Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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
24static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
25{
26 return (struct tcphdr *)skb_transport_header(skb);
27}
28
29static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
30{
31 return th->doff * 4;
32}
33
34static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
35{
36 return __tcp_hdrlen(tcp_hdr(skb));
37}
38
39static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
40{
41 return (struct tcphdr *)skb_inner_transport_header(skb);
42}
43
44static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
45{
46 return inner_tcp_hdr(skb)->doff * 4;
47}
48
49/**
50 * skb_tcp_all_headers - Returns size of all headers for a TCP packet
51 * @skb: buffer
52 *
53 * Used in TX path, for a packet known to be a TCP one.
54 *
55 * if (skb_is_gso(skb)) {
56 * int hlen = skb_tcp_all_headers(skb);
57 * ...
58 */
59static inline int skb_tcp_all_headers(const struct sk_buff *skb)
60{
61 return skb_transport_offset(skb) + tcp_hdrlen(skb);
62}
63
64/**
65 * skb_inner_tcp_all_headers - Returns size of all headers for an encap TCP packet
66 * @skb: buffer
67 *
68 * Used in TX path, for a packet known to be a TCP one.
69 *
70 * if (skb_is_gso(skb) && skb->encapsulation) {
71 * int hlen = skb_inner_tcp_all_headers(skb);
72 * ...
73 */
74static inline int skb_inner_tcp_all_headers(const struct sk_buff *skb)
75{
76 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
77}
78
79static inline unsigned int tcp_optlen(const struct sk_buff *skb)
80{
81 return (tcp_hdr(skb)->doff - 5) * 4;
82}
83
84/* TCP Fast Open */
85#define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */
86#define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */
87#define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */
88
89/* TCP Fast Open Cookie as stored in memory */
90struct tcp_fastopen_cookie {
91 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
92 s8 len;
93 bool exp; /* In RFC6994 experimental option format */
94};
95
96/* This defines a selective acknowledgement block. */
97struct tcp_sack_block_wire {
98 __be32 start_seq;
99 __be32 end_seq;
100};
101
102struct tcp_sack_block {
103 u32 start_seq;
104 u32 end_seq;
105};
106
107/*These are used to set the sack_ok field in struct tcp_options_received */
108#define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */
109#define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
110
111struct tcp_options_received {
112/* PAWS/RTTM data */
113 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
114 u32 ts_recent; /* Time stamp to echo next */
115 u32 rcv_tsval; /* Time stamp value */
116 u32 rcv_tsecr; /* Time stamp echo reply */
117 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
118 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
119 dsack : 1, /* D-SACK is scheduled */
120 wscale_ok : 1, /* Wscale seen on SYN packet */
121 sack_ok : 3, /* SACK seen on SYN packet */
122 smc_ok : 1, /* SMC seen on SYN packet */
123 snd_wscale : 4, /* Window scaling received from sender */
124 rcv_wscale : 4; /* Window scaling to send to receiver */
125 u8 accecn:6, /* AccECN index in header, 0=no options */
126 saw_unknown:1, /* Received unknown option */
127 unused:1;
128 u8 num_sacks; /* Number of SACK blocks */
129 u16 user_mss; /* mss requested by user in ioctl */
130 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
131};
132
133static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
134{
135 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
136 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
137#if IS_ENABLED(CONFIG_SMC)
138 rx_opt->smc_ok = 0;
139#endif
140}
141
142/* This is the max number of SACKS that we'll generate and process. It's safe
143 * to increase this, although since:
144 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
145 * only four options will fit in a standard TCP header */
146#define TCP_NUM_SACKS 4
147
148struct tcp_request_sock_ops;
149
150struct tcp_request_sock {
151 struct inet_request_sock req;
152 const struct tcp_request_sock_ops *af_specific;
153 u64 snt_synack; /* first SYNACK sent time */
154 bool tfo_listener;
155 bool is_mptcp;
156 bool req_usec_ts;
157#if IS_ENABLED(CONFIG_MPTCP)
158 bool drop_req;
159#endif
160 u32 txhash;
161 u32 rcv_isn;
162 u32 snt_isn;
163 u32 ts_off;
164 u32 snt_tsval_first;
165 u32 snt_tsval_last;
166 u32 last_oow_ack_time; /* last SYNACK */
167 u32 rcv_nxt; /* the ack # by SYNACK. For
168 * FastOpen it's the seq#
169 * after data-in-SYN.
170 */
171 u8 syn_tos;
172 bool accecn_ok;
173 u8 syn_ect_snt: 2,
174 syn_ect_rcv: 2,
175 accecn_fail_mode:4;
176 u8 saw_accecn_opt :2;
177#ifdef CONFIG_TCP_AO
178 u8 ao_keyid;
179 u8 ao_rcv_next;
180 bool used_tcp_ao;
181#endif
182};
183
184#define tcp_rsk(ptr) container_of_const(ptr, struct tcp_request_sock, req.req)
185
186static inline bool tcp_rsk_used_ao(const struct request_sock *req)
187{
188#ifndef CONFIG_TCP_AO
189 return false;
190#else
191 return tcp_rsk(req)->used_tcp_ao;
192#endif
193}
194
195#define TCP_RMEM_TO_WIN_SCALE 8
196
197struct tcp_sock {
198 /* Cacheline organization can be found documented in
199 * Documentation/networking/net_cachelines/tcp_sock.rst.
200 * Please update the document when adding new fields.
201 */
202
203 /* inet_connection_sock has to be the first member of tcp_sock */
204 struct inet_connection_sock inet_conn;
205
206 /* TX read-mostly hotpath cache lines */
207 __cacheline_group_begin(tcp_sock_read_tx);
208 u32 max_window; /* Maximal window ever seen from peer */
209 u32 rcv_ssthresh; /* Current window clamp */
210 u32 reordering; /* Packet reordering metric. */
211 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
212 u16 gso_segs; /* Max number of segs per GSO packet */
213 /* from STCP, retrans queue hinting */
214 struct sk_buff *retransmit_skb_hint;
215#if defined(CONFIG_TLS_DEVICE)
216 void (*tcp_clean_acked)(struct sock *sk, u32 acked_seq);
217#endif
218 __cacheline_group_end(tcp_sock_read_tx);
219
220 /* TXRX read-mostly hotpath cache lines */
221 __cacheline_group_begin(tcp_sock_read_txrx);
222 u32 tsoffset; /* timestamp offset */
223 u32 snd_wnd; /* The window we expect to receive */
224 u32 mss_cache; /* Cached effective mss, not including SACKS */
225 u32 snd_cwnd; /* Sending congestion window */
226 u32 prr_out; /* Total number of pkts sent during Recovery. */
227 u32 lost_out; /* Lost packets */
228 u32 sacked_out; /* SACK'd packets */
229 u16 tcp_header_len; /* Bytes of tcp header to send */
230 u8 scaling_ratio; /* see tcp_win_from_space() */
231 u8 repair : 1,
232 tcp_usec_ts : 1, /* TSval values in usec */
233 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
234 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
235 recvmsg_inq : 1;/* Indicate # of bytes in queue upon recvmsg */
236 __cacheline_group_end(tcp_sock_read_txrx);
237
238 /* RX read-mostly hotpath cache lines */
239 __cacheline_group_begin(tcp_sock_read_rx);
240 u32 copied_seq; /* Head of yet unread data */
241 u32 snd_wl1; /* Sequence for window update */
242 u32 tlp_high_seq; /* snd_nxt at the time of TLP */
243 u32 rttvar_us; /* smoothed mdev_max */
244 u32 retrans_out; /* Retransmitted packets out */
245 u16 advmss; /* Advertised MSS */
246 u16 urg_data; /* Saved octet of OOB data and control flags */
247 u32 lost; /* Total data packets lost incl. rexmits */
248 u32 snd_ssthresh; /* Slow start size threshold */
249 struct minmax rtt_min;
250 /* OOO segments go in this rbtree. Socket lock must be held. */
251 struct rb_root out_of_order_queue;
252 __cacheline_group_end(tcp_sock_read_rx);
253
254 /* TX read-write hotpath cache lines */
255 __cacheline_group_begin(tcp_sock_write_tx) ____cacheline_aligned;
256 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
257 * The total number of segments sent.
258 */
259 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
260 * total number of data segments sent.
261 */
262 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
263 * total number of data bytes sent.
264 */
265 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
266 u8 chrono_type; /* current chronograph type */
267 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
268 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
269 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
270 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
271 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
272 u32 mdev_us; /* medium deviation */
273 u32 rtt_seq; /* sequence number to update rttvar */
274 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
275 u64 accecn_opt_tstamp; /* Last AccECN option sent timestamp */
276 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
277 struct sk_buff *highest_sack; /* skb just after the highest
278 * skb with SACKed bit set
279 * (validity guaranteed only if
280 * sacked_out > 0)
281 */
282 u8 ecn_flags; /* ECN status bits. */
283 __cacheline_group_end(tcp_sock_write_tx);
284
285 /* TXRX read-write hotpath cache lines */
286 __cacheline_group_begin(tcp_sock_write_txrx);
287/*
288 * Header prediction flags
289 * 0x5?10 << 16 + snd_wnd in net byte order
290 */
291 u8 nonagle : 4,/* Disable Nagle algorithm? */
292 rate_app_limited:1; /* rate_{delivered,interval_us} limited? */
293 u8 received_ce_pending:4, /* Not yet transmit cnt of received_ce */
294 accecn_opt_sent_w_dsack:1,/* Sent ACCECN opt in previous ACK w/ D-SACK */
295 unused2:3;
296 u8 accecn_minlen:2,/* Minimum length of AccECN option sent */
297 est_ecnfield:2,/* ECN field for AccECN delivered estimates */
298 accecn_opt_demand:2,/* Demand AccECN option for n next ACKs */
299 prev_ecnfield:2; /* ECN bits from the previous segment */
300 __be32 pred_flags;
301 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
302 u64 tcp_mstamp; /* most recent packet received/sent */
303 u32 rcv_nxt; /* What we want to receive next */
304 u32 snd_nxt; /* Next sequence we send */
305 u32 snd_una; /* First byte we want an ack for */
306 u32 window_clamp; /* Maximal window to advertise */
307 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
308 u32 packets_out; /* Packets which are "in flight" */
309 u32 snd_up; /* Urgent pointer */
310 u32 delivered; /* Total data packets delivered incl. rexmits */
311 u32 delivered_ce; /* Like the above but only ECE marked packets */
312 u32 received_ce; /* Like the above but for rcvd CE marked pkts */
313 u32 received_ecn_bytes[3]; /* received byte counters for three ECN
314 * types: INET_ECN_ECT_1, INET_ECN_ECT_0,
315 * and INET_ECN_CE
316 */
317 u32 app_limited; /* limited until "delivered" reaches this val */
318 u32 rcv_wnd; /* Current receiver window */
319 u32 rcv_mwnd_seq; /* Maximum window sequence number (RFC 7323,
320 * section 2.4, receiver requirements)
321 */
322 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
323/*
324 * Options received (usually on last packet, some only on SYN packets).
325 */
326 struct tcp_options_received rx_opt;
327 __cacheline_group_end(tcp_sock_write_txrx);
328
329 /* RX read-write hotpath cache lines */
330 __cacheline_group_begin(tcp_sock_write_rx) __aligned(8);
331 u64 bytes_received;
332 /* RFC4898 tcpEStatsAppHCThruOctetsReceived
333 * sum(delta(rcv_nxt)), or how many bytes
334 * were acked.
335 */
336 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
337 * total number of segments in.
338 */
339 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
340 * total number of data segments in.
341 */
342 u32 rcv_wup; /* rcv_nxt on last window update sent */
343 u32 max_packets_out; /* max packets_out in last window */
344 u32 cwnd_usage_seq; /* right edge of cwnd usage tracking flight */
345 u32 rate_delivered; /* saved rate sample: packets delivered */
346 u32 rate_interval_us; /* saved rate sample: time elapsed */
347 u32 rcv_rtt_last_tsecr;
348 u32 delivered_ecn_bytes[3];
349 u16 pkts_acked_ewma;/* Pkts acked EWMA for AccECN cep heuristic */
350 u64 first_tx_mstamp; /* start of window send phase */
351 u64 delivered_mstamp; /* time we reached "delivered" */
352 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
353 * sum(delta(snd_una)), or how many bytes
354 * were acked.
355 */
356 struct {
357 u32 rtt_us;
358 u32 seq;
359 u64 time;
360 } rcv_rtt_est;
361/* Receiver queue space */
362 struct {
363 int space;
364 u32 seq;
365 u64 time;
366 } rcvq_space;
367 __cacheline_group_end(tcp_sock_write_rx);
368 /* End of Hot Path */
369
370/*
371 * RFC793 variables by their proper names. This means you can
372 * read the code and the spec side by side (and laugh ...)
373 * See RFC793 and RFC1122. The RFC writes these in capitals.
374 */
375 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
376 * total number of DSACK blocks received
377 */
378 u32 compressed_ack_rcv_nxt;
379 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
380
381 /* Information of the most recently (s)acked skb */
382 struct tcp_rack {
383 u64 mstamp; /* (Re)sent time of the skb */
384 u32 rtt_us; /* Associated RTT */
385 u32 end_seq; /* Ending TCP sequence of the skb */
386 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
387 u8 reo_wnd_steps; /* Allowed reordering window */
388#define TCP_RACK_RECOVERY_THRESH 16
389 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
390 dsack_seen:1, /* Whether DSACK seen after last adj */
391 advanced:1; /* mstamp advanced since last lost marking */
392 } rack;
393 u8 compressed_ack;
394 u8 dup_ack_counter:2,
395 tlp_retrans:1, /* TLP is a retransmission */
396 syn_ect_snt:2, /* AccECN ECT memory, only */
397 syn_ect_rcv:2; /* ... needed during 3WHS + first seqno */
398 u8 thin_lto : 1,/* Use linear timeouts for thin streams */
399 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
400 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
401 fastopen_client_fail:2, /* reason why fastopen failed */
402 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
403 u8 repair_queue;
404 u8 save_syn:2, /* Save headers of SYN packet */
405 syn_data:1, /* SYN includes data */
406 syn_fastopen:1, /* SYN includes Fast Open option */
407 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
408 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
409 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
410 syn_fastopen_child:1; /* created TFO passive child socket */
411
412 u8 keepalive_probes; /* num of allowed keep alive probes */
413 u8 accecn_fail_mode:4, /* AccECN failure handling */
414 saw_accecn_opt:2; /* An AccECN option was seen */
415 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
416
417/* RTT measurement */
418 u32 mdev_max_us; /* maximal mdev for the last rtt period */
419
420 u32 reord_seen; /* number of data packet reordering events */
421
422/*
423 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
424 */
425 u32 snd_cwnd_cnt; /* Linear increase counter */
426 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
427 u32 snd_cwnd_used;
428 u32 snd_cwnd_stamp;
429 u32 prior_cwnd; /* cwnd right before starting loss recovery */
430 u32 prr_delivered; /* Number of newly delivered packets to
431 * receiver in Recovery. */
432 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
433
434 struct hrtimer pacing_timer;
435 struct hrtimer compressed_ack_timer;
436
437 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
438
439 /* SACKs data, these 2 need to be together (see tcp_options_write) */
440 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
441 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
442
443 struct tcp_sack_block recv_sack_cache[4];
444
445 u32 prior_ssthresh; /* ssthresh saved at recovery start */
446 u32 high_seq; /* snd_nxt at onset of congestion */
447
448 u32 retrans_stamp; /* Timestamp of the last retransmit,
449 * also used in SYN-SENT to remember stamp of
450 * the first SYN. */
451 u32 undo_marker; /* snd_una upon a new recovery episode. */
452 int undo_retrans; /* number of undoable retransmissions. */
453 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
454 * while socket was owned by user.
455 */
456 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
457 * Total data bytes retransmitted
458 */
459 u32 total_retrans; /* Total retransmits for entire connection */
460 u32 rto_stamp; /* Start time (ms) of last CA_Loss recovery */
461 u16 total_rto; /* Total number of RTO timeouts, including
462 * SYN/SYN-ACK and recurring timeouts.
463 */
464 u16 total_rto_recoveries; /* Total number of RTO recoveries,
465 * including any unfinished recovery.
466 */
467 u32 total_rto_time; /* ms spent in (completed) RTO recoveries. */
468
469 u32 urg_seq; /* Seq of received urgent pointer */
470 unsigned int keepalive_time; /* time before keep alive takes place */
471 unsigned int keepalive_intvl; /* time interval between keep alive probes */
472
473 int linger2;
474
475
476/* Sock_ops bpf program related variables */
477#ifdef CONFIG_BPF
478 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
479 * values defined in uapi/linux/tcp.h
480 */
481 u8 bpf_chg_cc_inprogress:1; /* In the middle of
482 * bpf_setsockopt(TCP_CONGESTION),
483 * it is to avoid the bpf_tcp_cc->init()
484 * to recur itself by calling
485 * bpf_setsockopt(TCP_CONGESTION, "itself").
486 */
487#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
488#else
489#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
490#endif
491
492 u16 timeout_rehash; /* Timeout-triggered rehash attempts */
493
494 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
495
496/* TCP-specific MTU probe information. */
497 struct {
498 u32 probe_seq_start;
499 u32 probe_seq_end;
500 } mtu_probe;
501 u32 plb_rehash; /* PLB-triggered rehash attempts */
502#if IS_ENABLED(CONFIG_MPTCP)
503 bool is_mptcp;
504#endif
505#if IS_ENABLED(CONFIG_SMC)
506 bool syn_smc; /* SYN includes SMC */
507 bool (*smc_hs_congested)(const struct sock *sk);
508#endif
509
510#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
511/* TCP AF-Specific parts; only used by TCP-AO/MD5 Signature support so far */
512 const struct tcp_sock_af_ops *af_specific;
513
514#ifdef CONFIG_TCP_MD5SIG
515/* TCP MD5 Signature Option information */
516 struct tcp_md5sig_info __rcu *md5sig_info;
517#endif
518#ifdef CONFIG_TCP_AO
519 struct tcp_ao_info __rcu *ao_info;
520#endif
521#endif
522
523/* TCP fastopen related information */
524 struct tcp_fastopen_request *fastopen_req;
525 /* fastopen_rsk points to request_sock that resulted in this big
526 * socket. Used to retransmit SYNACKs etc.
527 */
528 struct request_sock __rcu *fastopen_rsk;
529 struct saved_syn *saved_syn;
530};
531
532enum tsq_enum {
533 TSQ_THROTTLED,
534 TSQ_QUEUED,
535 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
536 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
537 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
538 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
539 * tcp_v{4|6}_mtu_reduced()
540 */
541 TCP_ACK_DEFERRED, /* TX pure ack is deferred */
542};
543
544enum tsq_flags {
545 TSQF_THROTTLED = BIT(TSQ_THROTTLED),
546 TSQF_QUEUED = BIT(TSQ_QUEUED),
547 TCPF_TSQ_DEFERRED = BIT(TCP_TSQ_DEFERRED),
548 TCPF_WRITE_TIMER_DEFERRED = BIT(TCP_WRITE_TIMER_DEFERRED),
549 TCPF_DELACK_TIMER_DEFERRED = BIT(TCP_DELACK_TIMER_DEFERRED),
550 TCPF_MTU_REDUCED_DEFERRED = BIT(TCP_MTU_REDUCED_DEFERRED),
551 TCPF_ACK_DEFERRED = BIT(TCP_ACK_DEFERRED),
552};
553
554/* Flags of interest for tcp_release_cb() */
555#define TCP_DEFERRED_ALL (TCPF_TSQ_DEFERRED | \
556 TCPF_WRITE_TIMER_DEFERRED | \
557 TCPF_DELACK_TIMER_DEFERRED | \
558 TCPF_MTU_REDUCED_DEFERRED | \
559 TCPF_ACK_DEFERRED)
560
561#define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
562
563/* Variant of tcp_sk() upgrading a const sock to a read/write tcp socket.
564 * Used in context of (lockless) tcp listeners.
565 */
566#define tcp_sk_rw(ptr) container_of(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
567
568struct tcp_timewait_sock {
569 struct inet_timewait_sock tw_sk;
570#define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
571#define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
572 u32 tw_rcv_wnd;
573 u32 tw_ts_offset;
574 u32 tw_ts_recent;
575
576 /* The time we sent the last out-of-window ACK: */
577 u32 tw_last_oow_ack_time;
578
579 int tw_ts_recent_stamp;
580 u32 tw_tx_delay;
581#ifdef CONFIG_TCP_MD5SIG
582 struct tcp_md5sig_key *tw_md5_key;
583#endif
584#ifdef CONFIG_TCP_AO
585 struct tcp_ao_info __rcu *ao_info;
586#endif
587};
588
589static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
590{
591 return (struct tcp_timewait_sock *)sk;
592}
593
594static inline bool tcp_passive_fastopen(const struct sock *sk)
595{
596 return sk->sk_state == TCP_SYN_RECV &&
597 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
598}
599
600static inline void fastopen_queue_tune(struct sock *sk, int backlog)
601{
602 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
603 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
604
605 WRITE_ONCE(queue->fastopenq.max_qlen, min_t(unsigned int, backlog, somaxconn));
606}
607
608static inline void tcp_move_syn(struct tcp_sock *tp,
609 struct request_sock *req)
610{
611 tp->saved_syn = req->saved_syn;
612 req->saved_syn = NULL;
613}
614
615static inline void tcp_saved_syn_free(struct tcp_sock *tp)
616{
617 kfree(tp->saved_syn);
618 tp->saved_syn = NULL;
619}
620
621static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
622{
623 return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
624 saved_syn->tcp_hdrlen;
625}
626
627struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
628 const struct sk_buff *orig_skb,
629 const struct sk_buff *ack_skb);
630
631static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
632{
633 /* We use READ_ONCE() here because socket might not be locked.
634 * This happens for listeners.
635 */
636 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
637
638 return (user_mss && user_mss < mss) ? user_mss : mss;
639}
640
641int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
642 int shiftlen);
643
644void __tcp_sock_set_cork(struct sock *sk, bool on);
645void tcp_sock_set_cork(struct sock *sk, bool on);
646int tcp_sock_set_keepcnt(struct sock *sk, int val);
647int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
648int tcp_sock_set_keepidle(struct sock *sk, int val);
649int tcp_sock_set_keepintvl(struct sock *sk, int val);
650void __tcp_sock_set_nodelay(struct sock *sk, bool on);
651void tcp_sock_set_nodelay(struct sock *sk);
652void tcp_sock_set_quickack(struct sock *sk, int val);
653int tcp_sock_set_syncnt(struct sock *sk, int val);
654int tcp_sock_set_user_timeout(struct sock *sk, int val);
655int tcp_sock_set_maxseg(struct sock *sk, int val);
656
657static inline bool dst_tcp_usec_ts(const struct dst_entry *dst)
658{
659 return dst_feature(dst, RTAX_FEATURE_TCP_USEC_TS);
660}
661
662#endif /* _LINUX_TCP_H */