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/slab.h>
26 #include <linux/cache.h>
27 #include <linux/percpu.h>
28 #include <linux/skbuff.h>
29 #include <linux/dmaengine.h>
30 #include <linux/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
48 extern struct inet_hashinfo tcp_hashinfo;
50 extern struct percpu_counter tcp_orphan_count;
51 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55 #define TCP_MIN_SND_MSS 48
56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
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)) /* RFC2988bis 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 long sysctl_tcp_mem[3];
235 extern int sysctl_tcp_wmem[3];
236 extern int sysctl_tcp_rmem[3];
237 extern int sysctl_tcp_app_win;
238 extern int sysctl_tcp_adv_win_scale;
239 extern int sysctl_tcp_tw_reuse;
240 extern int sysctl_tcp_frto;
241 extern int sysctl_tcp_frto_response;
242 extern int sysctl_tcp_low_latency;
243 extern int sysctl_tcp_dma_copybreak;
244 extern int sysctl_tcp_nometrics_save;
245 extern int sysctl_tcp_moderate_rcvbuf;
246 extern int sysctl_tcp_tso_win_divisor;
247 extern int sysctl_tcp_abc;
248 extern int sysctl_tcp_mtu_probing;
249 extern int sysctl_tcp_base_mss;
250 extern int sysctl_tcp_min_snd_mss;
251 extern int sysctl_tcp_workaround_signed_windows;
252 extern int sysctl_tcp_slow_start_after_idle;
253 extern int sysctl_tcp_max_ssthresh;
254 extern int sysctl_tcp_cookie_size;
255 extern int sysctl_tcp_thin_linear_timeouts;
256 extern int sysctl_tcp_thin_dupack;
257 extern int sysctl_tcp_challenge_ack_limit;
259 extern atomic_long_t tcp_memory_allocated;
260 extern struct percpu_counter tcp_sockets_allocated;
261 extern int tcp_memory_pressure;
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
268 static inline int before(__u32 seq1, __u32 seq2)
270 return (__s32)(seq1-seq2) < 0;
272 #define after(seq2, seq1) before(seq1, seq2)
274 /* is s2<=s1<=s3 ? */
275 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
277 return seq3 - seq2 >= seq1 - seq2;
280 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
282 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
283 int orphans = percpu_counter_read_positive(ocp);
285 if (orphans << shift > sysctl_tcp_max_orphans) {
286 orphans = percpu_counter_sum_positive(ocp);
287 if (orphans << shift > sysctl_tcp_max_orphans)
291 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
292 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
297 /* syncookies: remember time of last synqueue overflow */
298 static inline void tcp_synq_overflow(struct sock *sk)
300 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
303 /* syncookies: no recent synqueue overflow on this listening socket? */
304 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
306 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
307 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
310 extern struct proto tcp_prot;
312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
315 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
316 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
318 extern void tcp_v4_err(struct sk_buff *skb, u32);
320 extern void tcp_shutdown (struct sock *sk, int how);
322 extern int tcp_v4_rcv(struct sk_buff *skb);
324 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
325 extern void *tcp_v4_tw_get_peer(struct sock *sk);
326 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
327 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
329 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
330 size_t size, int flags);
331 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
332 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
333 const struct tcphdr *th, unsigned int len);
334 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
335 const struct tcphdr *th, unsigned int len);
336 extern void tcp_rcv_space_adjust(struct sock *sk);
337 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
338 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
339 extern void tcp_twsk_destructor(struct sock *sk);
340 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
341 struct pipe_inode_info *pipe, size_t len,
344 static inline void tcp_dec_quickack_mode(struct sock *sk,
345 const unsigned int pkts)
347 struct inet_connection_sock *icsk = inet_csk(sk);
349 if (icsk->icsk_ack.quick) {
350 if (pkts >= icsk->icsk_ack.quick) {
351 icsk->icsk_ack.quick = 0;
352 /* Leaving quickack mode we deflate ATO. */
353 icsk->icsk_ack.ato = TCP_ATO_MIN;
355 icsk->icsk_ack.quick -= pkts;
360 #define TCP_ECN_QUEUE_CWR 2
361 #define TCP_ECN_DEMAND_CWR 4
362 #define TCP_ECN_SEEN 8
372 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
374 const struct tcphdr *th);
375 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
376 struct request_sock *req,
377 struct request_sock **prev);
378 extern int tcp_child_process(struct sock *parent, struct sock *child,
379 struct sk_buff *skb);
380 extern int tcp_use_frto(struct sock *sk);
381 extern void tcp_enter_frto(struct sock *sk);
382 extern void tcp_enter_loss(struct sock *sk, int how);
383 extern void tcp_clear_retrans(struct tcp_sock *tp);
384 extern void tcp_update_metrics(struct sock *sk);
385 extern void tcp_close(struct sock *sk, long timeout);
386 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
387 struct poll_table_struct *wait);
388 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
389 char __user *optval, int __user *optlen);
390 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
391 char __user *optval, unsigned int optlen);
392 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
393 char __user *optval, int __user *optlen);
394 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
395 char __user *optval, unsigned int optlen);
396 extern void tcp_set_keepalive(struct sock *sk, int val);
397 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
398 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
399 size_t len, int nonblock, int flags, int *addr_len);
400 extern void tcp_parse_options(const struct sk_buff *skb,
401 struct tcp_options_received *opt_rx, const u8 **hvpp,
403 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
406 * TCP v4 functions exported for the inet6 API
409 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
410 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
411 extern struct sock * tcp_create_openreq_child(struct sock *sk,
412 struct request_sock *req,
413 struct sk_buff *skb);
414 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
415 struct request_sock *req,
416 struct dst_entry *dst);
417 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
418 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
420 extern int tcp_connect(struct sock *sk);
421 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
422 struct request_sock *req,
423 struct request_values *rvp);
424 extern int tcp_disconnect(struct sock *sk, int flags);
427 /* From syncookies.c */
428 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
429 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
430 struct ip_options *opt);
431 #ifdef CONFIG_SYN_COOKIES
432 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
435 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
443 extern __u32 cookie_init_timestamp(struct request_sock *req);
444 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
446 /* From net/ipv6/syncookies.c */
447 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
448 #ifdef CONFIG_SYN_COOKIES
449 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
452 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
461 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
463 extern int tcp_may_send_now(struct sock *sk);
464 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
465 extern void tcp_retransmit_timer(struct sock *sk);
466 extern void tcp_xmit_retransmit_queue(struct sock *);
467 extern void tcp_simple_retransmit(struct sock *);
468 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
469 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
471 extern void tcp_send_probe0(struct sock *);
472 extern void tcp_send_partial(struct sock *);
473 extern int tcp_write_wakeup(struct sock *);
474 extern void tcp_send_fin(struct sock *sk);
475 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
476 extern int tcp_send_synack(struct sock *);
477 extern int tcp_syn_flood_action(struct sock *sk,
478 const struct sk_buff *skb,
480 extern void tcp_push_one(struct sock *, unsigned int mss_now);
481 extern void tcp_send_ack(struct sock *sk);
482 extern void tcp_send_delayed_ack(struct sock *sk);
485 extern void tcp_cwnd_application_limited(struct sock *sk);
488 extern void tcp_init_xmit_timers(struct sock *);
489 static inline void tcp_clear_xmit_timers(struct sock *sk)
491 inet_csk_clear_xmit_timers(sk);
494 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
495 extern unsigned int tcp_current_mss(struct sock *sk);
497 /* Bound MSS / TSO packet size with the half of the window */
498 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
502 /* When peer uses tiny windows, there is no use in packetizing
503 * to sub-MSS pieces for the sake of SWS or making sure there
504 * are enough packets in the pipe for fast recovery.
506 * On the other hand, for extremely large MSS devices, handling
507 * smaller than MSS windows in this way does make sense.
509 if (tp->max_window >= 512)
510 cutoff = (tp->max_window >> 1);
512 cutoff = tp->max_window;
514 if (cutoff && pktsize > cutoff)
515 return max_t(int, cutoff, 68U - tp->tcp_header_len);
521 extern void tcp_get_info(const struct sock *, struct tcp_info *);
523 /* Read 'sendfile()'-style from a TCP socket */
524 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
525 unsigned int, size_t);
526 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
527 sk_read_actor_t recv_actor);
529 extern void tcp_initialize_rcv_mss(struct sock *sk);
531 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu);
532 extern int tcp_mss_to_mtu(const struct sock *sk, int mss);
533 extern void tcp_mtup_init(struct sock *sk);
534 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
536 static inline void tcp_bound_rto(const struct sock *sk)
538 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
539 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
542 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
544 return (tp->srtt >> 3) + tp->rttvar;
547 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
549 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
550 ntohl(TCP_FLAG_ACK) |
554 static inline void tcp_fast_path_on(struct tcp_sock *tp)
556 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
559 static inline void tcp_fast_path_check(struct sock *sk)
561 struct tcp_sock *tp = tcp_sk(sk);
563 if (skb_queue_empty(&tp->out_of_order_queue) &&
565 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
567 tcp_fast_path_on(tp);
570 /* Compute the actual rto_min value */
571 static inline u32 tcp_rto_min(struct sock *sk)
573 const struct dst_entry *dst = __sk_dst_get(sk);
574 u32 rto_min = TCP_RTO_MIN;
576 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
577 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
581 /* Compute the actual receive window we are currently advertising.
582 * Rcv_nxt can be after the window if our peer push more data
583 * than the offered window.
585 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
587 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
594 /* Choose a new window, without checks for shrinking, and without
595 * scaling applied to the result. The caller does these things
596 * if necessary. This is a "raw" window selection.
598 extern u32 __tcp_select_window(struct sock *sk);
600 /* TCP timestamps are only 32-bits, this causes a slight
601 * complication on 64-bit systems since we store a snapshot
602 * of jiffies in the buffer control blocks below. We decided
603 * to use only the low 32-bits of jiffies and hide the ugly
604 * casts with the following macro.
606 #define tcp_time_stamp ((__u32)(jiffies))
608 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
610 #define TCPHDR_FIN 0x01
611 #define TCPHDR_SYN 0x02
612 #define TCPHDR_RST 0x04
613 #define TCPHDR_PSH 0x08
614 #define TCPHDR_ACK 0x10
615 #define TCPHDR_URG 0x20
616 #define TCPHDR_ECE 0x40
617 #define TCPHDR_CWR 0x80
619 /* This is what the send packet queuing engine uses to pass
620 * TCP per-packet control information to the transmission code.
621 * We also store the host-order sequence numbers in here too.
622 * This is 44 bytes if IPV6 is enabled.
623 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
627 struct inet_skb_parm h4;
628 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
629 struct inet6_skb_parm h6;
631 } header; /* For incoming frames */
632 __u32 seq; /* Starting sequence number */
633 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
634 __u32 when; /* used to compute rtt's */
635 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
636 __u8 sacked; /* State flags for SACK/FACK. */
637 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
638 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
639 #define TCPCB_LOST 0x04 /* SKB is lost */
640 #define TCPCB_TAGBITS 0x07 /* All tag bits */
641 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
643 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
644 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
646 __u32 ack_seq; /* Sequence number ACK'd */
649 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
651 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
653 * If we receive a SYN packet with these bits set, it means a network is
654 * playing bad games with TOS bits. In order to avoid possible false congestion
655 * notifications, we disable TCP ECN negociation.
658 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
660 const struct tcphdr *th = tcp_hdr(skb);
662 if (sysctl_tcp_ecn && th->ece && th->cwr &&
663 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
664 inet_rsk(req)->ecn_ok = 1;
667 /* Due to TSO, an SKB can be composed of multiple actual
668 * packets. To keep these tracked properly, we use this.
670 static inline int tcp_skb_pcount(const struct sk_buff *skb)
672 return skb_shinfo(skb)->gso_segs;
675 /* This is valid iff tcp_skb_pcount() > 1. */
676 static inline int tcp_skb_mss(const struct sk_buff *skb)
678 return skb_shinfo(skb)->gso_size;
681 /* Events passed to congestion control interface */
683 CA_EVENT_TX_START, /* first transmit when no packets in flight */
684 CA_EVENT_CWND_RESTART, /* congestion window restart */
685 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
686 CA_EVENT_FRTO, /* fast recovery timeout */
687 CA_EVENT_LOSS, /* loss timeout */
688 CA_EVENT_FAST_ACK, /* in sequence ack */
689 CA_EVENT_SLOW_ACK, /* other ack */
693 * Interface for adding new TCP congestion control handlers
695 #define TCP_CA_NAME_MAX 16
696 #define TCP_CA_MAX 128
697 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
699 #define TCP_CONG_NON_RESTRICTED 0x1
700 #define TCP_CONG_RTT_STAMP 0x2
702 struct tcp_congestion_ops {
703 struct list_head list;
706 /* initialize private data (optional) */
707 void (*init)(struct sock *sk);
708 /* cleanup private data (optional) */
709 void (*release)(struct sock *sk);
711 /* return slow start threshold (required) */
712 u32 (*ssthresh)(struct sock *sk);
713 /* lower bound for congestion window (optional) */
714 u32 (*min_cwnd)(const struct sock *sk);
715 /* do new cwnd calculation (required) */
716 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
717 /* call before changing ca_state (optional) */
718 void (*set_state)(struct sock *sk, u8 new_state);
719 /* call when cwnd event occurs (optional) */
720 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
721 /* new value of cwnd after loss (optional) */
722 u32 (*undo_cwnd)(struct sock *sk);
723 /* hook for packet ack accounting (optional) */
724 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
725 /* get info for inet_diag (optional) */
726 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
728 char name[TCP_CA_NAME_MAX];
729 struct module *owner;
732 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
733 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
735 extern void tcp_init_congestion_control(struct sock *sk);
736 extern void tcp_cleanup_congestion_control(struct sock *sk);
737 extern int tcp_set_default_congestion_control(const char *name);
738 extern void tcp_get_default_congestion_control(char *name);
739 extern void tcp_get_available_congestion_control(char *buf, size_t len);
740 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
741 extern int tcp_set_allowed_congestion_control(char *allowed);
742 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
743 extern void tcp_slow_start(struct tcp_sock *tp);
744 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
746 extern struct tcp_congestion_ops tcp_init_congestion_ops;
747 extern u32 tcp_reno_ssthresh(struct sock *sk);
748 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
749 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
750 extern struct tcp_congestion_ops tcp_reno;
752 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
754 struct inet_connection_sock *icsk = inet_csk(sk);
756 if (icsk->icsk_ca_ops->set_state)
757 icsk->icsk_ca_ops->set_state(sk, ca_state);
758 icsk->icsk_ca_state = ca_state;
761 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
763 const struct inet_connection_sock *icsk = inet_csk(sk);
765 if (icsk->icsk_ca_ops->cwnd_event)
766 icsk->icsk_ca_ops->cwnd_event(sk, event);
769 /* These functions determine how the current flow behaves in respect of SACK
770 * handling. SACK is negotiated with the peer, and therefore it can vary
771 * between different flows.
773 * tcp_is_sack - SACK enabled
774 * tcp_is_reno - No SACK
775 * tcp_is_fack - FACK enabled, implies SACK enabled
777 static inline int tcp_is_sack(const struct tcp_sock *tp)
779 return tp->rx_opt.sack_ok;
782 static inline int tcp_is_reno(const struct tcp_sock *tp)
784 return !tcp_is_sack(tp);
787 static inline int tcp_is_fack(const struct tcp_sock *tp)
789 return tp->rx_opt.sack_ok & 2;
792 static inline void tcp_enable_fack(struct tcp_sock *tp)
794 tp->rx_opt.sack_ok |= 2;
797 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
799 return tp->sacked_out + tp->lost_out;
802 /* This determines how many packets are "in the network" to the best
803 * of our knowledge. In many cases it is conservative, but where
804 * detailed information is available from the receiver (via SACK
805 * blocks etc.) we can make more aggressive calculations.
807 * Use this for decisions involving congestion control, use just
808 * tp->packets_out to determine if the send queue is empty or not.
810 * Read this equation as:
812 * "Packets sent once on transmission queue" MINUS
813 * "Packets left network, but not honestly ACKed yet" PLUS
814 * "Packets fast retransmitted"
816 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
818 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
821 #define TCP_INFINITE_SSTHRESH 0x7fffffff
823 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
825 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
828 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
829 * The exception is rate halving phase, when cwnd is decreasing towards
832 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
834 const struct tcp_sock *tp = tcp_sk(sk);
836 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
837 return tp->snd_ssthresh;
839 return max(tp->snd_ssthresh,
840 ((tp->snd_cwnd >> 1) +
841 (tp->snd_cwnd >> 2)));
844 /* Use define here intentionally to get WARN_ON location shown at the caller */
845 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
847 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
848 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
850 /* Slow start with delack produces 3 packets of burst, so that
851 * it is safe "de facto". This will be the default - same as
852 * the default reordering threshold - but if reordering increases,
853 * we must be able to allow cwnd to burst at least this much in order
854 * to not pull it back when holes are filled.
856 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
858 return tp->reordering;
861 /* Returns end sequence number of the receiver's advertised window */
862 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
864 return tp->snd_una + tp->snd_wnd;
866 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
868 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
869 const struct sk_buff *skb)
872 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
875 static inline void tcp_check_probe_timer(struct sock *sk)
877 const struct tcp_sock *tp = tcp_sk(sk);
878 const struct inet_connection_sock *icsk = inet_csk(sk);
880 if (!tp->packets_out && !icsk->icsk_pending)
881 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
882 icsk->icsk_rto, TCP_RTO_MAX);
885 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
890 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
896 * Calculate(/check) TCP checksum
898 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
899 __be32 daddr, __wsum base)
901 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
904 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
906 return __skb_checksum_complete(skb);
909 static inline int tcp_checksum_complete(struct sk_buff *skb)
911 return !skb_csum_unnecessary(skb) &&
912 __tcp_checksum_complete(skb);
915 /* Prequeue for VJ style copy to user, combined with checksumming. */
917 static inline void tcp_prequeue_init(struct tcp_sock *tp)
919 tp->ucopy.task = NULL;
921 tp->ucopy.memory = 0;
922 skb_queue_head_init(&tp->ucopy.prequeue);
923 #ifdef CONFIG_NET_DMA
924 tp->ucopy.dma_chan = NULL;
925 tp->ucopy.wakeup = 0;
926 tp->ucopy.pinned_list = NULL;
927 tp->ucopy.dma_cookie = 0;
931 /* Packet is added to VJ-style prequeue for processing in process
932 * context, if a reader task is waiting. Apparently, this exciting
933 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
934 * failed somewhere. Latency? Burstiness? Well, at least now we will
935 * see, why it failed. 8)8) --ANK
937 * NOTE: is this not too big to inline?
939 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
941 struct tcp_sock *tp = tcp_sk(sk);
943 if (sysctl_tcp_low_latency || !tp->ucopy.task)
947 __skb_queue_tail(&tp->ucopy.prequeue, skb);
948 tp->ucopy.memory += skb->truesize;
949 if (tp->ucopy.memory > sk->sk_rcvbuf) {
950 struct sk_buff *skb1;
952 BUG_ON(sock_owned_by_user(sk));
954 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
955 sk_backlog_rcv(sk, skb1);
956 NET_INC_STATS_BH(sock_net(sk),
957 LINUX_MIB_TCPPREQUEUEDROPPED);
960 tp->ucopy.memory = 0;
961 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
962 wake_up_interruptible_sync_poll(sk_sleep(sk),
963 POLLIN | POLLRDNORM | POLLRDBAND);
964 if (!inet_csk_ack_scheduled(sk))
965 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
966 (3 * tcp_rto_min(sk)) / 4,
972 int tcp_filter(struct sock *sk, struct sk_buff *skb);
977 static const char *statename[]={
978 "Unused","Established","Syn Sent","Syn Recv",
979 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
980 "Close Wait","Last ACK","Listen","Closing"
983 extern void tcp_set_state(struct sock *sk, int state);
985 extern void tcp_done(struct sock *sk);
987 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
990 rx_opt->num_sacks = 0;
993 /* Determine a window scaling and initial window to offer. */
994 extern void tcp_select_initial_window(int __space, __u32 mss,
995 __u32 *rcv_wnd, __u32 *window_clamp,
996 int wscale_ok, __u8 *rcv_wscale,
999 static inline int tcp_win_from_space(int space)
1001 return sysctl_tcp_adv_win_scale<=0 ?
1002 (space>>(-sysctl_tcp_adv_win_scale)) :
1003 space - (space>>sysctl_tcp_adv_win_scale);
1006 /* Note: caller must be prepared to deal with negative returns */
1007 static inline int tcp_space(const struct sock *sk)
1009 return tcp_win_from_space(sk->sk_rcvbuf -
1010 atomic_read(&sk->sk_rmem_alloc));
1013 static inline int tcp_full_space(const struct sock *sk)
1015 return tcp_win_from_space(sk->sk_rcvbuf);
1018 static inline void tcp_openreq_init(struct request_sock *req,
1019 struct tcp_options_received *rx_opt,
1020 struct sk_buff *skb)
1022 struct inet_request_sock *ireq = inet_rsk(req);
1024 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1026 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1027 req->mss = rx_opt->mss_clamp;
1028 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1029 ireq->tstamp_ok = rx_opt->tstamp_ok;
1030 ireq->sack_ok = rx_opt->sack_ok;
1031 ireq->snd_wscale = rx_opt->snd_wscale;
1032 ireq->wscale_ok = rx_opt->wscale_ok;
1035 ireq->rmt_port = tcp_hdr(skb)->source;
1036 ireq->loc_port = tcp_hdr(skb)->dest;
1039 extern void tcp_enter_memory_pressure(struct sock *sk);
1041 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1043 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1046 static inline int keepalive_time_when(const struct tcp_sock *tp)
1048 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1051 static inline int keepalive_probes(const struct tcp_sock *tp)
1053 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1056 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1058 const struct inet_connection_sock *icsk = &tp->inet_conn;
1060 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1061 tcp_time_stamp - tp->rcv_tstamp);
1064 static inline int tcp_fin_time(const struct sock *sk)
1066 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1067 const int rto = inet_csk(sk)->icsk_rto;
1069 if (fin_timeout < (rto << 2) - (rto >> 1))
1070 fin_timeout = (rto << 2) - (rto >> 1);
1075 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1078 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1080 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1083 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1084 * then following tcp messages have valid values. Ignore 0 value,
1085 * or else 'negative' tsval might forbid us to accept their packets.
1087 if (!rx_opt->ts_recent)
1092 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1095 if (tcp_paws_check(rx_opt, 0))
1098 /* RST segments are not recommended to carry timestamp,
1099 and, if they do, it is recommended to ignore PAWS because
1100 "their cleanup function should take precedence over timestamps."
1101 Certainly, it is mistake. It is necessary to understand the reasons
1102 of this constraint to relax it: if peer reboots, clock may go
1103 out-of-sync and half-open connections will not be reset.
1104 Actually, the problem would be not existing if all
1105 the implementations followed draft about maintaining clock
1106 via reboots. Linux-2.2 DOES NOT!
1108 However, we can relax time bounds for RST segments to MSL.
1110 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1115 static inline void tcp_mib_init(struct net *net)
1118 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1119 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1120 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1121 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1125 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1127 tp->lost_skb_hint = NULL;
1128 tp->scoreboard_skb_hint = NULL;
1131 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1133 tcp_clear_retrans_hints_partial(tp);
1134 tp->retransmit_skb_hint = NULL;
1140 /* - key database */
1141 struct tcp_md5sig_key {
1146 struct tcp4_md5sig_key {
1147 struct tcp_md5sig_key base;
1151 struct tcp6_md5sig_key {
1152 struct tcp_md5sig_key base;
1154 u32 scope_id; /* XXX */
1156 struct in6_addr addr;
1160 struct tcp_md5sig_info {
1161 struct tcp4_md5sig_key *keys4;
1162 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1163 struct tcp6_md5sig_key *keys6;
1171 /* - pseudo header */
1172 struct tcp4_pseudohdr {
1180 struct tcp6_pseudohdr {
1181 struct in6_addr saddr;
1182 struct in6_addr daddr;
1184 __be32 protocol; /* including padding */
1187 union tcp_md5sum_block {
1188 struct tcp4_pseudohdr ip4;
1189 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1190 struct tcp6_pseudohdr ip6;
1194 /* - pool: digest algorithm, hash description and scratch buffer */
1195 struct tcp_md5sig_pool {
1196 struct hash_desc md5_desc;
1197 union tcp_md5sum_block md5_blk;
1201 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1202 const struct sock *sk,
1203 const struct request_sock *req,
1204 const struct sk_buff *skb);
1205 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1206 struct sock *addr_sk);
1207 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1209 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1211 #ifdef CONFIG_TCP_MD5SIG
1212 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1213 &(struct tcp_md5sig_key) { \
1214 .key = (twsk)->tw_md5_key, \
1215 .keylen = (twsk)->tw_md5_keylen, \
1218 #define tcp_twsk_md5_key(twsk) NULL
1221 extern bool tcp_alloc_md5sig_pool(void);
1223 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1224 static inline void tcp_put_md5sig_pool(void)
1229 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1230 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1231 unsigned header_len);
1232 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1233 const struct tcp_md5sig_key *key);
1235 /* write queue abstraction */
1236 static inline void tcp_write_queue_purge(struct sock *sk)
1238 struct sk_buff *skb;
1240 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1241 sk_wmem_free_skb(sk, skb);
1243 tcp_clear_all_retrans_hints(tcp_sk(sk));
1246 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1248 return skb_peek(&sk->sk_write_queue);
1251 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1253 return skb_peek_tail(&sk->sk_write_queue);
1256 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1257 const struct sk_buff *skb)
1259 return skb_queue_next(&sk->sk_write_queue, skb);
1262 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1263 const struct sk_buff *skb)
1265 return skb_queue_prev(&sk->sk_write_queue, skb);
1268 #define tcp_for_write_queue(skb, sk) \
1269 skb_queue_walk(&(sk)->sk_write_queue, skb)
1271 #define tcp_for_write_queue_from(skb, sk) \
1272 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1274 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1275 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1277 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1279 return sk->sk_send_head;
1282 static inline bool tcp_skb_is_last(const struct sock *sk,
1283 const struct sk_buff *skb)
1285 return skb_queue_is_last(&sk->sk_write_queue, skb);
1288 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1290 if (tcp_skb_is_last(sk, skb))
1291 sk->sk_send_head = NULL;
1293 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1296 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1298 if (sk->sk_send_head == skb_unlinked)
1299 sk->sk_send_head = NULL;
1300 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1301 tcp_sk(sk)->highest_sack = NULL;
1304 static inline void tcp_init_send_head(struct sock *sk)
1306 sk->sk_send_head = NULL;
1309 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1311 __skb_queue_tail(&sk->sk_write_queue, skb);
1314 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1316 __tcp_add_write_queue_tail(sk, skb);
1318 /* Queue it, remembering where we must start sending. */
1319 if (sk->sk_send_head == NULL) {
1320 sk->sk_send_head = skb;
1322 if (tcp_sk(sk)->highest_sack == NULL)
1323 tcp_sk(sk)->highest_sack = skb;
1327 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1329 __skb_queue_head(&sk->sk_write_queue, skb);
1332 /* Insert buff after skb on the write queue of sk. */
1333 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1334 struct sk_buff *buff,
1337 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1340 /* Insert new before skb on the write queue of sk. */
1341 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1342 struct sk_buff *skb,
1345 __skb_queue_before(&sk->sk_write_queue, skb, new);
1347 if (sk->sk_send_head == skb)
1348 sk->sk_send_head = new;
1351 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1353 __skb_unlink(skb, &sk->sk_write_queue);
1356 static inline int tcp_write_queue_empty(struct sock *sk)
1358 return skb_queue_empty(&sk->sk_write_queue);
1361 static inline void tcp_push_pending_frames(struct sock *sk)
1363 if (tcp_send_head(sk)) {
1364 struct tcp_sock *tp = tcp_sk(sk);
1366 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1370 /* Start sequence of the highest skb with SACKed bit, valid only if
1371 * sacked > 0 or when the caller has ensured validity by itself.
1373 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1375 if (!tp->sacked_out)
1378 if (tp->highest_sack == NULL)
1381 return TCP_SKB_CB(tp->highest_sack)->seq;
1384 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1386 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1387 tcp_write_queue_next(sk, skb);
1390 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1392 return tcp_sk(sk)->highest_sack;
1395 static inline void tcp_highest_sack_reset(struct sock *sk)
1397 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1400 /* Called when old skb is about to be deleted and replaced by new skb */
1401 static inline void tcp_highest_sack_replace(struct sock *sk,
1402 struct sk_buff *old,
1403 struct sk_buff *new)
1405 if (old == tcp_highest_sack(sk))
1406 tcp_sk(sk)->highest_sack = new;
1409 /* Determines whether this is a thin stream (which may suffer from
1410 * increased latency). Used to trigger latency-reducing mechanisms.
1412 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1414 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1418 enum tcp_seq_states {
1419 TCP_SEQ_STATE_LISTENING,
1420 TCP_SEQ_STATE_OPENREQ,
1421 TCP_SEQ_STATE_ESTABLISHED,
1422 TCP_SEQ_STATE_TIME_WAIT,
1425 int tcp_seq_open(struct inode *inode, struct file *file);
1427 struct tcp_seq_afinfo {
1430 const struct file_operations *seq_fops;
1431 struct seq_operations seq_ops;
1434 struct tcp_iter_state {
1435 struct seq_net_private p;
1437 enum tcp_seq_states state;
1438 struct sock *syn_wait_sk;
1439 int bucket, offset, sbucket, num, uid;
1443 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1444 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1446 extern struct request_sock_ops tcp_request_sock_ops;
1447 extern struct request_sock_ops tcp6_request_sock_ops;
1449 extern void tcp_v4_destroy_sock(struct sock *sk);
1451 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1452 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features);
1453 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1454 struct sk_buff *skb);
1455 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1456 struct sk_buff *skb);
1457 extern int tcp_gro_complete(struct sk_buff *skb);
1458 extern int tcp4_gro_complete(struct sk_buff *skb);
1460 #ifdef CONFIG_PROC_FS
1461 extern int tcp4_proc_init(void);
1462 extern void tcp4_proc_exit(void);
1465 /* TCP af-specific functions */
1466 struct tcp_sock_af_ops {
1467 #ifdef CONFIG_TCP_MD5SIG
1468 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1469 struct sock *addr_sk);
1470 int (*calc_md5_hash) (char *location,
1471 struct tcp_md5sig_key *md5,
1472 const struct sock *sk,
1473 const struct request_sock *req,
1474 const struct sk_buff *skb);
1475 int (*md5_add) (struct sock *sk,
1476 struct sock *addr_sk,
1479 int (*md5_parse) (struct sock *sk,
1480 char __user *optval,
1485 struct tcp_request_sock_ops {
1486 #ifdef CONFIG_TCP_MD5SIG
1487 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1488 struct request_sock *req);
1489 int (*calc_md5_hash) (char *location,
1490 struct tcp_md5sig_key *md5,
1491 const struct sock *sk,
1492 const struct request_sock *req,
1493 const struct sk_buff *skb);
1497 /* Using SHA1 for now, define some constants.
1499 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1500 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1501 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1503 extern int tcp_cookie_generator(u32 *bakery);
1506 * struct tcp_cookie_values - each socket needs extra space for the
1507 * cookies, together with (optional) space for any SYN data.
1509 * A tcp_sock contains a pointer to the current value, and this is
1510 * cloned to the tcp_timewait_sock.
1512 * @cookie_pair: variable data from the option exchange.
1514 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1515 * indicates default (sysctl_tcp_cookie_size).
1516 * After cookie sent, remembers size of cookie.
1517 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1519 * @s_data_desired: user specified tcpct_s_data_desired. When the
1520 * constant payload is specified (@s_data_constant),
1521 * holds its length instead.
1522 * Range 0 to TCP_MSS_DESIRED.
1524 * @s_data_payload: constant data that is to be included in the
1525 * payload of SYN or SYNACK segments when the
1526 * cookie option is present.
1528 struct tcp_cookie_values {
1530 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1531 u8 cookie_pair_size;
1533 u16 s_data_desired:11,
1538 u8 s_data_payload[0];
1541 static inline void tcp_cookie_values_release(struct kref *kref)
1543 kfree(container_of(kref, struct tcp_cookie_values, kref));
1546 /* The length of constant payload data. Note that s_data_desired is
1547 * overloaded, depending on s_data_constant: either the length of constant
1548 * data (returned here) or the limit on variable data.
1550 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1552 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1553 ? tp->cookie_values->s_data_desired
1558 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1560 * As tcp_request_sock has already been extended in other places, the
1561 * only remaining method is to pass stack values along as function
1562 * parameters. These parameters are not needed after sending SYNACK.
1564 * @cookie_bakery: cryptographic secret and message workspace.
1566 * @cookie_plus: bytes in authenticator/cookie option, copied from
1567 * struct tcp_options_received (above).
1569 struct tcp_extend_values {
1570 struct request_values rv;
1571 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1577 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1579 return (struct tcp_extend_values *)rvp;
1582 extern void tcp_v4_init(void);
1583 extern void tcp_init(void);