /* (p[w]) for list iteration */
184 /* (p[r]/l) for addition/removal */
185 void *inp_ppcb; /* (i) pointer to per-protocol pcb */
186 struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */
187 struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */
188 LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */
189 struct socket *inp_socket; /* (i) back pointer to socket */
190 struct ucred *inp_cred; /* (c) cache of socket cred */
191 u_int32_t inp_flow; /* (i) IPv6 flow information */
192 int inp_flags; /* (i) generic IP/datagram flags */
193 int inp_flags2; /* (i) generic IP/datagram flags #2*/
194 u_char inp_vflag; /* (i) IP version flag (v4/v6) */
195 u_char inp_ip_ttl; /* (i) time to live proto */
196 u_char inp_ip_p; /* (c) protocol proto */
197 u_char inp_ip_minttl; /* (i) minimum TTL or drop */
198 uint32_t inp_flowid; /* (x) flow id / queue id */
199 u_int inp_refcount; /* (i) refcount */
200 void *inp_pspare[5]; /* (x) route caching / general use */
201 uint32_t inp_flowtype; /* (x) M_HASHTYPE value */
202 u_int inp_ispare[5]; /* (x) route caching / user cookie /
203 * general use */
204
205 /* Local and foreign ports, local and foreign addr. */
206 struct in_conninfo inp_inc; /* (i) list for PCB's local port */
207
208 /* MAC and IPSEC policy information. */
209 struct label *inp_label; /* (i) MAC label */
210 struct inpcbpolicy *inp_sp; /* (s) for IPSEC */
211
212 /* Protocol-dependent part; options. */
213 struct {
214 u_char inp4_ip_tos; /* (i) type of service proto */
215 struct mbuf *inp4_options; /* (i) IP options */
216 struct ip_moptions *inp4_moptions; /* (i) IP mcast options */
217 } inp_depend4;
218 struct {
219 /* (i) IP options */
220 struct mbuf *inp6_options;
221 /* (i) IP6 options for outgoing packets */
222 struct ip6_pktopts *inp6_outputopts;
223 /* (i) IP multicast options */
224 struct ip6_moptions *inp6_moptions;
225 /* (i) ICMPv6 code type filter */
226 struct icmp6_filter *inp6_icmp6filt;
227 /* (i) IPV6_CHECKSUM setsockopt */
228 int inp6_cksum;
229 short inp6_hops;
230 } inp_depend6;
231 LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */
232 struct inpcbport *inp_phd; /* (i/h) head of this list */
233 #define inp_zero_size offsetof(struct inpcb, inp_gencnt)
234 inp_gen_t inp_gencnt; /* (c) generation count */
235 struct llentry *inp_lle; /* cached L2 information */
236 struct rtentry *inp_rt; /* cached L3 information */
237 struct rwlock inp_lock;
238 };
239 #define inp_fport inp_inc.inc_fport
240 #define inp_lport inp_inc.inc_lport
241 #define inp_faddr inp_inc.inc_faddr
242 #define inp_laddr inp_inc.inc_laddr
243 #define inp_ip_tos inp_depend4.inp4_ip_tos
244 #define inp_options inp_depend4.inp4_options
245 #define inp_moptions inp_depend4.inp4_moptions
246
247 #define in6p_faddr inp_inc.inc6_faddr
248 #define in6p_laddr inp_inc.inc6_laddr
249 #define in6p_hops inp_depend6.inp6_hops /* default hop limit */
250 #define in6p_flowinfo inp_flow
251 #define in6p_options inp_depend6.inp6_options
252 #define in6p_outputopts inp_depend6.inp6_outputopts
253 #define in6p_moptions inp_depend6.inp6_moptions
254 #define in6p_icmp6filt inp_depend6.inp6_icmp6filt
255 #define in6p_cksum inp_depend6.inp6_cksum
256
257 #define inp_vnet inp_pcbinfo->ipi_vnet
258
259 /*
260 * The range of the generation count, as used in this implementation, is 9e19.
261 * We would have to create 300 billion connections per second for this number
262 * to roll over in a year. This seems sufficiently unlikely that we simply
263 * don't concern ourselves with that possibility.
264 */
265
266 /*
267 * Interface exported to userland by various protocols which use inpcbs. Hack
268 * alert -- only define if struct xsocket is in scope.
269 */
270 #ifdef _SYS_SOCKETVAR_H_
271 struct xinpcb {
272 size_t xi_len; /* length of this structure */
273 struct inpcb xi_inp;
274 struct xsocket xi_socket;
275 u_quad_t xi_alignment_hack;
276 };
277
278 struct xinpgen {
279 size_t xig_len; /* length of this structure */
280 u_int xig_count; /* number of PCBs at this time */
281 inp_gen_t xig_gen; /* generation count at this time */
282 so_gen_t xig_sogen; /* socket generation count at this time */
283 };
284 #endif /* _SYS_SOCKETVAR_H_ */
285
286 struct inpcbport {
287 LIST_ENTRY(inpcbport) phd_hash;
288 struct inpcbhead phd_pcblist;
289 u_short phd_port;
290 };
291
292 /*-
293 * Global data structure for each high-level protocol (UDP, TCP, ...) in both
294 * IPv4 and IPv6. Holds inpcb lists and information for managing them.
295 *
296 * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and
297 * ipi_list_lock:
298 * - ipi_lock covering the global pcb list stability during loop iteration,
299 * - ipi_hash_lock covering the hashed lookup tables,
300 * - ipi_list_lock covering mutable global fields (such as the global
301 * pcb list)
302 *
303 * The lock order is:
304 *
305 * ipi_lock (before)
306 * inpcb locks (before)
307 * ipi_list locks (before)
308 * {ipi_hash_lock, pcbgroup locks}
309 *
310 * Locking key:
311 *
312 * (c) Constant or nearly constant after initialisation
313 * (g) Locked by ipi_lock
314 * (l) Locked by ipi_list_lock
315 * (h) Read using either ipi_hash_lock or inpcb lock; write requires both
316 * (p) Protected by one or more pcbgroup locks
317 * (x) Synchronisation properties poorly defined
318 */
319 struct inpcbinfo {
320 /*
321 * Global lock protecting full inpcb list traversal
322 */
323 struct rwlock ipi_lock;
324
325 /*
326 * Global list of inpcbs on the protocol.
327 */
328 struct inpcbhead *ipi_listhead; /* (g/l) */
329 u_int ipi_count; /* (l) */
330
331 /*
332 * Generation count -- incremented each time a connection is allocated
333 * or freed.
334 */
335 u_quad_t ipi_gencnt; /* (l) */
336
337 /*
338 * Fields associated with port lookup and allocation.
339 */
340 u_short ipi_lastport; /* (x) */
341 u_short ipi_lastlow; /* (x) */
342 u_short ipi_lasthi; /* (x) */
343
344 /*
345 * UMA zone from which inpcbs are allocated for this protocol.
346 */
347 struct uma_zone *ipi_zone; /* (c) */
348
349 /*
350 * Connection groups associated with this protocol. These fields are
351 * constant, but pcbgroup structures themselves are protected by
352 * per-pcbgroup locks.
353 */
354 struct inpcbgroup *ipi_pcbgroups; /* (c) */
355 u_int ipi_npcbgroups; /* (c) */
356 u_int ipi_hashfields; /* (c) */
357
358 /*
359 * Global lock protecting non-pcbgroup hash lookup tables.
360 */
361 struct rwlock ipi_hash_lock;
362
363 /*
364 * Global hash of inpcbs, hashed by local and foreign addresses and
365 * port numbers.
366 */
367 struct inpcbhead *ipi_hashbase; /* (h) */
368 u_long ipi_hashmask; /* (h) */
369
370 /*
371 * Global hash of inpcbs, hashed by only local port number.
372 */
373 struct inpcbporthead *ipi_porthashbase; /* (h) */
374 u_long ipi_porthashmask; /* (h) */
375
376 /*
377 * List of wildcard inpcbs for use with pcbgroups. In the past, was
378 * per-pcbgroup but is now global. All pcbgroup locks must be held
379 * to modify the list, so any is sufficient to read it.
380 */
381 struct inpcbhead *ipi_wildbase; /* (p) */
382 u_long ipi_wildmask; /* (p) */
383
384 /*
385 * Pointer to network stack instance
386 */
387 struct vnet *ipi_vnet; /* (c) */
388
389 /*
390 * general use 2
391 */
392 void *ipi_pspare[2];
393
394 /*
395 * Global lock protecting global inpcb list, inpcb count, etc.
396 */
397 struct rwlock ipi_list_lock;
398 };
399
400 #ifdef _KERNEL
401 /*
402 * Connection groups hold sets of connections that have similar CPU/thread
403 * affinity. Each connection belongs to exactly one connection group.
404 */
405 struct inpcbgroup {
406 /*
407 * Per-connection group hash of inpcbs, hashed by local and foreign
408 * addresses and port numbers.
409 */
410 struct inpcbhead *ipg_hashbase; /* (c) */
411 u_long ipg_hashmask; /* (c) */
412
413 /*
414 * Notional affinity of this pcbgroup.
415 */
416 u_int ipg_cpu; /* (p) */
417
418 /*
419 * Per-connection group lock, not to be confused with ipi_lock.
420 * Protects the hash table hung off the group, but also the global
421 * wildcard list in inpcbinfo.
422 */
423 struct mtx ipg_lock;
424 } __aligned(CACHE_LINE_SIZE);
425
426 #define INP_LOCK_INIT(inp, d, t) \
427 rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK)
428 #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock)
429 #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock)
430 #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock)
431 #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock)
432 #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock)
433 #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock)
434 #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock)
435 #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock)
436 #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock)
437 #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock)
438 #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED)
439 #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED)
440 #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED)
441 #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED)
442
443 /*
444 * These locking functions are for inpcb consumers outside of sys/netinet,
445 * more specifically, they were added for the benefit of TOE drivers. The
446 * macros are reserved for use by the stack.
447 */
448 void inp_wlock(struct inpcb *);
449 void inp_wunlock(struct inpcb *);
450 void inp_rlock(struct inpcb *);
451 void inp_runlock(struct inpcb *);
452
453 #ifdef INVARIANTS
454 void inp_lock_assert(struct inpcb *);
455 void inp_unlock_assert(struct inpcb *);
456 #else
457 static __inline void
inp_lock_assert(struct inpcb * inp __unused)458 inp_lock_assert(struct inpcb *inp __unused)
459 {
460 }
461
462 static __inline void
inp_unlock_assert(struct inpcb * inp __unused)463 inp_unlock_assert(struct inpcb *inp __unused)
464 {
465 }
466
467 #endif
468
469 void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg);
470 int inp_ip_tos_get(const struct inpcb *inp);
471 void inp_ip_tos_set(struct inpcb *inp, int val);
472 struct socket *
473 inp_inpcbtosocket(struct inpcb *inp);
474 struct tcpcb *
475 inp_inpcbtotcpcb(struct inpcb *inp);
476 void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
477 uint32_t *faddr, uint16_t *fp);
478 short inp_so_options(const struct inpcb *inp);
479
480 #endif /* _KERNEL */
481
482 #define INP_INFO_LOCK_INIT(ipi, d) \
483 rw_init_flags(&(ipi)->ipi_lock, (d), RW_RECURSE)
484 #define INP_INFO_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_lock)
485 #define INP_INFO_RLOCK(ipi) rw_rlock(&(ipi)->ipi_lock)
486 #define INP_INFO_WLOCK(ipi) rw_wlock(&(ipi)->ipi_lock)
487 #define INP_INFO_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_lock)
488 #define INP_INFO_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_lock)
489 #define INP_INFO_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_lock)
490 #define INP_INFO_WLOCKED(ipi) rw_wowned(&(ipi)->ipi_lock)
491 #define INP_INFO_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_lock)
492 #define INP_INFO_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_lock)
493 #define INP_INFO_LOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_LOCKED)
494 #define INP_INFO_RLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_RLOCKED)
495 #define INP_INFO_WLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_WLOCKED)
496 #define INP_INFO_UNLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_lock, RA_UNLOCKED)
497
498 #define INP_LIST_LOCK_INIT(ipi, d) \
499 rw_init_flags(&(ipi)->ipi_list_lock, (d), 0)
500 #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock)
501 #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock)
502 #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock)
503 #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock)
504 #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock)
505 #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock)
506 #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock)
507 #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock)
508 #define INP_LIST_LOCK_ASSERT(ipi) \
509 rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED)
510 #define INP_LIST_RLOCK_ASSERT(ipi) \
511 rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED)
512 #define INP_LIST_WLOCK_ASSERT(ipi) \
513 rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED)
514 #define INP_LIST_UNLOCK_ASSERT(ipi) \
515 rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED)
516
517 #define INP_HASH_LOCK_INIT(ipi, d) \
518 rw_init_flags(&(ipi)->ipi_hash_lock, (d), 0)
519 #define INP_HASH_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_hash_lock)
520 #define INP_HASH_RLOCK(ipi) rw_rlock(&(ipi)->ipi_hash_lock)
521 #define INP_HASH_WLOCK(ipi) rw_wlock(&(ipi)->ipi_hash_lock)
522 #define INP_HASH_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_hash_lock)
523 #define INP_HASH_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_hash_lock)
524 #define INP_HASH_LOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_hash_lock, \
525 RA_LOCKED)
526 #define INP_HASH_WLOCK_ASSERT(ipi) rw_assert(&(ipi)->ipi_hash_lock, \
527 RA_WLOCKED)
528
529 #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \
530 MTX_DEF | MTX_DUPOK)
531 #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock)
532
533 #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock)
534 #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED)
535 #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock)
536
537 #define INP_PCBHASH(faddr, lport, fport, mask) \
538 (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask))
539 #define INP_PCBPORTHASH(lport, mask) \
540 (ntohs((lport)) & (mask))
541
542 /*
543 * Flags for inp_vflags -- historically version flags only
544 */
545 #define INP_IPV4 0x1
546 #define INP_IPV6 0x2
547 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */
548
549 /*
550 * Flags for inp_flags.
551 */
552 #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */
553 #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */
554 #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */
555 #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */
556 #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */
557 #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */
558 #define INP_ANONPORT 0x00000040 /* port chosen for user */
559 #define INP_RECVIF 0x00000080 /* receive incoming interface */
560 #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */
561 #define INP_FAITH 0x00000200 /* accept FAITH'ed connections */
562 #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */
563 #define INP_DONTFRAG 0x00000800 /* don't fragment packet */
564 #define INP_BINDANY 0x00001000 /* allow bind to any address */
565 #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */
566 #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */
567 #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */
568 #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */
569 #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */
570 #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */
571 #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */
572 #define IN6P_RTHDR 0x00100000 /* receive routing header */
573 #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */
574 #define IN6P_TCLASS 0x00400000 /* receive traffic class value */
575 #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */
576 #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */
577 #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */
578 #define INP_DROPPED 0x04000000 /* protocol drop flag */
579 #define INP_SOCKREF 0x08000000 /* strong socket reference */
580 #define INP_RESERVED_0 0x10000000 /* reserved field */
581 #define INP_RESERVED_1 0x20000000 /* reserved field */
582 #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */
583 #define IN6P_MTU 0x80000000 /* receive path MTU */
584
585 #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\
586 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\
587 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\
588 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\
589 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\
590 IN6P_MTU)
591
592 /*
593 * Flags for inp_flags2.
594 */
595 #define INP_LLE_VALID 0x00000001 /* cached lle is valid */
596 #define INP_RT_VALID 0x00000002 /* cached rtentry is valid */
597 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */
598 #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */
599 #define INP_FREED 0x00000010 /* inp itself is not valid */
600 #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */
601
602 /*
603 * Flags passed to in_pcblookup*() functions.
604 */
605 #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */
606 #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */
607 #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */
608
609 #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \
610 INPLOOKUP_WLOCKPCB)
611
612 #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb)
613 #define sotoin6pcb(so) sotoinpcb(so) /* for KAME src sync over BSD*'s */
614
615 #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family
616
617 #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af)
618
619 /*
620 * Constants for pcbinfo.ipi_hashfields.
621 */
622 #define IPI_HASHFIELDS_NONE 0
623 #define IPI_HASHFIELDS_2TUPLE 1
624 #define IPI_HASHFIELDS_4TUPLE 2
625
626 #ifdef _KERNEL
627 VNET_DECLARE(int, ipport_reservedhigh);
628 VNET_DECLARE(int, ipport_reservedlow);
629 VNET_DECLARE(int, ipport_lowfirstauto);
630 VNET_DECLARE(int, ipport_lowlastauto);
631 VNET_DECLARE(int, ipport_firstauto);
632 VNET_DECLARE(int, ipport_lastauto);
633 VNET_DECLARE(int, ipport_hifirstauto);
634 VNET_DECLARE(int, ipport_hilastauto);
635 VNET_DECLARE(int, ipport_randomized);
636 VNET_DECLARE(int, ipport_randomcps);
637 VNET_DECLARE(int, ipport_randomtime);
638 VNET_DECLARE(int, ipport_stoprandom);
639 VNET_DECLARE(int, ipport_tcpallocs);
640
641 #define V_ipport_reservedhigh VNET(ipport_reservedhigh)
642 #define V_ipport_reservedlow VNET(ipport_reservedlow)
643 #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto)
644 #define V_ipport_lowlastauto VNET(ipport_lowlastauto)
645 #define V_ipport_firstauto VNET(ipport_firstauto)
646 #define V_ipport_lastauto VNET(ipport_lastauto)
647 #define V_ipport_hifirstauto VNET(ipport_hifirstauto)
648 #define V_ipport_hilastauto VNET(ipport_hilastauto)
649 #define V_ipport_randomized VNET(ipport_randomized)
650 #define V_ipport_randomcps VNET(ipport_randomcps)
651 #define V_ipport_randomtime VNET(ipport_randomtime)
652 #define V_ipport_stoprandom VNET(ipport_stoprandom)
653 #define V_ipport_tcpallocs VNET(ipport_tcpallocs)
654
655 void in_pcbinfo_destroy(struct inpcbinfo *);
656 void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *,
657 int, int, char *, uma_init, uma_fini, uint32_t, u_int);
658
659 struct inpcbgroup *
660 in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t);
661 struct inpcbgroup *
662 in_pcbgroup_byinpcb(struct inpcb *);
663 struct inpcbgroup *
664 in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short,
665 struct in_addr, u_short);
666 void in_pcbgroup_destroy(struct inpcbinfo *);
667 int in_pcbgroup_enabled(struct inpcbinfo *);
668 void in_pcbgroup_init(struct inpcbinfo *, u_int, int);
669 void in_pcbgroup_remove(struct inpcb *);
670 void in_pcbgroup_update(struct inpcb *);
671 void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *);
672
673 void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *);
674 int in_pcballoc(struct socket *, struct inpcbinfo *);
675 int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *);
676 int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *,
677 struct ucred *, int);
678 int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
679 u_short *, struct ucred *);
680 int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *);
681 int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *,
682 struct mbuf *);
683 int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
684 u_short *, in_addr_t *, u_short *, struct inpcb **,
685 struct ucred *);
686 void in_pcbdetach(struct inpcb *);
687 void in_pcbdisconnect(struct inpcb *);
688 void in_pcbdrop(struct inpcb *);
689 void in_pcbfree(struct inpcb *);
690 int in_pcbinshash(struct inpcb *);
691 int in_pcbinshash_nopcbgroup(struct inpcb *);
692 int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *,
693 struct ucred *);
694 struct inpcb *
695 in_pcblookup_local(struct inpcbinfo *,
696 struct in_addr, u_short, int, struct ucred *);
697 struct inpcb *
698 in_pcblookup(struct inpcbinfo *, struct in_addr, u_int,
699 struct in_addr, u_int, int, struct ifnet *);
700 struct inpcb *
701 in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int,
702 struct in_addr, u_int, int, struct ifnet *, struct mbuf *);
703 void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr,
704 int, struct inpcb *(*)(struct inpcb *, int));
705 void in_pcbref(struct inpcb *);
706 void in_pcbrehash(struct inpcb *);
707 void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *);
708 int in_pcbrele(struct inpcb *);
709 int in_pcbrele_rlocked(struct inpcb *);
710 int in_pcbrele_wlocked(struct inpcb *);
711 void in_pcbsetsolabel(struct socket *so);
712 int in_getpeeraddr(struct socket *so, struct sockaddr **nam);
713 int in_getsockaddr(struct socket *so, struct sockaddr **nam);
714 struct sockaddr *
715 in_sockaddr(in_port_t port, struct in_addr *addr);
716 void in_pcbsosetlabel(struct socket *so);
717 #endif /* _KERNEL */
718
719 #endif /* !_NETINET_IN_PCB_H_ */
720