xref: /titanic_41/usr/src/uts/common/inet/tcp_impl.h (revision 6a1af1a67532df169a657cce07140be64bdea084)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, Joyent Inc. All rights reserved.
24  */
25 
26 #ifndef	_INET_TCP_IMPL_H
27 #define	_INET_TCP_IMPL_H
28 
29 /*
30  * TCP implementation private declarations.  These interfaces are
31  * used to build the IP module and are not meant to be accessed
32  * by any modules except IP itself.  They are undocumented and are
33  * subject to change without notice.
34  */
35 
36 #ifdef	__cplusplus
37 extern "C" {
38 #endif
39 
40 #ifdef _KERNEL
41 
42 #include <sys/cpuvar.h>
43 #include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */
44 #include <inet/optcom.h>
45 #include <inet/tcp.h>
46 #include <inet/tunables.h>
47 
48 #define	TCP_MOD_ID	5105
49 
50 extern struct qinit	tcp_sock_winit;
51 extern struct qinit	tcp_winit;
52 
53 extern sock_downcalls_t sock_tcp_downcalls;
54 
55 /*
56  * Bind hash list size and has function.  It has to be a power of 2 for
57  * hashing.
58  */
59 #define	TCP_BIND_FANOUT_SIZE	1024
60 #define	TCP_BIND_HASH(lport) (ntohs(lport) & (TCP_BIND_FANOUT_SIZE - 1))
61 
62 /*
63  * This implementation follows the 4.3BSD interpretation of the urgent
64  * pointer and not RFC 1122. Switching to RFC 1122 behavior would cause
65  * incompatible changes in protocols like telnet and rlogin.
66  */
67 #define	TCP_OLD_URP_INTERPRETATION	1
68 
69 /* TCP option length */
70 #define	TCPOPT_NOP_LEN		1
71 #define	TCPOPT_MAXSEG_LEN	4
72 #define	TCPOPT_WS_LEN		3
73 #define	TCPOPT_REAL_WS_LEN	(TCPOPT_WS_LEN+1)
74 #define	TCPOPT_TSTAMP_LEN	10
75 #define	TCPOPT_REAL_TS_LEN	(TCPOPT_TSTAMP_LEN+2)
76 #define	TCPOPT_SACK_OK_LEN	2
77 #define	TCPOPT_REAL_SACK_OK_LEN	(TCPOPT_SACK_OK_LEN+2)
78 #define	TCPOPT_REAL_SACK_LEN	4
79 #define	TCPOPT_MAX_SACK_LEN	36
80 #define	TCPOPT_HEADER_LEN	2
81 
82 /* Round up the value to the nearest mss. */
83 #define	MSS_ROUNDUP(value, mss)		((((value) - 1) / (mss) + 1) * (mss))
84 
85 /*
86  * Was this tcp created via socket() interface?
87  */
88 #define	TCP_IS_SOCKET(tcp)	((tcp)->tcp_issocket)
89 
90 /*
91  * Is this tcp not attached to any upper client?
92  */
93 #define	TCP_IS_DETACHED(tcp)	((tcp)->tcp_detached)
94 
95 /* TCP timers related data strucutres.  Refer to tcp_timers.c. */
96 typedef struct tcp_timer_s {
97 	conn_t	*connp;
98 	void 	(*tcpt_proc)(void *);
99 	callout_id_t   tcpt_tid;
100 } tcp_timer_t;
101 
102 extern kmem_cache_t *tcp_timercache;
103 
104 /*
105  * Macro for starting various timers.  Retransmission timer has its own macro,
106  * TCP_TIMER_RESTART().  tim is in millisec.
107  */
108 #define	TCP_TIMER(tcp, f, tim)		\
109 	tcp_timeout(tcp->tcp_connp, f, tim)
110 #define	TCP_TIMER_CANCEL(tcp, id)	\
111 	tcp_timeout_cancel(tcp->tcp_connp, id)
112 
113 /*
114  * To restart the TCP retransmission timer.  intvl is in millisec.
115  */
116 #define	TCP_TIMER_RESTART(tcp, intvl) {					\
117 	if ((tcp)->tcp_timer_tid != 0)					\
118 		(void) TCP_TIMER_CANCEL((tcp), (tcp)->tcp_timer_tid);	\
119 	(tcp)->tcp_timer_tid = TCP_TIMER((tcp), tcp_timer, (intvl));	\
120 }
121 
122 /*
123  * For scalability, we must not run a timer for every TCP connection
124  * in TIME_WAIT state.  To see why, consider (for time wait interval of
125  * 1 minutes):
126  *	10,000 connections/sec * 60 seconds/time wait = 600,000 active conn's
127  *
128  * This list is ordered by time, so you need only delete from the head
129  * until you get to entries which aren't old enough to delete yet.
130  * The list consists of only the detached TIME_WAIT connections.
131  *
132  * When a tcp_t enters TIME_WAIT state, a timer is started (timeout is
133  * tcps_time_wait_interval).  When the tcp_t is detached (upper layer closes
134  * the end point), it is moved to the time wait list and another timer is
135  * started (expiry time is set at tcp_time_wait_expire, which is
136  * also calculated using tcps_time_wait_interval).  This means that the
137  * TIME_WAIT state can be extended (up to doubled) if the tcp_t doesn't
138  * become detached for a long time.
139  *
140  * The list manipulations (including tcp_time_wait_next/prev)
141  * are protected by the tcp_time_wait_lock. The content of the
142  * detached TIME_WAIT connections is protected by the normal perimeters.
143  *
144  * This list is per squeue and squeues are shared across the tcp_stack_t's.
145  * Things on tcp_time_wait_head remain associated with the tcp_stack_t
146  * and conn_netstack.
147  * The tcp_t's that are added to tcp_free_list are disassociated and
148  * have NULL tcp_tcps and conn_netstack pointers.
149  */
150 typedef struct tcp_squeue_priv_s {
151 	kmutex_t	tcp_time_wait_lock;
152 	callout_id_t	tcp_time_wait_tid;
153 	tcp_t		*tcp_time_wait_head;
154 	tcp_t		*tcp_time_wait_tail;
155 	tcp_t		*tcp_free_list;
156 	uint_t		tcp_free_list_cnt;
157 #ifdef DEBUG
158 	/*
159 	 * For debugging purpose, true when tcp_time_wait_collector() is
160 	 * running.
161 	 */
162 	boolean_t	tcp_time_wait_running;
163 #endif
164 } tcp_squeue_priv_t;
165 
166 /*
167  * Parameters for TCP Initial Send Sequence number (ISS) generation.  When
168  * tcp_strong_iss is set to 1, which is the default, the ISS is calculated
169  * by adding three components: a time component which grows by 1 every 4096
170  * nanoseconds (versus every 4 microseconds suggested by RFC 793, page 27);
171  * a per-connection component which grows by 125000 for every new connection;
172  * and an "extra" component that grows by a random amount centered
173  * approximately on 64000.  This causes the ISS generator to cycle every
174  * 4.89 hours if no TCP connections are made, and faster if connections are
175  * made.
176  *
177  * When tcp_strong_iss is set to 0, ISS is calculated by adding two
178  * components: a time component which grows by 250000 every second; and
179  * a per-connection component which grows by 125000 for every new connections.
180  *
181  * A third method, when tcp_strong_iss is set to 2, for generating ISS is
182  * prescribed by Steve Bellovin.  This involves adding time, the 125000 per
183  * connection, and a one-way hash (MD5) of the connection ID <sport, dport,
184  * src, dst>, a "truly" random (per RFC 1750) number, and a console-entered
185  * password.
186  */
187 #define	ISS_INCR	250000
188 #define	ISS_NSEC_SHT	12
189 
190 /* Macros for timestamp comparisons */
191 #define	TSTMP_GEQ(a, b)	((int32_t)((a)-(b)) >= 0)
192 #define	TSTMP_LT(a, b)	((int32_t)((a)-(b)) < 0)
193 
194 /*
195  * Initialize cwnd according to RFC 3390.  def_max_init_cwnd is
196  * either tcp_slow_start_initial or tcp_slow_start_after idle
197  * depending on the caller.  If the upper layer has not used the
198  * TCP_INIT_CWND option to change the initial cwnd, tcp_init_cwnd
199  * should be 0 and we use the formula in RFC 3390 to set tcp_cwnd.
200  * If the upper layer has changed set the tcp_init_cwnd, just use
201  * it to calculate the tcp_cwnd.
202  */
203 #define	TCP_SET_INIT_CWND(tcp, mss, def_max_init_cwnd)			\
204 {									\
205 	if ((tcp)->tcp_init_cwnd == 0) {				\
206 		(tcp)->tcp_cwnd = MIN(def_max_init_cwnd * (mss),	\
207 		    MIN(4 * (mss), MAX(2 * (mss), 4380 / (mss) * (mss)))); \
208 	} else {							\
209 		(tcp)->tcp_cwnd = (tcp)->tcp_init_cwnd * (mss);		\
210 	}								\
211 	tcp->tcp_cwnd_cnt = 0;						\
212 }
213 
214 /*
215  * Set ECN capable transport (ECT) code point in IP header.
216  *
217  * Note that there are 2 ECT code points '01' and '10', which are called
218  * ECT(1) and ECT(0) respectively.  Here we follow the original ECT code
219  * point ECT(0) for TCP as described in RFC 2481.
220  */
221 #define	TCP_SET_ECT(tcp, iph) \
222 	if ((tcp)->tcp_connp->conn_ipversion == IPV4_VERSION) { \
223 		/* We need to clear the code point first. */ \
224 		((ipha_t *)(iph))->ipha_type_of_service &= 0xFC; \
225 		((ipha_t *)(iph))->ipha_type_of_service |= IPH_ECN_ECT0; \
226 	} else { \
227 		((ip6_t *)(iph))->ip6_vcf &= htonl(0xFFCFFFFF); \
228 		((ip6_t *)(iph))->ip6_vcf |= htonl(IPH_ECN_ECT0 << 20); \
229 	}
230 
231 /*
232  * Set tcp_rto with boundary checking.
233  */
234 #define	TCP_SET_RTO(tcp, rto) \
235 	if ((rto) < (tcp)->tcp_rto_min)			\
236 		(tcp)->tcp_rto = (tcp)->tcp_rto_min;	\
237 	else if ((rto) > (tcp)->tcp_rto_max)		\
238 		(tcp)->tcp_rto = (tcp)->tcp_rto_max;	\
239 	else						\
240 		(tcp)->tcp_rto = (rto);
241 
242 /*
243  * TCP options struct returned from tcp_parse_options.
244  */
245 typedef struct tcp_opt_s {
246 	uint32_t	tcp_opt_mss;
247 	uint32_t	tcp_opt_wscale;
248 	uint32_t	tcp_opt_ts_val;
249 	uint32_t	tcp_opt_ts_ecr;
250 	tcp_t		*tcp;
251 } tcp_opt_t;
252 
253 /*
254  * Write-side flow-control is implemented via the per instance STREAMS
255  * write-side Q by explicitly setting QFULL to stop the flow of mblk_t(s)
256  * and clearing QFULL and calling qbackenable() to restart the flow based
257  * on the number of TCP unsent bytes (i.e. those not on the wire waiting
258  * for a remote ACK).
259  *
260  * This is different than a standard STREAMS kmod which when using the
261  * STREAMS Q the framework would automatictly flow-control based on the
262  * defined hiwat/lowat values as mblk_t's are enqueued/dequeued.
263  *
264  * As of FireEngine TCP write-side flow-control needs to take into account
265  * both the unsent tcp_xmit list bytes but also any squeue_t enqueued bytes
266  * (i.e. from tcp_wput() -> tcp_output()).
267  *
268  * This is accomplished by adding a new tcp_t fields, tcp_squeue_bytes, to
269  * count the number of bytes enqueued by tcp_wput() and the number of bytes
270  * dequeued and processed by tcp_output().
271  *
272  * So, the total number of bytes unsent is (squeue_bytes + unsent) with all
273  * flow-control uses of unsent replaced with the macro TCP_UNSENT_BYTES.
274  */
275 extern void	tcp_clrqfull(tcp_t *);
276 extern void	tcp_setqfull(tcp_t *);
277 
278 #define	TCP_UNSENT_BYTES(tcp) \
279 	((tcp)->tcp_squeue_bytes + (tcp)->tcp_unsent)
280 
281 /*
282  * Linked list struct to store listener connection limit configuration per
283  * IP stack.  The list is stored at tcps_listener_conf in tcp_stack_t.
284  *
285  * tl_port: the listener port of this limit configuration
286  * tl_ratio: the maximum amount of memory consumed by all concurrent TCP
287  *           connections created by a listener does not exceed 1/tl_ratio
288  *           of the total system memory.  Note that this is only an
289  *           approximation.
290  * tl_link: linked list struct
291  */
292 typedef struct tcp_listener_s {
293 	in_port_t	tl_port;
294 	uint32_t	tl_ratio;
295 	list_node_t	tl_link;
296 } tcp_listener_t;
297 
298 /*
299  * If there is a limit set on the number of connections allowed per each
300  * listener, the following struct is used to store that counter.  It keeps
301  * the number of TCP connection created by a listener.  Note that this needs
302  * to be separated from the listener since the listener can go away before
303  * all the connections are gone.
304  *
305  * When the struct is allocated, tlc_cnt is set to 1.  When a new connection
306  * is created by the listener, tlc_cnt is incremented by 1.  When a connection
307  * created by the listener goes away, tlc_count is decremented by 1.  When the
308  * listener itself goes away, tlc_cnt is decremented  by one.  The last
309  * connection (or the listener) which decrements tlc_cnt to zero frees the
310  * struct.
311  *
312  * tlc_max is the maximum number of concurrent TCP connections created from a
313  * listner.  It is calculated when the tcp_listen_cnt_t is allocated.
314  *
315  * tlc_report_time stores the time when cmn_err() is called to report that the
316  * max has been exceeeded.  Report is done at most once every
317  * TCP_TLC_REPORT_INTERVAL mins for a listener.
318  *
319  * tlc_drop stores the number of connection attempt dropped because the
320  * limit has reached.
321  */
322 typedef struct tcp_listen_cnt_s {
323 	uint32_t	tlc_max;
324 	uint32_t	tlc_cnt;
325 	int64_t		tlc_report_time;
326 	uint32_t	tlc_drop;
327 } tcp_listen_cnt_t;
328 
329 #define	TCP_TLC_REPORT_INTERVAL	(30 * MINUTES)
330 
331 #define	TCP_DECR_LISTEN_CNT(tcp)					\
332 {									\
333 	ASSERT((tcp)->tcp_listen_cnt->tlc_cnt > 0);			\
334 	if (atomic_add_32_nv(&(tcp)->tcp_listen_cnt->tlc_cnt, -1) == 0) \
335 		kmem_free((tcp)->tcp_listen_cnt, sizeof (tcp_listen_cnt_t)); \
336 	(tcp)->tcp_listen_cnt = NULL;					\
337 }
338 
339 /* Increment and decrement the number of connections in tcp_stack_t. */
340 #define	TCPS_CONN_INC(tcps)						\
341 	atomic_inc_64(							\
342 	    (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
343 
344 #define	TCPS_CONN_DEC(tcps)						\
345 	atomic_dec_64(							\
346 	    (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
347 
348 /*
349  * When the system is under memory pressure, stack variable tcps_reclaim is
350  * true, we shorten the connection timeout abort interval to tcp_early_abort
351  * seconds.  Defined in tcp.c.
352  */
353 extern uint32_t tcp_early_abort;
354 
355 /*
356  * To reach to an eager in Q0 which can be dropped due to an incoming
357  * new SYN request when Q0 is full, a new doubly linked list is
358  * introduced. This list allows to select an eager from Q0 in O(1) time.
359  * This is needed to avoid spending too much time walking through the
360  * long list of eagers in Q0 when tcp_drop_q0() is called. Each member of
361  * this new list has to be a member of Q0.
362  * This list is headed by listener's tcp_t. When the list is empty,
363  * both the pointers - tcp_eager_next_drop_q0 and tcp_eager_prev_drop_q0,
364  * of listener's tcp_t point to listener's tcp_t itself.
365  *
366  * Given an eager in Q0 and a listener, MAKE_DROPPABLE() puts the eager
367  * in the list. MAKE_UNDROPPABLE() takes the eager out of the list.
368  * These macros do not affect the eager's membership to Q0.
369  */
370 #define	MAKE_DROPPABLE(listener, eager)					\
371 	if ((eager)->tcp_eager_next_drop_q0 == NULL) {			\
372 		(listener)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0\
373 		    = (eager);						\
374 		(eager)->tcp_eager_prev_drop_q0 = (listener);		\
375 		(eager)->tcp_eager_next_drop_q0 =			\
376 		    (listener)->tcp_eager_next_drop_q0;			\
377 		(listener)->tcp_eager_next_drop_q0 = (eager);		\
378 	}
379 
380 #define	MAKE_UNDROPPABLE(eager)						\
381 	if ((eager)->tcp_eager_next_drop_q0 != NULL) {			\
382 		(eager)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0	\
383 		    = (eager)->tcp_eager_prev_drop_q0;			\
384 		(eager)->tcp_eager_prev_drop_q0->tcp_eager_next_drop_q0	\
385 		    = (eager)->tcp_eager_next_drop_q0;			\
386 		(eager)->tcp_eager_prev_drop_q0 = NULL;			\
387 		(eager)->tcp_eager_next_drop_q0 = NULL;			\
388 	}
389 
390 /*
391  * The format argument to pass to tcp_display().
392  * DISP_PORT_ONLY means that the returned string has only port info.
393  * DISP_ADDR_AND_PORT means that the returned string also contains the
394  * remote and local IP address.
395  */
396 #define	DISP_PORT_ONLY		1
397 #define	DISP_ADDR_AND_PORT	2
398 
399 #define	IP_ADDR_CACHE_SIZE	2048
400 #define	IP_ADDR_CACHE_HASH(faddr)					\
401 	(ntohl(faddr) & (IP_ADDR_CACHE_SIZE -1))
402 
403 /* TCP cwnd burst factor. */
404 #define	TCP_CWND_INFINITE	65535
405 #define	TCP_CWND_SS		3
406 #define	TCP_CWND_NORMAL		5
407 
408 /*
409  * TCP reassembly macros.  We hide starting and ending sequence numbers in
410  * b_next and b_prev of messages on the reassembly queue.  The messages are
411  * chained using b_cont.  These macros are used in tcp_reass() so we don't
412  * have to see the ugly casts and assignments.
413  */
414 #define	TCP_REASS_SEQ(mp)		((uint32_t)(uintptr_t)((mp)->b_next))
415 #define	TCP_REASS_SET_SEQ(mp, u)	((mp)->b_next = \
416 					(mblk_t *)(uintptr_t)(u))
417 #define	TCP_REASS_END(mp)		((uint32_t)(uintptr_t)((mp)->b_prev))
418 #define	TCP_REASS_SET_END(mp, u)	((mp)->b_prev = \
419 					(mblk_t *)(uintptr_t)(u))
420 
421 #define	tcps_time_wait_interval		tcps_propinfo_tbl[0].prop_cur_uval
422 #define	tcps_conn_req_max_q		tcps_propinfo_tbl[1].prop_cur_uval
423 #define	tcps_conn_req_max_q0		tcps_propinfo_tbl[2].prop_cur_uval
424 #define	tcps_conn_req_min		tcps_propinfo_tbl[3].prop_cur_uval
425 #define	tcps_conn_grace_period		tcps_propinfo_tbl[4].prop_cur_uval
426 #define	tcps_cwnd_max_			tcps_propinfo_tbl[5].prop_cur_uval
427 #define	tcps_dbg			tcps_propinfo_tbl[6].prop_cur_uval
428 #define	tcps_smallest_nonpriv_port	tcps_propinfo_tbl[7].prop_cur_uval
429 #define	tcps_ip_abort_cinterval		tcps_propinfo_tbl[8].prop_cur_uval
430 #define	tcps_ip_abort_linterval		tcps_propinfo_tbl[9].prop_cur_uval
431 #define	tcps_ip_abort_interval		tcps_propinfo_tbl[10].prop_cur_uval
432 #define	tcps_ip_notify_cinterval	tcps_propinfo_tbl[11].prop_cur_uval
433 #define	tcps_ip_notify_interval		tcps_propinfo_tbl[12].prop_cur_uval
434 #define	tcps_ipv4_ttl			tcps_propinfo_tbl[13].prop_cur_uval
435 #define	tcps_keepalive_interval_high	tcps_propinfo_tbl[14].prop_max_uval
436 #define	tcps_keepalive_interval		tcps_propinfo_tbl[14].prop_cur_uval
437 #define	tcps_keepalive_interval_low	tcps_propinfo_tbl[14].prop_min_uval
438 #define	tcps_maxpsz_multiplier		tcps_propinfo_tbl[15].prop_cur_uval
439 #define	tcps_mss_def_ipv4		tcps_propinfo_tbl[16].prop_cur_uval
440 #define	tcps_mss_max_ipv4		tcps_propinfo_tbl[17].prop_cur_uval
441 #define	tcps_mss_min			tcps_propinfo_tbl[18].prop_cur_uval
442 #define	tcps_naglim_def			tcps_propinfo_tbl[19].prop_cur_uval
443 #define	tcps_rexmit_interval_initial_high	\
444 					tcps_propinfo_tbl[20].prop_max_uval
445 #define	tcps_rexmit_interval_initial	tcps_propinfo_tbl[20].prop_cur_uval
446 #define	tcps_rexmit_interval_initial_low	\
447 					tcps_propinfo_tbl[20].prop_min_uval
448 #define	tcps_rexmit_interval_max_high	tcps_propinfo_tbl[21].prop_max_uval
449 #define	tcps_rexmit_interval_max	tcps_propinfo_tbl[21].prop_cur_uval
450 #define	tcps_rexmit_interval_max_low	tcps_propinfo_tbl[21].prop_min_uval
451 #define	tcps_rexmit_interval_min_high	tcps_propinfo_tbl[22].prop_max_uval
452 #define	tcps_rexmit_interval_min	tcps_propinfo_tbl[22].prop_cur_uval
453 #define	tcps_rexmit_interval_min_low	tcps_propinfo_tbl[22].prop_min_uval
454 #define	tcps_deferred_ack_interval	tcps_propinfo_tbl[23].prop_cur_uval
455 #define	tcps_snd_lowat_fraction		tcps_propinfo_tbl[24].prop_cur_uval
456 #define	tcps_dupack_fast_retransmit	tcps_propinfo_tbl[25].prop_cur_uval
457 #define	tcps_ignore_path_mtu		tcps_propinfo_tbl[26].prop_cur_bval
458 #define	tcps_smallest_anon_port		tcps_propinfo_tbl[27].prop_cur_uval
459 #define	tcps_largest_anon_port		tcps_propinfo_tbl[28].prop_cur_uval
460 #define	tcps_xmit_hiwat			tcps_propinfo_tbl[29].prop_cur_uval
461 #define	tcps_xmit_lowat			tcps_propinfo_tbl[30].prop_cur_uval
462 #define	tcps_recv_hiwat			tcps_propinfo_tbl[31].prop_cur_uval
463 #define	tcps_recv_hiwat_minmss		tcps_propinfo_tbl[32].prop_cur_uval
464 #define	tcps_fin_wait_2_flush_interval_high	\
465 					tcps_propinfo_tbl[33].prop_max_uval
466 #define	tcps_fin_wait_2_flush_interval	tcps_propinfo_tbl[33].prop_cur_uval
467 #define	tcps_fin_wait_2_flush_interval_low	\
468 					tcps_propinfo_tbl[33].prop_min_uval
469 #define	tcps_max_buf			tcps_propinfo_tbl[34].prop_cur_uval
470 #define	tcps_strong_iss			tcps_propinfo_tbl[35].prop_cur_uval
471 #define	tcps_rtt_updates		tcps_propinfo_tbl[36].prop_cur_uval
472 #define	tcps_wscale_always		tcps_propinfo_tbl[37].prop_cur_bval
473 #define	tcps_tstamp_always		tcps_propinfo_tbl[38].prop_cur_bval
474 #define	tcps_tstamp_if_wscale		tcps_propinfo_tbl[39].prop_cur_bval
475 #define	tcps_rexmit_interval_extra	tcps_propinfo_tbl[40].prop_cur_uval
476 #define	tcps_deferred_acks_max		tcps_propinfo_tbl[41].prop_cur_uval
477 #define	tcps_slow_start_after_idle	tcps_propinfo_tbl[42].prop_cur_uval
478 #define	tcps_slow_start_initial		tcps_propinfo_tbl[43].prop_cur_uval
479 #define	tcps_sack_permitted		tcps_propinfo_tbl[44].prop_cur_uval
480 #define	tcps_ipv6_hoplimit		tcps_propinfo_tbl[45].prop_cur_uval
481 #define	tcps_mss_def_ipv6		tcps_propinfo_tbl[46].prop_cur_uval
482 #define	tcps_mss_max_ipv6		tcps_propinfo_tbl[47].prop_cur_uval
483 #define	tcps_rev_src_routes		tcps_propinfo_tbl[48].prop_cur_bval
484 #define	tcps_local_dack_interval	tcps_propinfo_tbl[49].prop_cur_uval
485 #define	tcps_local_dacks_max		tcps_propinfo_tbl[50].prop_cur_uval
486 #define	tcps_ecn_permitted		tcps_propinfo_tbl[51].prop_cur_uval
487 #define	tcps_rst_sent_rate_enabled	tcps_propinfo_tbl[52].prop_cur_bval
488 #define	tcps_rst_sent_rate		tcps_propinfo_tbl[53].prop_cur_uval
489 #define	tcps_push_timer_interval	tcps_propinfo_tbl[54].prop_cur_uval
490 #define	tcps_use_smss_as_mss_opt	tcps_propinfo_tbl[55].prop_cur_bval
491 #define	tcps_keepalive_abort_interval_high \
492 					tcps_propinfo_tbl[56].prop_max_uval
493 #define	tcps_keepalive_abort_interval \
494 					tcps_propinfo_tbl[56].prop_cur_uval
495 #define	tcps_keepalive_abort_interval_low \
496 					tcps_propinfo_tbl[56].prop_min_uval
497 #define	tcps_wroff_xtra			tcps_propinfo_tbl[57].prop_cur_uval
498 #define	tcps_dev_flow_ctl		tcps_propinfo_tbl[58].prop_cur_bval
499 #define	tcps_reass_timeout		tcps_propinfo_tbl[59].prop_cur_uval
500 
501 extern struct qinit tcp_rinitv4, tcp_rinitv6;
502 extern boolean_t do_tcp_fusion;
503 
504 /*
505  * Object to represent database of options to search passed to
506  * {sock,tpi}optcom_req() interface routine to take care of option
507  * management and associated methods.
508  */
509 extern optdb_obj_t	tcp_opt_obj;
510 extern uint_t		tcp_max_optsize;
511 
512 extern int tcp_squeue_flag;
513 
514 extern uint_t tcp_free_list_max_cnt;
515 
516 /*
517  * Functions in tcp.c.
518  */
519 extern void	tcp_acceptor_hash_insert(t_uscalar_t, tcp_t *);
520 extern tcp_t	*tcp_acceptor_hash_lookup(t_uscalar_t, tcp_stack_t *);
521 extern void	tcp_acceptor_hash_remove(tcp_t *);
522 extern mblk_t	*tcp_ack_mp(tcp_t *);
523 extern int	tcp_build_hdrs(tcp_t *);
524 extern void	tcp_cleanup(tcp_t *);
525 extern int	tcp_clean_death(tcp_t *, int);
526 extern void	tcp_clean_death_wrapper(void *, mblk_t *, void *,
527 		    ip_recv_attr_t *);
528 extern void	tcp_close_common(conn_t *, int);
529 extern void	tcp_close_detached(tcp_t *);
530 extern void	tcp_close_mpp(mblk_t **);
531 extern void	tcp_closei_local(tcp_t *);
532 extern sock_lower_handle_t tcp_create(int, int, int, sock_downcalls_t **,
533 		    uint_t *, int *, int, cred_t *);
534 extern conn_t	*tcp_create_common(cred_t *, boolean_t, boolean_t, int *);
535 extern void	tcp_disconnect(tcp_t *, mblk_t *);
536 extern char	*tcp_display(tcp_t *, char *, char);
537 extern int	tcp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
538 		    boolean_t);
539 extern int	tcp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
540 		    cred_t *, pid_t);
541 extern int	tcp_do_listen(conn_t *, struct sockaddr *, socklen_t, int,
542 		    cred_t *, boolean_t);
543 extern int	tcp_do_unbind(conn_t *);
544 extern boolean_t	tcp_eager_blowoff(tcp_t *, t_scalar_t);
545 extern void	tcp_eager_cleanup(tcp_t *, boolean_t);
546 extern void	tcp_eager_kill(void *, mblk_t *, void *, ip_recv_attr_t *);
547 extern void	tcp_eager_unlink(tcp_t *);
548 extern int	tcp_getpeername(sock_lower_handle_t, struct sockaddr *,
549 		    socklen_t *, cred_t *);
550 extern int	tcp_getsockname(sock_lower_handle_t, struct sockaddr *,
551 		    socklen_t *, cred_t *);
552 extern void	tcp_init_values(tcp_t *, tcp_t *);
553 extern void	tcp_ipsec_cleanup(tcp_t *);
554 extern int	tcp_maxpsz_set(tcp_t *, boolean_t);
555 extern void	tcp_mss_set(tcp_t *, uint32_t);
556 extern void	tcp_reinput(conn_t *, mblk_t *, ip_recv_attr_t *, ip_stack_t *);
557 extern void	tcp_rsrv(queue_t *);
558 extern uint_t	tcp_rwnd_reopen(tcp_t *);
559 extern int	tcp_rwnd_set(tcp_t *, uint32_t);
560 extern int	tcp_set_destination(tcp_t *);
561 extern void	tcp_set_ws_value(tcp_t *);
562 extern void	tcp_stop_lingering(tcp_t *);
563 extern void	tcp_update_pmtu(tcp_t *, boolean_t);
564 extern mblk_t	*tcp_zcopy_backoff(tcp_t *, mblk_t *, boolean_t);
565 extern boolean_t	tcp_zcopy_check(tcp_t *);
566 extern void	tcp_zcopy_notify(tcp_t *);
567 extern void	tcp_get_proto_props(tcp_t *, struct sock_proto_props *);
568 
569 /*
570  * Bind related functions in tcp_bind.c
571  */
572 extern int	tcp_bind_check(conn_t *, struct sockaddr *, socklen_t,
573 		    cred_t *, boolean_t);
574 extern void	tcp_bind_hash_insert(tf_t *, tcp_t *, int);
575 extern void	tcp_bind_hash_remove(tcp_t *);
576 extern in_port_t	tcp_bindi(tcp_t *, in_port_t, const in6_addr_t *,
577 			    int, boolean_t, boolean_t, boolean_t);
578 extern in_port_t	tcp_update_next_port(in_port_t, const tcp_t *,
579 			    boolean_t);
580 
581 /*
582  * Fusion related functions in tcp_fusion.c.
583  */
584 extern void	tcp_fuse(tcp_t *, uchar_t *, tcpha_t *);
585 extern void	tcp_unfuse(tcp_t *);
586 extern boolean_t tcp_fuse_output(tcp_t *, mblk_t *, uint32_t);
587 extern void	tcp_fuse_output_urg(tcp_t *, mblk_t *);
588 extern boolean_t tcp_fuse_rcv_drain(queue_t *, tcp_t *, mblk_t **);
589 extern size_t	tcp_fuse_set_rcv_hiwat(tcp_t *, size_t);
590 extern int	tcp_fuse_maxpsz(tcp_t *);
591 extern void	tcp_fuse_backenable(tcp_t *);
592 extern void	tcp_iss_key_init(uint8_t *, int, tcp_stack_t *);
593 
594 /*
595  * Output related functions in tcp_output.c.
596  */
597 extern void	tcp_close_output(void *, mblk_t *, void *, ip_recv_attr_t *);
598 extern void	tcp_output(void *, mblk_t *, void *, ip_recv_attr_t *);
599 extern void	tcp_output_urgent(void *, mblk_t *, void *, ip_recv_attr_t *);
600 extern void	tcp_rexmit_after_error(tcp_t *);
601 extern void	tcp_sack_rexmit(tcp_t *, uint_t *);
602 extern void	tcp_send_data(tcp_t *, mblk_t *);
603 extern void	tcp_send_synack(void *, mblk_t *, void *, ip_recv_attr_t *);
604 extern void	tcp_shutdown_output(void *, mblk_t *, void *, ip_recv_attr_t *);
605 extern void	tcp_ss_rexmit(tcp_t *);
606 extern void	tcp_update_xmit_tail(tcp_t *, uint32_t);
607 extern void	tcp_wput(queue_t *, mblk_t *);
608 extern void	tcp_wput_data(tcp_t *, mblk_t *, boolean_t);
609 extern void	tcp_wput_sock(queue_t *, mblk_t *);
610 extern void	tcp_wput_fallback(queue_t *, mblk_t *);
611 extern void	tcp_xmit_ctl(char *, tcp_t *, uint32_t, uint32_t, int);
612 extern void	tcp_xmit_listeners_reset(mblk_t *, ip_recv_attr_t *,
613 		    ip_stack_t *i, conn_t *);
614 extern mblk_t	*tcp_xmit_mp(tcp_t *, mblk_t *, int32_t, int32_t *,
615 		    mblk_t **, uint32_t, boolean_t, uint32_t *, boolean_t);
616 
617 /*
618  * Input related functions in tcp_input.c.
619  */
620 extern void	tcp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
621 extern void	tcp_input_data(void *, mblk_t *, void *, ip_recv_attr_t *);
622 extern void	tcp_input_listener_unbound(void *, mblk_t *, void *,
623 		    ip_recv_attr_t *);
624 extern boolean_t	tcp_paws_check(tcp_t *, tcpha_t *, tcp_opt_t *);
625 extern uint_t	tcp_rcv_drain(tcp_t *);
626 extern void	tcp_rcv_enqueue(tcp_t *, mblk_t *, uint_t, cred_t *);
627 extern boolean_t	tcp_verifyicmp(conn_t *, void *, icmph_t *, icmp6_t *,
628 			    ip_recv_attr_t *);
629 
630 /*
631  * Kernel socket related functions in tcp_socket.c.
632  */
633 extern int	tcp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
634 		    so_proto_quiesced_cb_t, sock_quiesce_arg_t *);
635 extern boolean_t tcp_newconn_notify(tcp_t *, ip_recv_attr_t *);
636 
637 /*
638  * Timer related functions in tcp_timers.c.
639  */
640 extern void	tcp_ack_timer(void *);
641 extern void	tcp_close_linger_timeout(void *);
642 extern void	tcp_keepalive_timer(void *);
643 extern void	tcp_push_timer(void *);
644 extern void	tcp_reass_timer(void *);
645 extern mblk_t	*tcp_timermp_alloc(int);
646 extern void	tcp_timermp_free(tcp_t *);
647 extern timeout_id_t tcp_timeout(conn_t *, void (*)(void *), hrtime_t);
648 extern clock_t	tcp_timeout_cancel(conn_t *, timeout_id_t);
649 extern void	tcp_timer(void *arg);
650 extern void	tcp_timers_stop(tcp_t *);
651 
652 /*
653  * TCP TPI related functions in tcp_tpi.c.
654  */
655 extern void	tcp_addr_req(tcp_t *, mblk_t *);
656 extern void	tcp_capability_req(tcp_t *, mblk_t *);
657 extern boolean_t	tcp_conn_con(tcp_t *, uchar_t *, mblk_t *,
658 			    mblk_t **, ip_recv_attr_t *);
659 extern void	tcp_err_ack(tcp_t *, mblk_t *, int, int);
660 extern void	tcp_err_ack_prim(tcp_t *, mblk_t *, int, int, int);
661 extern void	tcp_info_req(tcp_t *, mblk_t *);
662 extern void	tcp_send_conn_ind(void *, mblk_t *, void *);
663 extern void	tcp_send_pending(void *, mblk_t *, void *, ip_recv_attr_t *);
664 extern void	tcp_tpi_accept(queue_t *, mblk_t *);
665 extern void	tcp_tpi_bind(tcp_t *, mblk_t *);
666 extern int	tcp_tpi_close(queue_t *, int);
667 extern int	tcp_tpi_close_accept(queue_t *);
668 extern void	tcp_tpi_connect(tcp_t *, mblk_t *);
669 extern int	tcp_tpi_opt_get(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
670 extern int	tcp_tpi_opt_set(queue_t *, uint_t, int, int, uint_t, uchar_t *,
671 		    uint_t *, uchar_t *, void *, cred_t *);
672 extern void	tcp_tpi_unbind(tcp_t *, mblk_t *);
673 extern void	tcp_tli_accept(tcp_t *, mblk_t *);
674 extern void	tcp_use_pure_tpi(tcp_t *);
675 extern void	tcp_do_capability_ack(tcp_t *, struct T_capability_ack *,
676 		    t_uscalar_t);
677 
678 /*
679  * TCP option processing related functions in tcp_opt_data.c
680  */
681 extern int	tcp_opt_default(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
682 extern int	tcp_opt_get(conn_t *, int, int, uchar_t *);
683 extern int	tcp_opt_set(conn_t *, uint_t, int, int, uint_t, uchar_t *,
684 		    uint_t *, uchar_t *, void *, cred_t *);
685 
686 /*
687  * TCP time wait processing related functions in tcp_time_wait.c.
688  */
689 extern void		tcp_time_wait_append(tcp_t *);
690 extern void		tcp_time_wait_collector(void *);
691 extern boolean_t	tcp_time_wait_remove(tcp_t *, tcp_squeue_priv_t *);
692 extern void		tcp_time_wait_processing(tcp_t *, mblk_t *, uint32_t,
693 			    uint32_t, int, tcpha_t *, ip_recv_attr_t *);
694 
695 /*
696  * Misc functions in tcp_misc.c.
697  */
698 extern uint32_t	tcp_find_listener_conf(tcp_stack_t *, in_port_t);
699 extern void	tcp_ioctl_abort_conn(queue_t *, mblk_t *);
700 extern void	tcp_listener_conf_cleanup(tcp_stack_t *);
701 extern void	tcp_stack_cpu_add(tcp_stack_t *, processorid_t);
702 
703 #endif	/* _KERNEL */
704 
705 #ifdef	__cplusplus
706 }
707 #endif
708 
709 #endif	/* _INET_TCP_IMPL_H */
710