blob: bc6e353b60ef7f67d82adb384b972bd18737b0eb [file] [log] [blame]
Dave Barach68b0fb02017-02-28 15:15:56 -05001/*
Florin Coras222e1f412019-02-16 20:47:32 -08002 * Copyright (c) 2016-2019 Cisco and/or its affiliates.
Dave Barach68b0fb02017-02-28 15:15:56 -05003 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
6 *
7 * http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 */
15
16#ifndef _vnet_tcp_h_
17#define _vnet_tcp_h_
18
19#include <vnet/vnet.h>
20#include <vnet/ip/ip.h>
Dave Barach68b0fb02017-02-28 15:15:56 -050021#include <vnet/session/session.h>
Florin Coras999840c2020-03-18 20:31:34 +000022#include <vnet/tcp/tcp_types.h>
23#include <vnet/tcp/tcp_timer.h>
Florin Corase69f4952017-03-07 10:06:24 -080024#include <vnet/tcp/tcp_debug.h>
Florin Coras999840c2020-03-18 20:31:34 +000025#include <vnet/tcp/tcp_sack.h>
26#include <vnet/tcp/tcp_bt.h>
27#include <vnet/tcp/tcp_cc.h>
Florin Corasbbcfaac2019-10-10 13:52:04 -070028
Florin Corasaa388692020-02-14 23:41:25 +000029typedef void (timer_expiration_handler) (tcp_connection_t * tc);
30
Florin Corasaa388692020-02-14 23:41:25 +000031extern timer_expiration_handler tcp_timer_retransmit_handler;
32extern timer_expiration_handler tcp_timer_persist_handler;
33extern timer_expiration_handler tcp_timer_retransmit_syn_handler;
34
Dave Barach68b0fb02017-02-28 15:15:56 -050035typedef enum _tcp_error
36{
37#define tcp_error(n,s) TCP_ERROR_##n,
38#include <vnet/tcp/tcp_error.def>
39#undef tcp_error
40 TCP_N_ERROR,
41} tcp_error_t;
42
43typedef struct _tcp_lookup_dispatch
44{
45 u8 next, error;
46} tcp_lookup_dispatch_t;
47
Florin Coras5e6305f2020-02-12 07:42:01 +000048#define foreach_tcp_wrk_stat \
49 _(timer_expirations, u64, "timer expirations") \
50 _(rxt_segs, u64, "segments retransmitted") \
51 _(tr_events, u32, "timer retransmit events") \
52 _(to_closewait, u32, "timeout close-wait") \
Florin Corasaa388692020-02-14 23:41:25 +000053 _(to_closewait2, u32, "timeout close-wait w/data") \
Florin Coras5e6305f2020-02-12 07:42:01 +000054 _(to_finwait1, u32, "timeout fin-wait-1") \
Florin Coras3f78bbb2020-02-13 19:24:58 +000055 _(to_finwait2, u32, "timeout fin-wait-2") \
Florin Coras5e6305f2020-02-12 07:42:01 +000056 _(to_lastack, u32, "timeout last-ack") \
57 _(to_closing, u32, "timeout closing") \
58 _(tr_abort, u32, "timer retransmit abort") \
59 _(rst_unread, u32, "reset on close due to unread data") \
Florin Corasd4712a82020-05-22 21:51:30 +000060 _(no_buffer, u32, "out of buffers") \
Florin Coras5e6305f2020-02-12 07:42:01 +000061
62typedef struct tcp_wrk_stats_
63{
64#define _(name, type, str) type name;
65 foreach_tcp_wrk_stat
66#undef _
67} tcp_wrk_stats_t;
68
Florin Coras7a3a8662020-02-22 02:27:21 +000069typedef struct tcp_free_req_
70{
71 clib_time_type_t free_time;
72 u32 connection_index;
73} tcp_cleanup_req_t;
74
Florin Coras2c414432018-06-19 09:58:04 -070075typedef struct tcp_worker_ctx_
76{
77 CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
Florin Corasb11175d2018-11-09 14:34:08 -080078
Florin Coras04261852020-02-12 01:24:29 +000079 /** worker's pool of connections */
80 tcp_connection_t *connections;
Florin Corasb11175d2018-11-09 14:34:08 -080081
Florin Corasb11175d2018-11-09 14:34:08 -080082 /** vector of pending ack dequeues */
Florin Coras9ece3c02018-11-05 11:06:53 -080083 u32 *pending_deq_acked;
Florin Corasb11175d2018-11-09 14:34:08 -080084
Florin Corasb11175d2018-11-09 14:34:08 -080085 /** vector of pending disconnect notifications */
86 u32 *pending_disconnects;
87
Florin Coras04261852020-02-12 01:24:29 +000088 /** vector of pending reset notifications */
Florin Coras6939d5e2020-02-10 23:22:34 +000089 u32 *pending_resets;
90
Florin Corasb11175d2018-11-09 14:34:08 -080091 /** convenience pointer to this thread's vlib main */
92 vlib_main_t *vm;
Florin Corasbf4d5ce2018-10-19 16:26:24 -070093
Florin Coraseedc74b2020-07-31 12:32:40 -070094 /** Time measured in @ref TCP_TSTAMP_TICK used for time stamps */
Florin Coras6939d5e2020-02-10 23:22:34 +000095 u32 time_now;
96
Florin Corasa9d8cb42020-02-20 05:45:31 +000097 /* Max timers to be handled per dispatch loop */
98 u32 max_timers_per_loop;
99
Florin Coras5484daa2020-03-27 23:55:06 +0000100 /** Session layer edge indices to tcp output */
101 u32 tco_next_node[2];
102
103 /* Fifo of pending timer expirations */
104 u32 *pending_timers;
Florin Coras04261852020-02-12 01:24:29 +0000105
Florin Corasb26743d2018-06-26 09:31:04 -0700106 CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
Florin Corasb11175d2018-11-09 14:34:08 -0800107
108 /** cached 'on the wire' options for bursts */
109 u8 cached_opts[40];
Florin Corasbf4d5ce2018-10-19 16:26:24 -0700110
Florin Coras04261852020-02-12 01:24:29 +0000111 /** tx buffer free list */
112 u32 *tx_buffers;
113
Florin Coras7a3a8662020-02-22 02:27:21 +0000114 /* fifo of pending free requests */
115 tcp_cleanup_req_t *pending_cleanups;
116
Florin Coras6939d5e2020-02-10 23:22:34 +0000117 /** worker timer wheel */
Florin Coras0765d972020-03-18 21:26:41 +0000118 tcp_timer_wheel_t timer_wheel;
Florin Coras6939d5e2020-02-10 23:22:34 +0000119
Florin Coras5e6305f2020-02-12 07:42:01 +0000120 CLIB_CACHE_LINE_ALIGN_MARK (cacheline2);
121
122 tcp_wrk_stats_t stats;
Florin Coras2c414432018-06-19 09:58:04 -0700123} tcp_worker_ctx_t;
124
Florin Coras0765d972020-03-18 21:26:41 +0000125#define tcp_worker_stats_inc(_wrk,_stat,_val) \
Florin Coras5e6305f2020-02-12 07:42:01 +0000126 _wrk->stats._stat += _val
127
Florin Coras18e0d4f2019-01-02 12:22:02 -0800128typedef struct tcp_iss_seed_
129{
130 u64 first;
131 u64 second;
132} tcp_iss_seed_t;
133
Florin Coras9094b5c2019-08-12 14:17:47 -0700134typedef struct tcp_configuration_
135{
136 /** Max rx fifo size for a session (in bytes). It is used in to compute the
137 * rfc 7323 window scaling factor */
138 u32 max_rx_fifo;
139
140 /** Min rx fifo for a session (in bytes) */
141 u32 min_rx_fifo;
142
143 /** Default MTU to be used when establishing connections */
144 u16 default_mtu;
145
146 /** Initial CWND multiplier, which multiplies MSS to determine initial CWND.
147 * Set 0 to determine the initial CWND by another way */
148 u16 initial_cwnd_multiplier;
149
150 /** Enable tx pacing for new connections */
151 u8 enable_tx_pacing;
152
Florin Corasbbcfaac2019-10-10 13:52:04 -0700153 /** Allow use of TSO whenever available */
154 u8 allow_tso;
155
Florin Corasf4ce6ba2019-11-20 18:34:58 -0800156 /** Set if csum offloading is enabled */
157 u8 csum_offload;
158
Florin Coras9094b5c2019-08-12 14:17:47 -0700159 /** Default congestion control algorithm type */
160 tcp_cc_algorithm_type_e cc_algo;
161
Florin Coras017dc452019-08-30 11:06:35 -0700162 /** Min rwnd, as number of snd_mss segments, for update ack to be sent after
163 * a zero rwnd advertisement */
164 u32 rwnd_min_update_ack;
165
Florin Coras9094b5c2019-08-12 14:17:47 -0700166 /** Delayed ack time (disabled) */
167 u16 delack_time;
168
169 /** Timer ticks to wait for close from app */
170 u16 closewait_time;
171
172 /** Timer ticks to wait in time-wait. Also known as 2MSL */
173 u16 timewait_time;
174
175 /** Timer ticks to wait in fin-wait1 to send fin and rcv fin-ack */
176 u16 finwait1_time;
177
178 /** Timer ticks to wait in last ack for ack */
179 u16 lastack_time;
180
181 /** Timer ticks to wait in fin-wait2 for fin */
182 u16 finwait2_time;
183
184 /** Timer ticks to wait in closing for fin ack */
185 u16 closing_time;
186
Florin Coras7a3a8662020-02-22 02:27:21 +0000187 /** Time to wait (sec) before cleaning up the connection */
188 f32 cleanup_time;
Florin Coras9094b5c2019-08-12 14:17:47 -0700189
190 /** Number of preallocated connections */
191 u32 preallocated_connections;
192
193 /** Number of preallocated half-open connections */
194 u32 preallocated_half_open_connections;
195
196 /** Vectors of src addresses. Optional unless one needs > 63K active-opens */
197 ip4_address_t *ip4_src_addrs;
198 ip6_address_t *ip6_src_addrs;
199
200 /** Fault-injection. Debug only */
201 f64 buffer_fail_fraction;
202} tcp_configuration_t;
203
Dave Barach68b0fb02017-02-28 15:15:56 -0500204typedef struct _tcp_main
205{
Florin Coras04261852020-02-12 01:24:29 +0000206 /** per-worker context */
207 tcp_worker_ctx_t *wrk_ctx;
Dave Barach68b0fb02017-02-28 15:15:56 -0500208
209 /* Pool of listeners. */
210 tcp_connection_t *listener_pool;
211
Florin Corasca031862018-09-24 13:58:05 -0700212 /** vlib buffer size */
213 u32 bytes_per_buffer;
214
Florin Coras5484daa2020-03-27 23:55:06 +0000215 /** Session layer edge indices to ip lookup (syns, rst) */
216 u32 ipl_next_node[2];
217
Florin Coras04261852020-02-12 01:24:29 +0000218 /** Dispatch table by state and flags */
219 tcp_lookup_dispatch_t dispatch_table[TCP_N_STATES][64];
220
221 clib_spinlock_t half_open_lock;
222
223 /** Pool of half-open connections on which we've sent a SYN */
224 tcp_connection_t *half_open_connections;
225
Florin Coras9094b5c2019-08-12 14:17:47 -0700226 /** Seed used to generate random iss */
Florin Coras18e0d4f2019-01-02 12:22:02 -0800227 tcp_iss_seed_t iss_seed;
228
Florin Coras9094b5c2019-08-12 14:17:47 -0700229 /** Congestion control algorithms registered */
Florin Coras4e116fb2019-06-10 08:33:50 -0700230 tcp_cc_algorithm_t *cc_algos;
231
Florin Corasfbf278a2019-03-26 14:05:38 -0700232 /** Hash table of cc algorithms by name */
233 uword *cc_algo_by_name;
234
Florin Coras4e116fb2019-06-10 08:33:50 -0700235 /** Last cc algo registered */
236 tcp_cc_algorithm_type_e cc_last_type;
237
Florin Coras9094b5c2019-08-12 14:17:47 -0700238 /** Flag that indicates if stack is on or off */
Florin Corasa0b34a72017-03-07 01:20:52 -0800239 u8 is_enabled;
240
Florin Coras9094b5c2019-08-12 14:17:47 -0700241 /** Flag that indicates if v4 punting is enabled */
Pierre Pfister7fe51f32017-09-20 08:48:36 +0200242 u8 punt_unknown4;
Florin Coras9094b5c2019-08-12 14:17:47 -0700243
244 /** Flag that indicates if v6 punting is enabled */
Pierre Pfister7fe51f32017-09-20 08:48:36 +0200245 u8 punt_unknown6;
Florin Corasf988e692017-11-27 04:34:14 -0500246
Florin Coras9094b5c2019-08-12 14:17:47 -0700247 /** Rotor for v4 source addresses */
248 u32 last_v4_addr_rotor;
Florin Coras2e31cc32018-09-25 14:00:34 -0700249
Florin Coras9094b5c2019-08-12 14:17:47 -0700250 /** Rotor for v6 source addresses */
251 u32 last_v6_addr_rotor;
252
253 /** Protocol configuration */
254 tcp_configuration_t cfg;
Dave Barach68b0fb02017-02-28 15:15:56 -0500255} tcp_main_t;
256
257extern tcp_main_t tcp_main;
258extern vlib_node_registration_t tcp4_input_node;
259extern vlib_node_registration_t tcp6_input_node;
260extern vlib_node_registration_t tcp4_output_node;
261extern vlib_node_registration_t tcp6_output_node;
Filip Tehlare275bed2019-03-06 00:06:56 -0800262extern vlib_node_registration_t tcp4_established_node;
263extern vlib_node_registration_t tcp6_established_node;
264extern vlib_node_registration_t tcp4_syn_sent_node;
265extern vlib_node_registration_t tcp6_syn_sent_node;
266extern vlib_node_registration_t tcp4_rcv_process_node;
267extern vlib_node_registration_t tcp6_rcv_process_node;
268extern vlib_node_registration_t tcp4_listen_node;
269extern vlib_node_registration_t tcp6_listen_node;
Dave Barach68b0fb02017-02-28 15:15:56 -0500270
Florin Coras9094b5c2019-08-12 14:17:47 -0700271#define tcp_cfg tcp_main.cfg
Florin Coras1f421012019-07-26 10:18:51 -0700272#define tcp_node_index(node_id, is_ip4) \
273 ((is_ip4) ? tcp4_##node_id##_node.index : tcp6_##node_id##_node.index)
274
Dave Barach68b0fb02017-02-28 15:15:56 -0500275always_inline tcp_main_t *
276vnet_get_tcp_main ()
277{
278 return &tcp_main;
279}
280
Florin Corasbe72ae62018-11-01 11:23:03 -0700281always_inline tcp_worker_ctx_t *
282tcp_get_worker (u32 thread_index)
283{
Florin Coras04261852020-02-12 01:24:29 +0000284 ASSERT (thread_index < vec_len (tcp_main.wrk_ctx));
Florin Corasbe72ae62018-11-01 11:23:03 -0700285 return &tcp_main.wrk_ctx[thread_index];
286}
287
Florin Corasf988e692017-11-27 04:34:14 -0500288#if (VLIB_BUFFER_TRACE_TRAJECTORY)
289#define tcp_trajectory_add_start(b, start) \
290{ \
291 (*vlib_buffer_trace_trajectory_cb) (b, start); \
292}
293#else
294#define tcp_trajectory_add_start(b, start)
295#endif
296
Florin Coras8124cb72018-12-16 20:57:29 -0800297tcp_connection_t *tcp_connection_alloc (u8 thread_index);
Florin Coras12f69362019-08-16 09:44:00 -0700298tcp_connection_t *tcp_connection_alloc_w_base (u8 thread_index,
299 tcp_connection_t * base);
Florin Coras8124cb72018-12-16 20:57:29 -0800300void tcp_connection_free (tcp_connection_t * tc);
Florin Coras999840c2020-03-18 20:31:34 +0000301void tcp_connection_close (tcp_connection_t * tc);
302void tcp_connection_cleanup (tcp_connection_t * tc);
303void tcp_connection_del (tcp_connection_t * tc);
304int tcp_half_open_connection_cleanup (tcp_connection_t * tc);
Florin Coras3eb50622017-07-13 01:24:57 -0400305
Florin Coras1f152cd2017-08-18 19:28:03 -0700306void tcp_send_reset_w_pkt (tcp_connection_t * tc, vlib_buffer_t * pkt,
Florin Corasd4c49be2019-02-07 00:15:53 -0800307 u32 thread_index, u8 is_ip4);
Florin Coras1f152cd2017-08-18 19:28:03 -0700308void tcp_send_reset (tcp_connection_t * tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500309void tcp_send_syn (tcp_connection_t * tc);
Florin Coras7ac053b2018-11-05 15:57:21 -0800310void tcp_send_synack (tcp_connection_t * tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500311void tcp_send_fin (tcp_connection_t * tc);
Florin Corascb711a42019-10-16 19:28:17 -0700312void tcp_send_ack (tcp_connection_t * tc);
Vladimir Kropylev398afbd2019-06-25 00:06:52 +0300313void tcp_send_window_update_ack (tcp_connection_t * tc);
Florin Coras7ac053b2018-11-05 15:57:21 -0800314
Florin Coras26dd6de2019-07-23 23:54:47 -0700315void tcp_program_ack (tcp_connection_t * tc);
316void tcp_program_dupack (tcp_connection_t * tc);
Florin Coras36ebcff2019-09-12 18:36:44 -0700317void tcp_program_retransmit (tcp_connection_t * tc);
Florin Coras26dd6de2019-07-23 23:54:47 -0700318
Florin Coras999840c2020-03-18 20:31:34 +0000319void tcp_update_burst_snd_vars (tcp_connection_t * tc);
Florin Coras45ca73f2018-09-27 09:19:29 -0700320u32 tcp_snd_space (tcp_connection_t * tc);
Florin Corasbe237bf2019-09-27 08:16:40 -0700321int tcp_fastrecovery_prr_snd_space (tcp_connection_t * tc);
Florin Coras6080e0d2020-03-13 20:39:43 +0000322void tcp_reschedule (tcp_connection_t * tc);
Florin Corasf6359c82017-06-19 12:26:09 -0400323fib_node_index_t tcp_lookup_rmt_in_fib (tcp_connection_t * tc);
Florin Coras14ed6df2019-03-06 21:13:42 -0800324u32 tcp_session_push_header (transport_connection_t * tconn,
325 vlib_buffer_t * b);
Florin Coras9f86d222020-03-23 15:34:22 +0000326int tcp_session_custom_tx (void *conn, transport_send_params_t * sp);
Dave Barach68b0fb02017-02-28 15:15:56 -0500327
328void tcp_connection_timers_init (tcp_connection_t * tc);
329void tcp_connection_timers_reset (tcp_connection_t * tc);
Florin Coras4eeeaaf2017-09-05 14:03:37 -0400330void tcp_init_snd_vars (tcp_connection_t * tc);
Dave Barach68b0fb02017-02-28 15:15:56 -0500331void tcp_connection_init_vars (tcp_connection_t * tc);
Florin Corasc44a5582018-11-01 16:30:54 -0700332void tcp_connection_tx_pacer_update (tcp_connection_t * tc);
333void tcp_connection_tx_pacer_reset (tcp_connection_t * tc, u32 window,
334 u32 start_bucket);
Florin Coras7a3a8662020-02-22 02:27:21 +0000335void tcp_program_cleanup (tcp_worker_ctx_t * wrk, tcp_connection_t * tc);
Florin Corasd67f1122018-05-21 17:47:40 -0700336
Florin Coras999840c2020-03-18 20:31:34 +0000337void tcp_punt_unknown (vlib_main_t * vm, u8 is_ip4, u8 is_add);
338int tcp_configure_v4_source_address_range (vlib_main_t * vm,
339 ip4_address_t * start,
340 ip4_address_t * end, u32 table_id);
341int tcp_configure_v6_source_address_range (vlib_main_t * vm,
342 ip6_address_t * start,
343 ip6_address_t * end, u32 table_id);
Dave Barach68b0fb02017-02-28 15:15:56 -0500344
Florin Coras999840c2020-03-18 20:31:34 +0000345clib_error_t *vnet_tcp_enable_disable (vlib_main_t * vm, u8 is_en);
Florin Coras52814732019-06-12 15:38:19 -0700346
Florin Coras999840c2020-03-18 20:31:34 +0000347format_function_t format_tcp_state;
348format_function_t format_tcp_flags;
349format_function_t format_tcp_sacks;
350format_function_t format_tcp_rcv_sacks;
351format_function_t format_tcp_connection;
352format_function_t format_tcp_connection_id;
Florin Corasd2067242019-08-16 10:33:49 -0700353
Florin Coras93992a92017-05-24 18:03:56 -0700354#define tcp_validate_txf_size(_tc, _a) \
355 ASSERT(_tc->state != TCP_STATE_ESTABLISHED \
Florin Coras31c99552019-03-01 13:00:58 -0800356 || transport_max_tx_dequeue (&_tc->connection) >= _a)
Florin Coras93992a92017-05-24 18:03:56 -0700357
Dave Barach68b0fb02017-02-28 15:15:56 -0500358#endif /* _vnet_tcp_h_ */
359
360/*
361 * fd.io coding-style-patch-verification: ON
362 *
363 * Local Variables:
364 * eval: (c-set-style "gnu")
365 * End:
366 */