Print this page
DLPX-25998 TCP congestion control is inadequate
Reviewed at: http://reviews.delphix.com/r/34808/
DLPX-45697 Adding Avg. RTT to connstat
DLPX-43064 include high-resolution round-trip times in connstat (EP-652)
DLPX-37540 TCP per-connection kernel statistics DLPX-37544 connstat command to display per-connection TCP statistics
| 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 *
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.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
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 23 * Copyright 2015 Joyent, Inc.
24 24 * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
25 - * Copyright (c) 2014 by Delphix. All rights reserved.
25 + * Copyright (c) 2014, 2017 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>
|
↓ open down ↓ |
3 lines elided |
↑ open up ↑ |
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 +#include <inet/cc.h>
49 50
50 51 /* TCP states */
51 52 #define TCPS_CLOSED -6
52 53 #define TCPS_IDLE -5 /* idle (opened, but not bound) */
53 54 #define TCPS_BOUND -4 /* bound, ready to connect or accept */
54 55 #define TCPS_LISTEN -3 /* listening for connection */
55 56 #define TCPS_SYN_SENT -2 /* active, have sent syn */
56 57 #define TCPS_SYN_RCVD -1 /* have received syn (and sent ours) */
57 58 /* states < TCPS_ESTABLISHED are those where connections not established */
58 59 #define TCPS_ESTABLISHED 0 /* established */
59 60 #define TCPS_CLOSE_WAIT 1 /* rcvd fin, waiting for close */
60 61 /* states > TCPS_CLOSE_WAIT are those where user has closed */
61 62 #define TCPS_FIN_WAIT_1 2 /* have closed and sent fin */
62 63 #define TCPS_CLOSING 3 /* closed, xchd FIN, await FIN ACK */
63 64 #define TCPS_LAST_ACK 4 /* had fin and close; await FIN ACK */
64 65 /* states > TCPS_CLOSE_WAIT && < TCPS_FIN_WAIT_2 await ACK of FIN */
65 66 #define TCPS_FIN_WAIT_2 5 /* have closed, fin is acked */
66 67 #define TCPS_TIME_WAIT 6 /* in 2*msl quiet wait after close */
67 68
68 69 /*
69 70 * Internal flags used in conjunction with the packet header flags.
70 71 * Used in tcp_input_data to keep track of what needs to be done.
71 72 */
72 73 #define TH_LIMIT_XMIT 0x0400 /* Limited xmit is needed */
73 74 #define TH_XMIT_NEEDED 0x0800 /* Window opened - send queued data */
74 75 #define TH_REXMIT_NEEDED 0x1000 /* Time expired for unacked data */
75 76 #define TH_ACK_NEEDED 0x2000 /* Send an ack now. */
76 77 #define TH_NEED_SACK_REXMIT 0x4000 /* Use SACK info to retransmission */
77 78 #define TH_ACK_TIMER_NEEDED 0x8000 /* Start the delayed ACK timer */
78 79 #define TH_ORDREL_NEEDED 0x10000 /* Generate an ordrel indication */
79 80 #define TH_MARKNEXT_NEEDED 0x20000 /* Data should have MSGMARKNEXT */
80 81 #define TH_SEND_URP_MARK 0x40000 /* Send up tcp_urp_mark_mp */
81 82
82 83 /*
83 84 * TCP sequence numbers are 32 bit integers operated
84 85 * on with modular arithmetic. These macros can be
85 86 * used to compare such integers.
86 87 */
87 88 #define SEQ_LT(a, b) ((int32_t)((a)-(b)) < 0)
88 89 #define SEQ_LEQ(a, b) ((int32_t)((a)-(b)) <= 0)
89 90 #define SEQ_GT(a, b) ((int32_t)((a)-(b)) > 0)
90 91 #define SEQ_GEQ(a, b) ((int32_t)((a)-(b)) >= 0)
91 92
92 93 /* TCP Protocol header */
93 94 typedef struct tcphdr_s {
94 95 uint8_t th_lport[2]; /* Source port */
95 96 uint8_t th_fport[2]; /* Destination port */
96 97 uint8_t th_seq[4]; /* Sequence number */
97 98 uint8_t th_ack[4]; /* Acknowledgement number */
98 99 uint8_t th_offset_and_rsrvd[1]; /* Offset to the packet data */
99 100 uint8_t th_flags[1];
100 101 uint8_t th_win[2]; /* Allocation number */
101 102 uint8_t th_sum[2]; /* TCP checksum */
102 103 uint8_t th_urp[2]; /* Urgent pointer */
103 104 } tcph_t;
104 105
105 106 #define TCP_HDR_LENGTH(tcph) \
106 107 ((((tcph_t *)tcph)->th_offset_and_rsrvd[0] >>2) &(0xF << 2))
107 108 #define TCP_MAX_COMBINED_HEADER_LENGTH (60 + 60) /* Maxed out ip + tcp */
108 109 #define TCP_MAX_IP_OPTIONS_LENGTH (60 - IP_SIMPLE_HDR_LENGTH)
109 110 #define TCP_MAX_HDR_LENGTH 60
110 111 #define TCP_MAX_TCP_OPTIONS_LENGTH (60 - sizeof (tcpha_t))
111 112 #define TCP_MIN_HEADER_LENGTH 20
112 113 #define TCP_MAXWIN 65535
113 114 #define TCP_PORT_LEN sizeof (in_port_t)
114 115 #define TCP_MAX_WINSHIFT 14
115 116 #define TCP_MAX_LARGEWIN (TCP_MAXWIN << TCP_MAX_WINSHIFT)
116 117 #define TCP_MAX_LSO_LENGTH (IP_MAXPACKET - TCP_MAX_COMBINED_HEADER_LENGTH)
117 118
118 119 #define TCPIP_HDR_LENGTH(mp, n) \
119 120 (n) = IPH_HDR_LENGTH((mp)->b_rptr), \
120 121 (n) += TCP_HDR_LENGTH((tcpha_t *)&(mp)->b_rptr[(n)])
121 122
122 123 /* TCP Protocol header (used if the header is known to be 32-bit aligned) */
123 124 typedef struct tcphdra_s {
124 125 in_port_t tha_lport; /* Source port */
125 126 in_port_t tha_fport; /* Destination port */
126 127 uint32_t tha_seq; /* Sequence number */
127 128 uint32_t tha_ack; /* Acknowledgement number */
128 129 uint8_t tha_offset_and_reserved; /* Offset to the packet data */
129 130 uint8_t tha_flags;
130 131 uint16_t tha_win; /* Allocation number */
131 132 uint16_t tha_sum; /* TCP checksum */
132 133 uint16_t tha_urp; /* Urgent pointer */
133 134 } tcpha_t;
134 135
135 136 struct conn_s;
136 137 struct tcp_listen_cnt_s;
137 138 struct tcp_rg_s;
138 139
139 140 /*
140 141 * Control structure for each open TCP stream,
141 142 * defined only within the kernel or for a kmem user.
142 143 * NOTE: tcp_reinit_values MUST have a line for each field in this structure!
143 144 */
144 145 #if (defined(_KERNEL) || defined(_KMEMUSER))
145 146
|
↓ open down ↓ |
87 lines elided |
↑ open up ↑ |
146 147 typedef struct tcp_s {
147 148 struct tcp_s *tcp_time_wait_next;
148 149 /* Pointer to next T/W block */
149 150 struct tcp_s *tcp_time_wait_prev;
150 151 /* Pointer to previous T/W next */
151 152 int64_t tcp_time_wait_expire;
152 153
153 154 struct conn_s *tcp_connp; /* back pointer to conn_t */
154 155 tcp_stack_t *tcp_tcps; /* back pointer to tcp_stack_t */
155 156
157 + struct cc_algo *tcp_cc_algo; /* congestion control algorithm */
158 + struct cc_var tcp_ccv; /* congestion control specific vars */
159 +
156 160 int32_t tcp_state;
157 161 int32_t tcp_rcv_ws; /* My window scale power */
158 162 int32_t tcp_snd_ws; /* Sender's window scale power */
159 163 uint32_t tcp_ts_recent; /* Timestamp of earliest unacked */
160 164 /* data segment */
161 165 clock_t tcp_rto; /* Round trip timeout */
162 166 int64_t tcp_last_rcv_lbolt;
163 167 /* lbolt on last packet, used for PAWS */
164 168 uint32_t tcp_rto_initial; /* Initial RTO */
165 169 uint32_t tcp_rto_min; /* Minimum RTO */
166 170 uint32_t tcp_rto_max; /* Maximum RTO */
167 171
168 172 uint32_t tcp_snxt; /* Senders next seq num */
169 173 uint32_t tcp_swnd; /* Senders window (relative to suna) */
170 174 uint32_t tcp_mss; /* Max segment size */
|
↓ open down ↓ |
5 lines elided |
↑ open up ↑ |
171 175 uint32_t tcp_iss; /* Initial send seq num */
172 176 uint32_t tcp_rnxt; /* Seq we expect to recv next */
173 177 uint32_t tcp_rwnd;
174 178
175 179 /* Fields arranged in approximate access order along main paths */
176 180 mblk_t *tcp_xmit_head; /* Head of xmit/rexmit list */
177 181 mblk_t *tcp_xmit_last; /* Last valid data seen by tcp_wput */
178 182 mblk_t *tcp_xmit_tail; /* Last data sent */
179 183 uint32_t tcp_unsent; /* # of bytes in hand that are unsent */
180 184 uint32_t tcp_xmit_tail_unsent; /* # of unsent bytes in xmit_tail */
181 -
182 185 uint32_t tcp_suna; /* Sender unacknowledged */
183 186 uint32_t tcp_rexmit_nxt; /* Next rexmit seq num */
184 187 uint32_t tcp_rexmit_max; /* Max retran seq num */
185 188 uint32_t tcp_cwnd; /* Congestion window */
186 189 int32_t tcp_cwnd_cnt; /* cwnd cnt in congestion avoidance */
187 -
188 - uint32_t tcp_ibsegs; /* Inbound segments on this stream */
189 - uint32_t tcp_obsegs; /* Outbound segments on this stream */
190 -
191 190 uint32_t tcp_naglim; /* Tunable nagle limit */
192 191 uint32_t tcp_valid_bits;
193 192 #define TCP_ISS_VALID 0x1 /* Is the tcp_iss seq num active? */
194 193 #define TCP_FSS_VALID 0x2 /* Is the tcp_fss seq num active? */
195 194 #define TCP_URG_VALID 0x4 /* Is the tcp_urg seq num active? */
196 195 #define TCP_OFO_FIN_VALID 0x8 /* Has TCP received an out of order FIN? */
197 196
198 -
199 -
200 197 timeout_id_t tcp_timer_tid; /* Control block for timer service */
201 198 uchar_t tcp_timer_backoff; /* Backoff shift count. */
202 199 int64_t tcp_last_recv_time; /* Last time we receive a segment. */
203 200 uint32_t tcp_init_cwnd; /* Initial cwnd (start/restart) */
204 201
205 202 /* Following manipulated by TCP under squeue protection */
206 203 uint32_t
207 204 tcp_urp_last_valid : 1, /* Is tcp_urp_last valid? */
208 205 tcp_hard_binding : 1, /* TCP_DETACHED_NONEAGER */
209 206 tcp_fin_acked : 1, /* Has our FIN been acked? */
210 207 tcp_fin_rcvd : 1, /* Have we seen a FIN? */
211 208
212 209 tcp_fin_sent : 1, /* Have we sent our FIN yet? */
213 210 tcp_ordrel_done : 1, /* Have we sent the ord_rel upstream? */
214 211 tcp_detached : 1, /* If we're detached from a stream */
215 212 tcp_zero_win_probe: 1, /* Zero win probing is in progress */
216 213
217 214 tcp_loopback: 1, /* src and dst are the same machine */
218 215 tcp_localnet: 1, /* src and dst are on the same subnet */
219 216 tcp_syn_defense: 1, /* For defense against SYN attack */
220 217 #define tcp_dontdrop tcp_syn_defense
221 218 tcp_set_timer : 1,
222 219
223 220 tcp_active_open: 1, /* This is a active open */
224 221 tcp_rexmit : 1, /* TCP is retransmitting */
225 222 tcp_snd_sack_ok : 1, /* Can use SACK for this connection */
226 223 tcp_hwcksum : 1, /* The NIC is capable of hwcksum */
227 224
228 225 tcp_ip_forward_progress : 1,
229 226 tcp_ecn_ok : 1, /* Can use ECN for this connection */
230 227 tcp_ecn_echo_on : 1, /* Need to do ECN echo */
231 228 tcp_ecn_cwr_sent : 1, /* ECN_CWR has been sent */
232 229
233 230 tcp_cwr : 1, /* Cwnd has reduced recently */
234 231
235 232 tcp_pad_to_bit31 : 11;
236 233
237 234 /* Following manipulated by TCP under squeue protection */
238 235 uint32_t
239 236 tcp_snd_ts_ok : 1,
240 237 tcp_snd_ws_ok : 1,
241 238 tcp_reserved_port : 1,
242 239 tcp_in_free_list : 1,
243 240
244 241 tcp_snd_zcopy_on : 1, /* xmit zero-copy enabled */
245 242 tcp_snd_zcopy_aware : 1, /* client is zero-copy aware */
246 243 tcp_xmit_zc_clean : 1, /* the xmit list is free of zc-mblk */
247 244 tcp_wait_for_eagers : 1, /* Wait for eagers to disappear */
248 245
249 246 tcp_accept_error : 1, /* Error during TLI accept */
250 247 tcp_send_discon_ind : 1, /* TLI accept err, send discon ind */
251 248 tcp_cork : 1, /* tcp_cork option */
252 249 tcp_tconnind_started : 1, /* conn_ind message is being sent */
253 250
254 251 tcp_lso :1, /* Lower layer is capable of LSO */
255 252 tcp_is_wnd_shrnk : 1, /* Window has shrunk */
256 253
257 254 tcp_pad_to_bit_31 : 18;
258 255
259 256 uint32_t tcp_initial_pmtu; /* Initial outgoing Path MTU. */
260 257
261 258 mblk_t *tcp_reass_head; /* Out of order reassembly list head */
262 259 mblk_t *tcp_reass_tail; /* Out of order reassembly list tail */
263 260
264 261 /* SACK related info */
265 262 tcp_sack_info_t tcp_sack_info;
266 263
267 264 #define tcp_pipe tcp_sack_info.tcp_pipe
268 265 #define tcp_fack tcp_sack_info.tcp_fack
269 266 #define tcp_sack_snxt tcp_sack_info.tcp_sack_snxt
270 267 #define tcp_max_sack_blk tcp_sack_info.tcp_max_sack_blk
271 268 #define tcp_num_sack_blk tcp_sack_info.tcp_num_sack_blk
272 269 #define tcp_sack_list tcp_sack_info.tcp_sack_list
273 270 #define tcp_num_notsack_blk tcp_sack_info.tcp_num_notsack_blk
274 271 #define tcp_cnt_notsack_list tcp_sack_info.tcp_cnt_notsack_list
275 272 #define tcp_notsack_list tcp_sack_info.tcp_notsack_list
|
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
276 273
277 274 mblk_t *tcp_rcv_list; /* Queued until push, urgent data, */
278 275 mblk_t *tcp_rcv_last_head; /* optdata, or the count exceeds */
279 276 mblk_t *tcp_rcv_last_tail; /* tcp_rcv_push_wait. */
280 277 uint32_t tcp_rcv_cnt; /* tcp_rcv_list is b_next chain. */
281 278
282 279 uint32_t tcp_cwnd_ssthresh; /* Congestion window */
283 280 uint32_t tcp_cwnd_max;
284 281 uint32_t tcp_csuna; /* Clear (no rexmits in window) suna */
285 282
286 - clock_t tcp_rtt_sa; /* Round trip smoothed average */
287 - clock_t tcp_rtt_sd; /* Round trip smoothed deviation */
288 - clock_t tcp_rtt_update; /* Round trip update(s) */
283 + hrtime_t tcp_rtt_sum; /* Round trip sum */
284 + uint32_t tcp_rtt_cnt; /* Round trip count (non_dup ACKs) */
285 + hrtime_t tcp_rtt_sa; /* Round trip smoothed average */
286 + hrtime_t tcp_rtt_sd; /* Round trip smoothed deviation */
287 + uint32_t tcp_rtt_update; /* Round trip update(s) */
289 288 clock_t tcp_ms_we_have_waited; /* Total retrans time */
290 289
291 290 uint32_t tcp_swl1; /* These help us avoid using stale */
292 291 uint32_t tcp_swl2; /* packets to update state */
293 292
294 293 uint32_t tcp_rack; /* Seq # we have acked */
295 294 uint32_t tcp_rack_cnt; /* # of segs we have deferred ack */
296 295 uint32_t tcp_rack_cur_max; /* # of segs we may defer ack for now */
297 296 uint32_t tcp_rack_abs_max; /* # of segs we may defer ack ever */
298 297 timeout_id_t tcp_ack_tid; /* Delayed ACK timer ID */
299 298 timeout_id_t tcp_push_tid; /* Push timer ID */
300 299
301 300 uint32_t tcp_max_swnd; /* Maximum swnd we have seen */
302 301
303 302 struct tcp_s *tcp_listener; /* Our listener */
304 303
305 304 uint32_t tcp_irs; /* Initial recv seq num */
306 305 uint32_t tcp_fss; /* Final/fin send seq num */
307 306 uint32_t tcp_urg; /* Urgent data seq num */
308 307
309 308 clock_t tcp_first_timer_threshold; /* When to prod IP */
310 309 clock_t tcp_second_timer_threshold; /* When to give up completely */
311 310 clock_t tcp_first_ctimer_threshold; /* 1st threshold while connecting */
312 311 clock_t tcp_second_ctimer_threshold; /* 2nd ... while connecting */
313 312
314 313 uint32_t tcp_urp_last; /* Last urp for which signal sent */
315 314 mblk_t *tcp_urp_mp; /* T_EXDATA_IND for urgent byte */
316 315 mblk_t *tcp_urp_mark_mp; /* zero-length marked/unmarked msg */
317 316
318 317 int tcp_conn_req_cnt_q0; /* # of conn reqs in SYN_RCVD */
319 318 int tcp_conn_req_cnt_q; /* # of conn reqs in ESTABLISHED */
320 319 int tcp_conn_req_max; /* # of ESTABLISHED conn reqs allowed */
321 320 t_scalar_t tcp_conn_req_seqnum; /* Incrementing pending conn req ID */
322 321 #define tcp_ip_addr_cache tcp_reass_tail
323 322 /* Cache ip addresses that */
324 323 /* complete the 3-way handshake */
325 324 kmutex_t tcp_eager_lock;
326 325 struct tcp_s *tcp_eager_next_q; /* next eager in ESTABLISHED state */
327 326 struct tcp_s *tcp_eager_last_q; /* last eager in ESTABLISHED state */
328 327 struct tcp_s *tcp_eager_next_q0; /* next eager in SYN_RCVD state */
329 328 struct tcp_s *tcp_eager_prev_q0; /* prev eager in SYN_RCVD state */
330 329 /* all eagers form a circular list */
331 330 boolean_t tcp_conn_def_q0; /* move from q0 to q deferred */
332 331
333 332 union {
334 333 mblk_t *tcp_eager_conn_ind; /* T_CONN_IND waiting for 3rd ack. */
335 334 mblk_t *tcp_opts_conn_req; /* T_CONN_REQ w/ options processed */
336 335 } tcp_conn;
337 336 uint32_t tcp_syn_rcvd_timeout; /* How many SYN_RCVD timeout in q0 */
338 337
339 338 /*
340 339 * TCP Keepalive Timer members.
341 340 * All keepalive timer intervals are in milliseconds.
342 341 */
343 342 int32_t tcp_ka_last_intrvl; /* Last probe interval */
344 343 timeout_id_t tcp_ka_tid; /* Keepalive timer ID */
345 344 uint32_t tcp_ka_interval; /* Keepalive interval */
346 345
347 346 /*
348 347 * TCP connection is terminated if we don't hear back from the peer
349 348 * for tcp_ka_abort_thres milliseconds after the first keepalive probe.
350 349 * tcp_ka_rinterval is the interval in milliseconds between successive
351 350 * keepalive probes. tcp_ka_cnt is the number of keepalive probes to
352 351 * be sent before terminating the connection, if we don't hear back from
353 352 * peer.
354 353 * tcp_ka_abort_thres = tcp_ka_rinterval * tcp_ka_cnt
355 354 */
356 355 uint32_t tcp_ka_rinterval; /* keepalive retransmit interval */
357 356 uint32_t tcp_ka_abort_thres; /* Keepalive abort threshold */
358 357 uint32_t tcp_ka_cnt; /* count of keepalive probes */
359 358
360 359 int32_t tcp_client_errno; /* How the client screwed up */
361 360
362 361 /*
363 362 * The header template lives in conn_ht_iphc allocated by tcp_build_hdrs
364 363 * We maintain three pointers into conn_ht_iphc.
365 364 */
366 365 ipha_t *tcp_ipha; /* IPv4 header in conn_ht_iphc */
367 366 ip6_t *tcp_ip6h; /* IPv6 header in conn_ht_iphc */
368 367 tcpha_t *tcp_tcpha; /* TCP header in conn_ht_iphc */
369 368
370 369 uint16_t tcp_last_sent_len; /* Record length for nagle */
371 370 uint16_t tcp_last_recv_len; /* Used by DTrace */
372 371 uint16_t tcp_dupack_cnt; /* # of consequtive duplicate acks */
373 372
374 373 kmutex_t *tcp_acceptor_lockp; /* Ptr to tf_lock */
375 374
376 375 mblk_t *tcp_ordrel_mp; /* T_ordrel_ind mblk */
377 376 t_uscalar_t tcp_acceptor_id; /* ACCEPTOR_id */
378 377
379 378 int tcp_ipsec_overhead;
380 379
381 380 uint_t tcp_recvifindex; /* Last received IPV6_RCVPKTINFO */
382 381 uint_t tcp_recvhops; /* Last received IPV6_RECVHOPLIMIT */
383 382 uint_t tcp_recvtclass; /* Last received IPV6_RECVTCLASS */
384 383 ip6_hbh_t *tcp_hopopts; /* Last received IPV6_RECVHOPOPTS */
385 384 ip6_dest_t *tcp_dstopts; /* Last received IPV6_RECVDSTOPTS */
386 385 ip6_dest_t *tcp_rthdrdstopts; /* Last recv IPV6_RECVRTHDRDSTOPTS */
387 386 ip6_rthdr_t *tcp_rthdr; /* Last received IPV6_RECVRTHDR */
388 387 uint_t tcp_hopoptslen;
389 388 uint_t tcp_dstoptslen;
390 389 uint_t tcp_rthdrdstoptslen;
391 390 uint_t tcp_rthdrlen;
392 391
393 392 mblk_t *tcp_timercache;
394 393
395 394 kmutex_t tcp_closelock;
396 395 kcondvar_t tcp_closecv;
397 396 uint8_t tcp_closed;
398 397 uint8_t tcp_closeflags;
399 398 mblk_t tcp_closemp;
400 399 timeout_id_t tcp_linger_tid; /* Linger timer ID */
401 400
402 401 struct tcp_s *tcp_acceptor_hash; /* Acceptor hash chain */
403 402 struct tcp_s **tcp_ptpahn; /* Pointer to previous accept hash next. */
404 403 struct tcp_s *tcp_bind_hash; /* Bind hash chain */
405 404 struct tcp_s *tcp_bind_hash_port; /* tcp_t's bound to the same lport */
406 405 struct tcp_s **tcp_ptpbhn;
407 406
408 407 /*
409 408 * Group of tcp_t entries bound to the same adress and port via
410 409 * SO_REUSEPORT. The pointer itself is protected by tf_lock in the
411 410 * containing tcps_bind_fanout slot.
412 411 */
413 412 struct tcp_rg_s *tcp_rg_bind;
414 413
415 414 uint_t tcp_maxpsz_multiplier;
416 415
417 416 uint32_t tcp_lso_max; /* maximum LSO payload */
418 417
419 418 uint32_t tcp_ofo_fin_seq; /* Recv out of order FIN seq num */
420 419 uint32_t tcp_cwr_snd_max;
421 420
422 421 struct tcp_s *tcp_saved_listener; /* saved value of listener */
423 422
424 423 uint32_t tcp_in_ack_unsent; /* ACK for unsent data cnt. */
425 424
426 425 /*
427 426 * All fusion-related fields are protected by squeue.
428 427 */
429 428 struct tcp_s *tcp_loopback_peer; /* peer tcp for loopback */
430 429 mblk_t *tcp_fused_sigurg_mp; /* M_PCSIG mblk for SIGURG */
431 430
432 431 uint32_t
433 432 tcp_fused : 1, /* loopback tcp in fusion mode */
434 433 tcp_unfusable : 1, /* fusion not allowed on endpoint */
435 434 tcp_fused_sigurg : 1, /* send SIGURG upon draining */
436 435
437 436 tcp_fuse_to_bit_31 : 29;
438 437
439 438 kmutex_t tcp_non_sq_lock;
440 439
441 440 /*
442 441 * This variable is accessed without any lock protection
443 442 * and therefore must not be declared as a bit field along
444 443 * with the rest which require such condition.
445 444 */
446 445 boolean_t tcp_issocket; /* this is a socket tcp */
447 446
448 447 /* protected by the tcp_non_sq_lock lock */
449 448 uint32_t tcp_squeue_bytes;
450 449
451 450 /*
452 451 * tcp_closemp_used is protected by listener's tcp_eager_lock
453 452 * when used for eagers. When used for a tcp in TIME_WAIT state
454 453 * or in tcp_close(), it is not protected by any lock as we
455 454 * do not expect any other thread to use it concurrently.
456 455 * We do allow re-use of tcp_closemp in tcp_time_wait_collector()
457 456 * and tcp_close() but not concurrently.
458 457 */
459 458 boolean_t tcp_closemp_used;
460 459
461 460 /*
462 461 * previous and next eagers in the list of droppable eagers. See
463 462 * the comments before MAKE_DROPPABLE(). These pointers are
464 463 * protected by listener's tcp_eager_lock.
465 464 */
466 465 struct tcp_s *tcp_eager_prev_drop_q0;
467 466 struct tcp_s *tcp_eager_next_drop_q0;
468 467
469 468 /*
470 469 * Have we flow controlled xmitter?
471 470 * This variable can be modified outside the squeue and hence must
472 471 * not be declared as a bit field along with the rest that are
473 472 * modified only within the squeue.
474 473 * protected by the tcp_non_sq_lock lock.
475 474 */
476 475 boolean_t tcp_flow_stopped;
477 476
478 477 /*
479 478 * Sender's next sequence number at the time the window was shrunk.
480 479 */
481 480 uint32_t tcp_snxt_shrunk;
482 481
483 482 /*
484 483 * Socket generation number which is bumped when a connection attempt
485 484 * is initiated. Its main purpose is to ensure that the socket does not
486 485 * miss the asynchronous connected/disconnected notification.
487 486 */
488 487 sock_connid_t tcp_connid;
489 488
490 489 /* mblk_t used to enter TCP's squeue from the service routine. */
491 490 mblk_t *tcp_rsrv_mp;
492 491 /* Mutex for accessing tcp_rsrv_mp */
493 492 kmutex_t tcp_rsrv_mp_lock;
|
↓ open down ↓ |
195 lines elided |
↑ open up ↑ |
494 493
495 494 /* For connection counting. */
496 495 struct tcp_listen_cnt_s *tcp_listen_cnt;
497 496
498 497 /* Segment reassembly timer. */
499 498 timeout_id_t tcp_reass_tid;
500 499
501 500 /* FIN-WAIT-2 flush timeout */
502 501 uint32_t tcp_fin_wait_2_flush_interval;
503 502
503 + tcp_conn_stats_t tcp_cs;
504 +
504 505 #ifdef DEBUG
505 506 pc_t tcmp_stk[15];
506 507 #endif
507 508 } tcp_t;
508 509
509 510 #ifdef DEBUG
510 511 #define TCP_DEBUG_GETPCSTACK(buffer, depth) ((void) getpcstack(buffer, \
511 512 depth))
512 513 #else
513 514 #define TCP_DEBUG_GETPCSTACK(buffer, depth)
514 515 #endif
515 516
516 517 extern void tcp_conn_reclaim(void *);
517 -extern void tcp_free(tcp_t *tcp);
518 +extern void tcp_free(tcp_t *tcp);
518 519 extern void tcp_ddi_g_init(void);
519 520 extern void tcp_ddi_g_destroy(void);
520 -extern void *tcp_get_conn(void *arg, tcp_stack_t *);
521 +extern conn_t *tcp_get_conn(void *arg, tcp_stack_t *);
521 522 extern mblk_t *tcp_snmp_get(queue_t *, mblk_t *, boolean_t);
522 523 extern int tcp_snmp_set(queue_t *, int, int, uchar_t *, int len);
523 524
524 525 /* Pad for the tf_t structure to avoid false cache line sharing. */
525 526 #define TF_CACHEL_PAD 64
526 527
527 528 /*
528 529 * The TCP Fanout structure for bind and acceptor hashes.
529 530 * The hash tables and their linkage (tcp_*_hash, tcp_ptp*hn) are
530 531 * protected by the per-bucket tf_lock. Each tcp_t
531 532 * inserted in the list points back at this lock using tcp_*_lockp.
532 533 *
533 534 * The bind and acceptor hash queues are lists of tcp_t.
534 535 */
535 536 /* listener hash and acceptor hash queue head */
536 537 typedef struct tf_s {
537 538 tcp_t *tf_tcp;
538 539 kmutex_t tf_lock;
539 540 unsigned char tf_pad[TF_CACHEL_PAD -
540 541 (sizeof (tcp_t *) + sizeof (kmutex_t))];
541 542 } tf_t;
542 543
543 544
544 545 /* Also used in ipclassifier.c */
545 546 extern struct kmem_cache *tcp_sack_info_cache;
546 547
547 548 #endif /* (defined(_KERNEL) || defined(_KMEMUSER)) */
548 549
549 550 /* Contract private interface between TCP and Clustering. */
550 551
551 552 #define CL_TCPI_V1 1 /* cl_tcpi_version number */
552 553
553 554 typedef struct cl_tcp_info_s {
554 555 ushort_t cl_tcpi_version; /* cl_tcp_info_t's version no */
555 556 ushort_t cl_tcpi_ipversion; /* IP version */
556 557 int32_t cl_tcpi_state; /* TCP state */
557 558 in_port_t cl_tcpi_lport; /* Local port */
558 559 in_port_t cl_tcpi_fport; /* Remote port */
559 560 in6_addr_t cl_tcpi_laddr_v6; /* Local IP address */
560 561 in6_addr_t cl_tcpi_faddr_v6; /* Remote IP address */
561 562 #ifdef _KERNEL
562 563 /* Note: V4_PART_OF_V6 is meant to be used only for _KERNEL defined stuff */
563 564 #define cl_tcpi_laddr V4_PART_OF_V6(cl_tcpi_laddr_v6)
564 565 #define cl_tcpi_faddr V4_PART_OF_V6(cl_tcpi_faddr_v6)
565 566
566 567 #endif /* _KERNEL */
567 568 } cl_tcp_info_t;
568 569
569 570 /*
570 571 * Hook functions to enable cluster networking
571 572 * On non-clustered systems these vectors must always be NULL.
572 573 */
573 574 extern void (*cl_inet_listen)(netstackid_t, uint8_t, sa_family_t,
574 575 uint8_t *, in_port_t, void *);
575 576 extern void (*cl_inet_unlisten)(netstackid_t, uint8_t, sa_family_t,
576 577 uint8_t *, in_port_t, void *);
577 578
578 579 /*
579 580 * Contracted Consolidation Private ioctl for aborting TCP connections.
580 581 * In order to keep the offsets and size of the structure the same between
581 582 * a 32-bit application and a 64-bit amd64 kernel, we use a #pragma
582 583 * pack(4).
583 584 */
584 585 #define TCP_IOC_ABORT_CONN (('T' << 8) + 91)
585 586
586 587 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
587 588 #pragma pack(4)
588 589 #endif
589 590
590 591 typedef struct tcp_ioc_abort_conn_s {
591 592 struct sockaddr_storage ac_local; /* local addr and port */
592 593 struct sockaddr_storage ac_remote; /* remote addr and port */
593 594 int32_t ac_start; /* start state */
594 595 int32_t ac_end; /* end state */
595 596 int32_t ac_zoneid; /* zoneid */
596 597 } tcp_ioc_abort_conn_t;
597 598
598 599 #if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
599 600 #pragma pack()
600 601 #endif
601 602
602 603 #ifdef __cplusplus
603 604 }
604 605 #endif
605 606
606 607 #endif /* _INET_TCP_H */
|
↓ open down ↓ |
76 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX