]> git.ipfire.org Git - thirdparty/linux.git/blob - include/linux/tcp.h
io_uring: reset -EBUSY error when io sq thread is waken up
[thirdparty/linux.git] / include / linux / tcp.h
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the TCP protocol.
8 *
9 * Version: @(#)tcp.h 1.0.2 04/28/93
10 *
11 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 */
13 #ifndef _LINUX_TCP_H
14 #define _LINUX_TCP_H
15
16
17 #include <linux/skbuff.h>
18 #include <linux/win_minmax.h>
19 #include <net/sock.h>
20 #include <net/inet_connection_sock.h>
21 #include <net/inet_timewait_sock.h>
22 #include <uapi/linux/tcp.h>
23
24 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
25 {
26 return (struct tcphdr *)skb_transport_header(skb);
27 }
28
29 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
30 {
31 return th->doff * 4;
32 }
33
34 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
35 {
36 return __tcp_hdrlen(tcp_hdr(skb));
37 }
38
39 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
40 {
41 return (struct tcphdr *)skb_inner_transport_header(skb);
42 }
43
44 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
45 {
46 return inner_tcp_hdr(skb)->doff * 4;
47 }
48
49 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
50 {
51 return (tcp_hdr(skb)->doff - 5) * 4;
52 }
53
54 /* TCP Fast Open */
55 #define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */
56 #define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */
57 #define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */
58
59 /* TCP Fast Open Cookie as stored in memory */
60 struct tcp_fastopen_cookie {
61 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
62 s8 len;
63 bool exp; /* In RFC6994 experimental option format */
64 };
65
66 /* This defines a selective acknowledgement block. */
67 struct tcp_sack_block_wire {
68 __be32 start_seq;
69 __be32 end_seq;
70 };
71
72 struct tcp_sack_block {
73 u32 start_seq;
74 u32 end_seq;
75 };
76
77 /*These are used to set the sack_ok field in struct tcp_options_received */
78 #define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */
79 #define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
80
81 #if IS_ENABLED(CONFIG_MPTCP)
82 struct mptcp_options_received {
83 u64 sndr_key;
84 u64 rcvr_key;
85 u64 data_ack;
86 u64 data_seq;
87 u32 subflow_seq;
88 u16 data_len;
89 u16 mp_capable : 1,
90 mp_join : 1,
91 dss : 1,
92 add_addr : 1,
93 rm_addr : 1,
94 family : 4,
95 echo : 1,
96 backup : 1;
97 u32 token;
98 u32 nonce;
99 u64 thmac;
100 u8 hmac[20];
101 u8 join_id;
102 u8 use_map:1,
103 dsn64:1,
104 data_fin:1,
105 use_ack:1,
106 ack64:1,
107 mpc_map:1,
108 __unused:2;
109 u8 addr_id;
110 u8 rm_id;
111 union {
112 struct in_addr addr;
113 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
114 struct in6_addr addr6;
115 #endif
116 };
117 u64 ahmac;
118 u16 port;
119 };
120 #endif
121
122 struct tcp_options_received {
123 /* PAWS/RTTM data */
124 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
125 u32 ts_recent; /* Time stamp to echo next */
126 u32 rcv_tsval; /* Time stamp value */
127 u32 rcv_tsecr; /* Time stamp echo reply */
128 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
129 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
130 dsack : 1, /* D-SACK is scheduled */
131 wscale_ok : 1, /* Wscale seen on SYN packet */
132 sack_ok : 3, /* SACK seen on SYN packet */
133 smc_ok : 1, /* SMC seen on SYN packet */
134 snd_wscale : 4, /* Window scaling received from sender */
135 rcv_wscale : 4; /* Window scaling to send to receiver */
136 u8 num_sacks; /* Number of SACK blocks */
137 u16 user_mss; /* mss requested by user in ioctl */
138 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
139 #if IS_ENABLED(CONFIG_MPTCP)
140 struct mptcp_options_received mptcp;
141 #endif
142 };
143
144 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
145 {
146 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
147 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
148 #if IS_ENABLED(CONFIG_SMC)
149 rx_opt->smc_ok = 0;
150 #endif
151 #if IS_ENABLED(CONFIG_MPTCP)
152 rx_opt->mptcp.mp_capable = 0;
153 rx_opt->mptcp.mp_join = 0;
154 rx_opt->mptcp.add_addr = 0;
155 rx_opt->mptcp.rm_addr = 0;
156 rx_opt->mptcp.dss = 0;
157 #endif
158 }
159
160 /* This is the max number of SACKS that we'll generate and process. It's safe
161 * to increase this, although since:
162 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
163 * only four options will fit in a standard TCP header */
164 #define TCP_NUM_SACKS 4
165
166 struct tcp_request_sock_ops;
167
168 struct tcp_request_sock {
169 struct inet_request_sock req;
170 const struct tcp_request_sock_ops *af_specific;
171 u64 snt_synack; /* first SYNACK sent time */
172 bool tfo_listener;
173 bool is_mptcp;
174 u32 txhash;
175 u32 rcv_isn;
176 u32 snt_isn;
177 u32 ts_off;
178 u32 last_oow_ack_time; /* last SYNACK */
179 u32 rcv_nxt; /* the ack # by SYNACK. For
180 * FastOpen it's the seq#
181 * after data-in-SYN.
182 */
183 };
184
185 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
186 {
187 return (struct tcp_request_sock *)req;
188 }
189
190 struct tcp_sock {
191 /* inet_connection_sock has to be the first member of tcp_sock */
192 struct inet_connection_sock inet_conn;
193 u16 tcp_header_len; /* Bytes of tcp header to send */
194 u16 gso_segs; /* Max number of segs per GSO packet */
195
196 /*
197 * Header prediction flags
198 * 0x5?10 << 16 + snd_wnd in net byte order
199 */
200 __be32 pred_flags;
201
202 /*
203 * RFC793 variables by their proper names. This means you can
204 * read the code and the spec side by side (and laugh ...)
205 * See RFC793 and RFC1122. The RFC writes these in capitals.
206 */
207 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
208 * sum(delta(rcv_nxt)), or how many bytes
209 * were acked.
210 */
211 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
212 * total number of segments in.
213 */
214 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
215 * total number of data segments in.
216 */
217 u32 rcv_nxt; /* What we want to receive next */
218 u32 copied_seq; /* Head of yet unread data */
219 u32 rcv_wup; /* rcv_nxt on last window update sent */
220 u32 snd_nxt; /* Next sequence we send */
221 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
222 * The total number of segments sent.
223 */
224 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
225 * total number of data segments sent.
226 */
227 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
228 * total number of data bytes sent.
229 */
230 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
231 * sum(delta(snd_una)), or how many bytes
232 * were acked.
233 */
234 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
235 * total number of DSACK blocks received
236 */
237 u32 snd_una; /* First byte we want an ack for */
238 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
239 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
240 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
241 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
242 u32 compressed_ack_rcv_nxt;
243
244 u32 tsoffset; /* timestamp offset */
245
246 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
247 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
248
249 u32 snd_wl1; /* Sequence for window update */
250 u32 snd_wnd; /* The window we expect to receive */
251 u32 max_window; /* Maximal window ever seen from peer */
252 u32 mss_cache; /* Cached effective mss, not including SACKS */
253
254 u32 window_clamp; /* Maximal window to advertise */
255 u32 rcv_ssthresh; /* Current window clamp */
256
257 /* Information of the most recently (s)acked skb */
258 struct tcp_rack {
259 u64 mstamp; /* (Re)sent time of the skb */
260 u32 rtt_us; /* Associated RTT */
261 u32 end_seq; /* Ending TCP sequence of the skb */
262 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
263 u8 reo_wnd_steps; /* Allowed reordering window */
264 #define TCP_RACK_RECOVERY_THRESH 16
265 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
266 dsack_seen:1, /* Whether DSACK seen after last adj */
267 advanced:1; /* mstamp advanced since last lost marking */
268 } rack;
269 u16 advmss; /* Advertised MSS */
270 u8 compressed_ack;
271 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
272 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
273 u8 chrono_type:2, /* current chronograph type */
274 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */
275 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
276 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
277 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
278 fastopen_client_fail:2; /* reason why fastopen failed */
279 u8 nonagle : 4,/* Disable Nagle algorithm? */
280 thin_lto : 1,/* Use linear timeouts for thin streams */
281 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
282 repair : 1,
283 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
284 u8 repair_queue;
285 u8 syn_data:1, /* SYN includes data */
286 syn_fastopen:1, /* SYN includes Fast Open option */
287 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
288 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
289 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
290 save_syn:1, /* Save headers of SYN packet */
291 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
292 syn_smc:1; /* SYN includes SMC */
293 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */
294
295 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
296 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
297 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
298
299 /* RTT measurement */
300 u64 tcp_mstamp; /* most recent packet received/sent */
301 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
302 u32 mdev_us; /* medium deviation */
303 u32 mdev_max_us; /* maximal mdev for the last rtt period */
304 u32 rttvar_us; /* smoothed mdev_max */
305 u32 rtt_seq; /* sequence number to update rttvar */
306 struct minmax rtt_min;
307
308 u32 packets_out; /* Packets which are "in flight" */
309 u32 retrans_out; /* Retransmitted packets out */
310 u32 max_packets_out; /* max packets_out in last window */
311 u32 max_packets_seq; /* right edge of max_packets_out flight */
312
313 u16 urg_data; /* Saved octet of OOB data and control flags */
314 u8 ecn_flags; /* ECN status bits. */
315 u8 keepalive_probes; /* num of allowed keep alive probes */
316 u32 reordering; /* Packet reordering metric. */
317 u32 reord_seen; /* number of data packet reordering events */
318 u32 snd_up; /* Urgent pointer */
319
320 /*
321 * Options received (usually on last packet, some only on SYN packets).
322 */
323 struct tcp_options_received rx_opt;
324
325 /*
326 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
327 */
328 u32 snd_ssthresh; /* Slow start size threshold */
329 u32 snd_cwnd; /* Sending congestion window */
330 u32 snd_cwnd_cnt; /* Linear increase counter */
331 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
332 u32 snd_cwnd_used;
333 u32 snd_cwnd_stamp;
334 u32 prior_cwnd; /* cwnd right before starting loss recovery */
335 u32 prr_delivered; /* Number of newly delivered packets to
336 * receiver in Recovery. */
337 u32 prr_out; /* Total number of pkts sent during Recovery. */
338 u32 delivered; /* Total data packets delivered incl. rexmits */
339 u32 delivered_ce; /* Like the above but only ECE marked packets */
340 u32 lost; /* Total data packets lost incl. rexmits */
341 u32 app_limited; /* limited until "delivered" reaches this val */
342 u64 first_tx_mstamp; /* start of window send phase */
343 u64 delivered_mstamp; /* time we reached "delivered" */
344 u32 rate_delivered; /* saved rate sample: packets delivered */
345 u32 rate_interval_us; /* saved rate sample: time elapsed */
346
347 u32 rcv_wnd; /* Current receiver window */
348 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
349 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
350 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
351 u32 lost_out; /* Lost packets */
352 u32 sacked_out; /* SACK'd packets */
353
354 struct hrtimer pacing_timer;
355 struct hrtimer compressed_ack_timer;
356
357 /* from STCP, retrans queue hinting */
358 struct sk_buff* lost_skb_hint;
359 struct sk_buff *retransmit_skb_hint;
360
361 /* OOO segments go in this rbtree. Socket lock must be held. */
362 struct rb_root out_of_order_queue;
363 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
364
365 /* SACKs data, these 2 need to be together (see tcp_options_write) */
366 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
367 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
368
369 struct tcp_sack_block recv_sack_cache[4];
370
371 struct sk_buff *highest_sack; /* skb just after the highest
372 * skb with SACKed bit set
373 * (validity guaranteed only if
374 * sacked_out > 0)
375 */
376
377 int lost_cnt_hint;
378
379 u32 prior_ssthresh; /* ssthresh saved at recovery start */
380 u32 high_seq; /* snd_nxt at onset of congestion */
381
382 u32 retrans_stamp; /* Timestamp of the last retransmit,
383 * also used in SYN-SENT to remember stamp of
384 * the first SYN. */
385 u32 undo_marker; /* snd_una upon a new recovery episode. */
386 int undo_retrans; /* number of undoable retransmissions. */
387 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
388 * Total data bytes retransmitted
389 */
390 u32 total_retrans; /* Total retransmits for entire connection */
391
392 u32 urg_seq; /* Seq of received urgent pointer */
393 unsigned int keepalive_time; /* time before keep alive takes place */
394 unsigned int keepalive_intvl; /* time interval between keep alive probes */
395
396 int linger2;
397
398
399 /* Sock_ops bpf program related variables */
400 #ifdef CONFIG_BPF
401 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
402 * values defined in uapi/linux/tcp.h
403 */
404 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
405 #else
406 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
407 #endif
408
409 u16 timeout_rehash; /* Timeout-triggered rehash attempts */
410
411 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
412
413 /* Receiver side RTT estimation */
414 u32 rcv_rtt_last_tsecr;
415 struct {
416 u32 rtt_us;
417 u32 seq;
418 u64 time;
419 } rcv_rtt_est;
420
421 /* Receiver queue space */
422 struct {
423 u32 space;
424 u32 seq;
425 u64 time;
426 } rcvq_space;
427
428 /* TCP-specific MTU probe information. */
429 struct {
430 u32 probe_seq_start;
431 u32 probe_seq_end;
432 } mtu_probe;
433 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
434 * while socket was owned by user.
435 */
436 #if IS_ENABLED(CONFIG_MPTCP)
437 bool is_mptcp;
438 #endif
439
440 #ifdef CONFIG_TCP_MD5SIG
441 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
442 const struct tcp_sock_af_ops *af_specific;
443
444 /* TCP MD5 Signature Option information */
445 struct tcp_md5sig_info __rcu *md5sig_info;
446 #endif
447
448 /* TCP fastopen related information */
449 struct tcp_fastopen_request *fastopen_req;
450 /* fastopen_rsk points to request_sock that resulted in this big
451 * socket. Used to retransmit SYNACKs etc.
452 */
453 struct request_sock __rcu *fastopen_rsk;
454 u32 *saved_syn;
455 };
456
457 enum tsq_enum {
458 TSQ_THROTTLED,
459 TSQ_QUEUED,
460 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
461 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
462 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
463 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
464 * tcp_v{4|6}_mtu_reduced()
465 */
466 };
467
468 enum tsq_flags {
469 TSQF_THROTTLED = (1UL << TSQ_THROTTLED),
470 TSQF_QUEUED = (1UL << TSQ_QUEUED),
471 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED),
472 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
473 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
474 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
475 };
476
477 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
478 {
479 return (struct tcp_sock *)sk;
480 }
481
482 struct tcp_timewait_sock {
483 struct inet_timewait_sock tw_sk;
484 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
485 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
486 u32 tw_rcv_wnd;
487 u32 tw_ts_offset;
488 u32 tw_ts_recent;
489
490 /* The time we sent the last out-of-window ACK: */
491 u32 tw_last_oow_ack_time;
492
493 int tw_ts_recent_stamp;
494 u32 tw_tx_delay;
495 #ifdef CONFIG_TCP_MD5SIG
496 struct tcp_md5sig_key *tw_md5_key;
497 #endif
498 };
499
500 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
501 {
502 return (struct tcp_timewait_sock *)sk;
503 }
504
505 static inline bool tcp_passive_fastopen(const struct sock *sk)
506 {
507 return sk->sk_state == TCP_SYN_RECV &&
508 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
509 }
510
511 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
512 {
513 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
514 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
515
516 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
517 }
518
519 static inline void tcp_move_syn(struct tcp_sock *tp,
520 struct request_sock *req)
521 {
522 tp->saved_syn = req->saved_syn;
523 req->saved_syn = NULL;
524 }
525
526 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
527 {
528 kfree(tp->saved_syn);
529 tp->saved_syn = NULL;
530 }
531
532 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk);
533
534 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
535 {
536 /* We use READ_ONCE() here because socket might not be locked.
537 * This happens for listeners.
538 */
539 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
540
541 return (user_mss && user_mss < mss) ? user_mss : mss;
542 }
543
544 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
545 int shiftlen);
546
547 #endif /* _LINUX_TCP_H */