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.h
+++ new/usr/src/uts/common/inet/tcp.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 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
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) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
23 - * Copyright (c) 2011, Joyent, Inc. All rights reserved.
23 + * Copyright 2015 Joyent, Inc.
24 24 * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
25 25 * Copyright (c) 2014 by Delphix. All rights reserved.
26 26 */
27 27 /* Copyright (c) 1990 Mentat Inc. */
28 28
29 29 #ifndef _INET_TCP_H
30 30 #define _INET_TCP_H
31 31
32 32 #ifdef __cplusplus
33 33 extern "C" {
34 34 #endif
35 35
36 36 #include <sys/inttypes.h>
37 37 #include <netinet/ip6.h>
38 38 #include <netinet/tcp.h>
39 39 #include <sys/socket.h>
40 40 #include <sys/socket_proto.h>
41 41 #include <sys/md5.h>
42 42 #include <inet/common.h>
43 43 #include <inet/ip.h>
44 44 #include <inet/ip6.h>
45 45 #include <inet/mi.h>
46 46 #include <inet/mib2.h>
47 47 #include <inet/tcp_stack.h>
48 48 #include <inet/tcp_sack.h>
49 49
50 50 /* TCP states */
51 51 #define TCPS_CLOSED -6
52 52 #define TCPS_IDLE -5 /* idle (opened, but not bound) */
53 53 #define TCPS_BOUND -4 /* bound, ready to connect or accept */
54 54 #define TCPS_LISTEN -3 /* listening for connection */
55 55 #define TCPS_SYN_SENT -2 /* active, have sent syn */
56 56 #define TCPS_SYN_RCVD -1 /* have received syn (and sent ours) */
57 57 /* states < TCPS_ESTABLISHED are those where connections not established */
58 58 #define TCPS_ESTABLISHED 0 /* established */
59 59 #define TCPS_CLOSE_WAIT 1 /* rcvd fin, waiting for close */
60 60 /* states > TCPS_CLOSE_WAIT are those where user has closed */
61 61 #define TCPS_FIN_WAIT_1 2 /* have closed and sent fin */
62 62 #define TCPS_CLOSING 3 /* closed, xchd FIN, await FIN ACK */
63 63 #define TCPS_LAST_ACK 4 /* had fin and close; await FIN ACK */
64 64 /* states > TCPS_CLOSE_WAIT && < TCPS_FIN_WAIT_2 await ACK of FIN */
65 65 #define TCPS_FIN_WAIT_2 5 /* have closed, fin is acked */
66 66 #define TCPS_TIME_WAIT 6 /* in 2*msl quiet wait after close */
67 67
68 68 /*
69 69 * Internal flags used in conjunction with the packet header flags.
70 70 * Used in tcp_input_data to keep track of what needs to be done.
71 71 */
72 72 #define TH_LIMIT_XMIT 0x0400 /* Limited xmit is needed */
73 73 #define TH_XMIT_NEEDED 0x0800 /* Window opened - send queued data */
74 74 #define TH_REXMIT_NEEDED 0x1000 /* Time expired for unacked data */
75 75 #define TH_ACK_NEEDED 0x2000 /* Send an ack now. */
76 76 #define TH_NEED_SACK_REXMIT 0x4000 /* Use SACK info to retransmission */
77 77 #define TH_ACK_TIMER_NEEDED 0x8000 /* Start the delayed ACK timer */
78 78 #define TH_ORDREL_NEEDED 0x10000 /* Generate an ordrel indication */
79 79 #define TH_MARKNEXT_NEEDED 0x20000 /* Data should have MSGMARKNEXT */
80 80 #define TH_SEND_URP_MARK 0x40000 /* Send up tcp_urp_mark_mp */
81 81
82 82 /*
83 83 * TCP sequence numbers are 32 bit integers operated
84 84 * on with modular arithmetic. These macros can be
85 85 * used to compare such integers.
86 86 */
87 87 #define SEQ_LT(a, b) ((int32_t)((a)-(b)) < 0)
88 88 #define SEQ_LEQ(a, b) ((int32_t)((a)-(b)) <= 0)
89 89 #define SEQ_GT(a, b) ((int32_t)((a)-(b)) > 0)
90 90 #define SEQ_GEQ(a, b) ((int32_t)((a)-(b)) >= 0)
91 91
92 92 /* TCP Protocol header */
93 93 typedef struct tcphdr_s {
94 94 uint8_t th_lport[2]; /* Source port */
95 95 uint8_t th_fport[2]; /* Destination port */
96 96 uint8_t th_seq[4]; /* Sequence number */
97 97 uint8_t th_ack[4]; /* Acknowledgement number */
98 98 uint8_t th_offset_and_rsrvd[1]; /* Offset to the packet data */
99 99 uint8_t th_flags[1];
100 100 uint8_t th_win[2]; /* Allocation number */
101 101 uint8_t th_sum[2]; /* TCP checksum */
102 102 uint8_t th_urp[2]; /* Urgent pointer */
103 103 } tcph_t;
104 104
105 105 #define TCP_HDR_LENGTH(tcph) \
106 106 ((((tcph_t *)tcph)->th_offset_and_rsrvd[0] >>2) &(0xF << 2))
107 107 #define TCP_MAX_COMBINED_HEADER_LENGTH (60 + 60) /* Maxed out ip + tcp */
108 108 #define TCP_MAX_IP_OPTIONS_LENGTH (60 - IP_SIMPLE_HDR_LENGTH)
109 109 #define TCP_MAX_HDR_LENGTH 60
110 110 #define TCP_MAX_TCP_OPTIONS_LENGTH (60 - sizeof (tcpha_t))
111 111 #define TCP_MIN_HEADER_LENGTH 20
112 112 #define TCP_MAXWIN 65535
113 113 #define TCP_PORT_LEN sizeof (in_port_t)
114 114 #define TCP_MAX_WINSHIFT 14
115 115 #define TCP_MAX_LARGEWIN (TCP_MAXWIN << TCP_MAX_WINSHIFT)
116 116 #define TCP_MAX_LSO_LENGTH (IP_MAXPACKET - TCP_MAX_COMBINED_HEADER_LENGTH)
117 117
118 118 #define TCPIP_HDR_LENGTH(mp, n) \
119 119 (n) = IPH_HDR_LENGTH((mp)->b_rptr), \
120 120 (n) += TCP_HDR_LENGTH((tcpha_t *)&(mp)->b_rptr[(n)])
121 121
122 122 /* TCP Protocol header (used if the header is known to be 32-bit aligned) */
123 123 typedef struct tcphdra_s {
124 124 in_port_t tha_lport; /* Source port */
125 125 in_port_t tha_fport; /* Destination port */
126 126 uint32_t tha_seq; /* Sequence number */
|
↓ open down ↓ |
93 lines elided |
↑ open up ↑ |
127 127 uint32_t tha_ack; /* Acknowledgement number */
128 128 uint8_t tha_offset_and_reserved; /* Offset to the packet data */
129 129 uint8_t tha_flags;
130 130 uint16_t tha_win; /* Allocation number */
131 131 uint16_t tha_sum; /* TCP checksum */
132 132 uint16_t tha_urp; /* Urgent pointer */
133 133 } tcpha_t;
134 134
135 135 struct conn_s;
136 136 struct tcp_listen_cnt_s;
137 +struct tcp_rg_s;
137 138
138 139 /*
139 140 * Control structure for each open TCP stream,
140 141 * defined only within the kernel or for a kmem user.
141 142 * NOTE: tcp_reinit_values MUST have a line for each field in this structure!
142 143 */
143 144 #if (defined(_KERNEL) || defined(_KMEMUSER))
144 145
145 146 typedef struct tcp_s {
146 147 struct tcp_s *tcp_time_wait_next;
147 148 /* Pointer to next T/W block */
148 149 struct tcp_s *tcp_time_wait_prev;
149 150 /* Pointer to previous T/W next */
150 151 int64_t tcp_time_wait_expire;
151 152
152 153 struct conn_s *tcp_connp; /* back pointer to conn_t */
153 154 tcp_stack_t *tcp_tcps; /* back pointer to tcp_stack_t */
154 155
155 156 int32_t tcp_state;
156 157 int32_t tcp_rcv_ws; /* My window scale power */
157 158 int32_t tcp_snd_ws; /* Sender's window scale power */
158 159 uint32_t tcp_ts_recent; /* Timestamp of earliest unacked */
159 160 /* data segment */
160 161 clock_t tcp_rto; /* Round trip timeout */
161 162 int64_t tcp_last_rcv_lbolt;
162 163 /* lbolt on last packet, used for PAWS */
163 164 uint32_t tcp_rto_initial; /* Initial RTO */
164 165 uint32_t tcp_rto_min; /* Minimum RTO */
165 166 uint32_t tcp_rto_max; /* Maximum RTO */
166 167
167 168 uint32_t tcp_snxt; /* Senders next seq num */
168 169 uint32_t tcp_swnd; /* Senders window (relative to suna) */
169 170 uint32_t tcp_mss; /* Max segment size */
170 171 uint32_t tcp_iss; /* Initial send seq num */
171 172 uint32_t tcp_rnxt; /* Seq we expect to recv next */
172 173 uint32_t tcp_rwnd;
173 174
174 175 /* Fields arranged in approximate access order along main paths */
175 176 mblk_t *tcp_xmit_head; /* Head of xmit/rexmit list */
176 177 mblk_t *tcp_xmit_last; /* Last valid data seen by tcp_wput */
177 178 mblk_t *tcp_xmit_tail; /* Last data sent */
178 179 uint32_t tcp_unsent; /* # of bytes in hand that are unsent */
179 180 uint32_t tcp_xmit_tail_unsent; /* # of unsent bytes in xmit_tail */
180 181
181 182 uint32_t tcp_suna; /* Sender unacknowledged */
182 183 uint32_t tcp_rexmit_nxt; /* Next rexmit seq num */
183 184 uint32_t tcp_rexmit_max; /* Max retran seq num */
184 185 uint32_t tcp_cwnd; /* Congestion window */
185 186 int32_t tcp_cwnd_cnt; /* cwnd cnt in congestion avoidance */
186 187
187 188 uint32_t tcp_ibsegs; /* Inbound segments on this stream */
188 189 uint32_t tcp_obsegs; /* Outbound segments on this stream */
189 190
190 191 uint32_t tcp_naglim; /* Tunable nagle limit */
191 192 uint32_t tcp_valid_bits;
192 193 #define TCP_ISS_VALID 0x1 /* Is the tcp_iss seq num active? */
193 194 #define TCP_FSS_VALID 0x2 /* Is the tcp_fss seq num active? */
194 195 #define TCP_URG_VALID 0x4 /* Is the tcp_urg seq num active? */
195 196 #define TCP_OFO_FIN_VALID 0x8 /* Has TCP received an out of order FIN? */
196 197
197 198
198 199
199 200 timeout_id_t tcp_timer_tid; /* Control block for timer service */
200 201 uchar_t tcp_timer_backoff; /* Backoff shift count. */
201 202 int64_t tcp_last_recv_time; /* Last time we receive a segment. */
202 203 uint32_t tcp_init_cwnd; /* Initial cwnd (start/restart) */
203 204
204 205 /* Following manipulated by TCP under squeue protection */
205 206 uint32_t
206 207 tcp_urp_last_valid : 1, /* Is tcp_urp_last valid? */
207 208 tcp_hard_binding : 1, /* TCP_DETACHED_NONEAGER */
208 209 tcp_fin_acked : 1, /* Has our FIN been acked? */
209 210 tcp_fin_rcvd : 1, /* Have we seen a FIN? */
210 211
211 212 tcp_fin_sent : 1, /* Have we sent our FIN yet? */
212 213 tcp_ordrel_done : 1, /* Have we sent the ord_rel upstream? */
213 214 tcp_detached : 1, /* If we're detached from a stream */
214 215 tcp_zero_win_probe: 1, /* Zero win probing is in progress */
215 216
216 217 tcp_loopback: 1, /* src and dst are the same machine */
217 218 tcp_localnet: 1, /* src and dst are on the same subnet */
218 219 tcp_syn_defense: 1, /* For defense against SYN attack */
219 220 #define tcp_dontdrop tcp_syn_defense
220 221 tcp_set_timer : 1,
221 222
222 223 tcp_active_open: 1, /* This is a active open */
223 224 tcp_rexmit : 1, /* TCP is retransmitting */
224 225 tcp_snd_sack_ok : 1, /* Can use SACK for this connection */
225 226 tcp_hwcksum : 1, /* The NIC is capable of hwcksum */
226 227
227 228 tcp_ip_forward_progress : 1,
228 229 tcp_ecn_ok : 1, /* Can use ECN for this connection */
229 230 tcp_ecn_echo_on : 1, /* Need to do ECN echo */
230 231 tcp_ecn_cwr_sent : 1, /* ECN_CWR has been sent */
231 232
232 233 tcp_cwr : 1, /* Cwnd has reduced recently */
233 234
234 235 tcp_pad_to_bit31 : 11;
235 236
236 237 /* Following manipulated by TCP under squeue protection */
237 238 uint32_t
238 239 tcp_snd_ts_ok : 1,
239 240 tcp_snd_ws_ok : 1,
240 241 tcp_reserved_port : 1,
241 242 tcp_in_free_list : 1,
242 243
243 244 tcp_snd_zcopy_on : 1, /* xmit zero-copy enabled */
244 245 tcp_snd_zcopy_aware : 1, /* client is zero-copy aware */
245 246 tcp_xmit_zc_clean : 1, /* the xmit list is free of zc-mblk */
246 247 tcp_wait_for_eagers : 1, /* Wait for eagers to disappear */
247 248
248 249 tcp_accept_error : 1, /* Error during TLI accept */
249 250 tcp_send_discon_ind : 1, /* TLI accept err, send discon ind */
250 251 tcp_cork : 1, /* tcp_cork option */
251 252 tcp_tconnind_started : 1, /* conn_ind message is being sent */
252 253
253 254 tcp_lso :1, /* Lower layer is capable of LSO */
254 255 tcp_is_wnd_shrnk : 1, /* Window has shrunk */
255 256
256 257 tcp_pad_to_bit_31 : 18;
257 258
258 259 uint32_t tcp_initial_pmtu; /* Initial outgoing Path MTU. */
259 260
260 261 mblk_t *tcp_reass_head; /* Out of order reassembly list head */
261 262 mblk_t *tcp_reass_tail; /* Out of order reassembly list tail */
262 263
263 264 /* SACK related info */
264 265 tcp_sack_info_t tcp_sack_info;
265 266
266 267 #define tcp_pipe tcp_sack_info.tcp_pipe
267 268 #define tcp_fack tcp_sack_info.tcp_fack
268 269 #define tcp_sack_snxt tcp_sack_info.tcp_sack_snxt
269 270 #define tcp_max_sack_blk tcp_sack_info.tcp_max_sack_blk
270 271 #define tcp_num_sack_blk tcp_sack_info.tcp_num_sack_blk
271 272 #define tcp_sack_list tcp_sack_info.tcp_sack_list
272 273 #define tcp_num_notsack_blk tcp_sack_info.tcp_num_notsack_blk
273 274 #define tcp_cnt_notsack_list tcp_sack_info.tcp_cnt_notsack_list
274 275 #define tcp_notsack_list tcp_sack_info.tcp_notsack_list
275 276
276 277 mblk_t *tcp_rcv_list; /* Queued until push, urgent data, */
277 278 mblk_t *tcp_rcv_last_head; /* optdata, or the count exceeds */
278 279 mblk_t *tcp_rcv_last_tail; /* tcp_rcv_push_wait. */
279 280 uint32_t tcp_rcv_cnt; /* tcp_rcv_list is b_next chain. */
280 281
281 282 uint32_t tcp_cwnd_ssthresh; /* Congestion window */
282 283 uint32_t tcp_cwnd_max;
283 284 uint32_t tcp_csuna; /* Clear (no rexmits in window) suna */
284 285
285 286 clock_t tcp_rtt_sa; /* Round trip smoothed average */
286 287 clock_t tcp_rtt_sd; /* Round trip smoothed deviation */
287 288 clock_t tcp_rtt_update; /* Round trip update(s) */
288 289 clock_t tcp_ms_we_have_waited; /* Total retrans time */
289 290
290 291 uint32_t tcp_swl1; /* These help us avoid using stale */
291 292 uint32_t tcp_swl2; /* packets to update state */
292 293
293 294 uint32_t tcp_rack; /* Seq # we have acked */
294 295 uint32_t tcp_rack_cnt; /* # of segs we have deferred ack */
295 296 uint32_t tcp_rack_cur_max; /* # of segs we may defer ack for now */
296 297 uint32_t tcp_rack_abs_max; /* # of segs we may defer ack ever */
297 298 timeout_id_t tcp_ack_tid; /* Delayed ACK timer ID */
298 299 timeout_id_t tcp_push_tid; /* Push timer ID */
299 300
300 301 uint32_t tcp_max_swnd; /* Maximum swnd we have seen */
301 302
302 303 struct tcp_s *tcp_listener; /* Our listener */
303 304
304 305 uint32_t tcp_irs; /* Initial recv seq num */
305 306 uint32_t tcp_fss; /* Final/fin send seq num */
306 307 uint32_t tcp_urg; /* Urgent data seq num */
307 308
308 309 clock_t tcp_first_timer_threshold; /* When to prod IP */
309 310 clock_t tcp_second_timer_threshold; /* When to give up completely */
310 311 clock_t tcp_first_ctimer_threshold; /* 1st threshold while connecting */
311 312 clock_t tcp_second_ctimer_threshold; /* 2nd ... while connecting */
312 313
313 314 uint32_t tcp_urp_last; /* Last urp for which signal sent */
314 315 mblk_t *tcp_urp_mp; /* T_EXDATA_IND for urgent byte */
315 316 mblk_t *tcp_urp_mark_mp; /* zero-length marked/unmarked msg */
316 317
317 318 int tcp_conn_req_cnt_q0; /* # of conn reqs in SYN_RCVD */
318 319 int tcp_conn_req_cnt_q; /* # of conn reqs in ESTABLISHED */
319 320 int tcp_conn_req_max; /* # of ESTABLISHED conn reqs allowed */
320 321 t_scalar_t tcp_conn_req_seqnum; /* Incrementing pending conn req ID */
321 322 #define tcp_ip_addr_cache tcp_reass_tail
322 323 /* Cache ip addresses that */
323 324 /* complete the 3-way handshake */
324 325 kmutex_t tcp_eager_lock;
325 326 struct tcp_s *tcp_eager_next_q; /* next eager in ESTABLISHED state */
326 327 struct tcp_s *tcp_eager_last_q; /* last eager in ESTABLISHED state */
327 328 struct tcp_s *tcp_eager_next_q0; /* next eager in SYN_RCVD state */
328 329 struct tcp_s *tcp_eager_prev_q0; /* prev eager in SYN_RCVD state */
329 330 /* all eagers form a circular list */
330 331 boolean_t tcp_conn_def_q0; /* move from q0 to q deferred */
331 332
332 333 union {
333 334 mblk_t *tcp_eager_conn_ind; /* T_CONN_IND waiting for 3rd ack. */
334 335 mblk_t *tcp_opts_conn_req; /* T_CONN_REQ w/ options processed */
335 336 } tcp_conn;
336 337 uint32_t tcp_syn_rcvd_timeout; /* How many SYN_RCVD timeout in q0 */
337 338
338 339 /*
339 340 * TCP Keepalive Timer members.
340 341 * All keepalive timer intervals are in milliseconds.
341 342 */
342 343 int32_t tcp_ka_last_intrvl; /* Last probe interval */
343 344 timeout_id_t tcp_ka_tid; /* Keepalive timer ID */
344 345 uint32_t tcp_ka_interval; /* Keepalive interval */
345 346
346 347 /*
347 348 * TCP connection is terminated if we don't hear back from the peer
348 349 * for tcp_ka_abort_thres milliseconds after the first keepalive probe.
349 350 * tcp_ka_rinterval is the interval in milliseconds between successive
350 351 * keepalive probes. tcp_ka_cnt is the number of keepalive probes to
351 352 * be sent before terminating the connection, if we don't hear back from
352 353 * peer.
353 354 * tcp_ka_abort_thres = tcp_ka_rinterval * tcp_ka_cnt
354 355 */
355 356 uint32_t tcp_ka_rinterval; /* keepalive retransmit interval */
356 357 uint32_t tcp_ka_abort_thres; /* Keepalive abort threshold */
357 358 uint32_t tcp_ka_cnt; /* count of keepalive probes */
358 359
359 360 int32_t tcp_client_errno; /* How the client screwed up */
360 361
361 362 /*
362 363 * The header template lives in conn_ht_iphc allocated by tcp_build_hdrs
363 364 * We maintain three pointers into conn_ht_iphc.
364 365 */
365 366 ipha_t *tcp_ipha; /* IPv4 header in conn_ht_iphc */
366 367 ip6_t *tcp_ip6h; /* IPv6 header in conn_ht_iphc */
367 368 tcpha_t *tcp_tcpha; /* TCP header in conn_ht_iphc */
368 369
369 370 uint16_t tcp_last_sent_len; /* Record length for nagle */
370 371 uint16_t tcp_last_recv_len; /* Used by DTrace */
371 372 uint16_t tcp_dupack_cnt; /* # of consequtive duplicate acks */
372 373
373 374 kmutex_t *tcp_acceptor_lockp; /* Ptr to tf_lock */
374 375
375 376 mblk_t *tcp_ordrel_mp; /* T_ordrel_ind mblk */
376 377 t_uscalar_t tcp_acceptor_id; /* ACCEPTOR_id */
377 378
378 379 int tcp_ipsec_overhead;
379 380
380 381 uint_t tcp_recvifindex; /* Last received IPV6_RCVPKTINFO */
381 382 uint_t tcp_recvhops; /* Last received IPV6_RECVHOPLIMIT */
382 383 uint_t tcp_recvtclass; /* Last received IPV6_RECVTCLASS */
383 384 ip6_hbh_t *tcp_hopopts; /* Last received IPV6_RECVHOPOPTS */
384 385 ip6_dest_t *tcp_dstopts; /* Last received IPV6_RECVDSTOPTS */
385 386 ip6_dest_t *tcp_rthdrdstopts; /* Last recv IPV6_RECVRTHDRDSTOPTS */
386 387 ip6_rthdr_t *tcp_rthdr; /* Last received IPV6_RECVRTHDR */
387 388 uint_t tcp_hopoptslen;
388 389 uint_t tcp_dstoptslen;
389 390 uint_t tcp_rthdrdstoptslen;
390 391 uint_t tcp_rthdrlen;
391 392
392 393 mblk_t *tcp_timercache;
393 394
394 395 kmutex_t tcp_closelock;
395 396 kcondvar_t tcp_closecv;
396 397 uint8_t tcp_closed;
|
↓ open down ↓ |
250 lines elided |
↑ open up ↑ |
397 398 uint8_t tcp_closeflags;
398 399 mblk_t tcp_closemp;
399 400 timeout_id_t tcp_linger_tid; /* Linger timer ID */
400 401
401 402 struct tcp_s *tcp_acceptor_hash; /* Acceptor hash chain */
402 403 struct tcp_s **tcp_ptpahn; /* Pointer to previous accept hash next. */
403 404 struct tcp_s *tcp_bind_hash; /* Bind hash chain */
404 405 struct tcp_s *tcp_bind_hash_port; /* tcp_t's bound to the same lport */
405 406 struct tcp_s **tcp_ptpbhn;
406 407
408 + /*
409 + * Group of tcp_t entries bound to the same adress and port via
410 + * SO_REUSEPORT. The pointer itself is protected by tf_lock in the
411 + * containing tcps_bind_fanout slot.
412 + */
413 + struct tcp_rg_s *tcp_rg_bind;
414 +
407 415 uint_t tcp_maxpsz_multiplier;
408 416
409 417 uint32_t tcp_lso_max; /* maximum LSO payload */
410 418
411 419 uint32_t tcp_ofo_fin_seq; /* Recv out of order FIN seq num */
412 420 uint32_t tcp_cwr_snd_max;
413 421
414 422 struct tcp_s *tcp_saved_listener; /* saved value of listener */
415 423
416 424 uint32_t tcp_in_ack_unsent; /* ACK for unsent data cnt. */
417 425
418 426 /*
419 427 * All fusion-related fields are protected by squeue.
420 428 */
421 429 struct tcp_s *tcp_loopback_peer; /* peer tcp for loopback */
422 430 mblk_t *tcp_fused_sigurg_mp; /* M_PCSIG mblk for SIGURG */
423 431
424 432 uint32_t
425 433 tcp_fused : 1, /* loopback tcp in fusion mode */
426 434 tcp_unfusable : 1, /* fusion not allowed on endpoint */
427 435 tcp_fused_sigurg : 1, /* send SIGURG upon draining */
428 436
429 437 tcp_fuse_to_bit_31 : 29;
430 438
431 439 kmutex_t tcp_non_sq_lock;
432 440
433 441 /*
434 442 * This variable is accessed without any lock protection
435 443 * and therefore must not be declared as a bit field along
436 444 * with the rest which require such condition.
437 445 */
438 446 boolean_t tcp_issocket; /* this is a socket tcp */
439 447
440 448 /* protected by the tcp_non_sq_lock lock */
441 449 uint32_t tcp_squeue_bytes;
442 450
443 451 /*
444 452 * tcp_closemp_used is protected by listener's tcp_eager_lock
445 453 * when used for eagers. When used for a tcp in TIME_WAIT state
446 454 * or in tcp_close(), it is not protected by any lock as we
447 455 * do not expect any other thread to use it concurrently.
448 456 * We do allow re-use of tcp_closemp in tcp_time_wait_collector()
449 457 * and tcp_close() but not concurrently.
450 458 */
451 459 boolean_t tcp_closemp_used;
452 460
453 461 /*
454 462 * previous and next eagers in the list of droppable eagers. See
455 463 * the comments before MAKE_DROPPABLE(). These pointers are
456 464 * protected by listener's tcp_eager_lock.
457 465 */
458 466 struct tcp_s *tcp_eager_prev_drop_q0;
459 467 struct tcp_s *tcp_eager_next_drop_q0;
460 468
461 469 /*
462 470 * Have we flow controlled xmitter?
463 471 * This variable can be modified outside the squeue and hence must
464 472 * not be declared as a bit field along with the rest that are
465 473 * modified only within the squeue.
466 474 * protected by the tcp_non_sq_lock lock.
467 475 */
468 476 boolean_t tcp_flow_stopped;
469 477
470 478 /*
471 479 * Sender's next sequence number at the time the window was shrunk.
472 480 */
473 481 uint32_t tcp_snxt_shrunk;
474 482
475 483 /*
476 484 * Socket generation number which is bumped when a connection attempt
477 485 * is initiated. Its main purpose is to ensure that the socket does not
478 486 * miss the asynchronous connected/disconnected notification.
479 487 */
480 488 sock_connid_t tcp_connid;
481 489
482 490 /* mblk_t used to enter TCP's squeue from the service routine. */
483 491 mblk_t *tcp_rsrv_mp;
484 492 /* Mutex for accessing tcp_rsrv_mp */
485 493 kmutex_t tcp_rsrv_mp_lock;
486 494
487 495 /* For connection counting. */
488 496 struct tcp_listen_cnt_s *tcp_listen_cnt;
489 497
490 498 /* Segment reassembly timer. */
491 499 timeout_id_t tcp_reass_tid;
492 500
493 501 /* FIN-WAIT-2 flush timeout */
494 502 uint32_t tcp_fin_wait_2_flush_interval;
495 503
496 504 #ifdef DEBUG
497 505 pc_t tcmp_stk[15];
498 506 #endif
499 507 } tcp_t;
500 508
501 509 #ifdef DEBUG
502 510 #define TCP_DEBUG_GETPCSTACK(buffer, depth) ((void) getpcstack(buffer, \
503 511 depth))
504 512 #else
505 513 #define TCP_DEBUG_GETPCSTACK(buffer, depth)
506 514 #endif
507 515
508 516 extern void tcp_conn_reclaim(void *);
509 517 extern void tcp_free(tcp_t *tcp);
510 518 extern void tcp_ddi_g_init(void);
511 519 extern void tcp_ddi_g_destroy(void);
512 520 extern void *tcp_get_conn(void *arg, tcp_stack_t *);
513 521 extern mblk_t *tcp_snmp_get(queue_t *, mblk_t *, boolean_t);
514 522 extern int tcp_snmp_set(queue_t *, int, int, uchar_t *, int len);
515 523
516 524 /* Pad for the tf_t structure to avoid false cache line sharing. */
517 525 #define TF_CACHEL_PAD 64
518 526
519 527 /*
520 528 * The TCP Fanout structure for bind and acceptor hashes.
521 529 * The hash tables and their linkage (tcp_*_hash, tcp_ptp*hn) are
522 530 * protected by the per-bucket tf_lock. Each tcp_t
523 531 * inserted in the list points back at this lock using tcp_*_lockp.
524 532 *
525 533 * The bind and acceptor hash queues are lists of tcp_t.
526 534 */
527 535 /* listener hash and acceptor hash queue head */
528 536 typedef struct tf_s {
529 537 tcp_t *tf_tcp;
530 538 kmutex_t tf_lock;
531 539 unsigned char tf_pad[TF_CACHEL_PAD -
532 540 (sizeof (tcp_t *) + sizeof (kmutex_t))];
533 541 } tf_t;
534 542
535 543
536 544 /* Also used in ipclassifier.c */
537 545 extern struct kmem_cache *tcp_sack_info_cache;
538 546
539 547 #endif /* (defined(_KERNEL) || defined(_KMEMUSER)) */
540 548
541 549 /* Contract private interface between TCP and Clustering. */
542 550
543 551 #define CL_TCPI_V1 1 /* cl_tcpi_version number */
544 552
545 553 typedef struct cl_tcp_info_s {
546 554 ushort_t cl_tcpi_version; /* cl_tcp_info_t's version no */
547 555 ushort_t cl_tcpi_ipversion; /* IP version */
548 556 int32_t cl_tcpi_state; /* TCP state */
549 557 in_port_t cl_tcpi_lport; /* Local port */
550 558 in_port_t cl_tcpi_fport; /* Remote port */
551 559 in6_addr_t cl_tcpi_laddr_v6; /* Local IP address */
552 560 in6_addr_t cl_tcpi_faddr_v6; /* Remote IP address */
553 561 #ifdef _KERNEL
554 562 /* Note: V4_PART_OF_V6 is meant to be used only for _KERNEL defined stuff */
555 563 #define cl_tcpi_laddr V4_PART_OF_V6(cl_tcpi_laddr_v6)
556 564 #define cl_tcpi_faddr V4_PART_OF_V6(cl_tcpi_faddr_v6)
557 565
558 566 #endif /* _KERNEL */
559 567 } cl_tcp_info_t;
560 568
561 569 /*
562 570 * Hook functions to enable cluster networking
563 571 * On non-clustered systems these vectors must always be NULL.
564 572 */
565 573 extern void (*cl_inet_listen)(netstackid_t, uint8_t, sa_family_t,
566 574 uint8_t *, in_port_t, void *);
567 575 extern void (*cl_inet_unlisten)(netstackid_t, uint8_t, sa_family_t,
568 576 uint8_t *, in_port_t, void *);
569 577
570 578 /*
571 579 * Contracted Consolidation Private ioctl for aborting TCP connections.
572 580 * In order to keep the offsets and size of the structure the same between
573 581 * a 32-bit application and a 64-bit amd64 kernel, we use a #pragma
574 582 * pack(4).
575 583 */
576 584 #define TCP_IOC_ABORT_CONN (('T' << 8) + 91)
577 585
578 586 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
579 587 #pragma pack(4)
580 588 #endif
581 589
582 590 typedef struct tcp_ioc_abort_conn_s {
583 591 struct sockaddr_storage ac_local; /* local addr and port */
584 592 struct sockaddr_storage ac_remote; /* remote addr and port */
585 593 int32_t ac_start; /* start state */
586 594 int32_t ac_end; /* end state */
587 595 int32_t ac_zoneid; /* zoneid */
588 596 } tcp_ioc_abort_conn_t;
589 597
590 598 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
591 599 #pragma pack()
592 600 #endif
593 601
594 602 #ifdef __cplusplus
595 603 }
596 604 #endif
597 605
598 606 #endif /* _INET_TCP_H */
|
↓ open down ↓ |
182 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX