xref: /titanic_52/usr/src/uts/common/inet/tcp_impl.h (revision b49b27dcb66b2c7f4a23f7bc158e2dde5cd79030)
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  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
25  */
26 
27 #ifndef	_INET_TCP_IMPL_H
28 #define	_INET_TCP_IMPL_H
29 
30 /*
31  * TCP implementation private declarations.  These interfaces are
32  * used to build the IP module and are not meant to be accessed
33  * by any modules except IP itself.  They are undocumented and are
34  * subject to change without notice.
35  */
36 
37 #ifdef	__cplusplus
38 extern "C" {
39 #endif
40 
41 #ifdef _KERNEL
42 
43 #include <sys/cpuvar.h>
44 #include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */
45 #include <inet/optcom.h>
46 #include <inet/tcp.h>
47 #include <inet/tunables.h>
48 
49 #define	TCP_MOD_ID	5105
50 
51 extern struct qinit	tcp_sock_winit;
52 extern struct qinit	tcp_winit;
53 
54 extern sock_downcalls_t sock_tcp_downcalls;
55 
56 /*
57  * Bind hash list size and has function.  It has to be a power of 2 for
58  * hashing.
59  */
60 #define	TCP_BIND_FANOUT_SIZE	1024
61 #define	TCP_BIND_HASH(lport) (ntohs(lport) & (TCP_BIND_FANOUT_SIZE - 1))
62 
63 /*
64  * This implementation follows the 4.3BSD interpretation of the urgent
65  * pointer and not RFC 1122. Switching to RFC 1122 behavior would cause
66  * incompatible changes in protocols like telnet and rlogin.
67  */
68 #define	TCP_OLD_URP_INTERPRETATION	1
69 
70 /* TCP option length */
71 #define	TCPOPT_NOP_LEN		1
72 #define	TCPOPT_MAXSEG_LEN	4
73 #define	TCPOPT_WS_LEN		3
74 #define	TCPOPT_REAL_WS_LEN	(TCPOPT_WS_LEN+1)
75 #define	TCPOPT_TSTAMP_LEN	10
76 #define	TCPOPT_REAL_TS_LEN	(TCPOPT_TSTAMP_LEN+2)
77 #define	TCPOPT_SACK_OK_LEN	2
78 #define	TCPOPT_REAL_SACK_OK_LEN	(TCPOPT_SACK_OK_LEN+2)
79 #define	TCPOPT_REAL_SACK_LEN	4
80 #define	TCPOPT_MAX_SACK_LEN	36
81 #define	TCPOPT_HEADER_LEN	2
82 
83 /* Round up the value to the nearest mss. */
84 #define	MSS_ROUNDUP(value, mss)		((((value) - 1) / (mss) + 1) * (mss))
85 
86 /*
87  * Was this tcp created via socket() interface?
88  */
89 #define	TCP_IS_SOCKET(tcp)	((tcp)->tcp_issocket)
90 
91 /*
92  * Is this tcp not attached to any upper client?
93  */
94 #define	TCP_IS_DETACHED(tcp)	((tcp)->tcp_detached)
95 
96 /* TCP timers related data strucutres.  Refer to tcp_timers.c. */
97 typedef struct tcp_timer_s {
98 	conn_t	*connp;
99 	void 	(*tcpt_proc)(void *);
100 	callout_id_t   tcpt_tid;
101 } tcp_timer_t;
102 
103 extern kmem_cache_t *tcp_timercache;
104 
105 /*
106  * Macro for starting various timers.  Retransmission timer has its own macro,
107  * TCP_TIMER_RESTART().  tim is in millisec.
108  */
109 #define	TCP_TIMER(tcp, f, tim)		\
110 	tcp_timeout(tcp->tcp_connp, f, tim)
111 #define	TCP_TIMER_CANCEL(tcp, id)	\
112 	tcp_timeout_cancel(tcp->tcp_connp, id)
113 
114 /*
115  * To restart the TCP retransmission timer.  intvl is in millisec.
116  */
117 #define	TCP_TIMER_RESTART(tcp, intvl) {					\
118 	if ((tcp)->tcp_timer_tid != 0)					\
119 		(void) TCP_TIMER_CANCEL((tcp), (tcp)->tcp_timer_tid);	\
120 	(tcp)->tcp_timer_tid = TCP_TIMER((tcp), tcp_timer, (intvl));	\
121 }
122 
123 /*
124  * For scalability, we must not run a timer for every TCP connection
125  * in TIME_WAIT state.  To see why, consider (for time wait interval of
126  * 1 minutes):
127  *	10,000 connections/sec * 60 seconds/time wait = 600,000 active conn's
128  *
129  * This list is ordered by time, so you need only delete from the head
130  * until you get to entries which aren't old enough to delete yet.
131  * The list consists of only the detached TIME_WAIT connections.
132  *
133  * When a tcp_t enters TIME_WAIT state, a timer is started (timeout is
134  * tcps_time_wait_interval).  When the tcp_t is detached (upper layer closes
135  * the end point), it is moved to the time wait list and another timer is
136  * started (expiry time is set at tcp_time_wait_expire, which is
137  * also calculated using tcps_time_wait_interval).  This means that the
138  * TIME_WAIT state can be extended (up to doubled) if the tcp_t doesn't
139  * become detached for a long time.
140  *
141  * The list manipulations (including tcp_time_wait_next/prev)
142  * are protected by the tcp_time_wait_lock. The content of the
143  * detached TIME_WAIT connections is protected by the normal perimeters.
144  *
145  * This list is per squeue and squeues are shared across the tcp_stack_t's.
146  * Things on tcp_time_wait_head remain associated with the tcp_stack_t
147  * and conn_netstack.
148  * The tcp_t's that are added to tcp_free_list are disassociated and
149  * have NULL tcp_tcps and conn_netstack pointers.
150  */
151 typedef struct tcp_squeue_priv_s {
152 	kmutex_t	tcp_time_wait_lock;
153 	callout_id_t	tcp_time_wait_tid;
154 	tcp_t		*tcp_time_wait_head;
155 	tcp_t		*tcp_time_wait_tail;
156 	tcp_t		*tcp_free_list;
157 	uint_t		tcp_free_list_cnt;
158 #ifdef DEBUG
159 	/*
160 	 * For debugging purpose, true when tcp_time_wait_collector() is
161 	 * running.
162 	 */
163 	boolean_t	tcp_time_wait_running;
164 #endif
165 } tcp_squeue_priv_t;
166 
167 /*
168  * Parameters for TCP Initial Send Sequence number (ISS) generation.  When
169  * tcp_strong_iss is set to 1, which is the default, the ISS is calculated
170  * by adding three components: a time component which grows by 1 every 4096
171  * nanoseconds (versus every 4 microseconds suggested by RFC 793, page 27);
172  * a per-connection component which grows by 125000 for every new connection;
173  * and an "extra" component that grows by a random amount centered
174  * approximately on 64000.  This causes the ISS generator to cycle every
175  * 4.89 hours if no TCP connections are made, and faster if connections are
176  * made.
177  *
178  * When tcp_strong_iss is set to 0, ISS is calculated by adding two
179  * components: a time component which grows by 250000 every second; and
180  * a per-connection component which grows by 125000 for every new connections.
181  *
182  * A third method, when tcp_strong_iss is set to 2, for generating ISS is
183  * prescribed by Steve Bellovin.  This involves adding time, the 125000 per
184  * connection, and a one-way hash (MD5) of the connection ID <sport, dport,
185  * src, dst>, a "truly" random (per RFC 1750) number, and a console-entered
186  * password.
187  */
188 #define	ISS_INCR	250000
189 #define	ISS_NSEC_SHT	12
190 
191 /* Macros for timestamp comparisons */
192 #define	TSTMP_GEQ(a, b)	((int32_t)((a)-(b)) >= 0)
193 #define	TSTMP_LT(a, b)	((int32_t)((a)-(b)) < 0)
194 
195 /*
196  * Initialize cwnd according to RFC 3390.  def_max_init_cwnd is
197  * either tcp_slow_start_initial or tcp_slow_start_after idle
198  * depending on the caller.  If the upper layer has not used the
199  * TCP_INIT_CWND option to change the initial cwnd, tcp_init_cwnd
200  * should be 0 and we use the formula in RFC 3390 to set tcp_cwnd.
201  * If the upper layer has changed set the tcp_init_cwnd, just use
202  * it to calculate the tcp_cwnd.
203  *
204  * "An Argument for Increasing TCP's Initial Congestion Window"
205  * ACM SIGCOMM Computer Communications Review, vol. 40 (2010), pp. 27-33
206  *  -- Nandita Dukkipati, Tiziana Refice, Yuchung Cheng,
207  *     Hsiao-keng Jerry Chu, Tom Herbert, Amit Agarwal,
208  *     Arvind Jain, Natalia Sutin
209  *
210  *   "Based on the results from our experiments, we believe the
211  *    initial congestion window should be at least ten segments
212  *    and the same be investigated for standardization by the IETF."
213  *
214  * As such, the def_max_init_cwnd argument with which this macro is
215  * invoked is either the tcps_slow_start_initial or
216  * tcps_slow_start_after_idle which both default to 0 and will respect
217  * RFC 3390 exactly.  If the tunables are explicitly set by the operator,
218  * then the initial congestion window should be set as the operator
219  * demands, within reason. We shall arbitrarily define reason as a
220  * maximum of 16 (same as used by the TCP_INIT_CWND setsockopt).
221  */
222 
223 /* Maximum TCP initial cwin (start/restart). */
224 #define	TCP_MAX_INIT_CWND	16
225 
226 #define	TCP_SET_INIT_CWND(tcp, mss, def_max_init_cwnd)			\
227 {									\
228 	if ((tcp)->tcp_init_cwnd == 0) {				\
229 		if (def_max_init_cwnd == 0) {				\
230 			(tcp)->tcp_cwnd = MIN(4 * (mss),		\
231 			    MAX(2 * (mss), 4380 / (mss) * (mss)));	\
232 		} else {						\
233 			(tcp)->tcp_cwnd = MIN(TCP_MAX_INIT_CWND * (mss),\
234 			    def_max_init_cwnd * (mss));			\
235 		}							\
236 	} else {							\
237 		(tcp)->tcp_cwnd = (tcp)->tcp_init_cwnd * (mss);		\
238 	}								\
239 	tcp->tcp_cwnd_cnt = 0;						\
240 }
241 
242 /*
243  * Set ECN capable transport (ECT) code point in IP header.
244  *
245  * Note that there are 2 ECT code points '01' and '10', which are called
246  * ECT(1) and ECT(0) respectively.  Here we follow the original ECT code
247  * point ECT(0) for TCP as described in RFC 2481.
248  */
249 #define	TCP_SET_ECT(tcp, iph) \
250 	if ((tcp)->tcp_connp->conn_ipversion == IPV4_VERSION) { \
251 		/* We need to clear the code point first. */ \
252 		((ipha_t *)(iph))->ipha_type_of_service &= 0xFC; \
253 		((ipha_t *)(iph))->ipha_type_of_service |= IPH_ECN_ECT0; \
254 	} else { \
255 		((ip6_t *)(iph))->ip6_vcf &= htonl(0xFFCFFFFF); \
256 		((ip6_t *)(iph))->ip6_vcf |= htonl(IPH_ECN_ECT0 << 20); \
257 	}
258 
259 /*
260  * Set tcp_rto with boundary checking.
261  */
262 #define	TCP_SET_RTO(tcp, rto) \
263 	if ((rto) < (tcp)->tcp_rto_min)			\
264 		(tcp)->tcp_rto = (tcp)->tcp_rto_min;	\
265 	else if ((rto) > (tcp)->tcp_rto_max)		\
266 		(tcp)->tcp_rto = (tcp)->tcp_rto_max;	\
267 	else						\
268 		(tcp)->tcp_rto = (rto);
269 
270 /*
271  * TCP options struct returned from tcp_parse_options.
272  */
273 typedef struct tcp_opt_s {
274 	uint32_t	tcp_opt_mss;
275 	uint32_t	tcp_opt_wscale;
276 	uint32_t	tcp_opt_ts_val;
277 	uint32_t	tcp_opt_ts_ecr;
278 	tcp_t		*tcp;
279 } tcp_opt_t;
280 
281 /*
282  * Write-side flow-control is implemented via the per instance STREAMS
283  * write-side Q by explicitly setting QFULL to stop the flow of mblk_t(s)
284  * and clearing QFULL and calling qbackenable() to restart the flow based
285  * on the number of TCP unsent bytes (i.e. those not on the wire waiting
286  * for a remote ACK).
287  *
288  * This is different than a standard STREAMS kmod which when using the
289  * STREAMS Q the framework would automatictly flow-control based on the
290  * defined hiwat/lowat values as mblk_t's are enqueued/dequeued.
291  *
292  * As of FireEngine TCP write-side flow-control needs to take into account
293  * both the unsent tcp_xmit list bytes but also any squeue_t enqueued bytes
294  * (i.e. from tcp_wput() -> tcp_output()).
295  *
296  * This is accomplished by adding a new tcp_t fields, tcp_squeue_bytes, to
297  * count the number of bytes enqueued by tcp_wput() and the number of bytes
298  * dequeued and processed by tcp_output().
299  *
300  * So, the total number of bytes unsent is (squeue_bytes + unsent) with all
301  * flow-control uses of unsent replaced with the macro TCP_UNSENT_BYTES.
302  */
303 extern void	tcp_clrqfull(tcp_t *);
304 extern void	tcp_setqfull(tcp_t *);
305 
306 #define	TCP_UNSENT_BYTES(tcp) \
307 	((tcp)->tcp_squeue_bytes + (tcp)->tcp_unsent)
308 
309 /*
310  * Linked list struct to store listener connection limit configuration per
311  * IP stack.  The list is stored at tcps_listener_conf in tcp_stack_t.
312  *
313  * tl_port: the listener port of this limit configuration
314  * tl_ratio: the maximum amount of memory consumed by all concurrent TCP
315  *           connections created by a listener does not exceed 1/tl_ratio
316  *           of the total system memory.  Note that this is only an
317  *           approximation.
318  * tl_link: linked list struct
319  */
320 typedef struct tcp_listener_s {
321 	in_port_t	tl_port;
322 	uint32_t	tl_ratio;
323 	list_node_t	tl_link;
324 } tcp_listener_t;
325 
326 /*
327  * If there is a limit set on the number of connections allowed per each
328  * listener, the following struct is used to store that counter.  It keeps
329  * the number of TCP connection created by a listener.  Note that this needs
330  * to be separated from the listener since the listener can go away before
331  * all the connections are gone.
332  *
333  * When the struct is allocated, tlc_cnt is set to 1.  When a new connection
334  * is created by the listener, tlc_cnt is incremented by 1.  When a connection
335  * created by the listener goes away, tlc_count is decremented by 1.  When the
336  * listener itself goes away, tlc_cnt is decremented  by one.  The last
337  * connection (or the listener) which decrements tlc_cnt to zero frees the
338  * struct.
339  *
340  * tlc_max is the maximum number of concurrent TCP connections created from a
341  * listner.  It is calculated when the tcp_listen_cnt_t is allocated.
342  *
343  * tlc_report_time stores the time when cmn_err() is called to report that the
344  * max has been exceeeded.  Report is done at most once every
345  * TCP_TLC_REPORT_INTERVAL mins for a listener.
346  *
347  * tlc_drop stores the number of connection attempt dropped because the
348  * limit has reached.
349  */
350 typedef struct tcp_listen_cnt_s {
351 	uint32_t	tlc_max;
352 	uint32_t	tlc_cnt;
353 	int64_t		tlc_report_time;
354 	uint32_t	tlc_drop;
355 } tcp_listen_cnt_t;
356 
357 #define	TCP_TLC_REPORT_INTERVAL	(30 * MINUTES)
358 
359 #define	TCP_DECR_LISTEN_CNT(tcp)					\
360 {									\
361 	ASSERT((tcp)->tcp_listen_cnt->tlc_cnt > 0);			\
362 	if (atomic_add_32_nv(&(tcp)->tcp_listen_cnt->tlc_cnt, -1) == 0) \
363 		kmem_free((tcp)->tcp_listen_cnt, sizeof (tcp_listen_cnt_t)); \
364 	(tcp)->tcp_listen_cnt = NULL;					\
365 }
366 
367 /* Increment and decrement the number of connections in tcp_stack_t. */
368 #define	TCPS_CONN_INC(tcps)						\
369 	atomic_inc_64(							\
370 	    (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
371 
372 #define	TCPS_CONN_DEC(tcps)						\
373 	atomic_dec_64(							\
374 	    (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
375 
376 /*
377  * When the system is under memory pressure, stack variable tcps_reclaim is
378  * true, we shorten the connection timeout abort interval to tcp_early_abort
379  * seconds.  Defined in tcp.c.
380  */
381 extern uint32_t tcp_early_abort;
382 
383 /*
384  * To reach to an eager in Q0 which can be dropped due to an incoming
385  * new SYN request when Q0 is full, a new doubly linked list is
386  * introduced. This list allows to select an eager from Q0 in O(1) time.
387  * This is needed to avoid spending too much time walking through the
388  * long list of eagers in Q0 when tcp_drop_q0() is called. Each member of
389  * this new list has to be a member of Q0.
390  * This list is headed by listener's tcp_t. When the list is empty,
391  * both the pointers - tcp_eager_next_drop_q0 and tcp_eager_prev_drop_q0,
392  * of listener's tcp_t point to listener's tcp_t itself.
393  *
394  * Given an eager in Q0 and a listener, MAKE_DROPPABLE() puts the eager
395  * in the list. MAKE_UNDROPPABLE() takes the eager out of the list.
396  * These macros do not affect the eager's membership to Q0.
397  */
398 #define	MAKE_DROPPABLE(listener, eager)					\
399 	if ((eager)->tcp_eager_next_drop_q0 == NULL) {			\
400 		(listener)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0\
401 		    = (eager);						\
402 		(eager)->tcp_eager_prev_drop_q0 = (listener);		\
403 		(eager)->tcp_eager_next_drop_q0 =			\
404 		    (listener)->tcp_eager_next_drop_q0;			\
405 		(listener)->tcp_eager_next_drop_q0 = (eager);		\
406 	}
407 
408 #define	MAKE_UNDROPPABLE(eager)						\
409 	if ((eager)->tcp_eager_next_drop_q0 != NULL) {			\
410 		(eager)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0	\
411 		    = (eager)->tcp_eager_prev_drop_q0;			\
412 		(eager)->tcp_eager_prev_drop_q0->tcp_eager_next_drop_q0	\
413 		    = (eager)->tcp_eager_next_drop_q0;			\
414 		(eager)->tcp_eager_prev_drop_q0 = NULL;			\
415 		(eager)->tcp_eager_next_drop_q0 = NULL;			\
416 	}
417 
418 /*
419  * The format argument to pass to tcp_display().
420  * DISP_PORT_ONLY means that the returned string has only port info.
421  * DISP_ADDR_AND_PORT means that the returned string also contains the
422  * remote and local IP address.
423  */
424 #define	DISP_PORT_ONLY		1
425 #define	DISP_ADDR_AND_PORT	2
426 
427 #define	IP_ADDR_CACHE_SIZE	2048
428 #define	IP_ADDR_CACHE_HASH(faddr)					\
429 	(ntohl(faddr) & (IP_ADDR_CACHE_SIZE -1))
430 
431 /* TCP cwnd burst factor. */
432 #define	TCP_CWND_INFINITE	65535
433 #define	TCP_CWND_SS		3
434 #define	TCP_CWND_NORMAL		5
435 
436 /*
437  * TCP reassembly macros.  We hide starting and ending sequence numbers in
438  * b_next and b_prev of messages on the reassembly queue.  The messages are
439  * chained using b_cont.  These macros are used in tcp_reass() so we don't
440  * have to see the ugly casts and assignments.
441  */
442 #define	TCP_REASS_SEQ(mp)		((uint32_t)(uintptr_t)((mp)->b_next))
443 #define	TCP_REASS_SET_SEQ(mp, u)	((mp)->b_next = \
444 					(mblk_t *)(uintptr_t)(u))
445 #define	TCP_REASS_END(mp)		((uint32_t)(uintptr_t)((mp)->b_prev))
446 #define	TCP_REASS_SET_END(mp, u)	((mp)->b_prev = \
447 					(mblk_t *)(uintptr_t)(u))
448 
449 #define	tcps_time_wait_interval		tcps_propinfo_tbl[0].prop_cur_uval
450 #define	tcps_conn_req_max_q		tcps_propinfo_tbl[1].prop_cur_uval
451 #define	tcps_conn_req_max_q0		tcps_propinfo_tbl[2].prop_cur_uval
452 #define	tcps_conn_req_min		tcps_propinfo_tbl[3].prop_cur_uval
453 #define	tcps_conn_grace_period		tcps_propinfo_tbl[4].prop_cur_uval
454 #define	tcps_cwnd_max_			tcps_propinfo_tbl[5].prop_cur_uval
455 #define	tcps_dbg			tcps_propinfo_tbl[6].prop_cur_uval
456 #define	tcps_smallest_nonpriv_port	tcps_propinfo_tbl[7].prop_cur_uval
457 #define	tcps_ip_abort_cinterval		tcps_propinfo_tbl[8].prop_cur_uval
458 #define	tcps_ip_abort_linterval		tcps_propinfo_tbl[9].prop_cur_uval
459 #define	tcps_ip_abort_interval		tcps_propinfo_tbl[10].prop_cur_uval
460 #define	tcps_ip_notify_cinterval	tcps_propinfo_tbl[11].prop_cur_uval
461 #define	tcps_ip_notify_interval		tcps_propinfo_tbl[12].prop_cur_uval
462 #define	tcps_ipv4_ttl			tcps_propinfo_tbl[13].prop_cur_uval
463 #define	tcps_keepalive_interval_high	tcps_propinfo_tbl[14].prop_max_uval
464 #define	tcps_keepalive_interval		tcps_propinfo_tbl[14].prop_cur_uval
465 #define	tcps_keepalive_interval_low	tcps_propinfo_tbl[14].prop_min_uval
466 #define	tcps_maxpsz_multiplier		tcps_propinfo_tbl[15].prop_cur_uval
467 #define	tcps_mss_def_ipv4		tcps_propinfo_tbl[16].prop_cur_uval
468 #define	tcps_mss_max_ipv4		tcps_propinfo_tbl[17].prop_cur_uval
469 #define	tcps_mss_min			tcps_propinfo_tbl[18].prop_cur_uval
470 #define	tcps_naglim_def			tcps_propinfo_tbl[19].prop_cur_uval
471 #define	tcps_rexmit_interval_initial_high	\
472 					tcps_propinfo_tbl[20].prop_max_uval
473 #define	tcps_rexmit_interval_initial	tcps_propinfo_tbl[20].prop_cur_uval
474 #define	tcps_rexmit_interval_initial_low	\
475 					tcps_propinfo_tbl[20].prop_min_uval
476 #define	tcps_rexmit_interval_max_high	tcps_propinfo_tbl[21].prop_max_uval
477 #define	tcps_rexmit_interval_max	tcps_propinfo_tbl[21].prop_cur_uval
478 #define	tcps_rexmit_interval_max_low	tcps_propinfo_tbl[21].prop_min_uval
479 #define	tcps_rexmit_interval_min_high	tcps_propinfo_tbl[22].prop_max_uval
480 #define	tcps_rexmit_interval_min	tcps_propinfo_tbl[22].prop_cur_uval
481 #define	tcps_rexmit_interval_min_low	tcps_propinfo_tbl[22].prop_min_uval
482 #define	tcps_deferred_ack_interval	tcps_propinfo_tbl[23].prop_cur_uval
483 #define	tcps_snd_lowat_fraction		tcps_propinfo_tbl[24].prop_cur_uval
484 #define	tcps_dupack_fast_retransmit	tcps_propinfo_tbl[25].prop_cur_uval
485 #define	tcps_ignore_path_mtu		tcps_propinfo_tbl[26].prop_cur_bval
486 #define	tcps_smallest_anon_port		tcps_propinfo_tbl[27].prop_cur_uval
487 #define	tcps_largest_anon_port		tcps_propinfo_tbl[28].prop_cur_uval
488 #define	tcps_xmit_hiwat			tcps_propinfo_tbl[29].prop_cur_uval
489 #define	tcps_xmit_lowat			tcps_propinfo_tbl[30].prop_cur_uval
490 #define	tcps_recv_hiwat			tcps_propinfo_tbl[31].prop_cur_uval
491 #define	tcps_recv_hiwat_minmss		tcps_propinfo_tbl[32].prop_cur_uval
492 #define	tcps_fin_wait_2_flush_interval_high	\
493 					tcps_propinfo_tbl[33].prop_max_uval
494 #define	tcps_fin_wait_2_flush_interval	tcps_propinfo_tbl[33].prop_cur_uval
495 #define	tcps_fin_wait_2_flush_interval_low	\
496 					tcps_propinfo_tbl[33].prop_min_uval
497 #define	tcps_max_buf			tcps_propinfo_tbl[34].prop_cur_uval
498 #define	tcps_strong_iss			tcps_propinfo_tbl[35].prop_cur_uval
499 #define	tcps_rtt_updates		tcps_propinfo_tbl[36].prop_cur_uval
500 #define	tcps_wscale_always		tcps_propinfo_tbl[37].prop_cur_bval
501 #define	tcps_tstamp_always		tcps_propinfo_tbl[38].prop_cur_bval
502 #define	tcps_tstamp_if_wscale		tcps_propinfo_tbl[39].prop_cur_bval
503 #define	tcps_rexmit_interval_extra	tcps_propinfo_tbl[40].prop_cur_uval
504 #define	tcps_deferred_acks_max		tcps_propinfo_tbl[41].prop_cur_uval
505 #define	tcps_slow_start_after_idle	tcps_propinfo_tbl[42].prop_cur_uval
506 #define	tcps_slow_start_initial		tcps_propinfo_tbl[43].prop_cur_uval
507 #define	tcps_sack_permitted		tcps_propinfo_tbl[44].prop_cur_uval
508 #define	tcps_ipv6_hoplimit		tcps_propinfo_tbl[45].prop_cur_uval
509 #define	tcps_mss_def_ipv6		tcps_propinfo_tbl[46].prop_cur_uval
510 #define	tcps_mss_max_ipv6		tcps_propinfo_tbl[47].prop_cur_uval
511 #define	tcps_rev_src_routes		tcps_propinfo_tbl[48].prop_cur_bval
512 #define	tcps_local_dack_interval	tcps_propinfo_tbl[49].prop_cur_uval
513 #define	tcps_local_dacks_max		tcps_propinfo_tbl[50].prop_cur_uval
514 #define	tcps_ecn_permitted		tcps_propinfo_tbl[51].prop_cur_uval
515 #define	tcps_rst_sent_rate_enabled	tcps_propinfo_tbl[52].prop_cur_bval
516 #define	tcps_rst_sent_rate		tcps_propinfo_tbl[53].prop_cur_uval
517 #define	tcps_push_timer_interval	tcps_propinfo_tbl[54].prop_cur_uval
518 #define	tcps_use_smss_as_mss_opt	tcps_propinfo_tbl[55].prop_cur_bval
519 #define	tcps_keepalive_abort_interval_high \
520 					tcps_propinfo_tbl[56].prop_max_uval
521 #define	tcps_keepalive_abort_interval \
522 					tcps_propinfo_tbl[56].prop_cur_uval
523 #define	tcps_keepalive_abort_interval_low \
524 					tcps_propinfo_tbl[56].prop_min_uval
525 #define	tcps_wroff_xtra			tcps_propinfo_tbl[57].prop_cur_uval
526 #define	tcps_dev_flow_ctl		tcps_propinfo_tbl[58].prop_cur_bval
527 #define	tcps_reass_timeout		tcps_propinfo_tbl[59].prop_cur_uval
528 #define	tcps_iss_incr			tcps_propinfo_tbl[65].prop_cur_uval
529 
530 extern struct qinit tcp_rinitv4, tcp_rinitv6;
531 extern boolean_t do_tcp_fusion;
532 
533 /*
534  * Object to represent database of options to search passed to
535  * {sock,tpi}optcom_req() interface routine to take care of option
536  * management and associated methods.
537  */
538 extern optdb_obj_t	tcp_opt_obj;
539 extern uint_t		tcp_max_optsize;
540 
541 extern int tcp_squeue_flag;
542 
543 extern uint_t tcp_free_list_max_cnt;
544 
545 /*
546  * Functions in tcp.c.
547  */
548 extern void	tcp_acceptor_hash_insert(t_uscalar_t, tcp_t *);
549 extern tcp_t	*tcp_acceptor_hash_lookup(t_uscalar_t, tcp_stack_t *);
550 extern void	tcp_acceptor_hash_remove(tcp_t *);
551 extern mblk_t	*tcp_ack_mp(tcp_t *);
552 extern int	tcp_build_hdrs(tcp_t *);
553 extern void	tcp_cleanup(tcp_t *);
554 extern int	tcp_clean_death(tcp_t *, int);
555 extern void	tcp_clean_death_wrapper(void *, mblk_t *, void *,
556 		    ip_recv_attr_t *);
557 extern void	tcp_close_common(conn_t *, int);
558 extern void	tcp_close_detached(tcp_t *);
559 extern void	tcp_close_mpp(mblk_t **);
560 extern void	tcp_closei_local(tcp_t *);
561 extern sock_lower_handle_t tcp_create(int, int, int, sock_downcalls_t **,
562 		    uint_t *, int *, int, cred_t *);
563 extern conn_t	*tcp_create_common(cred_t *, boolean_t, boolean_t, int *);
564 extern void	tcp_disconnect(tcp_t *, mblk_t *);
565 extern char	*tcp_display(tcp_t *, char *, char);
566 extern int	tcp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
567 		    boolean_t);
568 extern int	tcp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
569 		    cred_t *, pid_t);
570 extern int	tcp_do_listen(conn_t *, struct sockaddr *, socklen_t, int,
571 		    cred_t *, boolean_t);
572 extern int	tcp_do_unbind(conn_t *);
573 extern boolean_t	tcp_eager_blowoff(tcp_t *, t_scalar_t);
574 extern void	tcp_eager_cleanup(tcp_t *, boolean_t);
575 extern void	tcp_eager_kill(void *, mblk_t *, void *, ip_recv_attr_t *);
576 extern void	tcp_eager_unlink(tcp_t *);
577 extern void	tcp_init_values(tcp_t *, tcp_t *);
578 extern void	tcp_ipsec_cleanup(tcp_t *);
579 extern int	tcp_maxpsz_set(tcp_t *, boolean_t);
580 extern void	tcp_mss_set(tcp_t *, uint32_t);
581 extern void	tcp_reinput(conn_t *, mblk_t *, ip_recv_attr_t *, ip_stack_t *);
582 extern void	tcp_rsrv(queue_t *);
583 extern uint_t	tcp_rwnd_reopen(tcp_t *);
584 extern int	tcp_rwnd_set(tcp_t *, uint32_t);
585 extern int	tcp_set_destination(tcp_t *);
586 extern void	tcp_set_ws_value(tcp_t *);
587 extern void	tcp_stop_lingering(tcp_t *);
588 extern void	tcp_update_pmtu(tcp_t *, boolean_t);
589 extern mblk_t	*tcp_zcopy_backoff(tcp_t *, mblk_t *, boolean_t);
590 extern boolean_t	tcp_zcopy_check(tcp_t *);
591 extern void	tcp_zcopy_notify(tcp_t *);
592 extern void	tcp_get_proto_props(tcp_t *, struct sock_proto_props *);
593 
594 /*
595  * Bind related functions in tcp_bind.c
596  */
597 extern int	tcp_bind_check(conn_t *, struct sockaddr *, socklen_t,
598 		    cred_t *, boolean_t);
599 extern void	tcp_bind_hash_insert(tf_t *, tcp_t *, int);
600 extern void	tcp_bind_hash_remove(tcp_t *);
601 extern in_port_t	tcp_bindi(tcp_t *, in_port_t, const in6_addr_t *,
602 			    int, boolean_t, boolean_t, boolean_t);
603 extern in_port_t	tcp_update_next_port(in_port_t, const tcp_t *,
604 			    boolean_t);
605 
606 /*
607  * Fusion related functions in tcp_fusion.c.
608  */
609 extern void	tcp_fuse(tcp_t *, uchar_t *, tcpha_t *);
610 extern void	tcp_unfuse(tcp_t *);
611 extern boolean_t tcp_fuse_output(tcp_t *, mblk_t *, uint32_t);
612 extern void	tcp_fuse_output_urg(tcp_t *, mblk_t *);
613 extern boolean_t tcp_fuse_rcv_drain(queue_t *, tcp_t *, mblk_t **);
614 extern size_t	tcp_fuse_set_rcv_hiwat(tcp_t *, size_t);
615 extern int	tcp_fuse_maxpsz(tcp_t *);
616 extern void	tcp_fuse_backenable(tcp_t *);
617 extern void	tcp_iss_key_init(uint8_t *, int, tcp_stack_t *);
618 
619 /*
620  * Output related functions in tcp_output.c.
621  */
622 extern void	tcp_close_output(void *, mblk_t *, void *, ip_recv_attr_t *);
623 extern void	tcp_output(void *, mblk_t *, void *, ip_recv_attr_t *);
624 extern void	tcp_output_urgent(void *, mblk_t *, void *, ip_recv_attr_t *);
625 extern void	tcp_rexmit_after_error(tcp_t *);
626 extern void	tcp_sack_rexmit(tcp_t *, uint_t *);
627 extern void	tcp_send_data(tcp_t *, mblk_t *);
628 extern void	tcp_send_synack(void *, mblk_t *, void *, ip_recv_attr_t *);
629 extern void	tcp_shutdown_output(void *, mblk_t *, void *, ip_recv_attr_t *);
630 extern void	tcp_ss_rexmit(tcp_t *);
631 extern void	tcp_update_xmit_tail(tcp_t *, uint32_t);
632 extern void	tcp_wput(queue_t *, mblk_t *);
633 extern void	tcp_wput_data(tcp_t *, mblk_t *, boolean_t);
634 extern void	tcp_wput_sock(queue_t *, mblk_t *);
635 extern void	tcp_wput_fallback(queue_t *, mblk_t *);
636 extern void	tcp_xmit_ctl(char *, tcp_t *, uint32_t, uint32_t, int);
637 extern void	tcp_xmit_listeners_reset(mblk_t *, ip_recv_attr_t *,
638 		    ip_stack_t *i, conn_t *);
639 extern mblk_t	*tcp_xmit_mp(tcp_t *, mblk_t *, int32_t, int32_t *,
640 		    mblk_t **, uint32_t, boolean_t, uint32_t *, boolean_t);
641 
642 /*
643  * Input related functions in tcp_input.c.
644  */
645 extern void	tcp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
646 extern void	tcp_input_data(void *, mblk_t *, void *, ip_recv_attr_t *);
647 extern void	tcp_input_listener_unbound(void *, mblk_t *, void *,
648 		    ip_recv_attr_t *);
649 extern boolean_t	tcp_paws_check(tcp_t *, tcpha_t *, tcp_opt_t *);
650 extern uint_t	tcp_rcv_drain(tcp_t *);
651 extern void	tcp_rcv_enqueue(tcp_t *, mblk_t *, uint_t, cred_t *);
652 extern boolean_t	tcp_verifyicmp(conn_t *, void *, icmph_t *, icmp6_t *,
653 			    ip_recv_attr_t *);
654 
655 /*
656  * Kernel socket related functions in tcp_socket.c.
657  */
658 extern int	tcp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
659 		    so_proto_quiesced_cb_t, sock_quiesce_arg_t *);
660 extern boolean_t tcp_newconn_notify(tcp_t *, ip_recv_attr_t *);
661 
662 /*
663  * Timer related functions in tcp_timers.c.
664  */
665 extern void	tcp_ack_timer(void *);
666 extern void	tcp_close_linger_timeout(void *);
667 extern void	tcp_keepalive_timer(void *);
668 extern void	tcp_push_timer(void *);
669 extern void	tcp_reass_timer(void *);
670 extern mblk_t	*tcp_timermp_alloc(int);
671 extern void	tcp_timermp_free(tcp_t *);
672 extern timeout_id_t tcp_timeout(conn_t *, void (*)(void *), hrtime_t);
673 extern clock_t	tcp_timeout_cancel(conn_t *, timeout_id_t);
674 extern void	tcp_timer(void *arg);
675 extern void	tcp_timers_stop(tcp_t *);
676 
677 /*
678  * TCP TPI related functions in tcp_tpi.c.
679  */
680 extern void	tcp_addr_req(tcp_t *, mblk_t *);
681 extern void	tcp_capability_req(tcp_t *, mblk_t *);
682 extern boolean_t	tcp_conn_con(tcp_t *, uchar_t *, mblk_t *,
683 			    mblk_t **, ip_recv_attr_t *);
684 extern void	tcp_err_ack(tcp_t *, mblk_t *, int, int);
685 extern void	tcp_err_ack_prim(tcp_t *, mblk_t *, int, int, int);
686 extern void	tcp_info_req(tcp_t *, mblk_t *);
687 extern void	tcp_send_conn_ind(void *, mblk_t *, void *);
688 extern void	tcp_send_pending(void *, mblk_t *, void *, ip_recv_attr_t *);
689 extern void	tcp_tpi_accept(queue_t *, mblk_t *);
690 extern void	tcp_tpi_bind(tcp_t *, mblk_t *);
691 extern int	tcp_tpi_close(queue_t *, int);
692 extern int	tcp_tpi_close_accept(queue_t *);
693 extern void	tcp_tpi_connect(tcp_t *, mblk_t *);
694 extern int	tcp_tpi_opt_get(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
695 extern int	tcp_tpi_opt_set(queue_t *, uint_t, int, int, uint_t, uchar_t *,
696 		    uint_t *, uchar_t *, void *, cred_t *);
697 extern void	tcp_tpi_unbind(tcp_t *, mblk_t *);
698 extern void	tcp_tli_accept(tcp_t *, mblk_t *);
699 extern void	tcp_use_pure_tpi(tcp_t *);
700 extern void	tcp_do_capability_ack(tcp_t *, struct T_capability_ack *,
701 		    t_uscalar_t);
702 
703 /*
704  * TCP option processing related functions in tcp_opt_data.c
705  */
706 extern int	tcp_opt_get(conn_t *, int, int, uchar_t *);
707 extern int	tcp_opt_set(conn_t *, uint_t, int, int, uint_t, uchar_t *,
708 		    uint_t *, uchar_t *, void *, cred_t *);
709 
710 /*
711  * TCP time wait processing related functions in tcp_time_wait.c.
712  */
713 extern void		tcp_time_wait_append(tcp_t *);
714 extern void		tcp_time_wait_collector(void *);
715 extern boolean_t	tcp_time_wait_remove(tcp_t *, tcp_squeue_priv_t *);
716 extern void		tcp_time_wait_processing(tcp_t *, mblk_t *, uint32_t,
717 			    uint32_t, int, tcpha_t *, ip_recv_attr_t *);
718 
719 /*
720  * Misc functions in tcp_misc.c.
721  */
722 extern uint32_t	tcp_find_listener_conf(tcp_stack_t *, in_port_t);
723 extern void	tcp_ioctl_abort_conn(queue_t *, mblk_t *);
724 extern void	tcp_listener_conf_cleanup(tcp_stack_t *);
725 extern void	tcp_stack_cpu_add(tcp_stack_t *, processorid_t);
726 
727 #endif	/* _KERNEL */
728 
729 #ifdef	__cplusplus
730 }
731 #endif
732 
733 #endif	/* _INET_TCP_IMPL_H */
734