xref: /titanic_52/usr/src/uts/common/inet/tcp.h (revision 9c468ea9d266203f8dac0165f60fc9b92d8aead3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /* Copyright (c) 1990 Mentat Inc. */
26 
27 #ifndef	_INET_TCP_H
28 #define	_INET_TCP_H
29 
30 #ifdef	__cplusplus
31 extern "C" {
32 #endif
33 
34 #include <sys/inttypes.h>
35 #include <netinet/ip6.h>
36 #include <netinet/tcp.h>
37 #include <sys/socket.h>
38 #include <sys/socket_proto.h>
39 #include <sys/md5.h>
40 #include <inet/common.h>
41 #include <inet/ip.h>
42 #include <inet/ip6.h>
43 #include <inet/mi.h>
44 #include <inet/mib2.h>
45 #include <inet/tcp_stack.h>
46 #include <inet/tcp_sack.h>
47 #include <inet/kssl/ksslapi.h>
48 
49 /* TCP states */
50 #define	TCPS_CLOSED		-6
51 #define	TCPS_IDLE		-5	/* idle (opened, but not bound) */
52 #define	TCPS_BOUND		-4	/* bound, ready to connect or accept */
53 #define	TCPS_LISTEN		-3	/* listening for connection */
54 #define	TCPS_SYN_SENT		-2	/* active, have sent syn */
55 #define	TCPS_SYN_RCVD		-1	/* have received syn (and sent ours) */
56 /* states < TCPS_ESTABLISHED are those where connections not established */
57 #define	TCPS_ESTABLISHED	0	/* established */
58 #define	TCPS_CLOSE_WAIT		1	/* rcvd fin, waiting for close */
59 /* states > TCPS_CLOSE_WAIT are those where user has closed */
60 #define	TCPS_FIN_WAIT_1		2	/* have closed and sent fin */
61 #define	TCPS_CLOSING		3	/* closed, xchd FIN, await FIN ACK */
62 #define	TCPS_LAST_ACK		4	/* had fin and close; await FIN ACK */
63 /* states > TCPS_CLOSE_WAIT && < TCPS_FIN_WAIT_2 await ACK of FIN */
64 #define	TCPS_FIN_WAIT_2		5	/* have closed, fin is acked */
65 #define	TCPS_TIME_WAIT		6	/* in 2*msl quiet wait after close */
66 
67 /*
68  * Internal flags used in conjunction with the packet header flags.
69  * Used in tcp_input_data to keep track of what needs to be done.
70  */
71 #define	TH_LIMIT_XMIT		0x0400	/* Limited xmit is needed */
72 #define	TH_XMIT_NEEDED		0x0800	/* Window opened - send queued data */
73 #define	TH_REXMIT_NEEDED	0x1000	/* Time expired for unacked data */
74 #define	TH_ACK_NEEDED		0x2000	/* Send an ack now. */
75 #define	TH_NEED_SACK_REXMIT	0x4000	/* Use SACK info to retransmission */
76 #define	TH_ACK_TIMER_NEEDED	0x8000	/* Start the delayed ACK timer */
77 #define	TH_ORDREL_NEEDED	0x10000	/* Generate an ordrel indication */
78 #define	TH_MARKNEXT_NEEDED	0x20000	/* Data should have MSGMARKNEXT */
79 #define	TH_SEND_URP_MARK	0x40000	/* Send up tcp_urp_mark_mp */
80 
81 /*
82  * TCP sequence numbers are 32 bit integers operated
83  * on with modular arithmetic.  These macros can be
84  * used to compare such integers.
85  */
86 #define	SEQ_LT(a, b)	((int32_t)((a)-(b)) < 0)
87 #define	SEQ_LEQ(a, b)	((int32_t)((a)-(b)) <= 0)
88 #define	SEQ_GT(a, b)	((int32_t)((a)-(b)) > 0)
89 #define	SEQ_GEQ(a, b)	((int32_t)((a)-(b)) >= 0)
90 
91 /* TCP Protocol header */
92 typedef	struct tcphdr_s {
93 	uint8_t		th_lport[2];	/* Source port */
94 	uint8_t		th_fport[2];	/* Destination port */
95 	uint8_t		th_seq[4];	/* Sequence number */
96 	uint8_t		th_ack[4];	/* Acknowledgement number */
97 	uint8_t		th_offset_and_rsrvd[1]; /* Offset to the packet data */
98 	uint8_t		th_flags[1];
99 	uint8_t		th_win[2];	/* Allocation number */
100 	uint8_t		th_sum[2];	/* TCP checksum */
101 	uint8_t		th_urp[2];	/* Urgent pointer */
102 } tcph_t;
103 
104 #define	TCP_HDR_LENGTH(tcph) \
105 	((((tcph_t *)tcph)->th_offset_and_rsrvd[0] >>2) &(0xF << 2))
106 #define	TCP_MAX_COMBINED_HEADER_LENGTH	(60 + 60) /* Maxed out ip + tcp */
107 #define	TCP_MAX_IP_OPTIONS_LENGTH	(60 - IP_SIMPLE_HDR_LENGTH)
108 #define	TCP_MAX_HDR_LENGTH		60
109 #define	TCP_MAX_TCP_OPTIONS_LENGTH	(60 - sizeof (tcpha_t))
110 #define	TCP_MIN_HEADER_LENGTH		20
111 #define	TCP_MAXWIN			65535
112 #define	TCP_PORT_LEN			sizeof (in_port_t)
113 #define	TCP_MAX_WINSHIFT		14
114 #define	TCP_MAX_LARGEWIN		(TCP_MAXWIN << TCP_MAX_WINSHIFT)
115 #define	TCP_MAX_LSO_LENGTH	(IP_MAXPACKET - TCP_MAX_COMBINED_HEADER_LENGTH)
116 
117 #define	TCPIP_HDR_LENGTH(mp, n)					\
118 	(n) = IPH_HDR_LENGTH((mp)->b_rptr),			\
119 	(n) += TCP_HDR_LENGTH((tcpha_t *)&(mp)->b_rptr[(n)])
120 
121 /* TCP Protocol header (used if the header is known to be 32-bit aligned) */
122 typedef	struct tcphdra_s {
123 	in_port_t	tha_lport;	/* Source port */
124 	in_port_t	tha_fport;	/* Destination port */
125 	uint32_t	tha_seq;	/* Sequence number */
126 	uint32_t	tha_ack;	/* Acknowledgement number */
127 	uint8_t tha_offset_and_reserved; /* Offset to the packet data */
128 	uint8_t		tha_flags;
129 	uint16_t	tha_win;	/* Allocation number */
130 	uint16_t	tha_sum;	/* TCP checksum */
131 	uint16_t	tha_urp;	/* Urgent pointer */
132 } tcpha_t;
133 
134 struct conn_s;
135 struct tcp_listen_cnt_s;
136 
137 /*
138  * Control structure for each open TCP stream,
139  * defined only within the kernel or for a kmem user.
140  * NOTE: tcp_reinit_values MUST have a line for each field in this structure!
141  */
142 #if (defined(_KERNEL) || defined(_KMEMUSER))
143 
144 typedef struct tcp_s {
145 				/* Pointer to previous bind hash next. */
146 	struct tcp_s	*tcp_time_wait_next;
147 				/* Pointer to next T/W block */
148 	struct tcp_s	*tcp_time_wait_prev;
149 				/* Pointer to previous T/W next */
150 	clock_t		tcp_time_wait_expire;
151 
152 	struct conn_s	*tcp_connp;
153 	tcp_stack_t	*tcp_tcps;	/* Shortcut via conn_netstack */
154 
155 	int32_t	tcp_state;
156 	int32_t	tcp_rcv_ws;		/* My window scale power */
157 	int32_t	tcp_snd_ws;		/* Sender's window scale power */
158 	uint32_t tcp_ts_recent;		/* Timestamp of earliest unacked */
159 					/*  data segment */
160 	clock_t	tcp_rto;		/* Round trip timeout */
161 	clock_t	tcp_last_rcv_lbolt;
162 				/* lbolt on last packet, used for PAWS */
163 
164 	uint32_t tcp_snxt;		/* Senders next seq num */
165 	uint32_t tcp_swnd;		/* Senders window (relative to suna) */
166 	uint32_t tcp_mss;		/* Max segment size */
167 	uint32_t tcp_iss;		/* Initial send seq num */
168 	uint32_t tcp_rnxt;		/* Seq we expect to recv next */
169 	uint32_t tcp_rwnd;
170 
171 	/* Fields arranged in approximate access order along main paths */
172 	mblk_t	*tcp_xmit_head;		/* Head of rexmit list */
173 	mblk_t	*tcp_xmit_last;		/* last valid data seen by tcp_wput */
174 	mblk_t	*tcp_xmit_tail;		/* Last rexmit data sent */
175 	uint32_t tcp_unsent;		/* # of bytes in hand that are unsent */
176 	uint32_t tcp_xmit_tail_unsent;	/* # of unsent bytes in xmit_tail */
177 
178 	uint32_t tcp_suna;		/* Sender unacknowledged */
179 	uint32_t tcp_rexmit_nxt;	/* Next rexmit seq num */
180 	uint32_t tcp_rexmit_max;	/* Max retran seq num */
181 	int32_t	tcp_snd_burst;		/* Send burst factor */
182 	uint32_t tcp_cwnd;		/* Congestion window */
183 	int32_t tcp_cwnd_cnt;		/* cwnd cnt in congestion avoidance */
184 
185 	uint32_t tcp_ibsegs;		/* Inbound segments on this stream */
186 	uint32_t tcp_obsegs;		/* Outbound segments on this stream */
187 
188 	uint32_t tcp_naglim;		/* Tunable nagle limit */
189 	uint32_t	tcp_valid_bits;
190 #define	TCP_ISS_VALID	0x1	/* Is the tcp_iss seq num active? */
191 #define	TCP_FSS_VALID	0x2	/* Is the tcp_fss seq num active? */
192 #define	TCP_URG_VALID	0x4	/* Is the tcp_urg seq num active? */
193 #define	TCP_OFO_FIN_VALID 0x8	/* Has TCP received an out of order FIN? */
194 
195 
196 
197 	timeout_id_t	tcp_timer_tid;	/* Control block for timer service */
198 	uchar_t	tcp_timer_backoff;	/* Backoff shift count. */
199 	int64_t tcp_last_recv_time;	/* Last time we receive a segment. */
200 	uint32_t tcp_init_cwnd;		/* Initial cwnd (start/restart) */
201 
202 	/* Following manipulated by TCP under squeue protection */
203 	uint32_t
204 		tcp_urp_last_valid : 1,	/* Is tcp_urp_last valid? */
205 		tcp_hard_binding : 1,	/* TCP_DETACHED_NONEAGER */
206 		tcp_fin_acked : 1,	/* Has our FIN been acked? */
207 		tcp_fin_rcvd : 1,	/* Have we seen a FIN? */
208 
209 		tcp_fin_sent : 1,	/* Have we sent our FIN yet? */
210 		tcp_ordrel_done : 1,	/* Have we sent the ord_rel upstream? */
211 		tcp_detached : 1,	/* If we're detached from a stream */
212 		tcp_zero_win_probe: 1,	/* Zero win probing is in progress */
213 
214 		tcp_loopback: 1,	/* src and dst are the same machine */
215 		tcp_localnet: 1,	/* src and dst are on the same subnet */
216 		tcp_syn_defense: 1,	/* For defense against SYN attack */
217 #define	tcp_dontdrop	tcp_syn_defense
218 		tcp_set_timer : 1,
219 
220 		tcp_active_open: 1,	/* This is a active open */
221 		tcp_rexmit : 1,		/* TCP is retransmitting */
222 		tcp_snd_sack_ok : 1,	/* Can use SACK for this connection */
223 		tcp_hwcksum : 1,	/* The NIC is capable of hwcksum */
224 
225 		tcp_ip_forward_progress : 1,
226 		tcp_ecn_ok : 1,		/* Can use ECN for this connection */
227 		tcp_ecn_echo_on : 1,	/* Need to do ECN echo */
228 		tcp_ecn_cwr_sent : 1,	/* ECN_CWR has been sent */
229 
230 		tcp_cwr : 1,		/* Cwnd has reduced recently */
231 
232 		tcp_pad_to_bit31 : 11;
233 
234 	/* Following manipulated by TCP under squeue protection */
235 	uint32_t
236 		tcp_snd_ts_ok  : 1,
237 		tcp_snd_ws_ok  : 1,
238 		tcp_reserved_port : 1,
239 		tcp_in_free_list : 1,
240 
241 		tcp_snd_zcopy_on : 1,	/* xmit zero-copy enabled */
242 		tcp_snd_zcopy_aware : 1, /* client is zero-copy aware */
243 		tcp_xmit_zc_clean : 1,	/* the xmit list is free of zc-mblk */
244 		tcp_wait_for_eagers : 1, /* Wait for eagers to disappear */
245 
246 		tcp_accept_error : 1,	/* Error during TLI accept */
247 		tcp_send_discon_ind : 1, /* TLI accept err, send discon ind */
248 		tcp_cork : 1,		/* tcp_cork option */
249 		tcp_tconnind_started : 1, /* conn_ind message is being sent */
250 
251 		tcp_lso :1,		/* Lower layer is capable of LSO */
252 		tcp_is_wnd_shrnk : 1,	/* Window has shrunk */
253 
254 		tcp_pad_to_bit_31 : 18;
255 
256 	uint32_t	tcp_initial_pmtu; /* Initial outgoing Path MTU. */
257 
258 	mblk_t	*tcp_reass_head;	/* Out of order reassembly list head */
259 	mblk_t	*tcp_reass_tail;	/* Out of order reassembly list tail */
260 
261 	tcp_sack_info_t	*tcp_sack_info;
262 
263 #define	tcp_pipe	tcp_sack_info->tcp_pipe
264 #define	tcp_fack	tcp_sack_info->tcp_fack
265 #define	tcp_sack_snxt	tcp_sack_info->tcp_sack_snxt
266 #define	tcp_max_sack_blk	tcp_sack_info->tcp_max_sack_blk
267 #define	tcp_num_sack_blk	tcp_sack_info->tcp_num_sack_blk
268 #define	tcp_sack_list		tcp_sack_info->tcp_sack_list
269 #define	tcp_num_notsack_blk	tcp_sack_info->tcp_num_notsack_blk
270 #define	tcp_cnt_notsack_list	tcp_sack_info->tcp_cnt_notsack_list
271 #define	tcp_notsack_list		tcp_sack_info->tcp_notsack_list
272 
273 	mblk_t	*tcp_rcv_list;		/* Queued until push, urgent data, */
274 	mblk_t	*tcp_rcv_last_head;	/* optdata, or the count exceeds */
275 	mblk_t	*tcp_rcv_last_tail;	/* tcp_rcv_push_wait. */
276 	uint32_t tcp_rcv_cnt;		/* tcp_rcv_list is b_next chain. */
277 
278 	uint32_t tcp_cwnd_ssthresh;	/* Congestion window */
279 	uint32_t tcp_cwnd_max;
280 	uint32_t tcp_csuna;		/* Clear (no rexmits in window) suna */
281 
282 	clock_t	tcp_rtt_sa;		/* Round trip smoothed average */
283 	clock_t	tcp_rtt_sd;		/* Round trip smoothed deviation */
284 	clock_t	tcp_rtt_update;		/* Round trip update(s) */
285 	clock_t tcp_ms_we_have_waited;	/* Total retrans time */
286 
287 	uint32_t tcp_swl1;		/* These help us avoid using stale */
288 	uint32_t tcp_swl2;		/*  packets to update state */
289 
290 	uint32_t tcp_rack;		/* Seq # we have acked */
291 	uint32_t tcp_rack_cnt;		/* # of segs we have deferred ack */
292 	uint32_t tcp_rack_cur_max;	/* # of segs we may defer ack for now */
293 	uint32_t tcp_rack_abs_max;	/* # of segs we may defer ack ever */
294 	timeout_id_t	tcp_ack_tid;	/* Delayed ACK timer ID */
295 	timeout_id_t	tcp_push_tid;	/* Push timer ID */
296 
297 	uint32_t tcp_max_swnd;		/* Maximum swnd we have seen */
298 
299 	struct tcp_s *tcp_listener;	/* Our listener */
300 
301 	uint32_t tcp_irs;		/* Initial recv seq num */
302 	uint32_t tcp_fss;		/* Final/fin send seq num */
303 	uint32_t tcp_urg;		/* Urgent data seq num */
304 
305 	clock_t	tcp_first_timer_threshold;  /* When to prod IP */
306 	clock_t	tcp_second_timer_threshold; /* When to give up completely */
307 	clock_t	tcp_first_ctimer_threshold; /* 1st threshold while connecting */
308 	clock_t tcp_second_ctimer_threshold; /* 2nd ... while connecting */
309 
310 	uint32_t tcp_urp_last;		/* Last urp for which signal sent */
311 	mblk_t	*tcp_urp_mp;		/* T_EXDATA_IND for urgent byte */
312 	mblk_t	*tcp_urp_mark_mp;	/* zero-length marked/unmarked msg */
313 
314 	int tcp_conn_req_cnt_q0;	/* # of conn reqs in SYN_RCVD */
315 	int tcp_conn_req_cnt_q;	/* # of conn reqs in ESTABLISHED */
316 	int tcp_conn_req_max;	/* # of ESTABLISHED conn reqs allowed */
317 	t_scalar_t tcp_conn_req_seqnum;	/* Incrementing pending conn req ID */
318 #define	tcp_ip_addr_cache	tcp_reass_tail
319 					/* Cache ip addresses that */
320 					/* complete the 3-way handshake */
321 	kmutex_t  tcp_eager_lock;
322 	struct tcp_s *tcp_eager_next_q; /* next eager in ESTABLISHED state */
323 	struct tcp_s *tcp_eager_last_q;	/* last eager in ESTABLISHED state */
324 	struct tcp_s *tcp_eager_next_q0; /* next eager in SYN_RCVD state */
325 	struct tcp_s *tcp_eager_prev_q0; /* prev eager in SYN_RCVD state */
326 					/* all eagers form a circular list */
327 	boolean_t tcp_conn_def_q0;	/* move from q0 to q deferred */
328 
329 	union {
330 	    mblk_t *tcp_eager_conn_ind; /* T_CONN_IND waiting for 3rd ack. */
331 	    mblk_t *tcp_opts_conn_req; /* T_CONN_REQ w/ options processed */
332 	} tcp_conn;
333 	uint32_t tcp_syn_rcvd_timeout;	/* How many SYN_RCVD timeout in q0 */
334 
335 	/* TCP Keepalive Timer members */
336 	int32_t	tcp_ka_last_intrvl;	/* Last probe interval */
337 	timeout_id_t tcp_ka_tid;	/* Keepalive timer ID */
338 	uint32_t tcp_ka_interval;	/* Keepalive interval */
339 	uint32_t tcp_ka_abort_thres;	/* Keepalive abort threshold */
340 
341 	int32_t	tcp_client_errno;	/* How the client screwed up */
342 
343 	/*
344 	 * The header template lives in conn_ht_iphc allocated by tcp_build_hdrs
345 	 * We maintain three pointers into conn_ht_iphc.
346 	 */
347 	ipha_t	*tcp_ipha;		/* IPv4 header in conn_ht_iphc */
348 	ip6_t	*tcp_ip6h;		/* IPv6 header in conn_ht_iphc */
349 	tcpha_t	*tcp_tcpha;		/* TCP header in conn_ht_iphc */
350 
351 	uint16_t tcp_last_sent_len;	/* Record length for nagle */
352 	uint16_t tcp_dupack_cnt;	/* # of consequtive duplicate acks */
353 
354 	kmutex_t	*tcp_acceptor_lockp;	/* Ptr to tf_lock */
355 
356 	mblk_t		*tcp_ordrel_mp;		/* T_ordrel_ind mblk */
357 	t_uscalar_t	tcp_acceptor_id;	/* ACCEPTOR_id */
358 
359 	int		tcp_ipsec_overhead;
360 
361 	uint_t		tcp_recvifindex; /* Last received IPV6_RCVPKTINFO */
362 	uint_t		tcp_recvhops;	/* Last received IPV6_RECVHOPLIMIT */
363 	uint_t		tcp_recvtclass;	/* Last received IPV6_RECVTCLASS */
364 	ip6_hbh_t	*tcp_hopopts;	/* Last received IPV6_RECVHOPOPTS */
365 	ip6_dest_t	*tcp_dstopts;	/* Last received IPV6_RECVDSTOPTS */
366 	ip6_dest_t	*tcp_rthdrdstopts; /* Last recv IPV6_RECVRTHDRDSTOPTS */
367 	ip6_rthdr_t	*tcp_rthdr;	/* Last received IPV6_RECVRTHDR */
368 	uint_t		tcp_hopoptslen;
369 	uint_t		tcp_dstoptslen;
370 	uint_t		tcp_rthdrdstoptslen;
371 	uint_t		tcp_rthdrlen;
372 
373 	mblk_t		*tcp_timercache;
374 
375 	kmutex_t	tcp_closelock;
376 	kcondvar_t	tcp_closecv;
377 	uint8_t		tcp_closed;
378 	uint8_t		tcp_closeflags;
379 	uint8_t		tcp_cleandeathtag;
380 	mblk_t		tcp_closemp;
381 	timeout_id_t	tcp_linger_tid;	/* Linger timer ID */
382 
383 	struct tcp_s *tcp_acceptor_hash; /* Acceptor hash chain */
384 	struct tcp_s **tcp_ptpahn; /* Pointer to previous accept hash next. */
385 	struct tcp_s *tcp_bind_hash; /* Bind hash chain */
386 	struct tcp_s *tcp_bind_hash_port; /* tcp_t's bound to the same lport */
387 	struct tcp_s **tcp_ptpbhn;
388 
389 	uint_t		tcp_maxpsz_multiplier;
390 
391 	uint32_t	tcp_lso_max; /* maximum LSO payload */
392 
393 	uint32_t	tcp_ofo_fin_seq; /* Recv out of order FIN seq num */
394 	uint32_t	tcp_cwr_snd_max;
395 
396 	struct tcp_s *tcp_saved_listener;	/* saved value of listener */
397 
398 	uint32_t	tcp_in_ack_unsent;	/* ACK for unsent data cnt. */
399 
400 	/*
401 	 * All fusion-related fields are protected by squeue.
402 	 */
403 	struct tcp_s *tcp_loopback_peer;	/* peer tcp for loopback */
404 	mblk_t	*tcp_fused_sigurg_mp;		/* M_PCSIG mblk for SIGURG */
405 
406 	uint32_t
407 		tcp_fused : 1,		/* loopback tcp in fusion mode */
408 		tcp_unfusable : 1,	/* fusion not allowed on endpoint */
409 		tcp_fused_sigurg : 1,	/* send SIGURG upon draining */
410 
411 		tcp_fuse_to_bit_31 : 29;
412 
413 	kmutex_t tcp_non_sq_lock;
414 
415 	/*
416 	 * This variable is accessed without any lock protection
417 	 * and therefore must not be declared as a bit field along
418 	 * with the rest which require such condition.
419 	 */
420 	boolean_t	tcp_issocket;	/* this is a socket tcp */
421 
422 	/* protected by the tcp_non_sq_lock lock */
423 	uint32_t	tcp_squeue_bytes;
424 	/*
425 	 * Kernel SSL session information
426 	 */
427 	boolean_t		tcp_kssl_pending; /* waiting for 1st SSL rec. */
428 	boolean_t		tcp_kssl_inhandshake; /* during SSL handshake */
429 	kssl_ent_t		tcp_kssl_ent;	/* SSL table entry */
430 	kssl_ctx_t		tcp_kssl_ctx;	/* SSL session */
431 
432 	/*
433 	 * tcp_closemp_used is protected by listener's tcp_eager_lock
434 	 * when used for eagers. When used for a tcp in TIME_WAIT state
435 	 * or in tcp_close(), it is not protected by any lock as we
436 	 * do not expect any other thread to use it concurrently.
437 	 * We do allow re-use of tcp_closemp in tcp_time_wait_collector()
438 	 * and tcp_close() but not concurrently.
439 	 */
440 	boolean_t tcp_closemp_used;
441 
442 	/*
443 	 * previous and next eagers in the list of droppable eagers. See
444 	 * the comments before MAKE_DROPPABLE(). These pointers are
445 	 * protected by listener's tcp_eager_lock.
446 	 */
447 	struct tcp_s	*tcp_eager_prev_drop_q0;
448 	struct tcp_s	*tcp_eager_next_drop_q0;
449 
450 	/*
451 	 * Have we flow controlled xmitter?
452 	 * This variable can be modified outside the squeue and hence must
453 	 * not be declared as a bit field along with the rest that are
454 	 * modified only within the squeue.
455 	 * protected by the tcp_non_sq_lock lock.
456 	 */
457 	boolean_t	tcp_flow_stopped;
458 
459 	/*
460 	 * Sender's next sequence number at the time the window was shrunk.
461 	 */
462 	uint32_t	tcp_snxt_shrunk;
463 
464 	/*
465 	 * Socket generation number which is bumped when a connection attempt
466 	 * is initiated. Its main purpose is to ensure that the socket does not
467 	 * miss the asynchronous connected/disconnected notification.
468 	 */
469 	sock_connid_t	tcp_connid;
470 
471 	/* mblk_t used to enter TCP's squeue from the service routine. */
472 	mblk_t		*tcp_rsrv_mp;
473 	/* Mutex for accessing tcp_rsrv_mp */
474 	kmutex_t	tcp_rsrv_mp_lock;
475 
476 	/* For connection counting. */
477 	struct tcp_listen_cnt_s	*tcp_listen_cnt;
478 
479 	/* Segment reassembly timer. */
480 	timeout_id_t		tcp_reass_tid;
481 
482 #ifdef DEBUG
483 	pc_t			tcmp_stk[15];
484 #endif
485 } tcp_t;
486 
487 #ifdef DEBUG
488 #define	TCP_DEBUG_GETPCSTACK(buffer, depth)	((void) getpcstack(buffer, \
489 						    depth))
490 #else
491 #define	TCP_DEBUG_GETPCSTACK(buffer, depth)
492 #endif
493 
494 extern void	tcp_conn_reclaim(void *);
495 extern void 	tcp_free(tcp_t *tcp);
496 extern void	tcp_ddi_g_init(void);
497 extern void	tcp_ddi_g_destroy(void);
498 extern void	tcp_xmit_listeners_reset(mblk_t *, ip_recv_attr_t *,
499     ip_stack_t *, conn_t *);
500 extern void	tcp_input_listener(void *arg, mblk_t *mp, void *arg2,
501     ip_recv_attr_t *);
502 extern void	tcp_input_listener_unbound(void *arg, mblk_t *mp, void *arg2,
503     ip_recv_attr_t *);
504 extern void	tcp_input_data(void *arg, mblk_t *mp, void *arg2,
505     ip_recv_attr_t *);
506 extern void 	*tcp_get_conn(void *arg, tcp_stack_t *);
507 extern void	tcp_time_wait_collector(void *arg);
508 extern mblk_t	*tcp_snmp_get(queue_t *, mblk_t *);
509 extern int	tcp_snmp_set(queue_t *, int, int, uchar_t *, int len);
510 extern mblk_t	*tcp_xmit_mp(tcp_t *tcp, mblk_t *mp, int32_t max_to_send,
511 		    int32_t *offset, mblk_t **end_mp, uint32_t seq,
512 		    boolean_t sendall, uint32_t *seg_len, boolean_t rexmit);
513 
514 /*
515  * The TCP Fanout structure.
516  * The hash tables and their linkage (tcp_*_hash_next, tcp_ptp*hn) are
517  * protected by the per-bucket tf_lock. Each tcp_t
518  * inserted in the list points back at this lock using tcp_*_lockp.
519  *
520  * The listener and acceptor hash queues are lists of tcp_t.
521  */
522 /* listener hash and acceptor hash queue head */
523 typedef struct tf_s {
524 	tcp_t		*tf_tcp;
525 	kmutex_t	tf_lock;
526 } tf_t;
527 #endif	/* (defined(_KERNEL) || defined(_KMEMUSER)) */
528 
529 /* Contract private interface between TCP and Clustering. */
530 
531 #define	CL_TCPI_V1	1	/* cl_tcpi_version number */
532 
533 typedef struct cl_tcp_info_s {
534 	ushort_t	cl_tcpi_version;	/* cl_tcp_info_t's version no */
535 	ushort_t	cl_tcpi_ipversion;	/* IP version */
536 	int32_t		cl_tcpi_state;		/* TCP state */
537 	in_port_t	cl_tcpi_lport;		/* Local port */
538 	in_port_t	cl_tcpi_fport;		/* Remote port */
539 	in6_addr_t	cl_tcpi_laddr_v6;	/* Local IP address */
540 	in6_addr_t	cl_tcpi_faddr_v6;	/* Remote IP address */
541 #ifdef _KERNEL
542 /* Note: V4_PART_OF_V6 is meant to be used only for _KERNEL defined stuff */
543 #define	cl_tcpi_laddr	V4_PART_OF_V6(cl_tcpi_laddr_v6)
544 #define	cl_tcpi_faddr	V4_PART_OF_V6(cl_tcpi_faddr_v6)
545 
546 #endif	/* _KERNEL */
547 } cl_tcp_info_t;
548 
549 /*
550  * Hook functions to enable cluster networking
551  * On non-clustered systems these vectors must always be NULL.
552  */
553 extern void	(*cl_inet_listen)(netstackid_t, uint8_t, sa_family_t,
554 		    uint8_t *, in_port_t, void *);
555 extern void	(*cl_inet_unlisten)(netstackid_t, uint8_t, sa_family_t,
556 		    uint8_t *, in_port_t, void *);
557 
558 /*
559  * Contracted Consolidation Private ioctl for aborting TCP connections.
560  * In order to keep the offsets and size of the structure the same between
561  * a 32-bit application and a 64-bit amd64 kernel, we use a #pragma
562  * pack(4).
563  */
564 #define	TCP_IOC_ABORT_CONN	(('T' << 8) + 91)
565 
566 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
567 #pragma pack(4)
568 #endif
569 
570 typedef struct tcp_ioc_abort_conn_s {
571 	struct sockaddr_storage ac_local;	/* local addr and port */
572 	struct sockaddr_storage ac_remote;	/* remote addr and port */
573 	int32_t ac_start;			/* start state */
574 	int32_t ac_end;				/* end state  */
575 	int32_t ac_zoneid;			/* zoneid */
576 } tcp_ioc_abort_conn_t;
577 
578 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
579 #pragma pack()
580 #endif
581 
582 #ifdef	__cplusplus
583 }
584 #endif
585 
586 #endif	/* _INET_TCP_H */
587