2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo
;
52 extern struct percpu_counter tcp_orphan_count
;
53 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
102 * connection: ~180sec is RFC minimum */
104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
105 * connection: ~180sec is RFC minimum */
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
128 * used as a fallback RTO for the
129 * initial data transmission if no
130 * valid RTT sample has been acquired,
131 * most likely due to retrans in 3WHS.
134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
135 * for local resources.
138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
140 #define TCP_KEEPALIVE_INTVL (75*HZ)
142 #define MAX_TCP_KEEPIDLE 32767
143 #define MAX_TCP_KEEPINTVL 32767
144 #define MAX_TCP_KEEPCNT 127
145 #define MAX_TCP_SYNCNT 127
147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
151 * after this time. It should be equal
152 * (or greater than) TCP_TIMEWAIT_LEN
153 * to provide reliability equal to one
154 * provided by timewait state.
156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
157 * timestamps. It must be less than
158 * minimal timewait lifetime.
164 #define TCPOPT_NOP 1 /* Padding */
165 #define TCPOPT_EOL 0 /* End of options */
166 #define TCPOPT_MSS 2 /* Segment size negotiating */
167 #define TCPOPT_WINDOW 3 /* Window scaling */
168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
169 #define TCPOPT_SACK 5 /* SACK Block */
170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
178 #define TCPOLEN_MSS 4
179 #define TCPOLEN_WINDOW 3
180 #define TCPOLEN_SACK_PERM 2
181 #define TCPOLEN_TIMESTAMP 10
182 #define TCPOLEN_MD5SIG 18
183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
188 /* But this is what stacks really send out. */
189 #define TCPOLEN_TSTAMP_ALIGNED 12
190 #define TCPOLEN_WSCALE_ALIGNED 4
191 #define TCPOLEN_SACKPERM_ALIGNED 4
192 #define TCPOLEN_SACK_BASE 2
193 #define TCPOLEN_SACK_BASE_ALIGNED 4
194 #define TCPOLEN_SACK_PERBLOCK 8
195 #define TCPOLEN_MD5SIG_ALIGNED 20
196 #define TCPOLEN_MSS_ALIGNED 4
198 /* Flags in tp->nonagle */
199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
200 #define TCP_NAGLE_CORK 2 /* Socket is corked */
201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
203 /* TCP thin-stream limits */
204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
207 #define TCP_INIT_CWND 10
209 extern struct inet_timewait_death_row tcp_death_row
;
211 /* sysctl variables for tcp */
212 extern int sysctl_tcp_timestamps
;
213 extern int sysctl_tcp_window_scaling
;
214 extern int sysctl_tcp_sack
;
215 extern int sysctl_tcp_fin_timeout
;
216 extern int sysctl_tcp_keepalive_time
;
217 extern int sysctl_tcp_keepalive_probes
;
218 extern int sysctl_tcp_keepalive_intvl
;
219 extern int sysctl_tcp_syn_retries
;
220 extern int sysctl_tcp_synack_retries
;
221 extern int sysctl_tcp_retries1
;
222 extern int sysctl_tcp_retries2
;
223 extern int sysctl_tcp_orphan_retries
;
224 extern int sysctl_tcp_syncookies
;
225 extern int sysctl_tcp_retrans_collapse
;
226 extern int sysctl_tcp_stdurg
;
227 extern int sysctl_tcp_rfc1337
;
228 extern int sysctl_tcp_abort_on_overflow
;
229 extern int sysctl_tcp_max_orphans
;
230 extern int sysctl_tcp_fack
;
231 extern int sysctl_tcp_reordering
;
232 extern int sysctl_tcp_ecn
;
233 extern int sysctl_tcp_dsack
;
234 extern int sysctl_tcp_wmem
[3];
235 extern int sysctl_tcp_rmem
[3];
236 extern int sysctl_tcp_app_win
;
237 extern int sysctl_tcp_adv_win_scale
;
238 extern int sysctl_tcp_tw_reuse
;
239 extern int sysctl_tcp_frto
;
240 extern int sysctl_tcp_frto_response
;
241 extern int sysctl_tcp_low_latency
;
242 extern int sysctl_tcp_dma_copybreak
;
243 extern int sysctl_tcp_nometrics_save
;
244 extern int sysctl_tcp_moderate_rcvbuf
;
245 extern int sysctl_tcp_tso_win_divisor
;
246 extern int sysctl_tcp_abc
;
247 extern int sysctl_tcp_mtu_probing
;
248 extern int sysctl_tcp_base_mss
;
249 extern int sysctl_tcp_workaround_signed_windows
;
250 extern int sysctl_tcp_slow_start_after_idle
;
251 extern int sysctl_tcp_max_ssthresh
;
252 extern int sysctl_tcp_cookie_size
;
253 extern int sysctl_tcp_thin_linear_timeouts
;
254 extern int sysctl_tcp_thin_dupack
;
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_out_of_memory(struct sock
*sk
)
279 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
280 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
285 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
287 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
288 int orphans
= percpu_counter_read_positive(ocp
);
290 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
291 orphans
= percpu_counter_sum_positive(ocp
);
292 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
298 extern bool tcp_check_oom(struct sock
*sk
, int shift
);
300 /* syncookies: remember time of last synqueue overflow */
301 static inline void tcp_synq_overflow(struct sock
*sk
)
303 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
306 /* syncookies: no recent synqueue overflow on this listening socket? */
307 static inline int tcp_synq_no_recent_overflow(const struct sock
*sk
)
309 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
310 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
313 extern struct proto tcp_prot
;
315 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
316 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
317 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
318 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
319 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
321 extern void tcp_init_mem(struct net
*net
);
323 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
325 extern void tcp_shutdown (struct sock
*sk
, int how
);
327 extern int tcp_v4_rcv(struct sk_buff
*skb
);
329 extern struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
);
330 extern void *tcp_v4_tw_get_peer(struct sock
*sk
);
331 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
332 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
334 extern int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
335 size_t size
, int flags
);
336 extern int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
337 extern int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
338 const struct tcphdr
*th
, unsigned int len
);
339 extern int tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
340 const struct tcphdr
*th
, unsigned int len
);
341 extern void tcp_rcv_space_adjust(struct sock
*sk
);
342 extern void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
343 extern int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
344 extern void tcp_twsk_destructor(struct sock
*sk
);
345 extern ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
346 struct pipe_inode_info
*pipe
, size_t len
,
349 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
350 const unsigned int pkts
)
352 struct inet_connection_sock
*icsk
= inet_csk(sk
);
354 if (icsk
->icsk_ack
.quick
) {
355 if (pkts
>= icsk
->icsk_ack
.quick
) {
356 icsk
->icsk_ack
.quick
= 0;
357 /* Leaving quickack mode we deflate ATO. */
358 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
360 icsk
->icsk_ack
.quick
-= pkts
;
365 #define TCP_ECN_QUEUE_CWR 2
366 #define TCP_ECN_DEMAND_CWR 4
367 #define TCP_ECN_SEEN 8
369 static __inline__
void
370 TCP_ECN_create_request(struct request_sock
*req
, struct tcphdr
*th
)
372 if (sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
)
373 inet_rsk(req
)->ecn_ok
= 1;
384 extern enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
386 const struct tcphdr
*th
);
387 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
388 struct request_sock
*req
,
389 struct request_sock
**prev
);
390 extern int tcp_child_process(struct sock
*parent
, struct sock
*child
,
391 struct sk_buff
*skb
);
392 extern int tcp_use_frto(struct sock
*sk
);
393 extern void tcp_enter_frto(struct sock
*sk
);
394 extern void tcp_enter_loss(struct sock
*sk
, int how
);
395 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
396 extern void tcp_update_metrics(struct sock
*sk
);
397 extern void tcp_close(struct sock
*sk
, long timeout
);
398 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
,
399 struct poll_table_struct
*wait
);
400 extern int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
401 char __user
*optval
, int __user
*optlen
);
402 extern int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
403 char __user
*optval
, unsigned int optlen
);
404 extern int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
405 char __user
*optval
, int __user
*optlen
);
406 extern int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
407 char __user
*optval
, unsigned int optlen
);
408 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
409 extern void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
410 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
411 size_t len
, int nonblock
, int flags
, int *addr_len
);
412 extern void tcp_parse_options(const struct sk_buff
*skb
,
413 struct tcp_options_received
*opt_rx
, const u8
**hvpp
,
415 extern const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
418 * TCP v4 functions exported for the inet6 API
421 extern void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
422 extern int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
423 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
424 struct request_sock
*req
,
425 struct sk_buff
*skb
);
426 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
427 struct request_sock
*req
,
428 struct dst_entry
*dst
);
429 extern int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
430 extern int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
432 extern int tcp_connect(struct sock
*sk
);
433 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
434 struct request_sock
*req
,
435 struct request_values
*rvp
);
436 extern int tcp_disconnect(struct sock
*sk
, int flags
);
438 void tcp_connect_init(struct sock
*sk
);
439 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
440 void tcp_queue_rcv(struct sock
*sk
, struct sk_buff
*skb
, int hdrlen
);
442 /* From syncookies.c */
443 extern __u32 syncookie_secret
[2][16-4+SHA_DIGEST_WORDS
];
444 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
445 struct ip_options
*opt
);
446 #ifdef CONFIG_SYN_COOKIES
447 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
450 static inline __u32
cookie_v4_init_sequence(struct sock
*sk
,
458 extern __u32
cookie_init_timestamp(struct request_sock
*req
);
459 extern bool cookie_check_timestamp(struct tcp_options_received
*opt
, bool *);
461 /* From net/ipv6/syncookies.c */
462 extern struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
463 #ifdef CONFIG_SYN_COOKIES
464 extern __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
467 static inline __u32
cookie_v6_init_sequence(struct sock
*sk
,
476 extern void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
478 extern int tcp_may_send_now(struct sock
*sk
);
479 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
480 extern void tcp_retransmit_timer(struct sock
*sk
);
481 extern void tcp_xmit_retransmit_queue(struct sock
*);
482 extern void tcp_simple_retransmit(struct sock
*);
483 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
484 extern int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
486 extern void tcp_send_probe0(struct sock
*);
487 extern void tcp_send_partial(struct sock
*);
488 extern int tcp_write_wakeup(struct sock
*);
489 extern void tcp_send_fin(struct sock
*sk
);
490 extern void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
491 extern int tcp_send_synack(struct sock
*);
492 extern int tcp_syn_flood_action(struct sock
*sk
,
493 const struct sk_buff
*skb
,
495 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
496 extern void tcp_send_ack(struct sock
*sk
);
497 extern void tcp_send_delayed_ack(struct sock
*sk
);
500 extern void tcp_cwnd_application_limited(struct sock
*sk
);
503 extern void tcp_init_xmit_timers(struct sock
*);
504 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
506 inet_csk_clear_xmit_timers(sk
);
509 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
510 extern unsigned int tcp_current_mss(struct sock
*sk
);
512 /* Bound MSS / TSO packet size with the half of the window */
513 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
517 /* When peer uses tiny windows, there is no use in packetizing
518 * to sub-MSS pieces for the sake of SWS or making sure there
519 * are enough packets in the pipe for fast recovery.
521 * On the other hand, for extremely large MSS devices, handling
522 * smaller than MSS windows in this way does make sense.
524 if (tp
->max_window
>= 512)
525 cutoff
= (tp
->max_window
>> 1);
527 cutoff
= tp
->max_window
;
529 if (cutoff
&& pktsize
> cutoff
)
530 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
536 extern void tcp_get_info(const struct sock
*, struct tcp_info
*);
538 /* Read 'sendfile()'-style from a TCP socket */
539 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
540 unsigned int, size_t);
541 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
542 sk_read_actor_t recv_actor
);
544 extern void tcp_initialize_rcv_mss(struct sock
*sk
);
546 extern int tcp_mtu_to_mss(const struct sock
*sk
, int pmtu
);
547 extern int tcp_mss_to_mtu(const struct sock
*sk
, int mss
);
548 extern void tcp_mtup_init(struct sock
*sk
);
549 extern void tcp_valid_rtt_meas(struct sock
*sk
, u32 seq_rtt
);
551 static inline void tcp_bound_rto(const struct sock
*sk
)
553 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
554 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
557 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
559 return (tp
->srtt
>> 3) + tp
->rttvar
;
562 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
564 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
565 ntohl(TCP_FLAG_ACK
) |
569 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
571 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
574 static inline void tcp_fast_path_check(struct sock
*sk
)
576 struct tcp_sock
*tp
= tcp_sk(sk
);
578 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
580 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
582 tcp_fast_path_on(tp
);
585 /* Compute the actual rto_min value */
586 static inline u32
tcp_rto_min(struct sock
*sk
)
588 const struct dst_entry
*dst
= __sk_dst_get(sk
);
589 u32 rto_min
= TCP_RTO_MIN
;
591 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
592 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
596 /* Compute the actual receive window we are currently advertising.
597 * Rcv_nxt can be after the window if our peer push more data
598 * than the offered window.
600 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
602 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
609 /* Choose a new window, without checks for shrinking, and without
610 * scaling applied to the result. The caller does these things
611 * if necessary. This is a "raw" window selection.
613 extern u32
__tcp_select_window(struct sock
*sk
);
615 void tcp_send_window_probe(struct sock
*sk
);
617 /* TCP timestamps are only 32-bits, this causes a slight
618 * complication on 64-bit systems since we store a snapshot
619 * of jiffies in the buffer control blocks below. We decided
620 * to use only the low 32-bits of jiffies and hide the ugly
621 * casts with the following macro.
623 #define tcp_time_stamp ((__u32)(jiffies))
625 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
627 #define TCPHDR_FIN 0x01
628 #define TCPHDR_SYN 0x02
629 #define TCPHDR_RST 0x04
630 #define TCPHDR_PSH 0x08
631 #define TCPHDR_ACK 0x10
632 #define TCPHDR_URG 0x20
633 #define TCPHDR_ECE 0x40
634 #define TCPHDR_CWR 0x80
636 /* This is what the send packet queuing engine uses to pass
637 * TCP per-packet control information to the transmission code.
638 * We also store the host-order sequence numbers in here too.
639 * This is 44 bytes if IPV6 is enabled.
640 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
644 struct inet_skb_parm h4
;
645 #if IS_ENABLED(CONFIG_IPV6)
646 struct inet6_skb_parm h6
;
648 } header
; /* For incoming frames */
649 __u32 seq
; /* Starting sequence number */
650 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
651 __u32 when
; /* used to compute rtt's */
652 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
654 __u8 sacked
; /* State flags for SACK/FACK. */
655 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
656 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
657 #define TCPCB_LOST 0x04 /* SKB is lost */
658 #define TCPCB_TAGBITS 0x07 /* All tag bits */
659 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
660 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
662 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
664 __u32 ack_seq
; /* Sequence number ACK'd */
667 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
669 /* Due to TSO, an SKB can be composed of multiple actual
670 * packets. To keep these tracked properly, we use this.
672 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
674 return skb_shinfo(skb
)->gso_segs
;
677 /* This is valid iff tcp_skb_pcount() > 1. */
678 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
680 return skb_shinfo(skb
)->gso_size
;
683 /* Events passed to congestion control interface */
685 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
686 CA_EVENT_CWND_RESTART
, /* congestion window restart */
687 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
688 CA_EVENT_FRTO
, /* fast recovery timeout */
689 CA_EVENT_LOSS
, /* loss timeout */
690 CA_EVENT_FAST_ACK
, /* in sequence ack */
691 CA_EVENT_SLOW_ACK
, /* other ack */
695 * Interface for adding new TCP congestion control handlers
697 #define TCP_CA_NAME_MAX 16
698 #define TCP_CA_MAX 128
699 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
701 #define TCP_CONG_NON_RESTRICTED 0x1
702 #define TCP_CONG_RTT_STAMP 0x2
704 struct tcp_congestion_ops
{
705 struct list_head list
;
708 /* initialize private data (optional) */
709 void (*init
)(struct sock
*sk
);
710 /* cleanup private data (optional) */
711 void (*release
)(struct sock
*sk
);
713 /* return slow start threshold (required) */
714 u32 (*ssthresh
)(struct sock
*sk
);
715 /* lower bound for congestion window (optional) */
716 u32 (*min_cwnd
)(const struct sock
*sk
);
717 /* do new cwnd calculation (required) */
718 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 in_flight
);
719 /* call before changing ca_state (optional) */
720 void (*set_state
)(struct sock
*sk
, u8 new_state
);
721 /* call when cwnd event occurs (optional) */
722 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
723 /* new value of cwnd after loss (optional) */
724 u32 (*undo_cwnd
)(struct sock
*sk
);
725 /* hook for packet ack accounting (optional) */
726 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
727 /* get info for inet_diag (optional) */
728 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
730 char name
[TCP_CA_NAME_MAX
];
731 struct module
*owner
;
734 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
735 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
737 extern void tcp_init_congestion_control(struct sock
*sk
);
738 extern void tcp_cleanup_congestion_control(struct sock
*sk
);
739 extern int tcp_set_default_congestion_control(const char *name
);
740 extern void tcp_get_default_congestion_control(char *name
);
741 extern void tcp_get_available_congestion_control(char *buf
, size_t len
);
742 extern void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
743 extern int tcp_set_allowed_congestion_control(char *allowed
);
744 extern int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
745 extern void tcp_slow_start(struct tcp_sock
*tp
);
746 extern void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
748 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
749 extern u32
tcp_reno_ssthresh(struct sock
*sk
);
750 extern void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 in_flight
);
751 extern u32
tcp_reno_min_cwnd(const struct sock
*sk
);
752 extern struct tcp_congestion_ops tcp_reno
;
754 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
756 struct inet_connection_sock
*icsk
= inet_csk(sk
);
758 if (icsk
->icsk_ca_ops
->set_state
)
759 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
760 icsk
->icsk_ca_state
= ca_state
;
763 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
765 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
767 if (icsk
->icsk_ca_ops
->cwnd_event
)
768 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
771 /* These functions determine how the current flow behaves in respect of SACK
772 * handling. SACK is negotiated with the peer, and therefore it can vary
773 * between different flows.
775 * tcp_is_sack - SACK enabled
776 * tcp_is_reno - No SACK
777 * tcp_is_fack - FACK enabled, implies SACK enabled
779 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
781 return tp
->rx_opt
.sack_ok
;
784 static inline int tcp_is_reno(const struct tcp_sock
*tp
)
786 return !tcp_is_sack(tp
);
789 static inline int tcp_is_fack(const struct tcp_sock
*tp
)
791 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
794 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
796 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
799 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
801 return tp
->sacked_out
+ tp
->lost_out
;
804 /* This determines how many packets are "in the network" to the best
805 * of our knowledge. In many cases it is conservative, but where
806 * detailed information is available from the receiver (via SACK
807 * blocks etc.) we can make more aggressive calculations.
809 * Use this for decisions involving congestion control, use just
810 * tp->packets_out to determine if the send queue is empty or not.
812 * Read this equation as:
814 * "Packets sent once on transmission queue" MINUS
815 * "Packets left network, but not honestly ACKed yet" PLUS
816 * "Packets fast retransmitted"
818 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
820 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
823 #define TCP_INFINITE_SSTHRESH 0x7fffffff
825 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
827 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
830 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
831 * The exception is rate halving phase, when cwnd is decreasing towards
834 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
836 const struct tcp_sock
*tp
= tcp_sk(sk
);
838 if ((1 << inet_csk(sk
)->icsk_ca_state
) & (TCPF_CA_CWR
| TCPF_CA_Recovery
))
839 return tp
->snd_ssthresh
;
841 return max(tp
->snd_ssthresh
,
842 ((tp
->snd_cwnd
>> 1) +
843 (tp
->snd_cwnd
>> 2)));
846 /* Use define here intentionally to get WARN_ON location shown at the caller */
847 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
849 extern void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
850 extern __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
852 /* The maximum number of MSS of available cwnd for which TSO defers
853 * sending if not using sysctl_tcp_tso_win_divisor.
855 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
860 /* Slow start with delack produces 3 packets of burst, so that
861 * it is safe "de facto". This will be the default - same as
862 * the default reordering threshold - but if reordering increases,
863 * we must be able to allow cwnd to burst at least this much in order
864 * to not pull it back when holes are filled.
866 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
868 return tp
->reordering
;
871 /* Returns end sequence number of the receiver's advertised window */
872 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
874 return tp
->snd_una
+ tp
->snd_wnd
;
876 extern int tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
878 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
879 const struct sk_buff
*skb
)
882 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
885 static inline void tcp_check_probe_timer(struct sock
*sk
)
887 const struct tcp_sock
*tp
= tcp_sk(sk
);
888 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
890 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
891 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
892 icsk
->icsk_rto
, TCP_RTO_MAX
);
895 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
900 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
906 * Calculate(/check) TCP checksum
908 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
909 __be32 daddr
, __wsum base
)
911 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
914 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
916 return __skb_checksum_complete(skb
);
919 static inline int tcp_checksum_complete(struct sk_buff
*skb
)
921 return !skb_csum_unnecessary(skb
) &&
922 __tcp_checksum_complete(skb
);
925 /* Prequeue for VJ style copy to user, combined with checksumming. */
927 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
929 tp
->ucopy
.task
= NULL
;
931 tp
->ucopy
.memory
= 0;
932 skb_queue_head_init(&tp
->ucopy
.prequeue
);
933 #ifdef CONFIG_NET_DMA
934 tp
->ucopy
.dma_chan
= NULL
;
935 tp
->ucopy
.wakeup
= 0;
936 tp
->ucopy
.pinned_list
= NULL
;
937 tp
->ucopy
.dma_cookie
= 0;
941 /* Packet is added to VJ-style prequeue for processing in process
942 * context, if a reader task is waiting. Apparently, this exciting
943 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
944 * failed somewhere. Latency? Burstiness? Well, at least now we will
945 * see, why it failed. 8)8) --ANK
947 * NOTE: is this not too big to inline?
949 static inline int tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
951 struct tcp_sock
*tp
= tcp_sk(sk
);
953 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
956 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
957 tp
->ucopy
.memory
+= skb
->truesize
;
958 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
959 struct sk_buff
*skb1
;
961 BUG_ON(sock_owned_by_user(sk
));
963 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
964 sk_backlog_rcv(sk
, skb1
);
965 NET_INC_STATS_BH(sock_net(sk
),
966 LINUX_MIB_TCPPREQUEUEDROPPED
);
969 tp
->ucopy
.memory
= 0;
970 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
971 wake_up_interruptible_sync_poll(sk_sleep(sk
),
972 POLLIN
| POLLRDNORM
| POLLRDBAND
);
973 if (!inet_csk_ack_scheduled(sk
))
974 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
975 (3 * tcp_rto_min(sk
)) / 4,
985 static const char *statename
[]={
986 "Unused","Established","Syn Sent","Syn Recv",
987 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
988 "Close Wait","Last ACK","Listen","Closing"
991 extern void tcp_set_state(struct sock
*sk
, int state
);
993 extern void tcp_done(struct sock
*sk
);
995 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
998 rx_opt
->num_sacks
= 0;
1001 /* Determine a window scaling and initial window to offer. */
1002 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1003 __u32
*rcv_wnd
, __u32
*window_clamp
,
1004 int wscale_ok
, __u8
*rcv_wscale
,
1005 __u32 init_rcv_wnd
);
1007 static inline int tcp_win_from_space(int space
)
1009 return sysctl_tcp_adv_win_scale
<=0 ?
1010 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1011 space
- (space
>>sysctl_tcp_adv_win_scale
);
1014 /* Note: caller must be prepared to deal with negative returns */
1015 static inline int tcp_space(const struct sock
*sk
)
1017 return tcp_win_from_space(sk
->sk_rcvbuf
-
1018 atomic_read(&sk
->sk_rmem_alloc
));
1021 static inline int tcp_full_space(const struct sock
*sk
)
1023 return tcp_win_from_space(sk
->sk_rcvbuf
);
1026 static inline void tcp_openreq_init(struct request_sock
*req
,
1027 struct tcp_options_received
*rx_opt
,
1028 struct sk_buff
*skb
)
1030 struct inet_request_sock
*ireq
= inet_rsk(req
);
1032 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1034 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1035 req
->mss
= rx_opt
->mss_clamp
;
1036 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1037 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1038 ireq
->sack_ok
= rx_opt
->sack_ok
;
1039 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1040 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1043 ireq
->rmt_port
= tcp_hdr(skb
)->source
;
1044 ireq
->loc_port
= tcp_hdr(skb
)->dest
;
1047 extern void tcp_enter_memory_pressure(struct sock
*sk
);
1049 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1051 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1054 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1056 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1059 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1061 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1064 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1066 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1068 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1069 tcp_time_stamp
- tp
->rcv_tstamp
);
1072 static inline int tcp_fin_time(const struct sock
*sk
)
1074 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1075 const int rto
= inet_csk(sk
)->icsk_rto
;
1077 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1078 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1083 static inline int tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1086 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1088 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1091 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1092 * then following tcp messages have valid values. Ignore 0 value,
1093 * or else 'negative' tsval might forbid us to accept their packets.
1095 if (!rx_opt
->ts_recent
)
1100 static inline int tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1103 if (tcp_paws_check(rx_opt
, 0))
1106 /* RST segments are not recommended to carry timestamp,
1107 and, if they do, it is recommended to ignore PAWS because
1108 "their cleanup function should take precedence over timestamps."
1109 Certainly, it is mistake. It is necessary to understand the reasons
1110 of this constraint to relax it: if peer reboots, clock may go
1111 out-of-sync and half-open connections will not be reset.
1112 Actually, the problem would be not existing if all
1113 the implementations followed draft about maintaining clock
1114 via reboots. Linux-2.2 DOES NOT!
1116 However, we can relax time bounds for RST segments to MSL.
1118 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1123 static inline void tcp_mib_init(struct net
*net
)
1126 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1127 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1128 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1129 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1133 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1135 tp
->lost_skb_hint
= NULL
;
1136 tp
->scoreboard_skb_hint
= NULL
;
1139 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1141 tcp_clear_retrans_hints_partial(tp
);
1142 tp
->retransmit_skb_hint
= NULL
;
1148 union tcp_md5_addr
{
1150 #if IS_ENABLED(CONFIG_IPV6)
1155 /* - key database */
1156 struct tcp_md5sig_key
{
1157 struct hlist_node node
;
1159 u8 family
; /* AF_INET or AF_INET6 */
1160 union tcp_md5_addr addr
;
1161 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1162 struct rcu_head rcu
;
1166 struct tcp_md5sig_info
{
1167 struct hlist_head head
;
1168 struct rcu_head rcu
;
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 IS_ENABLED(CONFIG_IPV6)
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 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1206 int family
, const u8
*newkey
,
1207 u8 newkeylen
, gfp_t gfp
);
1208 extern int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1210 extern struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1211 struct sock
*addr_sk
);
1213 #ifdef CONFIG_TCP_MD5SIG
1214 extern struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1215 const union tcp_md5_addr
*addr
, int family
);
1216 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1218 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1219 const union tcp_md5_addr
*addr
,
1224 #define tcp_twsk_md5_key(twsk) NULL
1227 extern struct tcp_md5sig_pool __percpu
*tcp_alloc_md5sig_pool(struct sock
*);
1228 extern void tcp_free_md5sig_pool(void);
1230 extern struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1231 extern void tcp_put_md5sig_pool(void);
1233 extern int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1234 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1235 unsigned int header_len
);
1236 extern int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1237 const struct tcp_md5sig_key
*key
);
1239 /* write queue abstraction */
1240 static inline void tcp_write_queue_purge(struct sock
*sk
)
1242 struct sk_buff
*skb
;
1244 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1245 sk_wmem_free_skb(sk
, skb
);
1247 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1250 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1252 return skb_peek(&sk
->sk_write_queue
);
1255 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1257 return skb_peek_tail(&sk
->sk_write_queue
);
1260 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1261 const struct sk_buff
*skb
)
1263 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1266 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1267 const struct sk_buff
*skb
)
1269 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1272 #define tcp_for_write_queue(skb, sk) \
1273 skb_queue_walk(&(sk)->sk_write_queue, skb)
1275 #define tcp_for_write_queue_from(skb, sk) \
1276 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1278 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1279 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1281 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1283 return sk
->sk_send_head
;
1286 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1287 const struct sk_buff
*skb
)
1289 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1292 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1294 if (tcp_skb_is_last(sk
, skb
))
1295 sk
->sk_send_head
= NULL
;
1297 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1300 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1302 if (sk
->sk_send_head
== skb_unlinked
)
1303 sk
->sk_send_head
= NULL
;
1306 static inline void tcp_init_send_head(struct sock
*sk
)
1308 sk
->sk_send_head
= NULL
;
1311 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1313 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1316 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1318 __tcp_add_write_queue_tail(sk
, skb
);
1320 /* Queue it, remembering where we must start sending. */
1321 if (sk
->sk_send_head
== NULL
) {
1322 sk
->sk_send_head
= skb
;
1324 if (tcp_sk(sk
)->highest_sack
== NULL
)
1325 tcp_sk(sk
)->highest_sack
= skb
;
1329 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1331 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1334 /* Insert buff after skb on the write queue of sk. */
1335 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1336 struct sk_buff
*buff
,
1339 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1342 /* Insert new before skb on the write queue of sk. */
1343 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1344 struct sk_buff
*skb
,
1347 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1349 if (sk
->sk_send_head
== skb
)
1350 sk
->sk_send_head
= new;
1353 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1355 __skb_unlink(skb
, &sk
->sk_write_queue
);
1358 static inline int tcp_write_queue_empty(struct sock
*sk
)
1360 return skb_queue_empty(&sk
->sk_write_queue
);
1363 static inline void tcp_push_pending_frames(struct sock
*sk
)
1365 if (tcp_send_head(sk
)) {
1366 struct tcp_sock
*tp
= tcp_sk(sk
);
1368 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1372 /* Start sequence of the skb just after the highest skb with SACKed
1373 * bit, valid only if sacked_out > 0 or when the caller has ensured
1374 * validity by itself.
1376 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1378 if (!tp
->sacked_out
)
1381 if (tp
->highest_sack
== NULL
)
1384 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1387 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1389 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1390 tcp_write_queue_next(sk
, skb
);
1393 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1395 return tcp_sk(sk
)->highest_sack
;
1398 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1400 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1403 /* Called when old skb is about to be deleted (to be combined with new skb) */
1404 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1405 struct sk_buff
*old
,
1406 struct sk_buff
*new)
1408 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1409 tcp_sk(sk
)->highest_sack
= new;
1412 /* Determines whether this is a thin stream (which may suffer from
1413 * increased latency). Used to trigger latency-reducing mechanisms.
1415 static inline unsigned int tcp_stream_is_thin(struct tcp_sock
*tp
)
1417 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1421 enum tcp_seq_states
{
1422 TCP_SEQ_STATE_LISTENING
,
1423 TCP_SEQ_STATE_OPENREQ
,
1424 TCP_SEQ_STATE_ESTABLISHED
,
1425 TCP_SEQ_STATE_TIME_WAIT
,
1428 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1430 struct tcp_seq_afinfo
{
1433 const struct file_operations
*seq_fops
;
1434 struct seq_operations seq_ops
;
1437 struct tcp_iter_state
{
1438 struct seq_net_private p
;
1440 enum tcp_seq_states state
;
1441 struct sock
*syn_wait_sk
;
1442 int bucket
, offset
, sbucket
, num
, uid
;
1446 extern int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1447 extern void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1449 extern struct request_sock_ops tcp_request_sock_ops
;
1450 extern struct request_sock_ops tcp6_request_sock_ops
;
1452 extern void tcp_v4_destroy_sock(struct sock
*sk
);
1454 extern int tcp_v4_gso_send_check(struct sk_buff
*skb
);
1455 extern struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
,
1456 netdev_features_t features
);
1457 extern struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
,
1458 struct sk_buff
*skb
);
1459 extern struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
,
1460 struct sk_buff
*skb
);
1461 extern int tcp_gro_complete(struct sk_buff
*skb
);
1462 extern int tcp4_gro_complete(struct sk_buff
*skb
);
1464 #ifdef CONFIG_PROC_FS
1465 extern int tcp4_proc_init(void);
1466 extern void tcp4_proc_exit(void);
1469 /* TCP af-specific functions */
1470 struct tcp_sock_af_ops
{
1471 #ifdef CONFIG_TCP_MD5SIG
1472 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1473 struct sock
*addr_sk
);
1474 int (*calc_md5_hash
) (char *location
,
1475 struct tcp_md5sig_key
*md5
,
1476 const struct sock
*sk
,
1477 const struct request_sock
*req
,
1478 const struct sk_buff
*skb
);
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);