Print this page
OS-4018 lxbrand support TCP SO_REUSEPORT
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Cody Mello <cody.mello@joyent.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/inet/tcp_impl.h
+++ new/usr/src/uts/common/inet/tcp_impl.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright 2016 Joyent, Inc.
24 24 * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
25 25 * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
26 26 */
27 27
28 28 #ifndef _INET_TCP_IMPL_H
29 29 #define _INET_TCP_IMPL_H
30 30
31 31 /*
32 32 * TCP implementation private declarations. These interfaces are
33 33 * used to build the IP module and are not meant to be accessed
34 34 * by any modules except IP itself. They are undocumented and are
35 35 * subject to change without notice.
36 36 */
37 37
38 38 #ifdef __cplusplus
39 39 extern "C" {
40 40 #endif
41 41
42 42 #ifdef _KERNEL
43 43
44 44 #include <sys/cpuvar.h>
45 45 #include <sys/clock_impl.h> /* For LBOLT_FASTPATH{,64} */
46 46 #include <inet/optcom.h>
47 47 #include <inet/tcp.h>
48 48 #include <inet/tunables.h>
49 49
50 50 #define TCP_MOD_ID 5105
51 51
52 52 extern struct qinit tcp_sock_winit;
53 53 extern struct qinit tcp_winit;
54 54
55 55 extern sock_downcalls_t sock_tcp_downcalls;
56 56
57 57 /*
58 58 * Note that by default, the _snd_lowat_fraction tunable controls the value of
59 59 * the transmit low water mark. TCP_XMIT_LOWATER (and thus the _xmit_lowat
60 60 * property) is only used if the administrator has disabled _snd_lowat_fraction
61 61 * by setting it to 0.
62 62 */
63 63 #define TCP_XMIT_LOWATER 4096
64 64 #define TCP_XMIT_HIWATER 49152
65 65 #define TCP_RECV_LOWATER 2048
66 66 #define TCP_RECV_HIWATER 128000
67 67
68 68 /*
69 69 * Bind hash list size and has function. It has to be a power of 2 for
70 70 * hashing.
71 71 */
72 72 #define TCP_BIND_FANOUT_SIZE 1024
73 73 #define TCP_BIND_HASH(lport) (ntohs(lport) & (TCP_BIND_FANOUT_SIZE - 1))
74 74
75 75 /*
76 76 * This implementation follows the 4.3BSD interpretation of the urgent
77 77 * pointer and not RFC 1122. Switching to RFC 1122 behavior would cause
78 78 * incompatible changes in protocols like telnet and rlogin.
79 79 */
80 80 #define TCP_OLD_URP_INTERPRETATION 1
81 81
82 82 /* TCP option length */
83 83 #define TCPOPT_NOP_LEN 1
84 84 #define TCPOPT_MAXSEG_LEN 4
85 85 #define TCPOPT_WS_LEN 3
86 86 #define TCPOPT_REAL_WS_LEN (TCPOPT_WS_LEN+1)
87 87 #define TCPOPT_TSTAMP_LEN 10
88 88 #define TCPOPT_REAL_TS_LEN (TCPOPT_TSTAMP_LEN+2)
89 89 #define TCPOPT_SACK_OK_LEN 2
90 90 #define TCPOPT_REAL_SACK_OK_LEN (TCPOPT_SACK_OK_LEN+2)
91 91 #define TCPOPT_REAL_SACK_LEN 4
92 92 #define TCPOPT_MAX_SACK_LEN 36
93 93 #define TCPOPT_HEADER_LEN 2
94 94
95 95 /* Round up the value to the nearest mss. */
96 96 #define MSS_ROUNDUP(value, mss) ((((value) - 1) / (mss) + 1) * (mss))
97 97
98 98 /*
99 99 * Was this tcp created via socket() interface?
100 100 */
101 101 #define TCP_IS_SOCKET(tcp) ((tcp)->tcp_issocket)
102 102
103 103 /*
104 104 * Is this tcp not attached to any upper client?
105 105 */
106 106 #define TCP_IS_DETACHED(tcp) ((tcp)->tcp_detached)
107 107
108 108 /* TCP timers related data structures. Refer to tcp_timers.c. */
109 109 typedef struct tcp_timer_s {
110 110 conn_t *connp;
111 111 void (*tcpt_proc)(void *);
112 112 callout_id_t tcpt_tid;
113 113 } tcp_timer_t;
114 114
115 115 extern kmem_cache_t *tcp_timercache;
116 116
117 117 /*
118 118 * Macro for starting various timers. Retransmission timer has its own macro,
119 119 * TCP_TIMER_RESTART(). tim is in millisec.
120 120 */
121 121 #define TCP_TIMER(tcp, f, tim) \
122 122 tcp_timeout(tcp->tcp_connp, f, tim)
123 123 #define TCP_TIMER_CANCEL(tcp, id) \
124 124 tcp_timeout_cancel(tcp->tcp_connp, id)
125 125
126 126 /*
127 127 * To restart the TCP retransmission timer. intvl is in millisec.
128 128 */
129 129 #define TCP_TIMER_RESTART(tcp, intvl) { \
130 130 if ((tcp)->tcp_timer_tid != 0) \
131 131 (void) TCP_TIMER_CANCEL((tcp), (tcp)->tcp_timer_tid); \
132 132 (tcp)->tcp_timer_tid = TCP_TIMER((tcp), tcp_timer, (intvl)); \
133 133 }
134 134
135 135
136 136 /*
137 137 * Maximum TIME_WAIT timeout. It is defined here (instead of tcp_tunables.c)
138 138 * so that other parameters can be derived from it.
139 139 */
140 140 #define TCP_TIME_WAIT_MAX (10 * MINUTES)
141 141
142 142 /*
143 143 * TCP_TIME_WAIT_DELAY governs how often the time_wait_collector runs.
144 144 * Running it every 5 seconds seems to yield a reasonable balance between
145 145 * cleanup liveliness and system load.
146 146 */
147 147 #define TCP_TIME_WAIT_DELAY (5 * SECONDS)
148 148
149 149 #define TCP_TIME_WAIT_BUCKETS ((TCP_TIME_WAIT_MAX / TCP_TIME_WAIT_DELAY) + 1)
150 150
151 151 /*
152 152 * For scalability, we must not run a timer for every TCP connection
153 153 * in TIME_WAIT state. To see why, consider (for time wait interval of
154 154 * 1 minutes):
155 155 * 10,000 connections/sec * 60 seconds/time wait = 600,000 active conn's
156 156 *
157 157 * Since TIME_WAIT expiration occurs on a per-squeue basis, handling
158 158 * connections from all netstacks on the system, a simple queue is inadequate
159 159 * for pending entries. This is because tcp_time_wait_interval may differ
160 160 * between connections, causing tail insertion to violate expiration order.
161 161 *
162 162 * Instead of performing expensive sorting or unnecessary list traversal to
163 163 * counteract interval variance between netstacks, a timing wheel structure is
164 164 * used. The duration covered by each bucket in the wheel is determined by the
165 165 * TCP_TIME_WAIT_DELAY (5 seconds). The number of buckets in the wheel is
166 166 * determined by dividing the maximum TIME_WAIT interval (10 minutes) by
167 167 * TCP_TIME_WAIT_DELAY, with one added bucket for rollover protection.
168 168 * (Yielding 121 buckets with the current parameters) When items are inserted
169 169 * into the set of buckets, they are indexed by using their expiration time
170 170 * divided by the bucket size, modulo the number of buckets. This means that
171 171 * when each bucket is processed, all items within should have expired within
172 172 * the last TCP_TIME_WAIT_DELAY interval.
173 173 *
174 174 * Since bucket timer schedules are rounded to the nearest TCP_TIME_WAIT_DELAY
175 175 * interval to ensure all connections in the pending bucket will be expired, a
176 176 * per-squeue offset is used when doing TIME_WAIT scheduling. This offset is
177 177 * between 0 and the TCP_TIME_WAIT_DELAY and is designed to avoid scheduling
178 178 * all of the tcp_time_wait_collector threads to run in lock-step. The offset
179 179 * is fixed while there are any connections present in the buckets.
180 180 *
181 181 * When a tcp_t enters TIME_WAIT state, a timer is started (timeout is
182 182 * tcps_time_wait_interval). When the tcp_t is detached (upper layer closes
183 183 * the end point), it is scheduled to be cleaned up by the squeue-driving
184 184 * tcp_time_wait_collector (also using tcps_time_wait_interval). This means
185 185 * that the TIME_WAIT state can be extended (up to doubled) if the tcp_t
186 186 * doesn't become detached for a long time.
187 187 *
188 188 * The list manipulations (including tcp_time_wait_next/prev)
189 189 * are protected by the tcp_time_wait_lock. The content of the
190 190 * detached TIME_WAIT connections is protected by the normal perimeters.
191 191 *
192 192 * These connection lists are per squeue and squeues are shared across the
193 193 * tcp_stack_t instances. Things in a tcp_time_wait_bucket remain associated
194 194 * with the tcp_stack_t and conn_netstack. Any tcp_t connections stored in the
195 195 * tcp_free_list are disassociated and have NULL tcp_tcps and conn_netstack
196 196 * pointers.
197 197 */
198 198 typedef struct tcp_squeue_priv_s {
199 199 kmutex_t tcp_time_wait_lock;
200 200 boolean_t tcp_time_wait_collector_active;
201 201 callout_id_t tcp_time_wait_tid;
202 202 uint64_t tcp_time_wait_cnt;
203 203 int64_t tcp_time_wait_schedule;
204 204 int64_t tcp_time_wait_offset;
205 205 tcp_t *tcp_time_wait_bucket[TCP_TIME_WAIT_BUCKETS];
206 206 tcp_t *tcp_free_list;
207 207 uint_t tcp_free_list_cnt;
208 208 } tcp_squeue_priv_t;
209 209
210 210 /*
211 211 * Parameters for TCP Initial Send Sequence number (ISS) generation. When
212 212 * tcp_strong_iss is set to 1, which is the default, the ISS is calculated
213 213 * by adding three components: a time component which grows by 1 every 4096
214 214 * nanoseconds (versus every 4 microseconds suggested by RFC 793, page 27);
215 215 * a per-connection component which grows by 125000 for every new connection;
216 216 * and an "extra" component that grows by a random amount centered
217 217 * approximately on 64000. This causes the ISS generator to cycle every
218 218 * 4.89 hours if no TCP connections are made, and faster if connections are
219 219 * made.
220 220 *
221 221 * When tcp_strong_iss is set to 0, ISS is calculated by adding two
222 222 * components: a time component which grows by 250000 every second; and
223 223 * a per-connection component which grows by 125000 for every new connections.
224 224 *
225 225 * A third method, when tcp_strong_iss is set to 2, for generating ISS is
226 226 * prescribed by Steve Bellovin. This involves adding time, the 125000 per
227 227 * connection, and a one-way hash (MD5) of the connection ID <sport, dport,
228 228 * src, dst>, a "truly" random (per RFC 1750) number, and a console-entered
229 229 * password.
230 230 */
231 231 #define ISS_INCR 250000
232 232 #define ISS_NSEC_SHT 12
233 233
234 234 /* Macros for timestamp comparisons */
235 235 #define TSTMP_GEQ(a, b) ((int32_t)((a)-(b)) >= 0)
236 236 #define TSTMP_LT(a, b) ((int32_t)((a)-(b)) < 0)
237 237
238 238 /*
239 239 * Initialize cwnd according to RFC 3390. def_max_init_cwnd is
240 240 * either tcp_slow_start_initial or tcp_slow_start_after idle
241 241 * depending on the caller. If the upper layer has not used the
242 242 * TCP_INIT_CWND option to change the initial cwnd, tcp_init_cwnd
243 243 * should be 0 and we use the formula in RFC 3390 to set tcp_cwnd.
244 244 * If the upper layer has changed set the tcp_init_cwnd, just use
245 245 * it to calculate the tcp_cwnd.
246 246 *
247 247 * "An Argument for Increasing TCP's Initial Congestion Window"
248 248 * ACM SIGCOMM Computer Communications Review, vol. 40 (2010), pp. 27-33
249 249 * -- Nandita Dukkipati, Tiziana Refice, Yuchung Cheng,
250 250 * Hsiao-keng Jerry Chu, Tom Herbert, Amit Agarwal,
251 251 * Arvind Jain, Natalia Sutin
252 252 *
253 253 * "Based on the results from our experiments, we believe the
254 254 * initial congestion window should be at least ten segments
255 255 * and the same be investigated for standardization by the IETF."
256 256 *
257 257 * As such, the def_max_init_cwnd argument with which this macro is
258 258 * invoked is either the tcps_slow_start_initial or
259 259 * tcps_slow_start_after_idle which both default to 0 and will respect
260 260 * RFC 3390 exactly. If the tunables are explicitly set by the operator,
261 261 * then the initial congestion window should be set as the operator
262 262 * demands, within reason. We shall arbitrarily define reason as a
263 263 * maximum of 16 (same as used by the TCP_INIT_CWND setsockopt).
264 264 */
265 265
266 266 /* Maximum TCP initial cwin (start/restart). */
267 267 #define TCP_MAX_INIT_CWND 16
268 268
269 269 #define TCP_SET_INIT_CWND(tcp, mss, def_max_init_cwnd) \
270 270 { \
271 271 if ((tcp)->tcp_init_cwnd == 0) { \
272 272 if (def_max_init_cwnd == 0) { \
273 273 (tcp)->tcp_cwnd = MIN(4 * (mss), \
274 274 MAX(2 * (mss), 4380 / (mss) * (mss))); \
275 275 } else { \
276 276 (tcp)->tcp_cwnd = MIN(TCP_MAX_INIT_CWND * (mss),\
277 277 def_max_init_cwnd * (mss)); \
278 278 } \
279 279 } else { \
280 280 (tcp)->tcp_cwnd = (tcp)->tcp_init_cwnd * (mss); \
281 281 } \
282 282 tcp->tcp_cwnd_cnt = 0; \
283 283 }
284 284
285 285 /*
286 286 * Set ECN capable transport (ECT) code point in IP header.
287 287 *
288 288 * Note that there are 2 ECT code points '01' and '10', which are called
289 289 * ECT(1) and ECT(0) respectively. Here we follow the original ECT code
290 290 * point ECT(0) for TCP as described in RFC 2481.
291 291 */
292 292 #define TCP_SET_ECT(tcp, iph) \
293 293 if ((tcp)->tcp_connp->conn_ipversion == IPV4_VERSION) { \
294 294 /* We need to clear the code point first. */ \
295 295 ((ipha_t *)(iph))->ipha_type_of_service &= 0xFC; \
296 296 ((ipha_t *)(iph))->ipha_type_of_service |= IPH_ECN_ECT0; \
297 297 } else { \
298 298 ((ip6_t *)(iph))->ip6_vcf &= htonl(0xFFCFFFFF); \
299 299 ((ip6_t *)(iph))->ip6_vcf |= htonl(IPH_ECN_ECT0 << 20); \
300 300 }
301 301
302 302 /*
303 303 * Set tcp_rto with boundary checking.
304 304 */
305 305 #define TCP_SET_RTO(tcp, rto) \
306 306 if ((rto) < (tcp)->tcp_rto_min) \
307 307 (tcp)->tcp_rto = (tcp)->tcp_rto_min; \
308 308 else if ((rto) > (tcp)->tcp_rto_max) \
309 309 (tcp)->tcp_rto = (tcp)->tcp_rto_max; \
310 310 else \
311 311 (tcp)->tcp_rto = (rto);
312 312
313 313 /*
314 314 * TCP options struct returned from tcp_parse_options.
315 315 */
316 316 typedef struct tcp_opt_s {
317 317 uint32_t tcp_opt_mss;
318 318 uint32_t tcp_opt_wscale;
319 319 uint32_t tcp_opt_ts_val;
320 320 uint32_t tcp_opt_ts_ecr;
321 321 tcp_t *tcp;
322 322 } tcp_opt_t;
323 323
324 324 /*
325 325 * Flags returned from tcp_parse_options.
326 326 */
327 327 #define TCP_OPT_MSS_PRESENT 1
328 328 #define TCP_OPT_WSCALE_PRESENT 2
329 329 #define TCP_OPT_TSTAMP_PRESENT 4
330 330 #define TCP_OPT_SACK_OK_PRESENT 8
331 331 #define TCP_OPT_SACK_PRESENT 16
332 332
333 333 /*
334 334 * Write-side flow-control is implemented via the per instance STREAMS
335 335 * write-side Q by explicitly setting QFULL to stop the flow of mblk_t(s)
336 336 * and clearing QFULL and calling qbackenable() to restart the flow based
337 337 * on the number of TCP unsent bytes (i.e. those not on the wire waiting
338 338 * for a remote ACK).
339 339 *
340 340 * This is different than a standard STREAMS kmod which when using the
341 341 * STREAMS Q the framework would automatictly flow-control based on the
342 342 * defined hiwat/lowat values as mblk_t's are enqueued/dequeued.
343 343 *
344 344 * As of FireEngine TCP write-side flow-control needs to take into account
345 345 * both the unsent tcp_xmit list bytes but also any squeue_t enqueued bytes
346 346 * (i.e. from tcp_wput() -> tcp_output()).
347 347 *
348 348 * This is accomplished by adding a new tcp_t fields, tcp_squeue_bytes, to
349 349 * count the number of bytes enqueued by tcp_wput() and the number of bytes
350 350 * dequeued and processed by tcp_output().
351 351 *
352 352 * So, the total number of bytes unsent is (squeue_bytes + unsent) with all
353 353 * flow-control uses of unsent replaced with the macro TCP_UNSENT_BYTES.
354 354 */
355 355 extern void tcp_clrqfull(tcp_t *);
356 356 extern void tcp_setqfull(tcp_t *);
357 357
358 358 #define TCP_UNSENT_BYTES(tcp) \
359 359 ((tcp)->tcp_squeue_bytes + (tcp)->tcp_unsent)
360 360
361 361 /*
362 362 * Linked list struct to store listener connection limit configuration per
363 363 * IP stack. The list is stored at tcps_listener_conf in tcp_stack_t.
364 364 *
365 365 * tl_port: the listener port of this limit configuration
366 366 * tl_ratio: the maximum amount of memory consumed by all concurrent TCP
367 367 * connections created by a listener does not exceed 1/tl_ratio
368 368 * of the total system memory. Note that this is only an
369 369 * approximation.
370 370 * tl_link: linked list struct
371 371 */
372 372 typedef struct tcp_listener_s {
373 373 in_port_t tl_port;
374 374 uint32_t tl_ratio;
375 375 list_node_t tl_link;
376 376 } tcp_listener_t;
377 377
378 378 /*
379 379 * If there is a limit set on the number of connections allowed per each
380 380 * listener, the following struct is used to store that counter. It keeps
381 381 * the number of TCP connection created by a listener. Note that this needs
382 382 * to be separated from the listener since the listener can go away before
383 383 * all the connections are gone.
384 384 *
385 385 * When the struct is allocated, tlc_cnt is set to 1. When a new connection
386 386 * is created by the listener, tlc_cnt is incremented by 1. When a connection
387 387 * created by the listener goes away, tlc_count is decremented by 1. When the
388 388 * listener itself goes away, tlc_cnt is decremented by one. The last
389 389 * connection (or the listener) which decrements tlc_cnt to zero frees the
390 390 * struct.
391 391 *
392 392 * tlc_max is the maximum number of concurrent TCP connections created from a
393 393 * listner. It is calculated when the tcp_listen_cnt_t is allocated.
394 394 *
395 395 * tlc_report_time stores the time when cmn_err() is called to report that the
396 396 * max has been exceeeded. Report is done at most once every
397 397 * TCP_TLC_REPORT_INTERVAL mins for a listener.
398 398 *
|
↓ open down ↓ |
398 lines elided |
↑ open up ↑ |
399 399 * tlc_drop stores the number of connection attempt dropped because the
400 400 * limit has reached.
401 401 */
402 402 typedef struct tcp_listen_cnt_s {
403 403 uint32_t tlc_max;
404 404 uint32_t tlc_cnt;
405 405 int64_t tlc_report_time;
406 406 uint32_t tlc_drop;
407 407 } tcp_listen_cnt_t;
408 408
409 +/*
410 + * Track tcp_t entities bound to the same port/address tuple via SO_REUSEPORT.
411 + * - tcprg_lock: Protects the other fields
412 + * - tcprg_size: Allocated size (in entries) of tcprg_members array
413 + * - tcprg_count: Count of occupied tcprg_members slots
414 + * - tcprg_active: Count of members which still have SO_REUSEPORT set
415 + * - tcprg_members: Connections associated with address/port group
416 + */
417 +typedef struct tcp_rg_s {
418 + kmutex_t tcprg_lock;
419 + unsigned int tcprg_size;
420 + unsigned int tcprg_count;
421 + unsigned int tcprg_active;
422 + tcp_t **tcprg_members;
423 +} tcp_rg_t;
424 +
409 425 #define TCP_TLC_REPORT_INTERVAL (30 * MINUTES)
410 426
411 427 #define TCP_DECR_LISTEN_CNT(tcp) \
412 428 { \
413 429 ASSERT((tcp)->tcp_listen_cnt->tlc_cnt > 0); \
414 430 if (atomic_dec_32_nv(&(tcp)->tcp_listen_cnt->tlc_cnt) == 0) \
415 431 kmem_free((tcp)->tcp_listen_cnt, sizeof (tcp_listen_cnt_t)); \
416 432 (tcp)->tcp_listen_cnt = NULL; \
417 433 }
418 434
419 435 /* Increment and decrement the number of connections in tcp_stack_t. */
420 436 #define TCPS_CONN_INC(tcps) \
421 437 atomic_inc_64( \
422 438 (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
423 439
424 440 #define TCPS_CONN_DEC(tcps) \
425 441 atomic_dec_64( \
426 442 (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
427 443
428 444 /*
429 445 * When the system is under memory pressure, stack variable tcps_reclaim is
430 446 * true, we shorten the connection timeout abort interval to tcp_early_abort
431 447 * seconds. Defined in tcp.c.
432 448 */
433 449 extern uint32_t tcp_early_abort;
434 450
435 451 /*
436 452 * To reach to an eager in Q0 which can be dropped due to an incoming
437 453 * new SYN request when Q0 is full, a new doubly linked list is
438 454 * introduced. This list allows to select an eager from Q0 in O(1) time.
439 455 * This is needed to avoid spending too much time walking through the
440 456 * long list of eagers in Q0 when tcp_drop_q0() is called. Each member of
441 457 * this new list has to be a member of Q0.
442 458 * This list is headed by listener's tcp_t. When the list is empty,
443 459 * both the pointers - tcp_eager_next_drop_q0 and tcp_eager_prev_drop_q0,
444 460 * of listener's tcp_t point to listener's tcp_t itself.
445 461 *
446 462 * Given an eager in Q0 and a listener, MAKE_DROPPABLE() puts the eager
447 463 * in the list. MAKE_UNDROPPABLE() takes the eager out of the list.
448 464 * These macros do not affect the eager's membership to Q0.
449 465 */
450 466 #define MAKE_DROPPABLE(listener, eager) \
451 467 if ((eager)->tcp_eager_next_drop_q0 == NULL) { \
452 468 (listener)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0\
453 469 = (eager); \
454 470 (eager)->tcp_eager_prev_drop_q0 = (listener); \
455 471 (eager)->tcp_eager_next_drop_q0 = \
456 472 (listener)->tcp_eager_next_drop_q0; \
457 473 (listener)->tcp_eager_next_drop_q0 = (eager); \
458 474 }
459 475
460 476 #define MAKE_UNDROPPABLE(eager) \
461 477 if ((eager)->tcp_eager_next_drop_q0 != NULL) { \
462 478 (eager)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0 \
463 479 = (eager)->tcp_eager_prev_drop_q0; \
464 480 (eager)->tcp_eager_prev_drop_q0->tcp_eager_next_drop_q0 \
465 481 = (eager)->tcp_eager_next_drop_q0; \
466 482 (eager)->tcp_eager_prev_drop_q0 = NULL; \
467 483 (eager)->tcp_eager_next_drop_q0 = NULL; \
468 484 }
469 485
470 486 /*
471 487 * The format argument to pass to tcp_display().
472 488 * DISP_PORT_ONLY means that the returned string has only port info.
473 489 * DISP_ADDR_AND_PORT means that the returned string also contains the
474 490 * remote and local IP address.
475 491 */
476 492 #define DISP_PORT_ONLY 1
477 493 #define DISP_ADDR_AND_PORT 2
478 494
479 495 #define IP_ADDR_CACHE_SIZE 2048
480 496 #define IP_ADDR_CACHE_HASH(faddr) \
481 497 (ntohl(faddr) & (IP_ADDR_CACHE_SIZE -1))
482 498
483 499 /*
484 500 * TCP reassembly macros. We hide starting and ending sequence numbers in
485 501 * b_next and b_prev of messages on the reassembly queue. The messages are
486 502 * chained using b_cont. These macros are used in tcp_reass() so we don't
487 503 * have to see the ugly casts and assignments.
488 504 */
489 505 #define TCP_REASS_SEQ(mp) ((uint32_t)(uintptr_t)((mp)->b_next))
490 506 #define TCP_REASS_SET_SEQ(mp, u) ((mp)->b_next = \
491 507 (mblk_t *)(uintptr_t)(u))
492 508 #define TCP_REASS_END(mp) ((uint32_t)(uintptr_t)((mp)->b_prev))
493 509 #define TCP_REASS_SET_END(mp, u) ((mp)->b_prev = \
494 510 (mblk_t *)(uintptr_t)(u))
495 511
496 512 #define tcps_time_wait_interval tcps_propinfo_tbl[0].prop_cur_uval
497 513 #define tcps_conn_req_max_q tcps_propinfo_tbl[1].prop_cur_uval
498 514 #define tcps_conn_req_max_q0 tcps_propinfo_tbl[2].prop_cur_uval
499 515 #define tcps_conn_req_min tcps_propinfo_tbl[3].prop_cur_uval
500 516 #define tcps_conn_grace_period tcps_propinfo_tbl[4].prop_cur_uval
501 517 #define tcps_cwnd_max_ tcps_propinfo_tbl[5].prop_cur_uval
502 518 #define tcps_dbg tcps_propinfo_tbl[6].prop_cur_uval
503 519 #define tcps_smallest_nonpriv_port tcps_propinfo_tbl[7].prop_cur_uval
504 520 #define tcps_ip_abort_cinterval tcps_propinfo_tbl[8].prop_cur_uval
505 521 #define tcps_ip_abort_linterval tcps_propinfo_tbl[9].prop_cur_uval
506 522 #define tcps_ip_abort_interval tcps_propinfo_tbl[10].prop_cur_uval
507 523 #define tcps_ip_notify_cinterval tcps_propinfo_tbl[11].prop_cur_uval
508 524 #define tcps_ip_notify_interval tcps_propinfo_tbl[12].prop_cur_uval
509 525 #define tcps_ipv4_ttl tcps_propinfo_tbl[13].prop_cur_uval
510 526 #define tcps_keepalive_interval_high tcps_propinfo_tbl[14].prop_max_uval
511 527 #define tcps_keepalive_interval tcps_propinfo_tbl[14].prop_cur_uval
512 528 #define tcps_keepalive_interval_low tcps_propinfo_tbl[14].prop_min_uval
513 529 #define tcps_maxpsz_multiplier tcps_propinfo_tbl[15].prop_cur_uval
514 530 #define tcps_mss_def_ipv4 tcps_propinfo_tbl[16].prop_cur_uval
515 531 #define tcps_mss_max_ipv4 tcps_propinfo_tbl[17].prop_cur_uval
516 532 #define tcps_mss_min tcps_propinfo_tbl[18].prop_cur_uval
517 533 #define tcps_naglim_def tcps_propinfo_tbl[19].prop_cur_uval
518 534 #define tcps_rexmit_interval_initial_high \
519 535 tcps_propinfo_tbl[20].prop_max_uval
520 536 #define tcps_rexmit_interval_initial tcps_propinfo_tbl[20].prop_cur_uval
521 537 #define tcps_rexmit_interval_initial_low \
522 538 tcps_propinfo_tbl[20].prop_min_uval
523 539 #define tcps_rexmit_interval_max_high tcps_propinfo_tbl[21].prop_max_uval
524 540 #define tcps_rexmit_interval_max tcps_propinfo_tbl[21].prop_cur_uval
525 541 #define tcps_rexmit_interval_max_low tcps_propinfo_tbl[21].prop_min_uval
526 542 #define tcps_rexmit_interval_min_high tcps_propinfo_tbl[22].prop_max_uval
527 543 #define tcps_rexmit_interval_min tcps_propinfo_tbl[22].prop_cur_uval
528 544 #define tcps_rexmit_interval_min_low tcps_propinfo_tbl[22].prop_min_uval
529 545 #define tcps_deferred_ack_interval tcps_propinfo_tbl[23].prop_cur_uval
530 546 #define tcps_snd_lowat_fraction tcps_propinfo_tbl[24].prop_cur_uval
531 547 #define tcps_dupack_fast_retransmit tcps_propinfo_tbl[25].prop_cur_uval
532 548 #define tcps_ignore_path_mtu tcps_propinfo_tbl[26].prop_cur_bval
533 549 #define tcps_smallest_anon_port tcps_propinfo_tbl[27].prop_cur_uval
534 550 #define tcps_largest_anon_port tcps_propinfo_tbl[28].prop_cur_uval
535 551 #define tcps_xmit_hiwat tcps_propinfo_tbl[29].prop_cur_uval
536 552 #define tcps_xmit_lowat tcps_propinfo_tbl[30].prop_cur_uval
537 553 #define tcps_recv_hiwat tcps_propinfo_tbl[31].prop_cur_uval
538 554 #define tcps_recv_hiwat_minmss tcps_propinfo_tbl[32].prop_cur_uval
539 555 #define tcps_fin_wait_2_flush_interval_high \
540 556 tcps_propinfo_tbl[33].prop_max_uval
541 557 #define tcps_fin_wait_2_flush_interval tcps_propinfo_tbl[33].prop_cur_uval
542 558 #define tcps_fin_wait_2_flush_interval_low \
543 559 tcps_propinfo_tbl[33].prop_min_uval
544 560 #define tcps_max_buf tcps_propinfo_tbl[34].prop_cur_uval
545 561 #define tcps_strong_iss tcps_propinfo_tbl[35].prop_cur_uval
546 562 #define tcps_rtt_updates tcps_propinfo_tbl[36].prop_cur_uval
547 563 #define tcps_wscale_always tcps_propinfo_tbl[37].prop_cur_bval
548 564 #define tcps_tstamp_always tcps_propinfo_tbl[38].prop_cur_bval
549 565 #define tcps_tstamp_if_wscale tcps_propinfo_tbl[39].prop_cur_bval
550 566 #define tcps_rexmit_interval_extra tcps_propinfo_tbl[40].prop_cur_uval
551 567 #define tcps_deferred_acks_max tcps_propinfo_tbl[41].prop_cur_uval
552 568 #define tcps_slow_start_after_idle tcps_propinfo_tbl[42].prop_cur_uval
553 569 #define tcps_slow_start_initial tcps_propinfo_tbl[43].prop_cur_uval
554 570 #define tcps_sack_permitted tcps_propinfo_tbl[44].prop_cur_uval
555 571 #define tcps_ipv6_hoplimit tcps_propinfo_tbl[45].prop_cur_uval
556 572 #define tcps_mss_def_ipv6 tcps_propinfo_tbl[46].prop_cur_uval
557 573 #define tcps_mss_max_ipv6 tcps_propinfo_tbl[47].prop_cur_uval
558 574 #define tcps_rev_src_routes tcps_propinfo_tbl[48].prop_cur_bval
559 575 #define tcps_local_dack_interval tcps_propinfo_tbl[49].prop_cur_uval
560 576 #define tcps_local_dacks_max tcps_propinfo_tbl[50].prop_cur_uval
561 577 #define tcps_ecn_permitted tcps_propinfo_tbl[51].prop_cur_uval
562 578 #define tcps_rst_sent_rate_enabled tcps_propinfo_tbl[52].prop_cur_bval
563 579 #define tcps_rst_sent_rate tcps_propinfo_tbl[53].prop_cur_uval
564 580 #define tcps_push_timer_interval tcps_propinfo_tbl[54].prop_cur_uval
565 581 #define tcps_use_smss_as_mss_opt tcps_propinfo_tbl[55].prop_cur_bval
566 582 #define tcps_keepalive_abort_interval_high \
567 583 tcps_propinfo_tbl[56].prop_max_uval
568 584 #define tcps_keepalive_abort_interval \
569 585 tcps_propinfo_tbl[56].prop_cur_uval
570 586 #define tcps_keepalive_abort_interval_low \
571 587 tcps_propinfo_tbl[56].prop_min_uval
572 588 #define tcps_wroff_xtra tcps_propinfo_tbl[57].prop_cur_uval
573 589 #define tcps_dev_flow_ctl tcps_propinfo_tbl[58].prop_cur_bval
574 590 #define tcps_reass_timeout tcps_propinfo_tbl[59].prop_cur_uval
575 591 #define tcps_iss_incr tcps_propinfo_tbl[65].prop_cur_uval
576 592
577 593 extern struct qinit tcp_rinitv4, tcp_rinitv6;
578 594 extern boolean_t do_tcp_fusion;
579 595
580 596 /*
581 597 * Object to represent database of options to search passed to
582 598 * {sock,tpi}optcom_req() interface routine to take care of option
583 599 * management and associated methods.
584 600 */
585 601 extern optdb_obj_t tcp_opt_obj;
586 602 extern uint_t tcp_max_optsize;
587 603
588 604 extern int tcp_squeue_flag;
589 605
590 606 extern uint_t tcp_free_list_max_cnt;
591 607
592 608 /*
593 609 * Functions in tcp.c.
594 610 */
595 611 extern void tcp_acceptor_hash_insert(t_uscalar_t, tcp_t *);
596 612 extern tcp_t *tcp_acceptor_hash_lookup(t_uscalar_t, tcp_stack_t *);
597 613 extern void tcp_acceptor_hash_remove(tcp_t *);
598 614 extern mblk_t *tcp_ack_mp(tcp_t *);
599 615 extern int tcp_build_hdrs(tcp_t *);
600 616 extern void tcp_cleanup(tcp_t *);
601 617 extern int tcp_clean_death(tcp_t *, int);
602 618 extern void tcp_clean_death_wrapper(void *, mblk_t *, void *,
603 619 ip_recv_attr_t *);
604 620 extern void tcp_close_common(conn_t *, int);
605 621 extern void tcp_close_detached(tcp_t *);
606 622 extern void tcp_close_mpp(mblk_t **);
607 623 extern void tcp_closei_local(tcp_t *);
608 624 extern sock_lower_handle_t tcp_create(int, int, int, sock_downcalls_t **,
609 625 uint_t *, int *, int, cred_t *);
610 626 extern conn_t *tcp_create_common(cred_t *, boolean_t, boolean_t, int *);
611 627 extern void tcp_disconnect(tcp_t *, mblk_t *);
612 628 extern char *tcp_display(tcp_t *, char *, char);
613 629 extern int tcp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
614 630 boolean_t);
615 631 extern int tcp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
616 632 cred_t *, pid_t);
617 633 extern int tcp_do_listen(conn_t *, struct sockaddr *, socklen_t, int,
618 634 cred_t *, boolean_t);
619 635 extern int tcp_do_unbind(conn_t *);
620 636 extern boolean_t tcp_eager_blowoff(tcp_t *, t_scalar_t);
621 637 extern void tcp_eager_cleanup(tcp_t *, boolean_t);
622 638 extern void tcp_eager_kill(void *, mblk_t *, void *, ip_recv_attr_t *);
623 639 extern void tcp_eager_unlink(tcp_t *);
624 640 extern void tcp_init_values(tcp_t *, tcp_t *);
625 641 extern void tcp_ipsec_cleanup(tcp_t *);
626 642 extern int tcp_maxpsz_set(tcp_t *, boolean_t);
627 643 extern void tcp_mss_set(tcp_t *, uint32_t);
628 644 extern void tcp_reinput(conn_t *, mblk_t *, ip_recv_attr_t *, ip_stack_t *);
629 645 extern void tcp_rsrv(queue_t *);
630 646 extern uint_t tcp_rwnd_reopen(tcp_t *);
631 647 extern int tcp_rwnd_set(tcp_t *, uint32_t);
632 648 extern int tcp_set_destination(tcp_t *);
633 649 extern void tcp_set_ws_value(tcp_t *);
634 650 extern void tcp_stop_lingering(tcp_t *);
635 651 extern void tcp_update_pmtu(tcp_t *, boolean_t);
636 652 extern mblk_t *tcp_zcopy_backoff(tcp_t *, mblk_t *, boolean_t);
637 653 extern boolean_t tcp_zcopy_check(tcp_t *);
638 654 extern void tcp_zcopy_notify(tcp_t *);
639 655 extern void tcp_get_proto_props(tcp_t *, struct sock_proto_props *);
640 656
641 657 /*
|
↓ open down ↓ |
223 lines elided |
↑ open up ↑ |
642 658 * Bind related functions in tcp_bind.c
643 659 */
644 660 extern int tcp_bind_check(conn_t *, struct sockaddr *, socklen_t,
645 661 cred_t *, boolean_t);
646 662 extern void tcp_bind_hash_insert(tf_t *, tcp_t *, int);
647 663 extern void tcp_bind_hash_remove(tcp_t *);
648 664 extern in_port_t tcp_bindi(tcp_t *, in_port_t, const in6_addr_t *,
649 665 int, boolean_t, boolean_t, boolean_t);
650 666 extern in_port_t tcp_update_next_port(in_port_t, const tcp_t *,
651 667 boolean_t);
668 +extern tcp_rg_t *tcp_rg_init(tcp_t *);
669 +extern boolean_t tcp_rg_remove(tcp_rg_t *, tcp_t *);
670 +extern void tcp_rg_destroy(tcp_rg_t *);
671 +extern void tcp_rg_setactive(tcp_rg_t *, boolean_t);
652 672
653 673 /*
654 674 * Fusion related functions in tcp_fusion.c.
655 675 */
656 676 extern void tcp_fuse(tcp_t *, uchar_t *, tcpha_t *);
657 677 extern void tcp_unfuse(tcp_t *);
658 678 extern boolean_t tcp_fuse_output(tcp_t *, mblk_t *, uint32_t);
659 679 extern void tcp_fuse_output_urg(tcp_t *, mblk_t *);
660 680 extern boolean_t tcp_fuse_rcv_drain(queue_t *, tcp_t *, mblk_t **);
661 681 extern size_t tcp_fuse_set_rcv_hiwat(tcp_t *, size_t);
662 682 extern int tcp_fuse_maxpsz(tcp_t *);
663 683 extern void tcp_fuse_backenable(tcp_t *);
664 684 extern void tcp_iss_key_init(uint8_t *, int, tcp_stack_t *);
665 685
666 686 /*
667 687 * Output related functions in tcp_output.c.
668 688 */
669 689 extern void tcp_close_output(void *, mblk_t *, void *, ip_recv_attr_t *);
670 690 extern void tcp_output(void *, mblk_t *, void *, ip_recv_attr_t *);
671 691 extern void tcp_output_urgent(void *, mblk_t *, void *, ip_recv_attr_t *);
672 692 extern void tcp_rexmit_after_error(tcp_t *);
673 693 extern void tcp_sack_rexmit(tcp_t *, uint_t *);
674 694 extern void tcp_send_data(tcp_t *, mblk_t *);
675 695 extern void tcp_send_synack(void *, mblk_t *, void *, ip_recv_attr_t *);
676 696 extern void tcp_shutdown_output(void *, mblk_t *, void *, ip_recv_attr_t *);
677 697 extern void tcp_ss_rexmit(tcp_t *);
678 698 extern void tcp_update_xmit_tail(tcp_t *, uint32_t);
679 699 extern void tcp_wput(queue_t *, mblk_t *);
680 700 extern void tcp_wput_data(tcp_t *, mblk_t *, boolean_t);
681 701 extern void tcp_wput_sock(queue_t *, mblk_t *);
682 702 extern void tcp_wput_fallback(queue_t *, mblk_t *);
683 703 extern void tcp_xmit_ctl(char *, tcp_t *, uint32_t, uint32_t, int);
684 704 extern void tcp_xmit_listeners_reset(mblk_t *, ip_recv_attr_t *,
685 705 ip_stack_t *i, conn_t *);
686 706 extern mblk_t *tcp_xmit_mp(tcp_t *, mblk_t *, int32_t, int32_t *,
687 707 mblk_t **, uint32_t, boolean_t, uint32_t *, boolean_t);
688 708
689 709 /*
690 710 * Input related functions in tcp_input.c.
691 711 */
692 712 extern void tcp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
693 713 extern void tcp_input_data(void *, mblk_t *, void *, ip_recv_attr_t *);
694 714 extern void tcp_input_listener_unbound(void *, mblk_t *, void *,
695 715 ip_recv_attr_t *);
696 716 extern boolean_t tcp_paws_check(tcp_t *, const tcp_opt_t *);
697 717 extern int tcp_parse_options(tcpha_t *, tcp_opt_t *);
698 718 extern uint_t tcp_rcv_drain(tcp_t *);
699 719 extern void tcp_rcv_enqueue(tcp_t *, mblk_t *, uint_t, cred_t *);
700 720 extern boolean_t tcp_verifyicmp(conn_t *, void *, icmph_t *, icmp6_t *,
701 721 ip_recv_attr_t *);
702 722
703 723 /*
704 724 * Kernel socket related functions in tcp_socket.c.
705 725 */
706 726 extern int tcp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
707 727 so_proto_quiesced_cb_t, sock_quiesce_arg_t *);
708 728 extern boolean_t tcp_newconn_notify(tcp_t *, ip_recv_attr_t *);
709 729
710 730 /*
711 731 * Timer related functions in tcp_timers.c.
712 732 */
713 733 extern void tcp_ack_timer(void *);
714 734 extern void tcp_close_linger_timeout(void *);
715 735 extern void tcp_keepalive_timer(void *);
716 736 extern void tcp_push_timer(void *);
717 737 extern void tcp_reass_timer(void *);
718 738 extern mblk_t *tcp_timermp_alloc(int);
719 739 extern void tcp_timermp_free(tcp_t *);
720 740 extern timeout_id_t tcp_timeout(conn_t *, void (*)(void *), hrtime_t);
721 741 extern clock_t tcp_timeout_cancel(conn_t *, timeout_id_t);
722 742 extern void tcp_timer(void *arg);
723 743 extern void tcp_timers_stop(tcp_t *);
724 744
725 745 /*
726 746 * TCP TPI related functions in tcp_tpi.c.
727 747 */
728 748 extern void tcp_addr_req(tcp_t *, mblk_t *);
729 749 extern void tcp_capability_req(tcp_t *, mblk_t *);
730 750 extern boolean_t tcp_conn_con(tcp_t *, uchar_t *, mblk_t *,
731 751 mblk_t **, ip_recv_attr_t *);
732 752 extern void tcp_err_ack(tcp_t *, mblk_t *, int, int);
733 753 extern void tcp_err_ack_prim(tcp_t *, mblk_t *, int, int, int);
734 754 extern void tcp_info_req(tcp_t *, mblk_t *);
735 755 extern void tcp_send_conn_ind(void *, mblk_t *, void *);
736 756 extern void tcp_send_pending(void *, mblk_t *, void *, ip_recv_attr_t *);
737 757 extern void tcp_tpi_accept(queue_t *, mblk_t *);
738 758 extern void tcp_tpi_bind(tcp_t *, mblk_t *);
739 759 extern int tcp_tpi_close(queue_t *, int);
740 760 extern int tcp_tpi_close_accept(queue_t *);
741 761 extern void tcp_tpi_connect(tcp_t *, mblk_t *);
742 762 extern int tcp_tpi_opt_get(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
743 763 extern int tcp_tpi_opt_set(queue_t *, uint_t, int, int, uint_t, uchar_t *,
744 764 uint_t *, uchar_t *, void *, cred_t *);
745 765 extern void tcp_tpi_unbind(tcp_t *, mblk_t *);
746 766 extern void tcp_tli_accept(tcp_t *, mblk_t *);
747 767 extern void tcp_use_pure_tpi(tcp_t *);
748 768 extern void tcp_do_capability_ack(tcp_t *, struct T_capability_ack *,
749 769 t_uscalar_t);
750 770
751 771 /*
752 772 * TCP option processing related functions in tcp_opt_data.c
753 773 */
754 774 extern int tcp_opt_get(conn_t *, int, int, uchar_t *);
755 775 extern int tcp_opt_set(conn_t *, uint_t, int, int, uint_t, uchar_t *,
756 776 uint_t *, uchar_t *, void *, cred_t *);
757 777
758 778 /*
759 779 * TCP time wait processing related functions in tcp_time_wait.c.
760 780 */
761 781 extern void tcp_time_wait_append(tcp_t *);
762 782 extern void tcp_time_wait_collector(void *);
763 783 extern boolean_t tcp_time_wait_remove(tcp_t *, tcp_squeue_priv_t *);
764 784 extern void tcp_time_wait_processing(tcp_t *, mblk_t *, uint32_t,
765 785 uint32_t, int, tcpha_t *, ip_recv_attr_t *);
766 786
767 787 /*
768 788 * Misc functions in tcp_misc.c.
769 789 */
770 790 extern uint32_t tcp_find_listener_conf(tcp_stack_t *, in_port_t);
771 791 extern void tcp_ioctl_abort_conn(queue_t *, mblk_t *);
772 792 extern void tcp_listener_conf_cleanup(tcp_stack_t *);
773 793 extern void tcp_stack_cpu_add(tcp_stack_t *, processorid_t);
774 794
775 795 #endif /* _KERNEL */
776 796
777 797 #ifdef __cplusplus
778 798 }
779 799 #endif
780 800
781 801 #endif /* _INET_TCP_IMPL_H */
|
↓ open down ↓ |
120 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX