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.
22 #define FASTRETRANS_DEBUG 1
24 #include <linux/list.h>
25 #include <linux/tcp.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>
49 extern struct inet_hashinfo tcp_hashinfo
;
51 extern struct percpu_counter tcp_orphan_count
;
52 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Offer an initial receive window of 10 mss. */
64 #define TCP_DEFAULT_INIT_RCVWND 10
66 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
67 #define TCP_MIN_MSS 88U
69 /* The least MTU to use for probing */
70 #define TCP_BASE_MSS 512
72 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
73 #define TCP_FASTRETRANS_THRESH 3
75 /* Maximal reordering. */
76 #define TCP_MAX_REORDERING 127
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
82 #define TCP_URG_VALID 0x0100
83 #define TCP_URG_NOTYET 0x0200
84 #define TCP_URG_READ 0x0400
86 #define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
93 #define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
100 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
101 * connection: ~180sec is RFC minimum */
103 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
104 * connection: ~180sec is RFC minimum */
106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118 #define TCP_ATO_MIN ((unsigned)(HZ/25))
120 #define TCP_DELACK_MIN 4U
121 #define TCP_ATO_MIN 4U
123 #define TCP_RTO_MAX ((unsigned)(120*HZ))
124 #define TCP_RTO_MIN ((unsigned)(HZ/5))
125 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */
126 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
127 * used as a fallback RTO for the
128 * initial data transmission if no
129 * valid RTT sample has been acquired,
130 * most likely due to retrans in 3WHS.
133 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
134 * for local resources.
137 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
138 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
139 #define TCP_KEEPALIVE_INTVL (75*HZ)
141 #define MAX_TCP_KEEPIDLE 32767
142 #define MAX_TCP_KEEPINTVL 32767
143 #define MAX_TCP_KEEPCNT 127
144 #define MAX_TCP_SYNCNT 127
146 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
148 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
149 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
150 * after this time. It should be equal
151 * (or greater than) TCP_TIMEWAIT_LEN
152 * to provide reliability equal to one
153 * provided by timewait state.
155 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
156 * timestamps. It must be less than
157 * minimal timewait lifetime.
163 #define TCPOPT_NOP 1 /* Padding */
164 #define TCPOPT_EOL 0 /* End of options */
165 #define TCPOPT_MSS 2 /* Segment size negotiating */
166 #define TCPOPT_WINDOW 3 /* Window scaling */
167 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
168 #define TCPOPT_SACK 5 /* SACK Block */
169 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
170 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
171 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
177 #define TCPOLEN_MSS 4
178 #define TCPOLEN_WINDOW 3
179 #define TCPOLEN_SACK_PERM 2
180 #define TCPOLEN_TIMESTAMP 10
181 #define TCPOLEN_MD5SIG 18
182 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
183 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
184 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
185 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
187 /* But this is what stacks really send out. */
188 #define TCPOLEN_TSTAMP_ALIGNED 12
189 #define TCPOLEN_WSCALE_ALIGNED 4
190 #define TCPOLEN_SACKPERM_ALIGNED 4
191 #define TCPOLEN_SACK_BASE 2
192 #define TCPOLEN_SACK_BASE_ALIGNED 4
193 #define TCPOLEN_SACK_PERBLOCK 8
194 #define TCPOLEN_MD5SIG_ALIGNED 20
195 #define TCPOLEN_MSS_ALIGNED 4
197 /* Flags in tp->nonagle */
198 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
199 #define TCP_NAGLE_CORK 2 /* Socket is corked */
200 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
202 /* TCP thin-stream limits */
203 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
205 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
206 #define TCP_INIT_CWND 10
208 extern struct inet_timewait_death_row tcp_death_row
;
210 /* sysctl variables for tcp */
211 extern int sysctl_tcp_timestamps
;
212 extern int sysctl_tcp_window_scaling
;
213 extern int sysctl_tcp_sack
;
214 extern int sysctl_tcp_fin_timeout
;
215 extern int sysctl_tcp_keepalive_time
;
216 extern int sysctl_tcp_keepalive_probes
;
217 extern int sysctl_tcp_keepalive_intvl
;
218 extern int sysctl_tcp_syn_retries
;
219 extern int sysctl_tcp_synack_retries
;
220 extern int sysctl_tcp_retries1
;
221 extern int sysctl_tcp_retries2
;
222 extern int sysctl_tcp_orphan_retries
;
223 extern int sysctl_tcp_syncookies
;
224 extern int sysctl_tcp_retrans_collapse
;
225 extern int sysctl_tcp_stdurg
;
226 extern int sysctl_tcp_rfc1337
;
227 extern int sysctl_tcp_abort_on_overflow
;
228 extern int sysctl_tcp_max_orphans
;
229 extern int sysctl_tcp_fack
;
230 extern int sysctl_tcp_reordering
;
231 extern int sysctl_tcp_ecn
;
232 extern int sysctl_tcp_dsack
;
233 extern long sysctl_tcp_mem
[3];
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
;
256 extern atomic_long_t tcp_memory_allocated
;
257 extern struct percpu_counter tcp_sockets_allocated
;
258 extern int tcp_memory_pressure
;
261 * The next routines deal with comparing 32 bit unsigned ints
262 * and worry about wraparound (automatic with unsigned arithmetic).
265 static inline int before(__u32 seq1
, __u32 seq2
)
267 return (__s32
)(seq1
-seq2
) < 0;
269 #define after(seq2, seq1) before(seq1, seq2)
271 /* is s2<=s1<=s3 ? */
272 static inline int between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
274 return seq3
- seq2
>= seq1
- seq2
;
277 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
279 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
280 int orphans
= percpu_counter_read_positive(ocp
);
282 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
283 orphans
= percpu_counter_sum_positive(ocp
);
284 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
288 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
289 atomic_long_read(&tcp_memory_allocated
) > sysctl_tcp_mem
[2])
294 /* syncookies: remember time of last synqueue overflow */
295 static inline void tcp_synq_overflow(struct sock
*sk
)
297 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
300 /* syncookies: no recent synqueue overflow on this listening socket? */
301 static inline int tcp_synq_no_recent_overflow(const struct sock
*sk
)
303 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
304 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
307 extern struct proto tcp_prot
;
309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
315 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
317 extern void tcp_shutdown (struct sock
*sk
, int how
);
319 extern int tcp_v4_rcv(struct sk_buff
*skb
);
321 extern struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
);
322 extern void *tcp_v4_tw_get_peer(struct sock
*sk
);
323 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
324 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
326 extern int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
327 size_t size
, int flags
);
328 extern int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
329 extern int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
330 const struct tcphdr
*th
, unsigned int len
);
331 extern int tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
332 const struct tcphdr
*th
, unsigned int len
);
333 extern void tcp_rcv_space_adjust(struct sock
*sk
);
334 extern void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
335 extern int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
336 extern void tcp_twsk_destructor(struct sock
*sk
);
337 extern ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
338 struct pipe_inode_info
*pipe
, size_t len
,
341 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
342 const unsigned int pkts
)
344 struct inet_connection_sock
*icsk
= inet_csk(sk
);
346 if (icsk
->icsk_ack
.quick
) {
347 if (pkts
>= icsk
->icsk_ack
.quick
) {
348 icsk
->icsk_ack
.quick
= 0;
349 /* Leaving quickack mode we deflate ATO. */
350 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
352 icsk
->icsk_ack
.quick
-= pkts
;
357 #define TCP_ECN_QUEUE_CWR 2
358 #define TCP_ECN_DEMAND_CWR 4
359 #define TCP_ECN_SEEN 8
361 static __inline__
void
362 TCP_ECN_create_request(struct request_sock
*req
, struct tcphdr
*th
)
364 if (sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
)
365 inet_rsk(req
)->ecn_ok
= 1;
376 extern enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
378 const struct tcphdr
*th
);
379 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
380 struct request_sock
*req
,
381 struct request_sock
**prev
);
382 extern int tcp_child_process(struct sock
*parent
, struct sock
*child
,
383 struct sk_buff
*skb
);
384 extern int tcp_use_frto(struct sock
*sk
);
385 extern void tcp_enter_frto(struct sock
*sk
);
386 extern void tcp_enter_loss(struct sock
*sk
, int how
);
387 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
388 extern void tcp_update_metrics(struct sock
*sk
);
389 extern void tcp_close(struct sock
*sk
, long timeout
);
390 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
,
391 struct poll_table_struct
*wait
);
392 extern int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
393 char __user
*optval
, int __user
*optlen
);
394 extern int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
395 char __user
*optval
, unsigned int optlen
);
396 extern int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
397 char __user
*optval
, int __user
*optlen
);
398 extern int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
399 char __user
*optval
, unsigned int optlen
);
400 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
401 extern void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
402 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
403 size_t len
, int nonblock
, int flags
, int *addr_len
);
404 extern void tcp_parse_options(const struct sk_buff
*skb
,
405 struct tcp_options_received
*opt_rx
, const u8
**hvpp
,
407 extern const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
410 * TCP v4 functions exported for the inet6 API
413 extern void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
414 extern int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
415 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
416 struct request_sock
*req
,
417 struct sk_buff
*skb
);
418 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
419 struct request_sock
*req
,
420 struct dst_entry
*dst
);
421 extern int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
422 extern int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
424 extern int tcp_connect(struct sock
*sk
);
425 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
426 struct request_sock
*req
,
427 struct request_values
*rvp
);
428 extern int tcp_disconnect(struct sock
*sk
, int flags
);
431 /* From syncookies.c */
432 extern __u32 syncookie_secret
[2][16-4+SHA_DIGEST_WORDS
];
433 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
434 struct ip_options
*opt
);
435 #ifdef CONFIG_SYN_COOKIES
436 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
439 static inline __u32
cookie_v4_init_sequence(struct sock
*sk
,
447 extern __u32
cookie_init_timestamp(struct request_sock
*req
);
448 extern bool cookie_check_timestamp(struct tcp_options_received
*opt
, bool *);
450 /* From net/ipv6/syncookies.c */
451 extern struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
452 #ifdef CONFIG_SYN_COOKIES
453 extern __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
456 static inline __u32
cookie_v6_init_sequence(struct sock
*sk
,
465 extern void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
467 extern int tcp_may_send_now(struct sock
*sk
);
468 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
469 extern void tcp_retransmit_timer(struct sock
*sk
);
470 extern void tcp_xmit_retransmit_queue(struct sock
*);
471 extern void tcp_simple_retransmit(struct sock
*);
472 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
473 extern int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
475 extern void tcp_send_probe0(struct sock
*);
476 extern void tcp_send_partial(struct sock
*);
477 extern int tcp_write_wakeup(struct sock
*);
478 extern void tcp_send_fin(struct sock
*sk
);
479 extern void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
480 extern int tcp_send_synack(struct sock
*);
481 extern int tcp_syn_flood_action(struct sock
*sk
,
482 const struct sk_buff
*skb
,
484 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
485 extern void tcp_send_ack(struct sock
*sk
);
486 extern void tcp_send_delayed_ack(struct sock
*sk
);
489 extern void tcp_cwnd_application_limited(struct sock
*sk
);
492 extern void tcp_init_xmit_timers(struct sock
*);
493 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
495 inet_csk_clear_xmit_timers(sk
);
498 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
499 extern unsigned int tcp_current_mss(struct sock
*sk
);
501 /* Bound MSS / TSO packet size with the half of the window */
502 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
506 /* When peer uses tiny windows, there is no use in packetizing
507 * to sub-MSS pieces for the sake of SWS or making sure there
508 * are enough packets in the pipe for fast recovery.
510 * On the other hand, for extremely large MSS devices, handling
511 * smaller than MSS windows in this way does make sense.
513 if (tp
->max_window
>= 512)
514 cutoff
= (tp
->max_window
>> 1);
516 cutoff
= tp
->max_window
;
518 if (cutoff
&& pktsize
> cutoff
)
519 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
525 extern void tcp_get_info(const struct sock
*, struct tcp_info
*);
527 /* Read 'sendfile()'-style from a TCP socket */
528 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
529 unsigned int, size_t);
530 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
531 sk_read_actor_t recv_actor
);
533 extern void tcp_initialize_rcv_mss(struct sock
*sk
);
535 extern int tcp_mtu_to_mss(const struct sock
*sk
, int pmtu
);
536 extern int tcp_mss_to_mtu(const struct sock
*sk
, int mss
);
537 extern void tcp_mtup_init(struct sock
*sk
);
538 extern void tcp_valid_rtt_meas(struct sock
*sk
, u32 seq_rtt
);
540 static inline void tcp_bound_rto(const struct sock
*sk
)
542 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
543 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
546 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
548 return (tp
->srtt
>> 3) + tp
->rttvar
;
551 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
553 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
554 ntohl(TCP_FLAG_ACK
) |
558 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
560 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
563 static inline void tcp_fast_path_check(struct sock
*sk
)
565 struct tcp_sock
*tp
= tcp_sk(sk
);
567 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
569 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
571 tcp_fast_path_on(tp
);
574 /* Compute the actual rto_min value */
575 static inline u32
tcp_rto_min(struct sock
*sk
)
577 const struct dst_entry
*dst
= __sk_dst_get(sk
);
578 u32 rto_min
= TCP_RTO_MIN
;
580 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
581 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
585 /* Compute the actual receive window we are currently advertising.
586 * Rcv_nxt can be after the window if our peer push more data
587 * than the offered window.
589 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
591 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
598 /* Choose a new window, without checks for shrinking, and without
599 * scaling applied to the result. The caller does these things
600 * if necessary. This is a "raw" window selection.
602 extern u32
__tcp_select_window(struct sock
*sk
);
604 /* TCP timestamps are only 32-bits, this causes a slight
605 * complication on 64-bit systems since we store a snapshot
606 * of jiffies in the buffer control blocks below. We decided
607 * to use only the low 32-bits of jiffies and hide the ugly
608 * casts with the following macro.
610 #define tcp_time_stamp ((__u32)(jiffies))
612 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
614 #define TCPHDR_FIN 0x01
615 #define TCPHDR_SYN 0x02
616 #define TCPHDR_RST 0x04
617 #define TCPHDR_PSH 0x08
618 #define TCPHDR_ACK 0x10
619 #define TCPHDR_URG 0x20
620 #define TCPHDR_ECE 0x40
621 #define TCPHDR_CWR 0x80
623 /* This is what the send packet queuing engine uses to pass
624 * TCP per-packet control information to the transmission code.
625 * We also store the host-order sequence numbers in here too.
626 * This is 44 bytes if IPV6 is enabled.
627 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
631 struct inet_skb_parm h4
;
632 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
633 struct inet6_skb_parm h6
;
635 } header
; /* For incoming frames */
636 __u32 seq
; /* Starting sequence number */
637 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
638 __u32 when
; /* used to compute rtt's */
639 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
640 __u8 sacked
; /* State flags for SACK/FACK. */
641 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
642 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
643 #define TCPCB_LOST 0x04 /* SKB is lost */
644 #define TCPCB_TAGBITS 0x07 /* All tag bits */
645 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
647 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
648 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
650 __u32 ack_seq
; /* Sequence number ACK'd */
653 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
655 /* Due to TSO, an SKB can be composed of multiple actual
656 * packets. To keep these tracked properly, we use this.
658 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
660 return skb_shinfo(skb
)->gso_segs
;
663 /* This is valid iff tcp_skb_pcount() > 1. */
664 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
666 return skb_shinfo(skb
)->gso_size
;
669 /* Events passed to congestion control interface */
671 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
672 CA_EVENT_CWND_RESTART
, /* congestion window restart */
673 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
674 CA_EVENT_FRTO
, /* fast recovery timeout */
675 CA_EVENT_LOSS
, /* loss timeout */
676 CA_EVENT_FAST_ACK
, /* in sequence ack */
677 CA_EVENT_SLOW_ACK
, /* other ack */
681 * Interface for adding new TCP congestion control handlers
683 #define TCP_CA_NAME_MAX 16
684 #define TCP_CA_MAX 128
685 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
687 #define TCP_CONG_NON_RESTRICTED 0x1
688 #define TCP_CONG_RTT_STAMP 0x2
690 struct tcp_congestion_ops
{
691 struct list_head list
;
694 /* initialize private data (optional) */
695 void (*init
)(struct sock
*sk
);
696 /* cleanup private data (optional) */
697 void (*release
)(struct sock
*sk
);
699 /* return slow start threshold (required) */
700 u32 (*ssthresh
)(struct sock
*sk
);
701 /* lower bound for congestion window (optional) */
702 u32 (*min_cwnd
)(const struct sock
*sk
);
703 /* do new cwnd calculation (required) */
704 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 in_flight
);
705 /* call before changing ca_state (optional) */
706 void (*set_state
)(struct sock
*sk
, u8 new_state
);
707 /* call when cwnd event occurs (optional) */
708 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
709 /* new value of cwnd after loss (optional) */
710 u32 (*undo_cwnd
)(struct sock
*sk
);
711 /* hook for packet ack accounting (optional) */
712 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
713 /* get info for inet_diag (optional) */
714 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
716 char name
[TCP_CA_NAME_MAX
];
717 struct module
*owner
;
720 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
721 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
723 extern void tcp_init_congestion_control(struct sock
*sk
);
724 extern void tcp_cleanup_congestion_control(struct sock
*sk
);
725 extern int tcp_set_default_congestion_control(const char *name
);
726 extern void tcp_get_default_congestion_control(char *name
);
727 extern void tcp_get_available_congestion_control(char *buf
, size_t len
);
728 extern void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
729 extern int tcp_set_allowed_congestion_control(char *allowed
);
730 extern int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
731 extern void tcp_slow_start(struct tcp_sock
*tp
);
732 extern void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
734 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
735 extern u32
tcp_reno_ssthresh(struct sock
*sk
);
736 extern void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 in_flight
);
737 extern u32
tcp_reno_min_cwnd(const struct sock
*sk
);
738 extern struct tcp_congestion_ops tcp_reno
;
740 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
742 struct inet_connection_sock
*icsk
= inet_csk(sk
);
744 if (icsk
->icsk_ca_ops
->set_state
)
745 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
746 icsk
->icsk_ca_state
= ca_state
;
749 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
751 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
753 if (icsk
->icsk_ca_ops
->cwnd_event
)
754 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
757 /* These functions determine how the current flow behaves in respect of SACK
758 * handling. SACK is negotiated with the peer, and therefore it can vary
759 * between different flows.
761 * tcp_is_sack - SACK enabled
762 * tcp_is_reno - No SACK
763 * tcp_is_fack - FACK enabled, implies SACK enabled
765 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
767 return tp
->rx_opt
.sack_ok
;
770 static inline int tcp_is_reno(const struct tcp_sock
*tp
)
772 return !tcp_is_sack(tp
);
775 static inline int tcp_is_fack(const struct tcp_sock
*tp
)
777 return tp
->rx_opt
.sack_ok
& 2;
780 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
782 tp
->rx_opt
.sack_ok
|= 2;
785 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
787 return tp
->sacked_out
+ tp
->lost_out
;
790 /* This determines how many packets are "in the network" to the best
791 * of our knowledge. In many cases it is conservative, but where
792 * detailed information is available from the receiver (via SACK
793 * blocks etc.) we can make more aggressive calculations.
795 * Use this for decisions involving congestion control, use just
796 * tp->packets_out to determine if the send queue is empty or not.
798 * Read this equation as:
800 * "Packets sent once on transmission queue" MINUS
801 * "Packets left network, but not honestly ACKed yet" PLUS
802 * "Packets fast retransmitted"
804 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
806 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
809 #define TCP_INFINITE_SSTHRESH 0x7fffffff
811 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
813 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
816 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
817 * The exception is rate halving phase, when cwnd is decreasing towards
820 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
822 const struct tcp_sock
*tp
= tcp_sk(sk
);
824 if ((1 << inet_csk(sk
)->icsk_ca_state
) & (TCPF_CA_CWR
| TCPF_CA_Recovery
))
825 return tp
->snd_ssthresh
;
827 return max(tp
->snd_ssthresh
,
828 ((tp
->snd_cwnd
>> 1) +
829 (tp
->snd_cwnd
>> 2)));
832 /* Use define here intentionally to get WARN_ON location shown at the caller */
833 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
835 extern void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
836 extern __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
838 /* Slow start with delack produces 3 packets of burst, so that
839 * it is safe "de facto". This will be the default - same as
840 * the default reordering threshold - but if reordering increases,
841 * we must be able to allow cwnd to burst at least this much in order
842 * to not pull it back when holes are filled.
844 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
846 return tp
->reordering
;
849 /* Returns end sequence number of the receiver's advertised window */
850 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
852 return tp
->snd_una
+ tp
->snd_wnd
;
854 extern int tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
856 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
857 const struct sk_buff
*skb
)
860 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
863 static inline void tcp_check_probe_timer(struct sock
*sk
)
865 const struct tcp_sock
*tp
= tcp_sk(sk
);
866 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
868 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
869 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
870 icsk
->icsk_rto
, TCP_RTO_MAX
);
873 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
878 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
884 * Calculate(/check) TCP checksum
886 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
887 __be32 daddr
, __wsum base
)
889 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
892 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
894 return __skb_checksum_complete(skb
);
897 static inline int tcp_checksum_complete(struct sk_buff
*skb
)
899 return !skb_csum_unnecessary(skb
) &&
900 __tcp_checksum_complete(skb
);
903 /* Prequeue for VJ style copy to user, combined with checksumming. */
905 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
907 tp
->ucopy
.task
= NULL
;
909 tp
->ucopy
.memory
= 0;
910 skb_queue_head_init(&tp
->ucopy
.prequeue
);
911 #ifdef CONFIG_NET_DMA
912 tp
->ucopy
.dma_chan
= NULL
;
913 tp
->ucopy
.wakeup
= 0;
914 tp
->ucopy
.pinned_list
= NULL
;
915 tp
->ucopy
.dma_cookie
= 0;
919 /* Packet is added to VJ-style prequeue for processing in process
920 * context, if a reader task is waiting. Apparently, this exciting
921 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
922 * failed somewhere. Latency? Burstiness? Well, at least now we will
923 * see, why it failed. 8)8) --ANK
925 * NOTE: is this not too big to inline?
927 static inline int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
929 struct tcp_sock
*tp
= tcp_sk(sk
);
931 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
934 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
935 tp
->ucopy
.memory
+= skb
->truesize
;
936 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
937 struct sk_buff
*skb1
;
939 BUG_ON(sock_owned_by_user(sk
));
941 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
942 sk_backlog_rcv(sk
, skb1
);
943 NET_INC_STATS_BH(sock_net(sk
),
944 LINUX_MIB_TCPPREQUEUEDROPPED
);
947 tp
->ucopy
.memory
= 0;
948 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
949 wake_up_interruptible_sync_poll(sk_sleep(sk
),
950 POLLIN
| POLLRDNORM
| POLLRDBAND
);
951 if (!inet_csk_ack_scheduled(sk
))
952 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
953 (3 * tcp_rto_min(sk
)) / 4,
963 static const char *statename
[]={
964 "Unused","Established","Syn Sent","Syn Recv",
965 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
966 "Close Wait","Last ACK","Listen","Closing"
969 extern void tcp_set_state(struct sock
*sk
, int state
);
971 extern void tcp_done(struct sock
*sk
);
973 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
976 rx_opt
->num_sacks
= 0;
979 /* Determine a window scaling and initial window to offer. */
980 extern void tcp_select_initial_window(int __space
, __u32 mss
,
981 __u32
*rcv_wnd
, __u32
*window_clamp
,
982 int wscale_ok
, __u8
*rcv_wscale
,
985 static inline int tcp_win_from_space(int space
)
987 return sysctl_tcp_adv_win_scale
<=0 ?
988 (space
>>(-sysctl_tcp_adv_win_scale
)) :
989 space
- (space
>>sysctl_tcp_adv_win_scale
);
992 /* Note: caller must be prepared to deal with negative returns */
993 static inline int tcp_space(const struct sock
*sk
)
995 return tcp_win_from_space(sk
->sk_rcvbuf
-
996 atomic_read(&sk
->sk_rmem_alloc
));
999 static inline int tcp_full_space(const struct sock
*sk
)
1001 return tcp_win_from_space(sk
->sk_rcvbuf
);
1004 static inline void tcp_openreq_init(struct request_sock
*req
,
1005 struct tcp_options_received
*rx_opt
,
1006 struct sk_buff
*skb
)
1008 struct inet_request_sock
*ireq
= inet_rsk(req
);
1010 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1012 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1013 req
->mss
= rx_opt
->mss_clamp
;
1014 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1015 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1016 ireq
->sack_ok
= rx_opt
->sack_ok
;
1017 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1018 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1021 ireq
->rmt_port
= tcp_hdr(skb
)->source
;
1022 ireq
->loc_port
= tcp_hdr(skb
)->dest
;
1025 extern void tcp_enter_memory_pressure(struct sock
*sk
);
1027 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1029 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1032 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1034 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1037 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1039 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1042 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1044 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1046 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1047 tcp_time_stamp
- tp
->rcv_tstamp
);
1050 static inline int tcp_fin_time(const struct sock
*sk
)
1052 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1053 const int rto
= inet_csk(sk
)->icsk_rto
;
1055 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1056 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1061 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1064 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1066 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1069 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1070 * then following tcp messages have valid values. Ignore 0 value,
1071 * or else 'negative' tsval might forbid us to accept their packets.
1073 if (!rx_opt
->ts_recent
)
1078 static inline int tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1081 if (tcp_paws_check(rx_opt
, 0))
1084 /* RST segments are not recommended to carry timestamp,
1085 and, if they do, it is recommended to ignore PAWS because
1086 "their cleanup function should take precedence over timestamps."
1087 Certainly, it is mistake. It is necessary to understand the reasons
1088 of this constraint to relax it: if peer reboots, clock may go
1089 out-of-sync and half-open connections will not be reset.
1090 Actually, the problem would be not existing if all
1091 the implementations followed draft about maintaining clock
1092 via reboots. Linux-2.2 DOES NOT!
1094 However, we can relax time bounds for RST segments to MSL.
1096 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1101 static inline void tcp_mib_init(struct net
*net
)
1104 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1105 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1106 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1107 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1111 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1113 tp
->lost_skb_hint
= NULL
;
1114 tp
->scoreboard_skb_hint
= NULL
;
1117 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1119 tcp_clear_retrans_hints_partial(tp
);
1120 tp
->retransmit_skb_hint
= NULL
;
1126 /* - key database */
1127 struct tcp_md5sig_key
{
1132 struct tcp4_md5sig_key
{
1133 struct tcp_md5sig_key base
;
1137 struct tcp6_md5sig_key
{
1138 struct tcp_md5sig_key base
;
1140 u32 scope_id
; /* XXX */
1142 struct in6_addr addr
;
1146 struct tcp_md5sig_info
{
1147 struct tcp4_md5sig_key
*keys4
;
1148 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1149 struct tcp6_md5sig_key
*keys6
;
1157 /* - pseudo header */
1158 struct tcp4_pseudohdr
{
1166 struct tcp6_pseudohdr
{
1167 struct in6_addr saddr
;
1168 struct in6_addr daddr
;
1170 __be32 protocol
; /* including padding */
1173 union tcp_md5sum_block
{
1174 struct tcp4_pseudohdr ip4
;
1175 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1176 struct tcp6_pseudohdr ip6
;
1180 /* - pool: digest algorithm, hash description and scratch buffer */
1181 struct tcp_md5sig_pool
{
1182 struct hash_desc md5_desc
;
1183 union tcp_md5sum_block md5_blk
;
1187 extern int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1188 struct sock
*sk
, struct request_sock
*req
,
1189 struct sk_buff
*skb
);
1190 extern struct tcp_md5sig_key
* tcp_v4_md5_lookup(struct sock
*sk
,
1191 struct sock
*addr_sk
);
1192 extern int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
, u8
*newkey
,
1194 extern int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
);
1196 #ifdef CONFIG_TCP_MD5SIG
1197 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1198 &(struct tcp_md5sig_key) { \
1199 .key = (twsk)->tw_md5_key, \
1200 .keylen = (twsk)->tw_md5_keylen, \
1203 #define tcp_twsk_md5_key(twsk) NULL
1206 extern struct tcp_md5sig_pool __percpu
*tcp_alloc_md5sig_pool(struct sock
*);
1207 extern void tcp_free_md5sig_pool(void);
1209 extern struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1210 extern void tcp_put_md5sig_pool(void);
1212 extern int tcp_md5_hash_header(struct tcp_md5sig_pool
*, struct tcphdr
*);
1213 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1214 unsigned header_len
);
1215 extern int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1216 const struct tcp_md5sig_key
*key
);
1218 /* write queue abstraction */
1219 static inline void tcp_write_queue_purge(struct sock
*sk
)
1221 struct sk_buff
*skb
;
1223 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1224 sk_wmem_free_skb(sk
, skb
);
1226 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1229 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1231 return skb_peek(&sk
->sk_write_queue
);
1234 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1236 return skb_peek_tail(&sk
->sk_write_queue
);
1239 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1240 const struct sk_buff
*skb
)
1242 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1245 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1246 const struct sk_buff
*skb
)
1248 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1251 #define tcp_for_write_queue(skb, sk) \
1252 skb_queue_walk(&(sk)->sk_write_queue, skb)
1254 #define tcp_for_write_queue_from(skb, sk) \
1255 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1257 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1258 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1260 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1262 return sk
->sk_send_head
;
1265 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1266 const struct sk_buff
*skb
)
1268 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1271 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1273 if (tcp_skb_is_last(sk
, skb
))
1274 sk
->sk_send_head
= NULL
;
1276 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1279 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1281 if (sk
->sk_send_head
== skb_unlinked
)
1282 sk
->sk_send_head
= NULL
;
1285 static inline void tcp_init_send_head(struct sock
*sk
)
1287 sk
->sk_send_head
= NULL
;
1290 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1292 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1295 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1297 __tcp_add_write_queue_tail(sk
, skb
);
1299 /* Queue it, remembering where we must start sending. */
1300 if (sk
->sk_send_head
== NULL
) {
1301 sk
->sk_send_head
= skb
;
1303 if (tcp_sk(sk
)->highest_sack
== NULL
)
1304 tcp_sk(sk
)->highest_sack
= skb
;
1308 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1310 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1313 /* Insert buff after skb on the write queue of sk. */
1314 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1315 struct sk_buff
*buff
,
1318 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1321 /* Insert new before skb on the write queue of sk. */
1322 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1323 struct sk_buff
*skb
,
1326 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1328 if (sk
->sk_send_head
== skb
)
1329 sk
->sk_send_head
= new;
1332 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1334 __skb_unlink(skb
, &sk
->sk_write_queue
);
1337 static inline int tcp_write_queue_empty(struct sock
*sk
)
1339 return skb_queue_empty(&sk
->sk_write_queue
);
1342 static inline void tcp_push_pending_frames(struct sock
*sk
)
1344 if (tcp_send_head(sk
)) {
1345 struct tcp_sock
*tp
= tcp_sk(sk
);
1347 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1351 /* Start sequence of the highest skb with SACKed bit, valid only if
1352 * sacked > 0 or when the caller has ensured validity by itself.
1354 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1356 if (!tp
->sacked_out
)
1359 if (tp
->highest_sack
== NULL
)
1362 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1365 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1367 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1368 tcp_write_queue_next(sk
, skb
);
1371 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1373 return tcp_sk(sk
)->highest_sack
;
1376 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1378 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1381 /* Called when old skb is about to be deleted (to be combined with new skb) */
1382 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1383 struct sk_buff
*old
,
1384 struct sk_buff
*new)
1386 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1387 tcp_sk(sk
)->highest_sack
= new;
1390 /* Determines whether this is a thin stream (which may suffer from
1391 * increased latency). Used to trigger latency-reducing mechanisms.
1393 static inline unsigned int tcp_stream_is_thin(struct tcp_sock
*tp
)
1395 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1399 enum tcp_seq_states
{
1400 TCP_SEQ_STATE_LISTENING
,
1401 TCP_SEQ_STATE_OPENREQ
,
1402 TCP_SEQ_STATE_ESTABLISHED
,
1403 TCP_SEQ_STATE_TIME_WAIT
,
1406 struct tcp_seq_afinfo
{
1409 struct file_operations seq_fops
;
1410 struct seq_operations seq_ops
;
1413 struct tcp_iter_state
{
1414 struct seq_net_private p
;
1416 enum tcp_seq_states state
;
1417 struct sock
*syn_wait_sk
;
1418 int bucket
, offset
, sbucket
, num
, uid
;
1422 extern int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1423 extern void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1425 extern struct request_sock_ops tcp_request_sock_ops
;
1426 extern struct request_sock_ops tcp6_request_sock_ops
;
1428 extern void tcp_v4_destroy_sock(struct sock
*sk
);
1430 extern int tcp_v4_gso_send_check(struct sk_buff
*skb
);
1431 extern struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
, u32 features
);
1432 extern struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
,
1433 struct sk_buff
*skb
);
1434 extern struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
,
1435 struct sk_buff
*skb
);
1436 extern int tcp_gro_complete(struct sk_buff
*skb
);
1437 extern int tcp4_gro_complete(struct sk_buff
*skb
);
1439 #ifdef CONFIG_PROC_FS
1440 extern int tcp4_proc_init(void);
1441 extern void tcp4_proc_exit(void);
1444 /* TCP af-specific functions */
1445 struct tcp_sock_af_ops
{
1446 #ifdef CONFIG_TCP_MD5SIG
1447 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1448 struct sock
*addr_sk
);
1449 int (*calc_md5_hash
) (char *location
,
1450 struct tcp_md5sig_key
*md5
,
1452 struct request_sock
*req
,
1453 struct sk_buff
*skb
);
1454 int (*md5_add
) (struct sock
*sk
,
1455 struct sock
*addr_sk
,
1458 int (*md5_parse
) (struct sock
*sk
,
1459 char __user
*optval
,
1464 struct tcp_request_sock_ops
{
1465 #ifdef CONFIG_TCP_MD5SIG
1466 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1467 struct request_sock
*req
);
1468 int (*calc_md5_hash
) (char *location
,
1469 struct tcp_md5sig_key
*md5
,
1471 struct request_sock
*req
,
1472 struct sk_buff
*skb
);
1476 /* Using SHA1 for now, define some constants.
1478 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1479 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1480 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1482 extern int tcp_cookie_generator(u32
*bakery
);
1485 * struct tcp_cookie_values - each socket needs extra space for the
1486 * cookies, together with (optional) space for any SYN data.
1488 * A tcp_sock contains a pointer to the current value, and this is
1489 * cloned to the tcp_timewait_sock.
1491 * @cookie_pair: variable data from the option exchange.
1493 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1494 * indicates default (sysctl_tcp_cookie_size).
1495 * After cookie sent, remembers size of cookie.
1496 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1498 * @s_data_desired: user specified tcpct_s_data_desired. When the
1499 * constant payload is specified (@s_data_constant),
1500 * holds its length instead.
1501 * Range 0 to TCP_MSS_DESIRED.
1503 * @s_data_payload: constant data that is to be included in the
1504 * payload of SYN or SYNACK segments when the
1505 * cookie option is present.
1507 struct tcp_cookie_values
{
1509 u8 cookie_pair
[TCP_COOKIE_PAIR_SIZE
];
1510 u8 cookie_pair_size
;
1512 u16 s_data_desired
:11,
1517 u8 s_data_payload
[0];
1520 static inline void tcp_cookie_values_release(struct kref
*kref
)
1522 kfree(container_of(kref
, struct tcp_cookie_values
, kref
));
1525 /* The length of constant payload data. Note that s_data_desired is
1526 * overloaded, depending on s_data_constant: either the length of constant
1527 * data (returned here) or the limit on variable data.
1529 static inline int tcp_s_data_size(const struct tcp_sock
*tp
)
1531 return (tp
->cookie_values
!= NULL
&& tp
->cookie_values
->s_data_constant
)
1532 ? tp
->cookie_values
->s_data_desired
1537 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1539 * As tcp_request_sock has already been extended in other places, the
1540 * only remaining method is to pass stack values along as function
1541 * parameters. These parameters are not needed after sending SYNACK.
1543 * @cookie_bakery: cryptographic secret and message workspace.
1545 * @cookie_plus: bytes in authenticator/cookie option, copied from
1546 * struct tcp_options_received (above).
1548 struct tcp_extend_values
{
1549 struct request_values rv
;
1550 u32 cookie_bakery
[COOKIE_WORKSPACE_WORDS
];
1556 static inline struct tcp_extend_values
*tcp_xv(struct request_values
*rvp
)
1558 return (struct tcp_extend_values
*)rvp
;
1561 extern void tcp_v4_init(void);
1562 extern void tcp_init(void);