xref: /freebsd/sys/netinet/in_pcb.h (revision 5bb3134a8c21cb87b30e135ef168483f0333dabb)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1990, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2010-2011 Juniper Networks, Inc.
7  * All rights reserved.
8  *
9  * Portions of this software were developed by Robert N. M. Watson under
10  * contract to Juniper Networks, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)in_pcb.h	8.1 (Berkeley) 6/10/93
37  * $FreeBSD$
38  */
39 
40 #ifndef _NETINET_IN_PCB_H_
41 #define _NETINET_IN_PCB_H_
42 
43 #include <sys/queue.h>
44 #include <sys/epoch.h>
45 #include <sys/_lock.h>
46 #include <sys/_mutex.h>
47 #include <sys/_rwlock.h>
48 #include <net/route.h>
49 
50 #ifdef _KERNEL
51 #include <sys/lock.h>
52 #include <sys/rwlock.h>
53 #include <net/vnet.h>
54 #include <vm/uma.h>
55 #endif
56 #include <sys/ck.h>
57 
58 /*
59  * struct inpcb is the common protocol control block structure used in most
60  * IP transport protocols.
61  *
62  * Pointers to local and foreign host table entries, local and foreign socket
63  * numbers, and pointers up (to a socket structure) and down (to a
64  * protocol-specific control block) are stored here.
65  */
66 CK_LIST_HEAD(inpcbhead, inpcb);
67 CK_LIST_HEAD(inpcbporthead, inpcbport);
68 CK_LIST_HEAD(inpcblbgrouphead, inpcblbgroup);
69 typedef	uint64_t	inp_gen_t;
70 
71 /*
72  * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet.
73  * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing
74  * the following structure.
75  */
76 struct in_addr_4in6 {
77 	u_int32_t	ia46_pad32[3];
78 	struct	in_addr	ia46_addr4;
79 };
80 
81 union in_dependaddr {
82 	struct in_addr_4in6 id46_addr;
83 	struct in6_addr	id6_addr;
84 };
85 
86 /*
87  * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553.  in_conninfo has
88  * some extra padding to accomplish this.
89  * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport,
90  * lport, faddr to generate hash, so these fields shouldn't be moved.
91  */
92 struct in_endpoints {
93 	u_int16_t	ie_fport;		/* foreign port */
94 	u_int16_t	ie_lport;		/* local port */
95 	/* protocol dependent part, local and foreign addr */
96 	union in_dependaddr ie_dependfaddr;	/* foreign host table entry */
97 	union in_dependaddr ie_dependladdr;	/* local host table entry */
98 #define	ie_faddr	ie_dependfaddr.id46_addr.ia46_addr4
99 #define	ie_laddr	ie_dependladdr.id46_addr.ia46_addr4
100 #define	ie6_faddr	ie_dependfaddr.id6_addr
101 #define	ie6_laddr	ie_dependladdr.id6_addr
102 	u_int32_t	ie6_zoneid;		/* scope zone id */
103 };
104 
105 /*
106  * XXX The defines for inc_* are hacks and should be changed to direct
107  * references.
108  */
109 struct in_conninfo {
110 	u_int8_t	inc_flags;
111 	u_int8_t	inc_len;
112 	u_int16_t	inc_fibnum;	/* XXX was pad, 16 bits is plenty */
113 	/* protocol dependent part */
114 	struct	in_endpoints inc_ie;
115 };
116 
117 /*
118  * Flags for inc_flags.
119  */
120 #define	INC_ISIPV6	0x01
121 #define	INC_IPV6MINMTU	0x02
122 
123 #define	inc_fport	inc_ie.ie_fport
124 #define	inc_lport	inc_ie.ie_lport
125 #define	inc_faddr	inc_ie.ie_faddr
126 #define	inc_laddr	inc_ie.ie_laddr
127 #define	inc6_faddr	inc_ie.ie6_faddr
128 #define	inc6_laddr	inc_ie.ie6_laddr
129 #define	inc6_zoneid	inc_ie.ie6_zoneid
130 
131 #if defined(_KERNEL) || defined(_WANT_INPCB)
132 /*
133  * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and
134  * IPv6 sockets.  In the case of TCP and UDP, further per-connection state is
135  * hung off of inp_ppcb most of the time.  Almost all fields of struct inpcb
136  * are static after creation or protected by a per-inpcb rwlock, inp_lock.  A
137  * few fields are protected by multiple locks as indicated in the locking notes
138  * below.  For these fields, all of the listed locks must be write-locked for
139  * any modifications.  However, these fields can be safely read while any one of
140  * the listed locks are read-locked.  This model can permit greater concurrency
141  * for read operations.  For example, connections can be looked up while only
142  * holding a read lock on the global pcblist lock.  This is important for
143  * performance when attempting to find the connection for a packet given its IP
144  * and port tuple.
145  *
146  * One noteworthy exception is that the global pcbinfo lock follows a different
147  * set of rules in relation to the inp_list field.  Rather than being
148  * write-locked for modifications and read-locked for list iterations, it must
149  * be read-locked during modifications and write-locked during list iterations.
150  * This ensures that the relatively rare global list iterations safely walk a
151  * stable snapshot of connections while allowing more common list modifications
152  * to safely grab the pcblist lock just while adding or removing a connection
153  * from the global list.
154  *
155  * Key:
156  * (b) - Protected by the hpts lock.
157  * (c) - Constant after initialization
158  * (e) - Protected by the net_epoch_prempt epoch
159  * (g) - Protected by the pcbgroup lock
160  * (i) - Protected by the inpcb lock
161  * (p) - Protected by the pcbinfo lock for the inpcb
162  * (l) - Protected by the pcblist lock for the inpcb
163  * (h) - Protected by the pcbhash lock for the inpcb
164  * (s) - Protected by another subsystem's locks
165  * (x) - Undefined locking
166  *
167  * Notes on the tcp_hpts:
168  *
169  * First Hpts lock order is
170  * 1) INP_WLOCK()
171  * 2) HPTS_LOCK() i.e. hpts->pmtx
172  *
173  * To insert a TCB on the hpts you *must* be holding the INP_WLOCK().
174  * You may check the inp->inp_in_hpts flag without the hpts lock.
175  * The hpts is the only one that will clear this flag holding
176  * only the hpts lock. This means that in your tcp_output()
177  * routine when you test for the inp_in_hpts flag to be 1
178  * it may be transitioning to 0 (by the hpts).
179  * That's ok since that will just mean an extra call to tcp_output
180  * that most likely will find the call you executed
181  * (when the mis-match occured) will have put the TCB back
182  * on the hpts and it will return. If your
183  * call did not add the inp back to the hpts then you will either
184  * over-send or the cwnd will block you from sending more.
185  *
186  * Note you should also be holding the INP_WLOCK() when you
187  * call the remove from the hpts as well. Though usually
188  * you are either doing this from a timer, where you need and have
189  * the INP_WLOCK() or from destroying your TCB where again
190  * you should already have the INP_WLOCK().
191  *
192  * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and
193  * inp_input_cpu_set fields are controlled completely by
194  * the hpts. Do not ever set these. The inp_hpts_cpu_set
195  * and inp_input_cpu_set fields indicate if the hpts has
196  * setup the respective cpu field. It is advised if this
197  * field is 0, to enqueue the packet with the appropriate
198  * hpts_immediate() call. If the _set field is 1, then
199  * you may compare the inp_*_cpu field to the curcpu and
200  * may want to again insert onto the hpts if these fields
201  * are not equal (i.e. you are not on the expected CPU).
202  *
203  * A note on inp_hpts_calls and inp_input_calls, these
204  * flags are set when the hpts calls either the output
205  * or do_segment routines respectively. If the routine
206  * being called wants to use this, then it needs to
207  * clear the flag before returning. The hpts will not
208  * clear the flag. The flags can be used to tell if
209  * the hpts is the function calling the respective
210  * routine.
211  *
212  * A few other notes:
213  *
214  * When a read lock is held, stability of the field is guaranteed; to write
215  * to a field, a write lock must generally be held.
216  *
217  * netinet/netinet6-layer code should not assume that the inp_socket pointer
218  * is safe to dereference without inp_lock being held, even for protocols
219  * other than TCP (where the inpcb persists during TIMEWAIT even after the
220  * socket has been freed), or there may be close(2)-related races.
221  *
222  * The inp_vflag field is overloaded, and would otherwise ideally be (c).
223  *
224  * TODO:  Currently only the TCP stack is leveraging the global pcbinfo lock
225  * read-lock usage during modification, this model can be applied to other
226  * protocols (especially SCTP).
227  */
228 struct icmp6_filter;
229 struct inpcbpolicy;
230 struct m_snd_tag;
231 struct inpcb {
232 	/* Cache line #1 (amd64) */
233 	CK_LIST_ENTRY(inpcb) inp_hash;	/* [w](h/i) [r](e/i)  hash list */
234 	CK_LIST_ENTRY(inpcb) inp_pcbgrouphash;	/* (g/i) hash list */
235 	struct rwlock	inp_lock;
236 	/* Cache line #2 (amd64) */
237 #define	inp_start_zero	inp_hpts
238 #define	inp_zero_size	(sizeof(struct inpcb) - \
239 			    offsetof(struct inpcb, inp_start_zero))
240 	TAILQ_ENTRY(inpcb) inp_hpts;	/* pacing out queue next lock(b) */
241 
242 	uint32_t inp_hpts_request;	/* Current hpts request, zero if
243 					 * fits in the pacing window (i&b). */
244 	/*
245 	 * Note the next fields are protected by a
246 	 * different lock (hpts-lock). This means that
247 	 * they must correspond in size to the smallest
248 	 * protectable bit field (uint8_t on x86, and
249 	 * other platfomrs potentially uint32_t?). Also
250 	 * since CPU switches can occur at different times the two
251 	 * fields can *not* be collapsed into a signal bit field.
252 	 */
253 #if defined(__amd64__) || defined(__i386__)
254 	volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */
255 	volatile uint8_t inp_in_input; /* on input hpts (lock b) */
256 #else
257 	volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */
258 	volatile uint32_t inp_in_input; /* on input hpts (lock b) */
259 #endif
260 	volatile uint16_t  inp_hpts_cpu; /* Lock (i) */
261 	volatile uint16_t  inp_irq_cpu;	/* Set by LRO in behalf of or the driver */
262 	u_int	inp_refcount;		/* (i) refcount */
263 	int	inp_flags;		/* (i) generic IP/datagram flags */
264 	int	inp_flags2;		/* (i) generic IP/datagram flags #2*/
265 	volatile uint16_t  inp_input_cpu; /* Lock (i) */
266 	volatile uint8_t inp_hpts_cpu_set :1,  /* on output hpts (i) */
267 			 inp_input_cpu_set : 1,	/* on input hpts (i) */
268 			 inp_hpts_calls :1,	/* (i) from output hpts */
269 			 inp_input_calls :1,	/* (i) from input hpts */
270 			 inp_irq_cpu_set :1,	/* (i) from LRO/Driver */
271 			 inp_spare_bits2 : 3;
272 	uint8_t inp_numa_domain;	/* numa domain */
273 	void	*inp_ppcb;		/* (i) pointer to per-protocol pcb */
274 	struct	socket *inp_socket;	/* (i) back pointer to socket */
275 	uint32_t 	 inp_hptsslot;	/* Hpts wheel slot this tcb is Lock(i&b) */
276 	uint32_t         inp_hpts_drop_reas;	/* reason we are dropping the PCB (lock i&b) */
277 	TAILQ_ENTRY(inpcb) inp_input;	/* pacing in  queue next lock(b) */
278 	struct	inpcbinfo *inp_pcbinfo;	/* (c) PCB list info */
279 	struct	inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */
280 	CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */
281 	struct	ucred	*inp_cred;	/* (c) cache of socket cred */
282 	u_int32_t inp_flow;		/* (i) IPv6 flow information */
283 	u_char	inp_vflag;		/* (i) IP version flag (v4/v6) */
284 	u_char	inp_ip_ttl;		/* (i) time to live proto */
285 	u_char	inp_ip_p;		/* (c) protocol proto */
286 	u_char	inp_ip_minttl;		/* (i) minimum TTL or drop */
287 	uint32_t inp_flowid;		/* (x) flow id / queue id */
288 	struct m_snd_tag *inp_snd_tag;	/* (i) send tag for outgoing mbufs */
289 	uint32_t inp_flowtype;		/* (x) M_HASHTYPE value */
290 	uint32_t inp_rss_listen_bucket;	/* (x) overridden RSS listen bucket */
291 
292 	/* Local and foreign ports, local and foreign addr. */
293 	struct	in_conninfo inp_inc;	/* (i) list for PCB's local port */
294 
295 	/* MAC and IPSEC policy information. */
296 	struct	label *inp_label;	/* (i) MAC label */
297 	struct	inpcbpolicy *inp_sp;    /* (s) for IPSEC */
298 
299 	/* Protocol-dependent part; options. */
300 	struct {
301 		u_char	inp_ip_tos;		/* (i) type of service proto */
302 		struct mbuf		*inp_options;	/* (i) IP options */
303 		struct ip_moptions	*inp_moptions;	/* (i) mcast options */
304 	};
305 	struct {
306 		/* (i) IP options */
307 		struct mbuf		*in6p_options;
308 		/* (i) IP6 options for outgoing packets */
309 		struct ip6_pktopts	*in6p_outputopts;
310 		/* (i) IP multicast options */
311 		struct ip6_moptions	*in6p_moptions;
312 		/* (i) ICMPv6 code type filter */
313 		struct icmp6_filter	*in6p_icmp6filt;
314 		/* (i) IPV6_CHECKSUM setsockopt */
315 		int	in6p_cksum;
316 		short	in6p_hops;
317 	};
318 	CK_LIST_ENTRY(inpcb) inp_portlist;	/* (i/h) */
319 	struct	inpcbport *inp_phd;	/* (i/h) head of this list */
320 	inp_gen_t	inp_gencnt;	/* (c) generation count */
321 	void		*spare_ptr;	/* Spare pointer. */
322 	rt_gen_t	inp_rt_cookie;	/* generation for route entry */
323 	union {				/* cached L3 information */
324 		struct route inp_route;
325 		struct route_in6 inp_route6;
326 	};
327 	CK_LIST_ENTRY(inpcb) inp_list;	/* (p/l) list for all PCBs for proto */
328 	                                /* (e[r]) for list iteration */
329 	                                /* (p[w]/l) for addition/removal */
330 	struct epoch_context inp_epoch_ctx;
331 };
332 #endif	/* _KERNEL */
333 
334 #define	inp_fport	inp_inc.inc_fport
335 #define	inp_lport	inp_inc.inc_lport
336 #define	inp_faddr	inp_inc.inc_faddr
337 #define	inp_laddr	inp_inc.inc_laddr
338 
339 #define	in6p_faddr	inp_inc.inc6_faddr
340 #define	in6p_laddr	inp_inc.inc6_laddr
341 #define	in6p_zoneid	inp_inc.inc6_zoneid
342 
343 #define	inp_vnet	inp_pcbinfo->ipi_vnet
344 
345 /*
346  * The range of the generation count, as used in this implementation, is 9e19.
347  * We would have to create 300 billion connections per second for this number
348  * to roll over in a year.  This seems sufficiently unlikely that we simply
349  * don't concern ourselves with that possibility.
350  */
351 
352 /*
353  * Interface exported to userland by various protocols which use inpcbs.  Hack
354  * alert -- only define if struct xsocket is in scope.
355  * Fields prefixed with "xi_" are unique to this structure, and the rest
356  * match fields in the struct inpcb, to ease coding and porting.
357  *
358  * Legend:
359  * (s) - used by userland utilities in src
360  * (p) - used by utilities in ports
361  * (3) - is known to be used by third party software not in ports
362  * (n) - no known usage
363  */
364 #ifdef _SYS_SOCKETVAR_H_
365 struct xinpcb {
366 	ksize_t		xi_len;			/* length of this structure */
367 	struct xsocket	xi_socket;		/* (s,p) */
368 	struct in_conninfo inp_inc;		/* (s,p) */
369 	uint64_t	inp_gencnt;		/* (s,p) */
370 	kvaddr_t	inp_ppcb;		/* (s) netstat(1) */
371 	int64_t		inp_spare64[4];
372 	uint32_t	inp_flow;		/* (s) */
373 	uint32_t	inp_flowid;		/* (s) */
374 	uint32_t	inp_flowtype;		/* (s) */
375 	int32_t		inp_flags;		/* (s,p) */
376 	int32_t		inp_flags2;		/* (s) */
377 	int32_t		inp_rss_listen_bucket;	/* (n) */
378 	int32_t		in6p_cksum;		/* (n) */
379 	int32_t		inp_spare32[4];
380 	uint16_t	in6p_hops;		/* (n) */
381 	uint8_t		inp_ip_tos;		/* (n) */
382 	int8_t		pad8;
383 	uint8_t		inp_vflag;		/* (s,p) */
384 	uint8_t		inp_ip_ttl;		/* (n) */
385 	uint8_t		inp_ip_p;		/* (n) */
386 	uint8_t		inp_ip_minttl;		/* (n) */
387 	int8_t		inp_spare8[4];
388 } __aligned(8);
389 
390 struct xinpgen {
391 	ksize_t	xig_len;	/* length of this structure */
392 	u_int		xig_count;	/* number of PCBs at this time */
393 	uint32_t	_xig_spare32;
394 	inp_gen_t	xig_gen;	/* generation count at this time */
395 	so_gen_t	xig_sogen;	/* socket generation count this time */
396 	uint64_t	_xig_spare64[4];
397 } __aligned(8);
398 #ifdef	_KERNEL
399 void	in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *);
400 #endif
401 #endif /* _SYS_SOCKETVAR_H_ */
402 
403 struct inpcbport {
404 	struct epoch_context phd_epoch_ctx;
405 	CK_LIST_ENTRY(inpcbport) phd_hash;
406 	struct inpcbhead phd_pcblist;
407 	u_short phd_port;
408 };
409 
410 /*-
411  * Global data structure for each high-level protocol (UDP, TCP, ...) in both
412  * IPv4 and IPv6.  Holds inpcb lists and information for managing them.
413  *
414  * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and
415  * ipi_list_lock:
416  *  - ipi_lock covering the global pcb list stability during loop iteration,
417  *  - ipi_hash_lock covering the hashed lookup tables,
418  *  - ipi_list_lock covering mutable global fields (such as the global
419  *    pcb list)
420  *
421  * The lock order is:
422  *
423  *    ipi_lock (before)
424  *        inpcb locks (before)
425  *            ipi_list locks (before)
426  *                {ipi_hash_lock, pcbgroup locks}
427  *
428  * Locking key:
429  *
430  * (c) Constant or nearly constant after initialisation
431  * (e) - Protected by the net_epoch_prempt epoch
432  * (g) Locked by ipi_lock
433  * (l) Locked by ipi_list_lock
434  * (h) Read using either net_epoch_preempt or inpcb lock; write requires both ipi_hash_lock and inpcb lock
435  * (p) Protected by one or more pcbgroup locks
436  * (x) Synchronisation properties poorly defined
437  */
438 struct inpcbinfo {
439 	/*
440 	 * Global lock protecting inpcb list modification
441 	 */
442 	struct mtx		 ipi_lock;
443 
444 	/*
445 	 * Global list of inpcbs on the protocol.
446 	 */
447 	struct inpcbhead	*ipi_listhead;		/* [r](e) [w](g/l) */
448 	u_int			 ipi_count;		/* (l) */
449 
450 	/*
451 	 * Generation count -- incremented each time a connection is allocated
452 	 * or freed.
453 	 */
454 	u_quad_t		 ipi_gencnt;		/* (l) */
455 
456 	/*
457 	 * Fields associated with port lookup and allocation.
458 	 */
459 	u_short			 ipi_lastport;		/* (x) */
460 	u_short			 ipi_lastlow;		/* (x) */
461 	u_short			 ipi_lasthi;		/* (x) */
462 
463 	/*
464 	 * UMA zone from which inpcbs are allocated for this protocol.
465 	 */
466 	struct	uma_zone	*ipi_zone;		/* (c) */
467 
468 	/*
469 	 * Connection groups associated with this protocol.  These fields are
470 	 * constant, but pcbgroup structures themselves are protected by
471 	 * per-pcbgroup locks.
472 	 */
473 	struct inpcbgroup	*ipi_pcbgroups;		/* (c) */
474 	u_int			 ipi_npcbgroups;	/* (c) */
475 	u_int			 ipi_hashfields;	/* (c) */
476 
477 	/*
478 	 * Global lock protecting modification non-pcbgroup hash lookup tables.
479 	 */
480 	struct mtx		 ipi_hash_lock;
481 
482 	/*
483 	 * Global hash of inpcbs, hashed by local and foreign addresses and
484 	 * port numbers.
485 	 */
486 	struct inpcbhead	*ipi_hashbase;		/* (h) */
487 	u_long			 ipi_hashmask;		/* (h) */
488 
489 	/*
490 	 * Global hash of inpcbs, hashed by only local port number.
491 	 */
492 	struct inpcbporthead	*ipi_porthashbase;	/* (h) */
493 	u_long			 ipi_porthashmask;	/* (h) */
494 
495 	/*
496 	 * List of wildcard inpcbs for use with pcbgroups.  In the past, was
497 	 * per-pcbgroup but is now global.  All pcbgroup locks must be held
498 	 * to modify the list, so any is sufficient to read it.
499 	 */
500 	struct inpcbhead	*ipi_wildbase;		/* (p) */
501 	u_long			 ipi_wildmask;		/* (p) */
502 
503 	/*
504 	 * Load balance groups used for the SO_REUSEPORT_LB option,
505 	 * hashed by local port.
506 	 */
507 	struct	inpcblbgrouphead *ipi_lbgrouphashbase;	/* (h) */
508 	u_long			 ipi_lbgrouphashmask;	/* (h) */
509 
510 	/*
511 	 * Pointer to network stack instance
512 	 */
513 	struct vnet		*ipi_vnet;		/* (c) */
514 
515 	/*
516 	 * general use 2
517 	 */
518 	void 			*ipi_pspare[2];
519 
520 	/*
521 	 * Global lock protecting global inpcb list, inpcb count, etc.
522 	 */
523 	struct rwlock		 ipi_list_lock;
524 };
525 
526 #ifdef _KERNEL
527 /*
528  * Connection groups hold sets of connections that have similar CPU/thread
529  * affinity.  Each connection belongs to exactly one connection group.
530  */
531 struct inpcbgroup {
532 	/*
533 	 * Per-connection group hash of inpcbs, hashed by local and foreign
534 	 * addresses and port numbers.
535 	 */
536 	struct inpcbhead	*ipg_hashbase;		/* (c) */
537 	u_long			 ipg_hashmask;		/* (c) */
538 
539 	/*
540 	 * Notional affinity of this pcbgroup.
541 	 */
542 	u_int			 ipg_cpu;		/* (p) */
543 
544 	/*
545 	 * Per-connection group lock, not to be confused with ipi_lock.
546 	 * Protects the hash table hung off the group, but also the global
547 	 * wildcard list in inpcbinfo.
548 	 */
549 	struct mtx		 ipg_lock;
550 } __aligned(CACHE_LINE_SIZE);
551 
552 /*
553  * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group
554  * (or unique address:port combination) can be re-used at most
555  * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which
556  * is dynamically resized as processes bind/unbind to that specific group.
557  */
558 struct inpcblbgroup {
559 	CK_LIST_ENTRY(inpcblbgroup) il_list;
560 	struct epoch_context il_epoch_ctx;
561 	uint16_t	il_lport;			/* (c) */
562 	u_char		il_vflag;			/* (c) */
563 	u_int8_t		il_numa_domain;
564 	uint32_t	il_pad2;
565 	union in_dependaddr il_dependladdr;		/* (c) */
566 #define	il_laddr	il_dependladdr.id46_addr.ia46_addr4
567 #define	il6_laddr	il_dependladdr.id6_addr
568 	uint32_t	il_inpsiz; /* max count in il_inp[] (h) */
569 	uint32_t	il_inpcnt; /* cur count in il_inp[] (h) */
570 	struct inpcb	*il_inp[];			/* (h) */
571 };
572 
573 #define INP_LOCK_INIT(inp, d, t) \
574 	rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE |  RW_DUPOK)
575 #define INP_LOCK_DESTROY(inp)	rw_destroy(&(inp)->inp_lock)
576 #define INP_RLOCK(inp)		rw_rlock(&(inp)->inp_lock)
577 #define INP_WLOCK(inp)		rw_wlock(&(inp)->inp_lock)
578 #define INP_TRY_RLOCK(inp)	rw_try_rlock(&(inp)->inp_lock)
579 #define INP_TRY_WLOCK(inp)	rw_try_wlock(&(inp)->inp_lock)
580 #define INP_RUNLOCK(inp)	rw_runlock(&(inp)->inp_lock)
581 #define INP_WUNLOCK(inp)	rw_wunlock(&(inp)->inp_lock)
582 #define INP_UNLOCK(inp)		rw_unlock(&(inp)->inp_lock)
583 #define	INP_TRY_UPGRADE(inp)	rw_try_upgrade(&(inp)->inp_lock)
584 #define	INP_DOWNGRADE(inp)	rw_downgrade(&(inp)->inp_lock)
585 #define	INP_WLOCKED(inp)	rw_wowned(&(inp)->inp_lock)
586 #define	INP_LOCK_ASSERT(inp)	rw_assert(&(inp)->inp_lock, RA_LOCKED)
587 #define	INP_RLOCK_ASSERT(inp)	rw_assert(&(inp)->inp_lock, RA_RLOCKED)
588 #define	INP_WLOCK_ASSERT(inp)	rw_assert(&(inp)->inp_lock, RA_WLOCKED)
589 #define	INP_UNLOCK_ASSERT(inp)	rw_assert(&(inp)->inp_lock, RA_UNLOCKED)
590 
591 /*
592  * These locking functions are for inpcb consumers outside of sys/netinet,
593  * more specifically, they were added for the benefit of TOE drivers. The
594  * macros are reserved for use by the stack.
595  */
596 void inp_wlock(struct inpcb *);
597 void inp_wunlock(struct inpcb *);
598 void inp_rlock(struct inpcb *);
599 void inp_runlock(struct inpcb *);
600 
601 #ifdef INVARIANT_SUPPORT
602 void inp_lock_assert(struct inpcb *);
603 void inp_unlock_assert(struct inpcb *);
604 #else
605 #define	inp_lock_assert(inp)	do {} while (0)
606 #define	inp_unlock_assert(inp)	do {} while (0)
607 #endif
608 
609 void	inp_apply_all(void (*func)(struct inpcb *, void *), void *arg);
610 int 	inp_ip_tos_get(const struct inpcb *inp);
611 void 	inp_ip_tos_set(struct inpcb *inp, int val);
612 struct socket *
613 	inp_inpcbtosocket(struct inpcb *inp);
614 struct tcpcb *
615 	inp_inpcbtotcpcb(struct inpcb *inp);
616 void 	inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
617 		uint32_t *faddr, uint16_t *fp);
618 int	inp_so_options(const struct inpcb *inp);
619 
620 #endif /* _KERNEL */
621 
622 #define INP_INFO_LOCK_INIT(ipi, d) \
623 	mtx_init(&(ipi)->ipi_lock, (d), NULL, MTX_DEF| MTX_RECURSE)
624 #define INP_INFO_LOCK_DESTROY(ipi)  mtx_destroy(&(ipi)->ipi_lock)
625 #define INP_INFO_WLOCK(ipi) mtx_lock(&(ipi)->ipi_lock)
626 #define INP_INFO_TRY_WLOCK(ipi)	mtx_trylock(&(ipi)->ipi_lock)
627 #define INP_INFO_WLOCKED(ipi)	mtx_owned(&(ipi)->ipi_lock)
628 #define INP_INFO_WUNLOCK(ipi)	mtx_unlock(&(ipi)->ipi_lock)
629 #define	INP_INFO_LOCK_ASSERT(ipi)	MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_lock))
630 #define INP_INFO_WLOCK_ASSERT(ipi)	mtx_assert(&(ipi)->ipi_lock, MA_OWNED)
631 #define INP_INFO_WUNLOCK_ASSERT(ipi)	\
632 	mtx_assert(&(ipi)->ipi_lock, MA_NOTOWNED)
633 
634 #define INP_LIST_LOCK_INIT(ipi, d) \
635         rw_init_flags(&(ipi)->ipi_list_lock, (d), 0)
636 #define INP_LIST_LOCK_DESTROY(ipi)  rw_destroy(&(ipi)->ipi_list_lock)
637 #define INP_LIST_RLOCK(ipi)     rw_rlock(&(ipi)->ipi_list_lock)
638 #define INP_LIST_WLOCK(ipi)     rw_wlock(&(ipi)->ipi_list_lock)
639 #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock)
640 #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock)
641 #define INP_LIST_TRY_UPGRADE(ipi)       rw_try_upgrade(&(ipi)->ipi_list_lock)
642 #define INP_LIST_RUNLOCK(ipi)   rw_runlock(&(ipi)->ipi_list_lock)
643 #define INP_LIST_WUNLOCK(ipi)   rw_wunlock(&(ipi)->ipi_list_lock)
644 #define INP_LIST_LOCK_ASSERT(ipi) \
645 	rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED)
646 #define INP_LIST_RLOCK_ASSERT(ipi) \
647 	rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED)
648 #define INP_LIST_WLOCK_ASSERT(ipi) \
649 	rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED)
650 #define INP_LIST_UNLOCK_ASSERT(ipi) \
651 	rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED)
652 
653 #define	INP_HASH_LOCK_INIT(ipi, d) mtx_init(&(ipi)->ipi_hash_lock, (d), NULL, MTX_DEF)
654 #define	INP_HASH_LOCK_DESTROY(ipi)	mtx_destroy(&(ipi)->ipi_hash_lock)
655 #define	INP_HASH_WLOCK(ipi)		mtx_lock(&(ipi)->ipi_hash_lock)
656 #define	INP_HASH_WUNLOCK(ipi)		mtx_unlock(&(ipi)->ipi_hash_lock)
657 #define	INP_HASH_LOCK_ASSERT(ipi)	MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_hash_lock))
658 #define	INP_HASH_WLOCK_ASSERT(ipi)	mtx_assert(&(ipi)->ipi_hash_lock, MA_OWNED);
659 
660 #define	INP_GROUP_LOCK_INIT(ipg, d)	mtx_init(&(ipg)->ipg_lock, (d), NULL, \
661 					    MTX_DEF | MTX_DUPOK)
662 #define	INP_GROUP_LOCK_DESTROY(ipg)	mtx_destroy(&(ipg)->ipg_lock)
663 
664 #define	INP_GROUP_LOCK(ipg)		mtx_lock(&(ipg)->ipg_lock)
665 #define	INP_GROUP_LOCK_ASSERT(ipg)	mtx_assert(&(ipg)->ipg_lock, MA_OWNED)
666 #define	INP_GROUP_UNLOCK(ipg)		mtx_unlock(&(ipg)->ipg_lock)
667 
668 #define INP_PCBHASH(faddr, lport, fport, mask) \
669 	(((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask))
670 #define INP_PCBPORTHASH(lport, mask) \
671 	(ntohs((lport)) & (mask))
672 #define	INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \
673 	((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport)))
674 #define	INP6_PCBHASHKEY(faddr)	((faddr)->s6_addr32[3])
675 
676 /*
677  * Flags for inp_vflags -- historically version flags only
678  */
679 #define	INP_IPV4	0x1
680 #define	INP_IPV6	0x2
681 #define	INP_IPV6PROTO	0x4		/* opened under IPv6 protocol */
682 
683 /*
684  * Flags for inp_flags.
685  */
686 #define	INP_RECVOPTS		0x00000001 /* receive incoming IP options */
687 #define	INP_RECVRETOPTS		0x00000002 /* receive IP options for reply */
688 #define	INP_RECVDSTADDR		0x00000004 /* receive IP dst address */
689 #define	INP_HDRINCL		0x00000008 /* user supplies entire IP header */
690 #define	INP_HIGHPORT		0x00000010 /* user wants "high" port binding */
691 #define	INP_LOWPORT		0x00000020 /* user wants "low" port binding */
692 #define	INP_ANONPORT		0x00000040 /* port chosen for user */
693 #define	INP_RECVIF		0x00000080 /* receive incoming interface */
694 #define	INP_MTUDISC		0x00000100 /* user can do MTU discovery */
695 				   	   /* 0x000200 unused: was INP_FAITH */
696 #define	INP_RECVTTL		0x00000400 /* receive incoming IP TTL */
697 #define	INP_DONTFRAG		0x00000800 /* don't fragment packet */
698 #define	INP_BINDANY		0x00001000 /* allow bind to any address */
699 #define	INP_INHASHLIST		0x00002000 /* in_pcbinshash() has been called */
700 #define	INP_RECVTOS		0x00004000 /* receive incoming IP TOS */
701 #define	IN6P_IPV6_V6ONLY	0x00008000 /* restrict AF_INET6 socket for v6 */
702 #define	IN6P_PKTINFO		0x00010000 /* receive IP6 dst and I/F */
703 #define	IN6P_HOPLIMIT		0x00020000 /* receive hoplimit */
704 #define	IN6P_HOPOPTS		0x00040000 /* receive hop-by-hop options */
705 #define	IN6P_DSTOPTS		0x00080000 /* receive dst options after rthdr */
706 #define	IN6P_RTHDR		0x00100000 /* receive routing header */
707 #define	IN6P_RTHDRDSTOPTS	0x00200000 /* receive dstoptions before rthdr */
708 #define	IN6P_TCLASS		0x00400000 /* receive traffic class value */
709 #define	IN6P_AUTOFLOWLABEL	0x00800000 /* attach flowlabel automatically */
710 #define	INP_TIMEWAIT		0x01000000 /* in TIMEWAIT, ppcb is tcptw */
711 #define	INP_ONESBCAST		0x02000000 /* send all-ones broadcast */
712 #define	INP_DROPPED		0x04000000 /* protocol drop flag */
713 #define	INP_SOCKREF		0x08000000 /* strong socket reference */
714 #define	INP_RESERVED_0          0x10000000 /* reserved field */
715 #define	INP_RESERVED_1          0x20000000 /* reserved field */
716 #define	IN6P_RFC2292		0x40000000 /* used RFC2292 API on the socket */
717 #define	IN6P_MTU		0x80000000 /* receive path MTU */
718 
719 #define	INP_CONTROLOPTS		(INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\
720 				 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\
721 				 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\
722 				 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\
723 				 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\
724 				 IN6P_MTU)
725 
726 /*
727  * Flags for inp_flags2.
728  */
729 #define	INP_MBUF_L_ACKS		0x00000001 /* We need large mbufs for ack compression */
730 #define	INP_MBUF_ACKCMP		0x00000002 /* TCP mbuf ack compression ok */
731 #define	INP_PCBGROUPWILD	0x00000004 /* in pcbgroup wildcard list */
732 #define	INP_REUSEPORT		0x00000008 /* SO_REUSEPORT option is set */
733 #define	INP_FREED		0x00000010 /* inp itself is not valid */
734 #define	INP_REUSEADDR		0x00000020 /* SO_REUSEADDR option is set */
735 #define	INP_BINDMULTI		0x00000040 /* IP_BINDMULTI option is set */
736 #define	INP_RSS_BUCKET_SET	0x00000080 /* IP_RSS_LISTEN_BUCKET is set */
737 #define	INP_RECVFLOWID		0x00000100 /* populate recv datagram with flow info */
738 #define	INP_RECVRSSBUCKETID	0x00000200 /* populate recv datagram with bucket id */
739 #define	INP_RATE_LIMIT_CHANGED	0x00000400 /* rate limit needs attention */
740 #define	INP_ORIGDSTADDR		0x00000800 /* receive IP dst address/port */
741 #define INP_CANNOT_DO_ECN	0x00001000 /* The stack does not do ECN */
742 #define	INP_REUSEPORT_LB	0x00002000 /* SO_REUSEPORT_LB option is set */
743 #define INP_SUPPORTS_MBUFQ	0x00004000 /* Supports the mbuf queue method of LRO */
744 #define INP_MBUF_QUEUE_READY	0x00008000 /* The transport is pacing, inputs can be queued */
745 #define INP_DONT_SACK_QUEUE	0x00010000 /* If a sack arrives do not wake me */
746 #define INP_2PCP_SET		0x00020000 /* If the Eth PCP should be set explicitly */
747 #define INP_2PCP_BIT0		0x00040000 /* Eth PCP Bit 0 */
748 #define INP_2PCP_BIT1		0x00080000 /* Eth PCP Bit 1 */
749 #define INP_2PCP_BIT2		0x00100000 /* Eth PCP Bit 2 */
750 #define INP_2PCP_BASE	INP_2PCP_BIT0
751 #define INP_2PCP_MASK	(INP_2PCP_BIT0 | INP_2PCP_BIT1 | INP_2PCP_BIT2)
752 #define INP_2PCP_SHIFT		18         /* shift PCP field in/out of inp_flags2 */
753 /*
754  * Flags passed to in_pcblookup*() functions.
755  */
756 #define	INPLOOKUP_WILDCARD	0x00000001	/* Allow wildcard sockets. */
757 #define	INPLOOKUP_RLOCKPCB	0x00000002	/* Return inpcb read-locked. */
758 #define	INPLOOKUP_WLOCKPCB	0x00000004	/* Return inpcb write-locked. */
759 
760 #define	INPLOOKUP_MASK	(INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \
761 	    INPLOOKUP_WLOCKPCB)
762 
763 #define	sotoinpcb(so)	((struct inpcb *)(so)->so_pcb)
764 
765 #define	INP_SOCKAF(so) so->so_proto->pr_domain->dom_family
766 
767 #define	INP_CHECK_SOCKAF(so, af)	(INP_SOCKAF(so) == af)
768 
769 /*
770  * Constants for pcbinfo.ipi_hashfields.
771  */
772 #define	IPI_HASHFIELDS_NONE	0
773 #define	IPI_HASHFIELDS_2TUPLE	1
774 #define	IPI_HASHFIELDS_4TUPLE	2
775 
776 #ifdef _KERNEL
777 VNET_DECLARE(int, ipport_reservedhigh);
778 VNET_DECLARE(int, ipport_reservedlow);
779 VNET_DECLARE(int, ipport_lowfirstauto);
780 VNET_DECLARE(int, ipport_lowlastauto);
781 VNET_DECLARE(int, ipport_firstauto);
782 VNET_DECLARE(int, ipport_lastauto);
783 VNET_DECLARE(int, ipport_hifirstauto);
784 VNET_DECLARE(int, ipport_hilastauto);
785 VNET_DECLARE(int, ipport_randomized);
786 VNET_DECLARE(int, ipport_randomcps);
787 VNET_DECLARE(int, ipport_randomtime);
788 VNET_DECLARE(int, ipport_stoprandom);
789 VNET_DECLARE(int, ipport_tcpallocs);
790 
791 #define	V_ipport_reservedhigh	VNET(ipport_reservedhigh)
792 #define	V_ipport_reservedlow	VNET(ipport_reservedlow)
793 #define	V_ipport_lowfirstauto	VNET(ipport_lowfirstauto)
794 #define	V_ipport_lowlastauto	VNET(ipport_lowlastauto)
795 #define	V_ipport_firstauto	VNET(ipport_firstauto)
796 #define	V_ipport_lastauto	VNET(ipport_lastauto)
797 #define	V_ipport_hifirstauto	VNET(ipport_hifirstauto)
798 #define	V_ipport_hilastauto	VNET(ipport_hilastauto)
799 #define	V_ipport_randomized	VNET(ipport_randomized)
800 #define	V_ipport_randomcps	VNET(ipport_randomcps)
801 #define	V_ipport_randomtime	VNET(ipport_randomtime)
802 #define	V_ipport_stoprandom	VNET(ipport_stoprandom)
803 #define	V_ipport_tcpallocs	VNET(ipport_tcpallocs)
804 
805 void	in_pcbinfo_destroy(struct inpcbinfo *);
806 void	in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *,
807 	    int, int, char *, uma_init, u_int);
808 
809 int	in_pcbbind_check_bindmulti(const struct inpcb *ni,
810 	    const struct inpcb *oi);
811 
812 struct inpcbgroup *
813 	in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t);
814 struct inpcbgroup *
815 	in_pcbgroup_byinpcb(struct inpcb *);
816 struct inpcbgroup *
817 	in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short,
818 	    struct in_addr, u_short);
819 void	in_pcbgroup_destroy(struct inpcbinfo *);
820 int	in_pcbgroup_enabled(struct inpcbinfo *);
821 void	in_pcbgroup_init(struct inpcbinfo *, u_int, int);
822 void	in_pcbgroup_remove(struct inpcb *);
823 void	in_pcbgroup_update(struct inpcb *);
824 void	in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *);
825 
826 void	in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *);
827 int	in_pcballoc(struct socket *, struct inpcbinfo *);
828 int	in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *);
829 int	in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
830 	    u_short *, struct ucred *);
831 int	in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *);
832 int	in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *,
833 	    struct mbuf *, bool);
834 int	in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
835 	    u_short *, in_addr_t *, u_short *, struct inpcb **,
836 	    struct ucred *);
837 void	in_pcbdetach(struct inpcb *);
838 void	in_pcbdisconnect(struct inpcb *);
839 void	in_pcbdrop(struct inpcb *);
840 void	in_pcbfree(struct inpcb *);
841 int	in_pcbinshash(struct inpcb *);
842 int	in_pcbinshash_mbuf(struct inpcb *, struct mbuf *);
843 int	in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *,
844 	    struct ucred *);
845 int	in_pcblbgroup_numa(struct inpcb *, int arg);
846 struct inpcb *
847 	in_pcblookup(struct inpcbinfo *, struct in_addr, u_int,
848 	    struct in_addr, u_int, int, struct ifnet *);
849 struct inpcb *
850 	in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int,
851 	    struct in_addr, u_int, int, struct ifnet *, struct mbuf *);
852 void	in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr,
853 	    int, struct inpcb *(*)(struct inpcb *, int));
854 void	in_pcbref(struct inpcb *);
855 void	in_pcbrehash(struct inpcb *);
856 void	in_pcbrehash_mbuf(struct inpcb *, struct mbuf *);
857 int	in_pcbrele_rlocked(struct inpcb *);
858 int	in_pcbrele_wlocked(struct inpcb *);
859 void	in_losing(struct inpcb *);
860 void	in_pcbsetsolabel(struct socket *so);
861 int	in_getpeeraddr(struct socket *so, struct sockaddr **nam);
862 int	in_getsockaddr(struct socket *so, struct sockaddr **nam);
863 struct sockaddr *
864 	in_sockaddr(in_port_t port, struct in_addr *addr);
865 void	in_pcbsosetlabel(struct socket *so);
866 #ifdef RATELIMIT
867 int
868 in_pcboutput_txrtlmt_locked(struct inpcb *, struct ifnet *,
869 	    struct mbuf *, uint32_t);
870 int	in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t,
871 	    uint32_t, struct m_snd_tag **);
872 void	in_pcbdetach_txrtlmt(struct inpcb *);
873 void    in_pcbdetach_tag(struct m_snd_tag *);
874 int	in_pcbmodify_txrtlmt(struct inpcb *, uint32_t);
875 int	in_pcbquery_txrtlmt(struct inpcb *, uint32_t *);
876 int	in_pcbquery_txrlevel(struct inpcb *, uint32_t *);
877 void	in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *);
878 void	in_pcboutput_eagain(struct inpcb *);
879 #endif
880 #endif /* _KERNEL */
881 
882 #endif /* !_NETINET_IN_PCB_H_ */
883