xref: /titanic_50/usr/src/uts/common/inet/tcp_impl.h (revision 3e5bc1d795e8c41f3680a71e3954e72d079ee46d)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
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 <inet/optcom.h>
43 #include <inet/tcp.h>
44 
45 #define	TCP_MOD_ID	5105
46 
47 /*
48  * Was this tcp created via socket() interface?
49  */
50 #define	TCP_IS_SOCKET(tcp)	((tcp)->tcp_issocket)
51 
52 /*
53  * Is this tcp not attached to any upper client?
54  */
55 #define	TCP_IS_DETACHED(tcp)	((tcp)->tcp_detached)
56 
57 #define	TCP_TIMER(tcp, f, tim)		\
58 	tcp_timeout(tcp->tcp_connp, f, tim)
59 #define	TCP_TIMER_CANCEL(tcp, id)	\
60 	tcp_timeout_cancel(tcp->tcp_connp, id)
61 
62 /*
63  * To restart the TCP retransmission timer.
64  */
65 #define	TCP_TIMER_RESTART(tcp, intvl) {					\
66 	if ((tcp)->tcp_timer_tid != 0)					\
67 		(void) TCP_TIMER_CANCEL((tcp), (tcp)->tcp_timer_tid);	\
68 	(tcp)->tcp_timer_tid = TCP_TIMER((tcp), tcp_timer,		\
69 	    MSEC_TO_TICK(intvl));					\
70 }
71 
72 /*
73  * This stops synchronous streams for a fused tcp endpoint
74  * and prevents tcp_fuse_rrw() from pulling data from it.
75  */
76 #define	TCP_FUSE_SYNCSTR_STOP(tcp) {				\
77 	if ((tcp)->tcp_direct_sockfs) {				\
78 		mutex_enter(&(tcp)->tcp_non_sq_lock);		\
79 		(tcp)->tcp_fuse_syncstr_stopped = B_TRUE;	\
80 		mutex_exit(&(tcp)->tcp_non_sq_lock);		\
81 	}							\
82 }
83 
84 /*
85  * This causes all calls to tcp_fuse_rrw() to block until
86  * TCP_FUSE_SYNCSTR_UNPLUG_DRAIN() is called.
87  */
88 #define	TCP_FUSE_SYNCSTR_PLUG_DRAIN(tcp) {			\
89 	if ((tcp)->tcp_direct_sockfs) {				\
90 		mutex_enter(&(tcp)->tcp_non_sq_lock);		\
91 		ASSERT(!(tcp)->tcp_fuse_syncstr_plugged);	\
92 		(tcp)->tcp_fuse_syncstr_plugged = B_TRUE;	\
93 		mutex_exit(&(tcp)->tcp_non_sq_lock);		\
94 	}							\
95 }
96 
97 /*
98  * This unplugs the draining of data through tcp_fuse_rrw(); see
99  * the comments in tcp_fuse_rrw() for how we preserve ordering.
100  */
101 #define	TCP_FUSE_SYNCSTR_UNPLUG_DRAIN(tcp) {			\
102 	if ((tcp)->tcp_direct_sockfs) {				\
103 		mutex_enter(&(tcp)->tcp_non_sq_lock);		\
104 		(tcp)->tcp_fuse_syncstr_plugged = B_FALSE;	\
105 		(void) cv_broadcast(&(tcp)->tcp_fuse_plugcv);	\
106 		mutex_exit(&(tcp)->tcp_non_sq_lock);		\
107 	}							\
108 }
109 
110 /*
111  * Before caching the conn IRE, we need to make sure certain TCP
112  * states are in sync with the ire. The mismatch could occur if the
113  * TCP state has been set in tcp_adapt_ire() using a different IRE,
114  * e.g if an address was not present during an initial connect(),
115  * tcp_adapt_ire() will set the state using the interface route.
116  * Subsequently, if the address is added to the local machine, the
117  * retransmitted SYN will get the correct (loopback) IRE, but the TCP
118  * state (tcp_loopback and tcp_localnet) will remain out of sync.
119  * This is especially an issue with TCP fusion which relies on the
120  * TCP state to be accurate.
121  *
122  * This check/change should be made only if the TCP is not yet in
123  * the established state, else it would lead to inconsistencies.
124  */
125 #define	TCP_CHECK_IREINFO(tcp, ire) {					\
126 	if ((tcp)->tcp_state < TCPS_ESTABLISHED) {			\
127 		if (((ire)->ire_type & (IRE_LOOPBACK | 			\
128 		    IRE_LOCAL)) && !(tcp)->tcp_loopback) {		\
129 			(tcp)->tcp_loopback = B_TRUE;			\
130 		} else if ((tcp)->tcp_loopback && 			\
131 		    !((ire)->ire_type & (IRE_LOOPBACK | IRE_LOCAL))) {	\
132 			(tcp)->tcp_loopback = B_FALSE;			\
133 		}							\
134 		if ((tcp)->tcp_ipversion == IPV4_VERSION) {		\
135 			(tcp)->tcp_localnet =				\
136 			    ((ire)->ire_gateway_addr == 0);		\
137 		} else {						\
138 			(tcp)->tcp_localnet =				\
139 			    IN6_IS_ADDR_UNSPECIFIED(			\
140 			    &(ire)->ire_gateway_addr_v6);		\
141 		}							\
142 	}								\
143 }
144 
145 /*
146  * Write-side flow-control is implemented via the per instance STREAMS
147  * write-side Q by explicitly setting QFULL to stop the flow of mblk_t(s)
148  * and clearing QFULL and calling qbackenable() to restart the flow based
149  * on the number of TCP unsent bytes (i.e. those not on the wire waiting
150  * for a remote ACK).
151  *
152  * This is different than a standard STREAMS kmod which when using the
153  * STREAMS Q the framework would automatictly flow-control based on the
154  * defined hiwat/lowat values as mblk_t's are enqueued/dequeued.
155  *
156  * As of FireEngine TCP write-side flow-control needs to take into account
157  * both the unsent tcp_xmit list bytes but also any squeue_t enqueued bytes
158  * (i.e. from tcp_wput() -> tcp_output()).
159  *
160  * This is accomplished by adding a new tcp_t fields, tcp_squeue_bytes, to
161  * count the number of bytes enqueued by tcp_wput() and the number of bytes
162  * dequeued and processed by tcp_output().
163  *
164  * So, the total number of bytes unsent is (squeue_bytes + unsent) with all
165  * flow-control uses of unsent replaced with the macro TCP_UNSENT_BYTES.
166  */
167 extern void	tcp_clrqfull(tcp_t *);
168 extern void	tcp_setqfull(tcp_t *);
169 
170 #define	TCP_UNSENT_BYTES(tcp) \
171 	((tcp)->tcp_squeue_bytes + (tcp)->tcp_unsent)
172 
173 /* Named Dispatch Parameter Management Structure */
174 typedef struct tcpparam_s {
175 	uint32_t	tcp_param_min;
176 	uint32_t	tcp_param_max;
177 	uint32_t	tcp_param_val;
178 	char		*tcp_param_name;
179 } tcpparam_t;
180 
181 
182 #define	tcps_time_wait_interval		tcps_params[0].tcp_param_val
183 #define	tcps_conn_req_max_q		tcps_params[1].tcp_param_val
184 #define	tcps_conn_req_max_q0		tcps_params[2].tcp_param_val
185 #define	tcps_conn_req_min		tcps_params[3].tcp_param_val
186 #define	tcps_conn_grace_period		tcps_params[4].tcp_param_val
187 #define	tcps_cwnd_max_			tcps_params[5].tcp_param_val
188 #define	tcps_dbg			tcps_params[6].tcp_param_val
189 #define	tcps_smallest_nonpriv_port	tcps_params[7].tcp_param_val
190 #define	tcps_ip_abort_cinterval		tcps_params[8].tcp_param_val
191 #define	tcps_ip_abort_linterval		tcps_params[9].tcp_param_val
192 #define	tcps_ip_abort_interval		tcps_params[10].tcp_param_val
193 #define	tcps_ip_notify_cinterval	tcps_params[11].tcp_param_val
194 #define	tcps_ip_notify_interval		tcps_params[12].tcp_param_val
195 #define	tcps_ipv4_ttl			tcps_params[13].tcp_param_val
196 #define	tcps_keepalive_interval_high	tcps_params[14].tcp_param_max
197 #define	tcps_keepalive_interval		tcps_params[14].tcp_param_val
198 #define	tcps_keepalive_interval_low	tcps_params[14].tcp_param_min
199 #define	tcps_maxpsz_multiplier		tcps_params[15].tcp_param_val
200 #define	tcps_mss_def_ipv4		tcps_params[16].tcp_param_val
201 #define	tcps_mss_max_ipv4		tcps_params[17].tcp_param_val
202 #define	tcps_mss_min			tcps_params[18].tcp_param_val
203 #define	tcps_naglim_def			tcps_params[19].tcp_param_val
204 #define	tcps_rexmit_interval_initial	tcps_params[20].tcp_param_val
205 #define	tcps_rexmit_interval_max	tcps_params[21].tcp_param_val
206 #define	tcps_rexmit_interval_min	tcps_params[22].tcp_param_val
207 #define	tcps_deferred_ack_interval	tcps_params[23].tcp_param_val
208 #define	tcps_snd_lowat_fraction		tcps_params[24].tcp_param_val
209 #define	tcps_sth_rcv_hiwat		tcps_params[25].tcp_param_val
210 #define	__tcps_not_used1		tcps_params[26].tcp_param_val
211 #define	tcps_dupack_fast_retransmit	tcps_params[27].tcp_param_val
212 #define	tcps_ignore_path_mtu		tcps_params[28].tcp_param_val
213 #define	tcps_smallest_anon_port		tcps_params[29].tcp_param_val
214 #define	tcps_largest_anon_port		tcps_params[30].tcp_param_val
215 #define	tcps_xmit_hiwat			tcps_params[31].tcp_param_val
216 #define	tcps_xmit_lowat			tcps_params[32].tcp_param_val
217 #define	tcps_recv_hiwat			tcps_params[33].tcp_param_val
218 #define	tcps_recv_hiwat_minmss		tcps_params[34].tcp_param_val
219 #define	tcps_fin_wait_2_flush_interval	tcps_params[35].tcp_param_val
220 #define	tcps_max_buf			tcps_params[36].tcp_param_val
221 #define	tcps_strong_iss			tcps_params[37].tcp_param_val
222 #define	tcps_rtt_updates		tcps_params[38].tcp_param_val
223 #define	tcps_wscale_always		tcps_params[39].tcp_param_val
224 #define	tcps_tstamp_always		tcps_params[40].tcp_param_val
225 #define	tcps_tstamp_if_wscale		tcps_params[41].tcp_param_val
226 #define	tcps_rexmit_interval_extra	tcps_params[42].tcp_param_val
227 #define	tcps_deferred_acks_max		tcps_params[43].tcp_param_val
228 #define	tcps_slow_start_after_idle	tcps_params[44].tcp_param_val
229 #define	tcps_slow_start_initial		tcps_params[45].tcp_param_val
230 #define	tcps_sack_permitted		tcps_params[46].tcp_param_val
231 #define	__tcps_not_used2		tcps_params[47].tcp_param_val
232 #define	tcps_ipv6_hoplimit		tcps_params[48].tcp_param_val
233 #define	tcps_mss_def_ipv6		tcps_params[49].tcp_param_val
234 #define	tcps_mss_max_ipv6		tcps_params[50].tcp_param_val
235 #define	tcps_rev_src_routes		tcps_params[51].tcp_param_val
236 #define	tcps_local_dack_interval	tcps_params[52].tcp_param_val
237 #define	tcps_ndd_get_info_interval	tcps_params[53].tcp_param_val
238 #define	tcps_local_dacks_max		tcps_params[54].tcp_param_val
239 #define	tcps_ecn_permitted		tcps_params[55].tcp_param_val
240 #define	tcps_rst_sent_rate_enabled	tcps_params[56].tcp_param_val
241 #define	tcps_rst_sent_rate		tcps_params[57].tcp_param_val
242 #define	tcps_push_timer_interval	tcps_params[58].tcp_param_val
243 #define	tcps_use_smss_as_mss_opt	tcps_params[59].tcp_param_val
244 #define	tcps_keepalive_abort_interval_high	tcps_params[60].tcp_param_max
245 #define	tcps_keepalive_abort_interval		tcps_params[60].tcp_param_val
246 #define	tcps_keepalive_abort_interval_low	tcps_params[60].tcp_param_min
247 
248 extern struct qinit tcp_loopback_rinit, tcp_rinitv4, tcp_rinitv6;
249 extern boolean_t do_tcp_fusion;
250 
251 extern int	tcp_maxpsz_set(tcp_t *, boolean_t);
252 extern void	tcp_timers_stop(tcp_t *);
253 extern void	tcp_rcv_enqueue(tcp_t *, mblk_t *, uint_t);
254 extern void	tcp_push_timer(void *);
255 extern timeout_id_t tcp_timeout(conn_t *, void (*)(void *), clock_t);
256 extern clock_t	tcp_timeout_cancel(conn_t *, timeout_id_t);
257 
258 extern void	tcp_fuse(tcp_t *, uchar_t *, tcph_t *);
259 extern void	tcp_unfuse(tcp_t *);
260 extern boolean_t tcp_fuse_output(tcp_t *, mblk_t *, uint32_t);
261 extern void	tcp_fuse_output_urg(tcp_t *, mblk_t *);
262 extern boolean_t tcp_fuse_rcv_drain(queue_t *, tcp_t *, mblk_t **);
263 extern void	tcp_fuse_syncstr_enable_pair(tcp_t *);
264 extern void	tcp_fuse_disable_pair(tcp_t *, boolean_t);
265 extern int	tcp_fuse_rrw(queue_t *, struiod_t *);
266 extern int	tcp_fuse_rinfop(queue_t *, infod_t *);
267 extern size_t	tcp_fuse_set_rcv_hiwat(tcp_t *, size_t);
268 extern int	tcp_fuse_maxpsz_set(tcp_t *);
269 
270 /*
271  * Object to represent database of options to search passed to
272  * {sock,tpi}optcom_req() interface routine to take care of option
273  * management and associated methods.
274  */
275 extern optdb_obj_t	tcp_opt_obj;
276 extern uint_t		tcp_max_optsize;
277 
278 extern sock_lower_handle_t tcp_create(int, int, int, sock_downcalls_t **,
279     uint_t *, int *, int, cred_t *);
280 extern void tcp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
281     so_proto_quiesced_cb_t);
282 
283 extern sock_downcalls_t sock_tcp_downcalls;
284 
285 
286 #endif	/* _KERNEL */
287 
288 #ifdef	__cplusplus
289 }
290 #endif
291 
292 #endif	/* _INET_TCP_IMPL_H */
293