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