2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
102 * connection: ~180sec is RFC minimum */
104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
105 * connection: ~180sec is RFC minimum */
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
128 * used as a fallback RTO for the
129 * initial data transmission if no
130 * valid RTT sample has been acquired,
131 * most likely due to retrans in 3WHS.
134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
135 * for local resources.
138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
140 #define TCP_KEEPALIVE_INTVL (75*HZ)
142 #define MAX_TCP_KEEPIDLE 32767
143 #define MAX_TCP_KEEPINTVL 32767
144 #define MAX_TCP_KEEPCNT 127
145 #define MAX_TCP_SYNCNT 127
147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
151 * after this time. It should be equal
152 * (or greater than) TCP_TIMEWAIT_LEN
153 * to provide reliability equal to one
154 * provided by timewait state.
156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
157 * timestamps. It must be less than
158 * minimal timewait lifetime.
164 #define TCPOPT_NOP 1 /* Padding */
165 #define TCPOPT_EOL 0 /* End of options */
166 #define TCPOPT_MSS 2 /* Segment size negotiating */
167 #define TCPOPT_WINDOW 3 /* Window scaling */
168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
169 #define TCPOPT_SACK 5 /* SACK Block */
170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
178 #define TCPOLEN_MSS 4
179 #define TCPOLEN_WINDOW 3
180 #define TCPOLEN_SACK_PERM 2
181 #define TCPOLEN_TIMESTAMP 10
182 #define TCPOLEN_MD5SIG 18
183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
188 /* But this is what stacks really send out. */
189 #define TCPOLEN_TSTAMP_ALIGNED 12
190 #define TCPOLEN_WSCALE_ALIGNED 4
191 #define TCPOLEN_SACKPERM_ALIGNED 4
192 #define TCPOLEN_SACK_BASE 2
193 #define TCPOLEN_SACK_BASE_ALIGNED 4
194 #define TCPOLEN_SACK_PERBLOCK 8
195 #define TCPOLEN_MD5SIG_ALIGNED 20
196 #define TCPOLEN_MSS_ALIGNED 4
198 /* Flags in tp->nonagle */
199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
200 #define TCP_NAGLE_CORK 2 /* Socket is corked */
201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
203 /* TCP thin-stream limits */
204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
207 #define TCP_INIT_CWND 10
209 extern struct inet_timewait_death_row tcp_death_row;
211 /* sysctl variables for tcp */
212 extern int sysctl_tcp_timestamps;
213 extern int sysctl_tcp_window_scaling;
214 extern int sysctl_tcp_sack;
215 extern int sysctl_tcp_fin_timeout;
216 extern int sysctl_tcp_keepalive_time;
217 extern int sysctl_tcp_keepalive_probes;
218 extern int sysctl_tcp_keepalive_intvl;
219 extern int sysctl_tcp_syn_retries;
220 extern int sysctl_tcp_synack_retries;
221 extern int sysctl_tcp_retries1;
222 extern int sysctl_tcp_retries2;
223 extern int sysctl_tcp_orphan_retries;
224 extern int sysctl_tcp_syncookies;
225 extern int sysctl_tcp_retrans_collapse;
226 extern int sysctl_tcp_stdurg;
227 extern int sysctl_tcp_rfc1337;
228 extern int sysctl_tcp_abort_on_overflow;
229 extern int sysctl_tcp_max_orphans;
230 extern int sysctl_tcp_fack;
231 extern int sysctl_tcp_reordering;
232 extern int sysctl_tcp_ecn;
233 extern int sysctl_tcp_dsack;
234 extern int sysctl_tcp_wmem[3];
235 extern int sysctl_tcp_rmem[3];
236 extern int sysctl_tcp_app_win;
237 extern int sysctl_tcp_adv_win_scale;
238 extern int sysctl_tcp_tw_reuse;
239 extern int sysctl_tcp_frto;
240 extern int sysctl_tcp_frto_response;
241 extern int sysctl_tcp_low_latency;
242 extern int sysctl_tcp_dma_copybreak;
243 extern int sysctl_tcp_nometrics_save;
244 extern int sysctl_tcp_moderate_rcvbuf;
245 extern int sysctl_tcp_tso_win_divisor;
246 extern int sysctl_tcp_abc;
247 extern int sysctl_tcp_mtu_probing;
248 extern int sysctl_tcp_base_mss;
249 extern int sysctl_tcp_workaround_signed_windows;
250 extern int sysctl_tcp_slow_start_after_idle;
251 extern int sysctl_tcp_max_ssthresh;
252 extern int sysctl_tcp_cookie_size;
253 extern int sysctl_tcp_thin_linear_timeouts;
254 extern int sysctl_tcp_thin_dupack;
255 extern int sysctl_tcp_early_retrans;
257 extern atomic_long_t tcp_memory_allocated;
258 extern struct percpu_counter tcp_sockets_allocated;
259 extern int tcp_memory_pressure;
262 * The next routines deal with comparing 32 bit unsigned ints
263 * and worry about wraparound (automatic with unsigned arithmetic).
266 static inline int before(__u32 seq1, __u32 seq2)
268 return (__s32)(seq1-seq2) < 0;
270 #define after(seq2, seq1) before(seq1, seq2)
272 /* is s2<=s1<=s3 ? */
273 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
275 return seq3 - seq2 >= seq1 - seq2;
278 static inline bool tcp_out_of_memory(struct sock *sk)
280 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
281 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
286 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
288 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
289 int orphans = percpu_counter_read_positive(ocp);
291 if (orphans << shift > sysctl_tcp_max_orphans) {
292 orphans = percpu_counter_sum_positive(ocp);
293 if (orphans << shift > sysctl_tcp_max_orphans)
299 extern bool tcp_check_oom(struct sock *sk, int shift);
301 /* syncookies: remember time of last synqueue overflow */
302 static inline void tcp_synq_overflow(struct sock *sk)
304 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
307 /* syncookies: no recent synqueue overflow on this listening socket? */
308 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
310 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
311 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
314 extern struct proto tcp_prot;
316 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
317 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
318 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
319 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
320 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
322 extern void tcp_init_mem(struct net *net);
324 extern void tcp_v4_err(struct sk_buff *skb, u32);
326 extern void tcp_shutdown (struct sock *sk, int how);
328 extern int tcp_v4_rcv(struct sk_buff *skb);
330 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
331 extern void *tcp_v4_tw_get_peer(struct sock *sk);
332 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
333 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
335 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
336 size_t size, int flags);
337 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
338 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
339 const struct tcphdr *th, unsigned int len);
340 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
341 const struct tcphdr *th, unsigned int len);
342 extern void tcp_rcv_space_adjust(struct sock *sk);
343 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
344 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
345 extern void tcp_twsk_destructor(struct sock *sk);
346 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
347 struct pipe_inode_info *pipe, size_t len,
350 static inline void tcp_dec_quickack_mode(struct sock *sk,
351 const unsigned int pkts)
353 struct inet_connection_sock *icsk = inet_csk(sk);
355 if (icsk->icsk_ack.quick) {
356 if (pkts >= icsk->icsk_ack.quick) {
357 icsk->icsk_ack.quick = 0;
358 /* Leaving quickack mode we deflate ATO. */
359 icsk->icsk_ack.ato = TCP_ATO_MIN;
361 icsk->icsk_ack.quick -= pkts;
366 #define TCP_ECN_QUEUE_CWR 2
367 #define TCP_ECN_DEMAND_CWR 4
368 #define TCP_ECN_SEEN 8
370 static __inline__ void
371 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
373 if (sysctl_tcp_ecn && th->ece && th->cwr)
374 inet_rsk(req)->ecn_ok = 1;
385 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
387 const struct tcphdr *th);
388 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
389 struct request_sock *req,
390 struct request_sock **prev);
391 extern int tcp_child_process(struct sock *parent, struct sock *child,
392 struct sk_buff *skb);
393 extern int tcp_use_frto(struct sock *sk);
394 extern void tcp_enter_frto(struct sock *sk);
395 extern void tcp_enter_loss(struct sock *sk, int how);
396 extern void tcp_clear_retrans(struct tcp_sock *tp);
397 extern void tcp_update_metrics(struct sock *sk);
398 extern void tcp_close(struct sock *sk, long timeout);
399 extern void tcp_init_sock(struct sock *sk);
400 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
401 struct poll_table_struct *wait);
402 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
403 char __user *optval, int __user *optlen);
404 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
405 char __user *optval, unsigned int optlen);
406 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
407 char __user *optval, int __user *optlen);
408 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
409 char __user *optval, unsigned int optlen);
410 extern void tcp_set_keepalive(struct sock *sk, int val);
411 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
412 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
413 size_t len, int nonblock, int flags, int *addr_len);
414 extern void tcp_parse_options(const struct sk_buff *skb,
415 struct tcp_options_received *opt_rx, const u8 **hvpp,
417 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
420 * TCP v4 functions exported for the inet6 API
423 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
424 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
425 extern struct sock * tcp_create_openreq_child(struct sock *sk,
426 struct request_sock *req,
427 struct sk_buff *skb);
428 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
429 struct request_sock *req,
430 struct dst_entry *dst);
431 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
432 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
434 extern int tcp_connect(struct sock *sk);
435 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
436 struct request_sock *req,
437 struct request_values *rvp);
438 extern int tcp_disconnect(struct sock *sk, int flags);
440 void tcp_connect_init(struct sock *sk);
441 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
442 int __must_check tcp_queue_rcv(struct sock *sk, struct sk_buff *skb,
443 int hdrlen, bool *fragstolen);
445 /* From syncookies.c */
446 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
447 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
448 struct ip_options *opt);
449 #ifdef CONFIG_SYN_COOKIES
450 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
453 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
461 extern __u32 cookie_init_timestamp(struct request_sock *req);
462 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
464 /* From net/ipv6/syncookies.c */
465 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
466 #ifdef CONFIG_SYN_COOKIES
467 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
470 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
479 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
481 extern int tcp_may_send_now(struct sock *sk);
482 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
483 extern void tcp_retransmit_timer(struct sock *sk);
484 extern void tcp_xmit_retransmit_queue(struct sock *);
485 extern void tcp_simple_retransmit(struct sock *);
486 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
487 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
489 extern void tcp_send_probe0(struct sock *);
490 extern void tcp_send_partial(struct sock *);
491 extern int tcp_write_wakeup(struct sock *);
492 extern void tcp_send_fin(struct sock *sk);
493 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
494 extern int tcp_send_synack(struct sock *);
495 extern int tcp_syn_flood_action(struct sock *sk,
496 const struct sk_buff *skb,
498 extern void tcp_push_one(struct sock *, unsigned int mss_now);
499 extern void tcp_send_ack(struct sock *sk);
500 extern void tcp_send_delayed_ack(struct sock *sk);
503 extern void tcp_cwnd_application_limited(struct sock *sk);
504 extern void tcp_resume_early_retransmit(struct sock *sk);
505 extern void tcp_rearm_rto(struct sock *sk);
508 extern void tcp_init_xmit_timers(struct sock *);
509 static inline void tcp_clear_xmit_timers(struct sock *sk)
511 inet_csk_clear_xmit_timers(sk);
514 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
515 extern unsigned int tcp_current_mss(struct sock *sk);
517 /* Bound MSS / TSO packet size with the half of the window */
518 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
522 /* When peer uses tiny windows, there is no use in packetizing
523 * to sub-MSS pieces for the sake of SWS or making sure there
524 * are enough packets in the pipe for fast recovery.
526 * On the other hand, for extremely large MSS devices, handling
527 * smaller than MSS windows in this way does make sense.
529 if (tp->max_window >= 512)
530 cutoff = (tp->max_window >> 1);
532 cutoff = tp->max_window;
534 if (cutoff && pktsize > cutoff)
535 return max_t(int, cutoff, 68U - tp->tcp_header_len);
541 extern void tcp_get_info(const struct sock *, struct tcp_info *);
543 /* Read 'sendfile()'-style from a TCP socket */
544 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
545 unsigned int, size_t);
546 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
547 sk_read_actor_t recv_actor);
549 extern void tcp_initialize_rcv_mss(struct sock *sk);
551 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
552 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
553 extern void tcp_mtup_init(struct sock *sk);
554 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
556 static inline void tcp_bound_rto(const struct sock *sk)
558 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
559 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
562 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
564 return (tp->srtt >> 3) + tp->rttvar;
567 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
569 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
570 ntohl(TCP_FLAG_ACK) |
574 static inline void tcp_fast_path_on(struct tcp_sock *tp)
576 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
579 static inline void tcp_fast_path_check(struct sock *sk)
581 struct tcp_sock *tp = tcp_sk(sk);
583 if (skb_queue_empty(&tp->out_of_order_queue) &&
585 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
587 tcp_fast_path_on(tp);
590 /* Compute the actual rto_min value */
591 static inline u32 tcp_rto_min(struct sock *sk)
593 const struct dst_entry *dst = __sk_dst_get(sk);
594 u32 rto_min = TCP_RTO_MIN;
596 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
597 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
601 /* Compute the actual receive window we are currently advertising.
602 * Rcv_nxt can be after the window if our peer push more data
603 * than the offered window.
605 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
607 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
614 /* Choose a new window, without checks for shrinking, and without
615 * scaling applied to the result. The caller does these things
616 * if necessary. This is a "raw" window selection.
618 extern u32 __tcp_select_window(struct sock *sk);
620 void tcp_send_window_probe(struct sock *sk);
622 /* TCP timestamps are only 32-bits, this causes a slight
623 * complication on 64-bit systems since we store a snapshot
624 * of jiffies in the buffer control blocks below. We decided
625 * to use only the low 32-bits of jiffies and hide the ugly
626 * casts with the following macro.
628 #define tcp_time_stamp ((__u32)(jiffies))
630 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
632 #define TCPHDR_FIN 0x01
633 #define TCPHDR_SYN 0x02
634 #define TCPHDR_RST 0x04
635 #define TCPHDR_PSH 0x08
636 #define TCPHDR_ACK 0x10
637 #define TCPHDR_URG 0x20
638 #define TCPHDR_ECE 0x40
639 #define TCPHDR_CWR 0x80
641 /* This is what the send packet queuing engine uses to pass
642 * TCP per-packet control information to the transmission code.
643 * We also store the host-order sequence numbers in here too.
644 * This is 44 bytes if IPV6 is enabled.
645 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
649 struct inet_skb_parm h4;
650 #if IS_ENABLED(CONFIG_IPV6)
651 struct inet6_skb_parm h6;
653 } header; /* For incoming frames */
654 __u32 seq; /* Starting sequence number */
655 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
656 __u32 when; /* used to compute rtt's */
657 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
659 __u8 sacked; /* State flags for SACK/FACK. */
660 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
661 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
662 #define TCPCB_LOST 0x04 /* SKB is lost */
663 #define TCPCB_TAGBITS 0x07 /* All tag bits */
664 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
665 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
667 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
669 __u32 ack_seq; /* Sequence number ACK'd */
672 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
674 /* Due to TSO, an SKB can be composed of multiple actual
675 * packets. To keep these tracked properly, we use this.
677 static inline int tcp_skb_pcount(const struct sk_buff *skb)
679 return skb_shinfo(skb)->gso_segs;
682 /* This is valid iff tcp_skb_pcount() > 1. */
683 static inline int tcp_skb_mss(const struct sk_buff *skb)
685 return skb_shinfo(skb)->gso_size;
688 /* Events passed to congestion control interface */
690 CA_EVENT_TX_START, /* first transmit when no packets in flight */
691 CA_EVENT_CWND_RESTART, /* congestion window restart */
692 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
693 CA_EVENT_FRTO, /* fast recovery timeout */
694 CA_EVENT_LOSS, /* loss timeout */
695 CA_EVENT_FAST_ACK, /* in sequence ack */
696 CA_EVENT_SLOW_ACK, /* other ack */
700 * Interface for adding new TCP congestion control handlers
702 #define TCP_CA_NAME_MAX 16
703 #define TCP_CA_MAX 128
704 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
706 #define TCP_CONG_NON_RESTRICTED 0x1
707 #define TCP_CONG_RTT_STAMP 0x2
709 struct tcp_congestion_ops {
710 struct list_head list;
713 /* initialize private data (optional) */
714 void (*init)(struct sock *sk);
715 /* cleanup private data (optional) */
716 void (*release)(struct sock *sk);
718 /* return slow start threshold (required) */
719 u32 (*ssthresh)(struct sock *sk);
720 /* lower bound for congestion window (optional) */
721 u32 (*min_cwnd)(const struct sock *sk);
722 /* do new cwnd calculation (required) */
723 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
724 /* call before changing ca_state (optional) */
725 void (*set_state)(struct sock *sk, u8 new_state);
726 /* call when cwnd event occurs (optional) */
727 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
728 /* new value of cwnd after loss (optional) */
729 u32 (*undo_cwnd)(struct sock *sk);
730 /* hook for packet ack accounting (optional) */
731 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
732 /* get info for inet_diag (optional) */
733 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
735 char name[TCP_CA_NAME_MAX];
736 struct module *owner;
739 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
740 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
742 extern void tcp_init_congestion_control(struct sock *sk);
743 extern void tcp_cleanup_congestion_control(struct sock *sk);
744 extern int tcp_set_default_congestion_control(const char *name);
745 extern void tcp_get_default_congestion_control(char *name);
746 extern void tcp_get_available_congestion_control(char *buf, size_t len);
747 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
748 extern int tcp_set_allowed_congestion_control(char *allowed);
749 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
750 extern void tcp_slow_start(struct tcp_sock *tp);
751 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
753 extern struct tcp_congestion_ops tcp_init_congestion_ops;
754 extern u32 tcp_reno_ssthresh(struct sock *sk);
755 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
756 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
757 extern struct tcp_congestion_ops tcp_reno;
759 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
761 struct inet_connection_sock *icsk = inet_csk(sk);
763 if (icsk->icsk_ca_ops->set_state)
764 icsk->icsk_ca_ops->set_state(sk, ca_state);
765 icsk->icsk_ca_state = ca_state;
768 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
770 const struct inet_connection_sock *icsk = inet_csk(sk);
772 if (icsk->icsk_ca_ops->cwnd_event)
773 icsk->icsk_ca_ops->cwnd_event(sk, event);
776 /* These functions determine how the current flow behaves in respect of SACK
777 * handling. SACK is negotiated with the peer, and therefore it can vary
778 * between different flows.
780 * tcp_is_sack - SACK enabled
781 * tcp_is_reno - No SACK
782 * tcp_is_fack - FACK enabled, implies SACK enabled
784 static inline int tcp_is_sack(const struct tcp_sock *tp)
786 return tp->rx_opt.sack_ok;
789 static inline int tcp_is_reno(const struct tcp_sock *tp)
791 return !tcp_is_sack(tp);
794 static inline int tcp_is_fack(const struct tcp_sock *tp)
796 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
799 static inline void tcp_enable_fack(struct tcp_sock *tp)
801 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
804 /* TCP early-retransmit (ER) is similar to but more conservative than
805 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
807 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
809 tp->do_early_retrans = sysctl_tcp_early_retrans &&
810 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3;
811 tp->early_retrans_delayed = 0;
814 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
816 tp->do_early_retrans = 0;
819 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
821 return tp->sacked_out + tp->lost_out;
824 /* This determines how many packets are "in the network" to the best
825 * of our knowledge. In many cases it is conservative, but where
826 * detailed information is available from the receiver (via SACK
827 * blocks etc.) we can make more aggressive calculations.
829 * Use this for decisions involving congestion control, use just
830 * tp->packets_out to determine if the send queue is empty or not.
832 * Read this equation as:
834 * "Packets sent once on transmission queue" MINUS
835 * "Packets left network, but not honestly ACKed yet" PLUS
836 * "Packets fast retransmitted"
838 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
840 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
843 #define TCP_INFINITE_SSTHRESH 0x7fffffff
845 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
847 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
850 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
851 * The exception is rate halving phase, when cwnd is decreasing towards
854 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
856 const struct tcp_sock *tp = tcp_sk(sk);
858 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
859 return tp->snd_ssthresh;
861 return max(tp->snd_ssthresh,
862 ((tp->snd_cwnd >> 1) +
863 (tp->snd_cwnd >> 2)));
866 /* Use define here intentionally to get WARN_ON location shown at the caller */
867 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
869 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
870 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
872 /* The maximum number of MSS of available cwnd for which TSO defers
873 * sending if not using sysctl_tcp_tso_win_divisor.
875 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
880 /* Slow start with delack produces 3 packets of burst, so that
881 * it is safe "de facto". This will be the default - same as
882 * the default reordering threshold - but if reordering increases,
883 * we must be able to allow cwnd to burst at least this much in order
884 * to not pull it back when holes are filled.
886 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
888 return tp->reordering;
891 /* Returns end sequence number of the receiver's advertised window */
892 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
894 return tp->snd_una + tp->snd_wnd;
896 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
898 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
899 const struct sk_buff *skb)
902 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
905 static inline void tcp_check_probe_timer(struct sock *sk)
907 const struct tcp_sock *tp = tcp_sk(sk);
908 const struct inet_connection_sock *icsk = inet_csk(sk);
910 if (!tp->packets_out && !icsk->icsk_pending)
911 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
912 icsk->icsk_rto, TCP_RTO_MAX);
915 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
920 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
926 * Calculate(/check) TCP checksum
928 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
929 __be32 daddr, __wsum base)
931 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
934 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
936 return __skb_checksum_complete(skb);
939 static inline int tcp_checksum_complete(struct sk_buff *skb)
941 return !skb_csum_unnecessary(skb) &&
942 __tcp_checksum_complete(skb);
945 /* Prequeue for VJ style copy to user, combined with checksumming. */
947 static inline void tcp_prequeue_init(struct tcp_sock *tp)
949 tp->ucopy.task = NULL;
951 tp->ucopy.memory = 0;
952 skb_queue_head_init(&tp->ucopy.prequeue);
953 #ifdef CONFIG_NET_DMA
954 tp->ucopy.dma_chan = NULL;
955 tp->ucopy.wakeup = 0;
956 tp->ucopy.pinned_list = NULL;
957 tp->ucopy.dma_cookie = 0;
961 /* Packet is added to VJ-style prequeue for processing in process
962 * context, if a reader task is waiting. Apparently, this exciting
963 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
964 * failed somewhere. Latency? Burstiness? Well, at least now we will
965 * see, why it failed. 8)8) --ANK
967 * NOTE: is this not too big to inline?
969 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
971 struct tcp_sock *tp = tcp_sk(sk);
973 if (sysctl_tcp_low_latency || !tp->ucopy.task)
976 __skb_queue_tail(&tp->ucopy.prequeue, skb);
977 tp->ucopy.memory += skb->truesize;
978 if (tp->ucopy.memory > sk->sk_rcvbuf) {
979 struct sk_buff *skb1;
981 BUG_ON(sock_owned_by_user(sk));
983 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
984 sk_backlog_rcv(sk, skb1);
985 NET_INC_STATS_BH(sock_net(sk),
986 LINUX_MIB_TCPPREQUEUEDROPPED);
989 tp->ucopy.memory = 0;
990 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
991 wake_up_interruptible_sync_poll(sk_sleep(sk),
992 POLLIN | POLLRDNORM | POLLRDBAND);
993 if (!inet_csk_ack_scheduled(sk))
994 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
995 (3 * tcp_rto_min(sk)) / 4,
1005 static const char *statename[]={
1006 "Unused","Established","Syn Sent","Syn Recv",
1007 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1008 "Close Wait","Last ACK","Listen","Closing"
1011 extern void tcp_set_state(struct sock *sk, int state);
1013 extern void tcp_done(struct sock *sk);
1015 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1018 rx_opt->num_sacks = 0;
1021 /* Determine a window scaling and initial window to offer. */
1022 extern void tcp_select_initial_window(int __space, __u32 mss,
1023 __u32 *rcv_wnd, __u32 *window_clamp,
1024 int wscale_ok, __u8 *rcv_wscale,
1025 __u32 init_rcv_wnd);
1027 static inline int tcp_win_from_space(int space)
1029 return sysctl_tcp_adv_win_scale<=0 ?
1030 (space>>(-sysctl_tcp_adv_win_scale)) :
1031 space - (space>>sysctl_tcp_adv_win_scale);
1034 /* Note: caller must be prepared to deal with negative returns */
1035 static inline int tcp_space(const struct sock *sk)
1037 return tcp_win_from_space(sk->sk_rcvbuf -
1038 atomic_read(&sk->sk_rmem_alloc));
1041 static inline int tcp_full_space(const struct sock *sk)
1043 return tcp_win_from_space(sk->sk_rcvbuf);
1046 static inline void tcp_openreq_init(struct request_sock *req,
1047 struct tcp_options_received *rx_opt,
1048 struct sk_buff *skb)
1050 struct inet_request_sock *ireq = inet_rsk(req);
1052 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1054 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1055 req->mss = rx_opt->mss_clamp;
1056 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1057 ireq->tstamp_ok = rx_opt->tstamp_ok;
1058 ireq->sack_ok = rx_opt->sack_ok;
1059 ireq->snd_wscale = rx_opt->snd_wscale;
1060 ireq->wscale_ok = rx_opt->wscale_ok;
1063 ireq->rmt_port = tcp_hdr(skb)->source;
1064 ireq->loc_port = tcp_hdr(skb)->dest;
1067 extern void tcp_enter_memory_pressure(struct sock *sk);
1069 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1071 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1074 static inline int keepalive_time_when(const struct tcp_sock *tp)
1076 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1079 static inline int keepalive_probes(const struct tcp_sock *tp)
1081 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1084 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1086 const struct inet_connection_sock *icsk = &tp->inet_conn;
1088 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1089 tcp_time_stamp - tp->rcv_tstamp);
1092 static inline int tcp_fin_time(const struct sock *sk)
1094 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1095 const int rto = inet_csk(sk)->icsk_rto;
1097 if (fin_timeout < (rto << 2) - (rto >> 1))
1098 fin_timeout = (rto << 2) - (rto >> 1);
1103 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1106 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1108 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1111 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1112 * then following tcp messages have valid values. Ignore 0 value,
1113 * or else 'negative' tsval might forbid us to accept their packets.
1115 if (!rx_opt->ts_recent)
1120 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1123 if (tcp_paws_check(rx_opt, 0))
1126 /* RST segments are not recommended to carry timestamp,
1127 and, if they do, it is recommended to ignore PAWS because
1128 "their cleanup function should take precedence over timestamps."
1129 Certainly, it is mistake. It is necessary to understand the reasons
1130 of this constraint to relax it: if peer reboots, clock may go
1131 out-of-sync and half-open connections will not be reset.
1132 Actually, the problem would be not existing if all
1133 the implementations followed draft about maintaining clock
1134 via reboots. Linux-2.2 DOES NOT!
1136 However, we can relax time bounds for RST segments to MSL.
1138 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1143 static inline void tcp_mib_init(struct net *net)
1146 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1147 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1148 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1149 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1153 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1155 tp->lost_skb_hint = NULL;
1156 tp->scoreboard_skb_hint = NULL;
1159 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1161 tcp_clear_retrans_hints_partial(tp);
1162 tp->retransmit_skb_hint = NULL;
1168 union tcp_md5_addr {
1170 #if IS_ENABLED(CONFIG_IPV6)
1175 /* - key database */
1176 struct tcp_md5sig_key {
1177 struct hlist_node node;
1179 u8 family; /* AF_INET or AF_INET6 */
1180 union tcp_md5_addr addr;
1181 u8 key[TCP_MD5SIG_MAXKEYLEN];
1182 struct rcu_head rcu;
1186 struct tcp_md5sig_info {
1187 struct hlist_head head;
1188 struct rcu_head rcu;
1191 /* - pseudo header */
1192 struct tcp4_pseudohdr {
1200 struct tcp6_pseudohdr {
1201 struct in6_addr saddr;
1202 struct in6_addr daddr;
1204 __be32 protocol; /* including padding */
1207 union tcp_md5sum_block {
1208 struct tcp4_pseudohdr ip4;
1209 #if IS_ENABLED(CONFIG_IPV6)
1210 struct tcp6_pseudohdr ip6;
1214 /* - pool: digest algorithm, hash description and scratch buffer */
1215 struct tcp_md5sig_pool {
1216 struct hash_desc md5_desc;
1217 union tcp_md5sum_block md5_blk;
1221 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1222 const struct sock *sk,
1223 const struct request_sock *req,
1224 const struct sk_buff *skb);
1225 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1226 int family, const u8 *newkey,
1227 u8 newkeylen, gfp_t gfp);
1228 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1230 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1231 struct sock *addr_sk);
1233 #ifdef CONFIG_TCP_MD5SIG
1234 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1235 const union tcp_md5_addr *addr, int family);
1236 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1238 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1239 const union tcp_md5_addr *addr,
1244 #define tcp_twsk_md5_key(twsk) NULL
1247 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
1248 extern void tcp_free_md5sig_pool(void);
1250 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1251 extern void tcp_put_md5sig_pool(void);
1253 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1254 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1255 unsigned int header_len);
1256 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1257 const struct tcp_md5sig_key *key);
1259 /* write queue abstraction */
1260 static inline void tcp_write_queue_purge(struct sock *sk)
1262 struct sk_buff *skb;
1264 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1265 sk_wmem_free_skb(sk, skb);
1267 tcp_clear_all_retrans_hints(tcp_sk(sk));
1270 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1272 return skb_peek(&sk->sk_write_queue);
1275 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1277 return skb_peek_tail(&sk->sk_write_queue);
1280 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1281 const struct sk_buff *skb)
1283 return skb_queue_next(&sk->sk_write_queue, skb);
1286 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1287 const struct sk_buff *skb)
1289 return skb_queue_prev(&sk->sk_write_queue, skb);
1292 #define tcp_for_write_queue(skb, sk) \
1293 skb_queue_walk(&(sk)->sk_write_queue, skb)
1295 #define tcp_for_write_queue_from(skb, sk) \
1296 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1298 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1299 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1301 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1303 return sk->sk_send_head;
1306 static inline bool tcp_skb_is_last(const struct sock *sk,
1307 const struct sk_buff *skb)
1309 return skb_queue_is_last(&sk->sk_write_queue, skb);
1312 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1314 if (tcp_skb_is_last(sk, skb))
1315 sk->sk_send_head = NULL;
1317 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1320 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1322 if (sk->sk_send_head == skb_unlinked)
1323 sk->sk_send_head = NULL;
1326 static inline void tcp_init_send_head(struct sock *sk)
1328 sk->sk_send_head = NULL;
1331 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1333 __skb_queue_tail(&sk->sk_write_queue, skb);
1336 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1338 __tcp_add_write_queue_tail(sk, skb);
1340 /* Queue it, remembering where we must start sending. */
1341 if (sk->sk_send_head == NULL) {
1342 sk->sk_send_head = skb;
1344 if (tcp_sk(sk)->highest_sack == NULL)
1345 tcp_sk(sk)->highest_sack = skb;
1349 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1351 __skb_queue_head(&sk->sk_write_queue, skb);
1354 /* Insert buff after skb on the write queue of sk. */
1355 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1356 struct sk_buff *buff,
1359 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1362 /* Insert new before skb on the write queue of sk. */
1363 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1364 struct sk_buff *skb,
1367 __skb_queue_before(&sk->sk_write_queue, skb, new);
1369 if (sk->sk_send_head == skb)
1370 sk->sk_send_head = new;
1373 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1375 __skb_unlink(skb, &sk->sk_write_queue);
1378 static inline int tcp_write_queue_empty(struct sock *sk)
1380 return skb_queue_empty(&sk->sk_write_queue);
1383 static inline void tcp_push_pending_frames(struct sock *sk)
1385 if (tcp_send_head(sk)) {
1386 struct tcp_sock *tp = tcp_sk(sk);
1388 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1392 /* Start sequence of the skb just after the highest skb with SACKed
1393 * bit, valid only if sacked_out > 0 or when the caller has ensured
1394 * validity by itself.
1396 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1398 if (!tp->sacked_out)
1401 if (tp->highest_sack == NULL)
1404 return TCP_SKB_CB(tp->highest_sack)->seq;
1407 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1409 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1410 tcp_write_queue_next(sk, skb);
1413 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1415 return tcp_sk(sk)->highest_sack;
1418 static inline void tcp_highest_sack_reset(struct sock *sk)
1420 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1423 /* Called when old skb is about to be deleted (to be combined with new skb) */
1424 static inline void tcp_highest_sack_combine(struct sock *sk,
1425 struct sk_buff *old,
1426 struct sk_buff *new)
1428 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1429 tcp_sk(sk)->highest_sack = new;
1432 /* Determines whether this is a thin stream (which may suffer from
1433 * increased latency). Used to trigger latency-reducing mechanisms.
1435 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1437 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1441 enum tcp_seq_states {
1442 TCP_SEQ_STATE_LISTENING,
1443 TCP_SEQ_STATE_OPENREQ,
1444 TCP_SEQ_STATE_ESTABLISHED,
1445 TCP_SEQ_STATE_TIME_WAIT,
1448 int tcp_seq_open(struct inode *inode, struct file *file);
1450 struct tcp_seq_afinfo {
1453 const struct file_operations *seq_fops;
1454 struct seq_operations seq_ops;
1457 struct tcp_iter_state {
1458 struct seq_net_private p;
1460 enum tcp_seq_states state;
1461 struct sock *syn_wait_sk;
1462 int bucket, offset, sbucket, num, uid;
1466 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1467 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1469 extern struct request_sock_ops tcp_request_sock_ops;
1470 extern struct request_sock_ops tcp6_request_sock_ops;
1472 extern void tcp_v4_destroy_sock(struct sock *sk);
1474 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1475 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1476 netdev_features_t features);
1477 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1478 struct sk_buff *skb);
1479 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1480 struct sk_buff *skb);
1481 extern int tcp_gro_complete(struct sk_buff *skb);
1482 extern int tcp4_gro_complete(struct sk_buff *skb);
1484 #ifdef CONFIG_PROC_FS
1485 extern int tcp4_proc_init(void);
1486 extern void tcp4_proc_exit(void);
1489 /* TCP af-specific functions */
1490 struct tcp_sock_af_ops {
1491 #ifdef CONFIG_TCP_MD5SIG
1492 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1493 struct sock *addr_sk);
1494 int (*calc_md5_hash) (char *location,
1495 struct tcp_md5sig_key *md5,
1496 const struct sock *sk,
1497 const struct request_sock *req,
1498 const struct sk_buff *skb);
1499 int (*md5_parse) (struct sock *sk,
1500 char __user *optval,
1505 struct tcp_request_sock_ops {
1506 #ifdef CONFIG_TCP_MD5SIG
1507 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1508 struct request_sock *req);
1509 int (*calc_md5_hash) (char *location,
1510 struct tcp_md5sig_key *md5,
1511 const struct sock *sk,
1512 const struct request_sock *req,
1513 const struct sk_buff *skb);
1517 /* Using SHA1 for now, define some constants.
1519 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1520 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1521 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1523 extern int tcp_cookie_generator(u32 *bakery);
1526 * struct tcp_cookie_values - each socket needs extra space for the
1527 * cookies, together with (optional) space for any SYN data.
1529 * A tcp_sock contains a pointer to the current value, and this is
1530 * cloned to the tcp_timewait_sock.
1532 * @cookie_pair: variable data from the option exchange.
1534 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1535 * indicates default (sysctl_tcp_cookie_size).
1536 * After cookie sent, remembers size of cookie.
1537 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1539 * @s_data_desired: user specified tcpct_s_data_desired. When the
1540 * constant payload is specified (@s_data_constant),
1541 * holds its length instead.
1542 * Range 0 to TCP_MSS_DESIRED.
1544 * @s_data_payload: constant data that is to be included in the
1545 * payload of SYN or SYNACK segments when the
1546 * cookie option is present.
1548 struct tcp_cookie_values {
1550 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1551 u8 cookie_pair_size;
1553 u16 s_data_desired:11,
1558 u8 s_data_payload[0];
1561 static inline void tcp_cookie_values_release(struct kref *kref)
1563 kfree(container_of(kref, struct tcp_cookie_values, kref));
1566 /* The length of constant payload data. Note that s_data_desired is
1567 * overloaded, depending on s_data_constant: either the length of constant
1568 * data (returned here) or the limit on variable data.
1570 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1572 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1573 ? tp->cookie_values->s_data_desired
1578 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1580 * As tcp_request_sock has already been extended in other places, the
1581 * only remaining method is to pass stack values along as function
1582 * parameters. These parameters are not needed after sending SYNACK.
1584 * @cookie_bakery: cryptographic secret and message workspace.
1586 * @cookie_plus: bytes in authenticator/cookie option, copied from
1587 * struct tcp_options_received (above).
1589 struct tcp_extend_values {
1590 struct request_values rv;
1591 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1597 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1599 return (struct tcp_extend_values *)rvp;
1602 extern void tcp_v4_init(void);
1603 extern void tcp_init(void);