2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
50 extern struct inet_hashinfo tcp_hashinfo
;
52 extern struct percpu_counter tcp_orphan_count
;
53 extern void tcp_time_wait(struct sock
*sk
, int state
, int timeo
);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
102 * connection: ~180sec is RFC minimum */
104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
105 * connection: ~180sec is RFC minimum */
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
128 * used as a fallback RTO for the
129 * initial data transmission if no
130 * valid RTT sample has been acquired,
131 * most likely due to retrans in 3WHS.
134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
135 * for local resources.
138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
140 #define TCP_KEEPALIVE_INTVL (75*HZ)
142 #define MAX_TCP_KEEPIDLE 32767
143 #define MAX_TCP_KEEPINTVL 32767
144 #define MAX_TCP_KEEPCNT 127
145 #define MAX_TCP_SYNCNT 127
147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
151 * after this time. It should be equal
152 * (or greater than) TCP_TIMEWAIT_LEN
153 * to provide reliability equal to one
154 * provided by timewait state.
156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
157 * timestamps. It must be less than
158 * minimal timewait lifetime.
164 #define TCPOPT_NOP 1 /* Padding */
165 #define TCPOPT_EOL 0 /* End of options */
166 #define TCPOPT_MSS 2 /* Segment size negotiating */
167 #define TCPOPT_WINDOW 3 /* Window scaling */
168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
169 #define TCPOPT_SACK 5 /* SACK Block */
170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
178 #define TCPOLEN_MSS 4
179 #define TCPOLEN_WINDOW 3
180 #define TCPOLEN_SACK_PERM 2
181 #define TCPOLEN_TIMESTAMP 10
182 #define TCPOLEN_MD5SIG 18
183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
188 /* But this is what stacks really send out. */
189 #define TCPOLEN_TSTAMP_ALIGNED 12
190 #define TCPOLEN_WSCALE_ALIGNED 4
191 #define TCPOLEN_SACKPERM_ALIGNED 4
192 #define TCPOLEN_SACK_BASE 2
193 #define TCPOLEN_SACK_BASE_ALIGNED 4
194 #define TCPOLEN_SACK_PERBLOCK 8
195 #define TCPOLEN_MD5SIG_ALIGNED 20
196 #define TCPOLEN_MSS_ALIGNED 4
198 /* Flags in tp->nonagle */
199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
200 #define TCP_NAGLE_CORK 2 /* Socket is corked */
201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
203 /* TCP thin-stream limits */
204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
207 #define TCP_INIT_CWND 10
209 extern struct inet_timewait_death_row tcp_death_row
;
211 /* sysctl variables for tcp */
212 extern int sysctl_tcp_timestamps
;
213 extern int sysctl_tcp_window_scaling
;
214 extern int sysctl_tcp_sack
;
215 extern int sysctl_tcp_fin_timeout
;
216 extern int sysctl_tcp_keepalive_time
;
217 extern int sysctl_tcp_keepalive_probes
;
218 extern int sysctl_tcp_keepalive_intvl
;
219 extern int sysctl_tcp_syn_retries
;
220 extern int sysctl_tcp_synack_retries
;
221 extern int sysctl_tcp_retries1
;
222 extern int sysctl_tcp_retries2
;
223 extern int sysctl_tcp_orphan_retries
;
224 extern int sysctl_tcp_syncookies
;
225 extern int sysctl_tcp_retrans_collapse
;
226 extern int sysctl_tcp_stdurg
;
227 extern int sysctl_tcp_rfc1337
;
228 extern int sysctl_tcp_abort_on_overflow
;
229 extern int sysctl_tcp_max_orphans
;
230 extern int sysctl_tcp_fack
;
231 extern int sysctl_tcp_reordering
;
232 extern int sysctl_tcp_ecn
;
233 extern int sysctl_tcp_dsack
;
234 extern int sysctl_tcp_wmem
[3];
235 extern int sysctl_tcp_rmem
[3];
236 extern int sysctl_tcp_app_win
;
237 extern int sysctl_tcp_adv_win_scale
;
238 extern int sysctl_tcp_tw_reuse
;
239 extern int sysctl_tcp_frto
;
240 extern int sysctl_tcp_frto_response
;
241 extern int sysctl_tcp_low_latency
;
242 extern int sysctl_tcp_dma_copybreak
;
243 extern int sysctl_tcp_nometrics_save
;
244 extern int sysctl_tcp_moderate_rcvbuf
;
245 extern int sysctl_tcp_tso_win_divisor
;
246 extern int sysctl_tcp_abc
;
247 extern int sysctl_tcp_mtu_probing
;
248 extern int sysctl_tcp_base_mss
;
249 extern int sysctl_tcp_workaround_signed_windows
;
250 extern int sysctl_tcp_slow_start_after_idle
;
251 extern int sysctl_tcp_max_ssthresh
;
252 extern int sysctl_tcp_cookie_size
;
253 extern int sysctl_tcp_thin_linear_timeouts
;
254 extern int sysctl_tcp_thin_dupack
;
255 extern int sysctl_tcp_early_retrans
;
257 extern atomic_long_t tcp_memory_allocated
;
258 extern struct percpu_counter tcp_sockets_allocated
;
259 extern int tcp_memory_pressure
;
262 * The next routines deal with comparing 32 bit unsigned ints
263 * and worry about wraparound (automatic with unsigned arithmetic).
266 static inline bool before(__u32 seq1
, __u32 seq2
)
268 return (__s32
)(seq1
-seq2
) < 0;
270 #define after(seq2, seq1) before(seq1, seq2)
272 /* is s2<=s1<=s3 ? */
273 static inline bool between(__u32 seq1
, __u32 seq2
, __u32 seq3
)
275 return seq3
- seq2
>= seq1
- seq2
;
278 static inline bool tcp_out_of_memory(struct sock
*sk
)
280 if (sk
->sk_wmem_queued
> SOCK_MIN_SNDBUF
&&
281 sk_memory_allocated(sk
) > sk_prot_mem_limits(sk
, 2))
286 static inline bool tcp_too_many_orphans(struct sock
*sk
, int shift
)
288 struct percpu_counter
*ocp
= sk
->sk_prot
->orphan_count
;
289 int orphans
= percpu_counter_read_positive(ocp
);
291 if (orphans
<< shift
> sysctl_tcp_max_orphans
) {
292 orphans
= percpu_counter_sum_positive(ocp
);
293 if (orphans
<< shift
> sysctl_tcp_max_orphans
)
299 extern bool tcp_check_oom(struct sock
*sk
, int shift
);
301 /* syncookies: remember time of last synqueue overflow */
302 static inline void tcp_synq_overflow(struct sock
*sk
)
304 tcp_sk(sk
)->rx_opt
.ts_recent_stamp
= jiffies
;
307 /* syncookies: no recent synqueue overflow on this listening socket? */
308 static inline bool tcp_synq_no_recent_overflow(const struct sock
*sk
)
310 unsigned long last_overflow
= tcp_sk(sk
)->rx_opt
.ts_recent_stamp
;
311 return time_after(jiffies
, last_overflow
+ TCP_TIMEOUT_FALLBACK
);
314 extern struct proto tcp_prot
;
316 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
317 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
318 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
319 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
320 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
322 extern void tcp_init_mem(struct net
*net
);
324 extern void tcp_v4_err(struct sk_buff
*skb
, u32
);
326 extern void tcp_shutdown (struct sock
*sk
, int how
);
328 extern void tcp_v4_early_demux(struct sk_buff
*skb
);
329 extern int tcp_v4_rcv(struct sk_buff
*skb
);
331 extern struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
);
332 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
);
333 extern int tcp_sendmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
335 extern int tcp_sendpage(struct sock
*sk
, struct page
*page
, int offset
,
336 size_t size
, int flags
);
337 extern int tcp_ioctl(struct sock
*sk
, int cmd
, unsigned long arg
);
338 extern int tcp_rcv_state_process(struct sock
*sk
, struct sk_buff
*skb
,
339 const struct tcphdr
*th
, unsigned int len
);
340 extern int tcp_rcv_established(struct sock
*sk
, struct sk_buff
*skb
,
341 const struct tcphdr
*th
, unsigned int len
);
342 extern void tcp_rcv_space_adjust(struct sock
*sk
);
343 extern void tcp_cleanup_rbuf(struct sock
*sk
, int copied
);
344 extern int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
);
345 extern void tcp_twsk_destructor(struct sock
*sk
);
346 extern ssize_t
tcp_splice_read(struct socket
*sk
, loff_t
*ppos
,
347 struct pipe_inode_info
*pipe
, size_t len
,
350 static inline void tcp_dec_quickack_mode(struct sock
*sk
,
351 const unsigned int pkts
)
353 struct inet_connection_sock
*icsk
= inet_csk(sk
);
355 if (icsk
->icsk_ack
.quick
) {
356 if (pkts
>= icsk
->icsk_ack
.quick
) {
357 icsk
->icsk_ack
.quick
= 0;
358 /* Leaving quickack mode we deflate ATO. */
359 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
361 icsk
->icsk_ack
.quick
-= pkts
;
366 #define TCP_ECN_QUEUE_CWR 2
367 #define TCP_ECN_DEMAND_CWR 4
368 #define TCP_ECN_SEEN 8
378 extern enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock
*tw
,
380 const struct tcphdr
*th
);
381 extern struct sock
* tcp_check_req(struct sock
*sk
,struct sk_buff
*skb
,
382 struct request_sock
*req
,
383 struct request_sock
**prev
);
384 extern int tcp_child_process(struct sock
*parent
, struct sock
*child
,
385 struct sk_buff
*skb
);
386 extern bool tcp_use_frto(struct sock
*sk
);
387 extern void tcp_enter_frto(struct sock
*sk
);
388 extern void tcp_enter_loss(struct sock
*sk
, int how
);
389 extern void tcp_clear_retrans(struct tcp_sock
*tp
);
390 extern void tcp_update_metrics(struct sock
*sk
);
391 extern void tcp_init_metrics(struct sock
*sk
);
392 extern void tcp_disable_fack(struct tcp_sock
*tp
);
393 extern void tcp_close(struct sock
*sk
, long timeout
);
394 extern void tcp_init_sock(struct sock
*sk
);
395 extern unsigned int tcp_poll(struct file
* file
, struct socket
*sock
,
396 struct poll_table_struct
*wait
);
397 extern int tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
398 char __user
*optval
, int __user
*optlen
);
399 extern int tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
400 char __user
*optval
, unsigned int optlen
);
401 extern int compat_tcp_getsockopt(struct sock
*sk
, int level
, int optname
,
402 char __user
*optval
, int __user
*optlen
);
403 extern int compat_tcp_setsockopt(struct sock
*sk
, int level
, int optname
,
404 char __user
*optval
, unsigned int optlen
);
405 extern void tcp_set_keepalive(struct sock
*sk
, int val
);
406 extern void tcp_syn_ack_timeout(struct sock
*sk
, struct request_sock
*req
);
407 extern int tcp_recvmsg(struct kiocb
*iocb
, struct sock
*sk
, struct msghdr
*msg
,
408 size_t len
, int nonblock
, int flags
, int *addr_len
);
409 extern void tcp_parse_options(const struct sk_buff
*skb
,
410 struct tcp_options_received
*opt_rx
, const u8
**hvpp
,
412 extern const u8
*tcp_parse_md5sig_option(const struct tcphdr
*th
);
415 * TCP v4 functions exported for the inet6 API
418 extern void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
);
419 extern int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
);
420 extern struct sock
* tcp_create_openreq_child(struct sock
*sk
,
421 struct request_sock
*req
,
422 struct sk_buff
*skb
);
423 extern struct sock
* tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
424 struct request_sock
*req
,
425 struct dst_entry
*dst
);
426 extern int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
);
427 extern int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
,
429 extern int tcp_connect(struct sock
*sk
);
430 extern struct sk_buff
* tcp_make_synack(struct sock
*sk
, struct dst_entry
*dst
,
431 struct request_sock
*req
,
432 struct request_values
*rvp
);
433 extern int tcp_disconnect(struct sock
*sk
, int flags
);
435 void tcp_connect_init(struct sock
*sk
);
436 void tcp_finish_connect(struct sock
*sk
, struct sk_buff
*skb
);
437 int tcp_send_rcvq(struct sock
*sk
, struct msghdr
*msg
, size_t size
);
439 /* From syncookies.c */
440 extern __u32 syncookie_secret
[2][16-4+SHA_DIGEST_WORDS
];
441 extern struct sock
*cookie_v4_check(struct sock
*sk
, struct sk_buff
*skb
,
442 struct ip_options
*opt
);
443 #ifdef CONFIG_SYN_COOKIES
444 extern __u32
cookie_v4_init_sequence(struct sock
*sk
, struct sk_buff
*skb
,
447 static inline __u32
cookie_v4_init_sequence(struct sock
*sk
,
455 extern __u32
cookie_init_timestamp(struct request_sock
*req
);
456 extern bool cookie_check_timestamp(struct tcp_options_received
*opt
, bool *);
458 /* From net/ipv6/syncookies.c */
459 extern struct sock
*cookie_v6_check(struct sock
*sk
, struct sk_buff
*skb
);
460 #ifdef CONFIG_SYN_COOKIES
461 extern __u32
cookie_v6_init_sequence(struct sock
*sk
, const struct sk_buff
*skb
,
464 static inline __u32
cookie_v6_init_sequence(struct sock
*sk
,
473 extern void __tcp_push_pending_frames(struct sock
*sk
, unsigned int cur_mss
,
475 extern bool tcp_may_send_now(struct sock
*sk
);
476 extern int tcp_retransmit_skb(struct sock
*, struct sk_buff
*);
477 extern void tcp_retransmit_timer(struct sock
*sk
);
478 extern void tcp_xmit_retransmit_queue(struct sock
*);
479 extern void tcp_simple_retransmit(struct sock
*);
480 extern int tcp_trim_head(struct sock
*, struct sk_buff
*, u32
);
481 extern int tcp_fragment(struct sock
*, struct sk_buff
*, u32
, unsigned int);
483 extern void tcp_send_probe0(struct sock
*);
484 extern void tcp_send_partial(struct sock
*);
485 extern int tcp_write_wakeup(struct sock
*);
486 extern void tcp_send_fin(struct sock
*sk
);
487 extern void tcp_send_active_reset(struct sock
*sk
, gfp_t priority
);
488 extern int tcp_send_synack(struct sock
*);
489 extern bool tcp_syn_flood_action(struct sock
*sk
,
490 const struct sk_buff
*skb
,
492 extern void tcp_push_one(struct sock
*, unsigned int mss_now
);
493 extern void tcp_send_ack(struct sock
*sk
);
494 extern void tcp_send_delayed_ack(struct sock
*sk
);
497 extern void tcp_cwnd_application_limited(struct sock
*sk
);
498 extern void tcp_resume_early_retransmit(struct sock
*sk
);
499 extern void tcp_rearm_rto(struct sock
*sk
);
502 extern void tcp_init_xmit_timers(struct sock
*);
503 static inline void tcp_clear_xmit_timers(struct sock
*sk
)
505 inet_csk_clear_xmit_timers(sk
);
508 extern unsigned int tcp_sync_mss(struct sock
*sk
, u32 pmtu
);
509 extern unsigned int tcp_current_mss(struct sock
*sk
);
511 /* Bound MSS / TSO packet size with the half of the window */
512 static inline int tcp_bound_to_half_wnd(struct tcp_sock
*tp
, int pktsize
)
516 /* When peer uses tiny windows, there is no use in packetizing
517 * to sub-MSS pieces for the sake of SWS or making sure there
518 * are enough packets in the pipe for fast recovery.
520 * On the other hand, for extremely large MSS devices, handling
521 * smaller than MSS windows in this way does make sense.
523 if (tp
->max_window
>= 512)
524 cutoff
= (tp
->max_window
>> 1);
526 cutoff
= tp
->max_window
;
528 if (cutoff
&& pktsize
> cutoff
)
529 return max_t(int, cutoff
, 68U - tp
->tcp_header_len
);
535 extern void tcp_get_info(const struct sock
*, struct tcp_info
*);
537 /* Read 'sendfile()'-style from a TCP socket */
538 typedef int (*sk_read_actor_t
)(read_descriptor_t
*, struct sk_buff
*,
539 unsigned int, size_t);
540 extern int tcp_read_sock(struct sock
*sk
, read_descriptor_t
*desc
,
541 sk_read_actor_t recv_actor
);
543 extern void tcp_initialize_rcv_mss(struct sock
*sk
);
545 extern int tcp_mtu_to_mss(struct sock
*sk
, int pmtu
);
546 extern int tcp_mss_to_mtu(struct sock
*sk
, int mss
);
547 extern void tcp_mtup_init(struct sock
*sk
);
548 extern void tcp_valid_rtt_meas(struct sock
*sk
, u32 seq_rtt
);
550 static inline void tcp_bound_rto(const struct sock
*sk
)
552 if (inet_csk(sk
)->icsk_rto
> TCP_RTO_MAX
)
553 inet_csk(sk
)->icsk_rto
= TCP_RTO_MAX
;
556 static inline u32
__tcp_set_rto(const struct tcp_sock
*tp
)
558 return (tp
->srtt
>> 3) + tp
->rttvar
;
561 extern void tcp_set_rto(struct sock
*sk
);
563 static inline void __tcp_fast_path_on(struct tcp_sock
*tp
, u32 snd_wnd
)
565 tp
->pred_flags
= htonl((tp
->tcp_header_len
<< 26) |
566 ntohl(TCP_FLAG_ACK
) |
570 static inline void tcp_fast_path_on(struct tcp_sock
*tp
)
572 __tcp_fast_path_on(tp
, tp
->snd_wnd
>> tp
->rx_opt
.snd_wscale
);
575 static inline void tcp_fast_path_check(struct sock
*sk
)
577 struct tcp_sock
*tp
= tcp_sk(sk
);
579 if (skb_queue_empty(&tp
->out_of_order_queue
) &&
581 atomic_read(&sk
->sk_rmem_alloc
) < sk
->sk_rcvbuf
&&
583 tcp_fast_path_on(tp
);
586 /* Compute the actual rto_min value */
587 static inline u32
tcp_rto_min(struct sock
*sk
)
589 const struct dst_entry
*dst
= __sk_dst_get(sk
);
590 u32 rto_min
= TCP_RTO_MIN
;
592 if (dst
&& dst_metric_locked(dst
, RTAX_RTO_MIN
))
593 rto_min
= dst_metric_rtt(dst
, RTAX_RTO_MIN
);
597 /* Compute the actual receive window we are currently advertising.
598 * Rcv_nxt can be after the window if our peer push more data
599 * than the offered window.
601 static inline u32
tcp_receive_window(const struct tcp_sock
*tp
)
603 s32 win
= tp
->rcv_wup
+ tp
->rcv_wnd
- tp
->rcv_nxt
;
610 /* Choose a new window, without checks for shrinking, and without
611 * scaling applied to the result. The caller does these things
612 * if necessary. This is a "raw" window selection.
614 extern u32
__tcp_select_window(struct sock
*sk
);
616 void tcp_send_window_probe(struct sock
*sk
);
618 /* TCP timestamps are only 32-bits, this causes a slight
619 * complication on 64-bit systems since we store a snapshot
620 * of jiffies in the buffer control blocks below. We decided
621 * to use only the low 32-bits of jiffies and hide the ugly
622 * casts with the following macro.
624 #define tcp_time_stamp ((__u32)(jiffies))
626 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
628 #define TCPHDR_FIN 0x01
629 #define TCPHDR_SYN 0x02
630 #define TCPHDR_RST 0x04
631 #define TCPHDR_PSH 0x08
632 #define TCPHDR_ACK 0x10
633 #define TCPHDR_URG 0x20
634 #define TCPHDR_ECE 0x40
635 #define TCPHDR_CWR 0x80
637 /* This is what the send packet queuing engine uses to pass
638 * TCP per-packet control information to the transmission code.
639 * We also store the host-order sequence numbers in here too.
640 * This is 44 bytes if IPV6 is enabled.
641 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
645 struct inet_skb_parm h4
;
646 #if IS_ENABLED(CONFIG_IPV6)
647 struct inet6_skb_parm h6
;
649 } header
; /* For incoming frames */
650 __u32 seq
; /* Starting sequence number */
651 __u32 end_seq
; /* SEQ + FIN + SYN + datalen */
652 __u32 when
; /* used to compute rtt's */
653 __u8 tcp_flags
; /* TCP header flags. (tcp[13]) */
655 __u8 sacked
; /* State flags for SACK/FACK. */
656 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
657 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
658 #define TCPCB_LOST 0x04 /* SKB is lost */
659 #define TCPCB_TAGBITS 0x07 /* All tag bits */
660 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
661 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
663 __u8 ip_dsfield
; /* IPv4 tos or IPv6 dsfield */
665 __u32 ack_seq
; /* Sequence number ACK'd */
668 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
670 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
672 * If we receive a SYN packet with these bits set, it means a network is
673 * playing bad games with TOS bits. In order to avoid possible false congestion
674 * notifications, we disable TCP ECN negociation.
677 TCP_ECN_create_request(struct request_sock
*req
, const struct sk_buff
*skb
)
679 const struct tcphdr
*th
= tcp_hdr(skb
);
681 if (sysctl_tcp_ecn
&& th
->ece
&& th
->cwr
&&
682 INET_ECN_is_not_ect(TCP_SKB_CB(skb
)->ip_dsfield
))
683 inet_rsk(req
)->ecn_ok
= 1;
686 /* Due to TSO, an SKB can be composed of multiple actual
687 * packets. To keep these tracked properly, we use this.
689 static inline int tcp_skb_pcount(const struct sk_buff
*skb
)
691 return skb_shinfo(skb
)->gso_segs
;
694 /* This is valid iff tcp_skb_pcount() > 1. */
695 static inline int tcp_skb_mss(const struct sk_buff
*skb
)
697 return skb_shinfo(skb
)->gso_size
;
700 /* Events passed to congestion control interface */
702 CA_EVENT_TX_START
, /* first transmit when no packets in flight */
703 CA_EVENT_CWND_RESTART
, /* congestion window restart */
704 CA_EVENT_COMPLETE_CWR
, /* end of congestion recovery */
705 CA_EVENT_FRTO
, /* fast recovery timeout */
706 CA_EVENT_LOSS
, /* loss timeout */
707 CA_EVENT_FAST_ACK
, /* in sequence ack */
708 CA_EVENT_SLOW_ACK
, /* other ack */
712 * Interface for adding new TCP congestion control handlers
714 #define TCP_CA_NAME_MAX 16
715 #define TCP_CA_MAX 128
716 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
718 #define TCP_CONG_NON_RESTRICTED 0x1
719 #define TCP_CONG_RTT_STAMP 0x2
721 struct tcp_congestion_ops
{
722 struct list_head list
;
725 /* initialize private data (optional) */
726 void (*init
)(struct sock
*sk
);
727 /* cleanup private data (optional) */
728 void (*release
)(struct sock
*sk
);
730 /* return slow start threshold (required) */
731 u32 (*ssthresh
)(struct sock
*sk
);
732 /* lower bound for congestion window (optional) */
733 u32 (*min_cwnd
)(const struct sock
*sk
);
734 /* do new cwnd calculation (required) */
735 void (*cong_avoid
)(struct sock
*sk
, u32 ack
, u32 in_flight
);
736 /* call before changing ca_state (optional) */
737 void (*set_state
)(struct sock
*sk
, u8 new_state
);
738 /* call when cwnd event occurs (optional) */
739 void (*cwnd_event
)(struct sock
*sk
, enum tcp_ca_event ev
);
740 /* new value of cwnd after loss (optional) */
741 u32 (*undo_cwnd
)(struct sock
*sk
);
742 /* hook for packet ack accounting (optional) */
743 void (*pkts_acked
)(struct sock
*sk
, u32 num_acked
, s32 rtt_us
);
744 /* get info for inet_diag (optional) */
745 void (*get_info
)(struct sock
*sk
, u32 ext
, struct sk_buff
*skb
);
747 char name
[TCP_CA_NAME_MAX
];
748 struct module
*owner
;
751 extern int tcp_register_congestion_control(struct tcp_congestion_ops
*type
);
752 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops
*type
);
754 extern void tcp_init_congestion_control(struct sock
*sk
);
755 extern void tcp_cleanup_congestion_control(struct sock
*sk
);
756 extern int tcp_set_default_congestion_control(const char *name
);
757 extern void tcp_get_default_congestion_control(char *name
);
758 extern void tcp_get_available_congestion_control(char *buf
, size_t len
);
759 extern void tcp_get_allowed_congestion_control(char *buf
, size_t len
);
760 extern int tcp_set_allowed_congestion_control(char *allowed
);
761 extern int tcp_set_congestion_control(struct sock
*sk
, const char *name
);
762 extern void tcp_slow_start(struct tcp_sock
*tp
);
763 extern void tcp_cong_avoid_ai(struct tcp_sock
*tp
, u32 w
);
765 extern struct tcp_congestion_ops tcp_init_congestion_ops
;
766 extern u32
tcp_reno_ssthresh(struct sock
*sk
);
767 extern void tcp_reno_cong_avoid(struct sock
*sk
, u32 ack
, u32 in_flight
);
768 extern u32
tcp_reno_min_cwnd(const struct sock
*sk
);
769 extern struct tcp_congestion_ops tcp_reno
;
771 static inline void tcp_set_ca_state(struct sock
*sk
, const u8 ca_state
)
773 struct inet_connection_sock
*icsk
= inet_csk(sk
);
775 if (icsk
->icsk_ca_ops
->set_state
)
776 icsk
->icsk_ca_ops
->set_state(sk
, ca_state
);
777 icsk
->icsk_ca_state
= ca_state
;
780 static inline void tcp_ca_event(struct sock
*sk
, const enum tcp_ca_event event
)
782 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
784 if (icsk
->icsk_ca_ops
->cwnd_event
)
785 icsk
->icsk_ca_ops
->cwnd_event(sk
, event
);
788 /* These functions determine how the current flow behaves in respect of SACK
789 * handling. SACK is negotiated with the peer, and therefore it can vary
790 * between different flows.
792 * tcp_is_sack - SACK enabled
793 * tcp_is_reno - No SACK
794 * tcp_is_fack - FACK enabled, implies SACK enabled
796 static inline int tcp_is_sack(const struct tcp_sock
*tp
)
798 return tp
->rx_opt
.sack_ok
;
801 static inline bool tcp_is_reno(const struct tcp_sock
*tp
)
803 return !tcp_is_sack(tp
);
806 static inline bool tcp_is_fack(const struct tcp_sock
*tp
)
808 return tp
->rx_opt
.sack_ok
& TCP_FACK_ENABLED
;
811 static inline void tcp_enable_fack(struct tcp_sock
*tp
)
813 tp
->rx_opt
.sack_ok
|= TCP_FACK_ENABLED
;
816 /* TCP early-retransmit (ER) is similar to but more conservative than
817 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
819 static inline void tcp_enable_early_retrans(struct tcp_sock
*tp
)
821 tp
->do_early_retrans
= sysctl_tcp_early_retrans
&&
822 !sysctl_tcp_thin_dupack
&& sysctl_tcp_reordering
== 3;
823 tp
->early_retrans_delayed
= 0;
826 static inline void tcp_disable_early_retrans(struct tcp_sock
*tp
)
828 tp
->do_early_retrans
= 0;
831 static inline unsigned int tcp_left_out(const struct tcp_sock
*tp
)
833 return tp
->sacked_out
+ tp
->lost_out
;
836 /* This determines how many packets are "in the network" to the best
837 * of our knowledge. In many cases it is conservative, but where
838 * detailed information is available from the receiver (via SACK
839 * blocks etc.) we can make more aggressive calculations.
841 * Use this for decisions involving congestion control, use just
842 * tp->packets_out to determine if the send queue is empty or not.
844 * Read this equation as:
846 * "Packets sent once on transmission queue" MINUS
847 * "Packets left network, but not honestly ACKed yet" PLUS
848 * "Packets fast retransmitted"
850 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock
*tp
)
852 return tp
->packets_out
- tcp_left_out(tp
) + tp
->retrans_out
;
855 #define TCP_INFINITE_SSTHRESH 0x7fffffff
857 static inline bool tcp_in_initial_slowstart(const struct tcp_sock
*tp
)
859 return tp
->snd_ssthresh
>= TCP_INFINITE_SSTHRESH
;
862 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
863 * The exception is rate halving phase, when cwnd is decreasing towards
866 static inline __u32
tcp_current_ssthresh(const struct sock
*sk
)
868 const struct tcp_sock
*tp
= tcp_sk(sk
);
870 if ((1 << inet_csk(sk
)->icsk_ca_state
) & (TCPF_CA_CWR
| TCPF_CA_Recovery
))
871 return tp
->snd_ssthresh
;
873 return max(tp
->snd_ssthresh
,
874 ((tp
->snd_cwnd
>> 1) +
875 (tp
->snd_cwnd
>> 2)));
878 /* Use define here intentionally to get WARN_ON location shown at the caller */
879 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
881 extern void tcp_enter_cwr(struct sock
*sk
, const int set_ssthresh
);
882 extern __u32
tcp_init_cwnd(const struct tcp_sock
*tp
, const struct dst_entry
*dst
);
884 /* The maximum number of MSS of available cwnd for which TSO defers
885 * sending if not using sysctl_tcp_tso_win_divisor.
887 static inline __u32
tcp_max_tso_deferred_mss(const struct tcp_sock
*tp
)
892 /* Slow start with delack produces 3 packets of burst, so that
893 * it is safe "de facto". This will be the default - same as
894 * the default reordering threshold - but if reordering increases,
895 * we must be able to allow cwnd to burst at least this much in order
896 * to not pull it back when holes are filled.
898 static __inline__ __u32
tcp_max_burst(const struct tcp_sock
*tp
)
900 return tp
->reordering
;
903 /* Returns end sequence number of the receiver's advertised window */
904 static inline u32
tcp_wnd_end(const struct tcp_sock
*tp
)
906 return tp
->snd_una
+ tp
->snd_wnd
;
908 extern bool tcp_is_cwnd_limited(const struct sock
*sk
, u32 in_flight
);
910 static inline void tcp_minshall_update(struct tcp_sock
*tp
, unsigned int mss
,
911 const struct sk_buff
*skb
)
914 tp
->snd_sml
= TCP_SKB_CB(skb
)->end_seq
;
917 static inline void tcp_check_probe_timer(struct sock
*sk
)
919 const struct tcp_sock
*tp
= tcp_sk(sk
);
920 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
922 if (!tp
->packets_out
&& !icsk
->icsk_pending
)
923 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_PROBE0
,
924 icsk
->icsk_rto
, TCP_RTO_MAX
);
927 static inline void tcp_init_wl(struct tcp_sock
*tp
, u32 seq
)
932 static inline void tcp_update_wl(struct tcp_sock
*tp
, u32 seq
)
938 * Calculate(/check) TCP checksum
940 static inline __sum16
tcp_v4_check(int len
, __be32 saddr
,
941 __be32 daddr
, __wsum base
)
943 return csum_tcpudp_magic(saddr
,daddr
,len
,IPPROTO_TCP
,base
);
946 static inline __sum16
__tcp_checksum_complete(struct sk_buff
*skb
)
948 return __skb_checksum_complete(skb
);
951 static inline bool tcp_checksum_complete(struct sk_buff
*skb
)
953 return !skb_csum_unnecessary(skb
) &&
954 __tcp_checksum_complete(skb
);
957 /* Prequeue for VJ style copy to user, combined with checksumming. */
959 static inline void tcp_prequeue_init(struct tcp_sock
*tp
)
961 tp
->ucopy
.task
= NULL
;
963 tp
->ucopy
.memory
= 0;
964 skb_queue_head_init(&tp
->ucopy
.prequeue
);
965 #ifdef CONFIG_NET_DMA
966 tp
->ucopy
.dma_chan
= NULL
;
967 tp
->ucopy
.wakeup
= 0;
968 tp
->ucopy
.pinned_list
= NULL
;
969 tp
->ucopy
.dma_cookie
= 0;
973 /* Packet is added to VJ-style prequeue for processing in process
974 * context, if a reader task is waiting. Apparently, this exciting
975 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
976 * failed somewhere. Latency? Burstiness? Well, at least now we will
977 * see, why it failed. 8)8) --ANK
979 * NOTE: is this not too big to inline?
981 static inline bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
983 struct tcp_sock
*tp
= tcp_sk(sk
);
985 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
988 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
989 tp
->ucopy
.memory
+= skb
->truesize
;
990 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
991 struct sk_buff
*skb1
;
993 BUG_ON(sock_owned_by_user(sk
));
995 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
996 sk_backlog_rcv(sk
, skb1
);
997 NET_INC_STATS_BH(sock_net(sk
),
998 LINUX_MIB_TCPPREQUEUEDROPPED
);
1001 tp
->ucopy
.memory
= 0;
1002 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1003 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1004 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1005 if (!inet_csk_ack_scheduled(sk
))
1006 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1007 (3 * tcp_rto_min(sk
)) / 4,
1017 static const char *statename
[]={
1018 "Unused","Established","Syn Sent","Syn Recv",
1019 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1020 "Close Wait","Last ACK","Listen","Closing"
1023 extern void tcp_set_state(struct sock
*sk
, int state
);
1025 extern void tcp_done(struct sock
*sk
);
1027 static inline void tcp_sack_reset(struct tcp_options_received
*rx_opt
)
1030 rx_opt
->num_sacks
= 0;
1033 /* Determine a window scaling and initial window to offer. */
1034 extern void tcp_select_initial_window(int __space
, __u32 mss
,
1035 __u32
*rcv_wnd
, __u32
*window_clamp
,
1036 int wscale_ok
, __u8
*rcv_wscale
,
1037 __u32 init_rcv_wnd
);
1039 static inline int tcp_win_from_space(int space
)
1041 return sysctl_tcp_adv_win_scale
<=0 ?
1042 (space
>>(-sysctl_tcp_adv_win_scale
)) :
1043 space
- (space
>>sysctl_tcp_adv_win_scale
);
1046 /* Note: caller must be prepared to deal with negative returns */
1047 static inline int tcp_space(const struct sock
*sk
)
1049 return tcp_win_from_space(sk
->sk_rcvbuf
-
1050 atomic_read(&sk
->sk_rmem_alloc
));
1053 static inline int tcp_full_space(const struct sock
*sk
)
1055 return tcp_win_from_space(sk
->sk_rcvbuf
);
1058 static inline void tcp_openreq_init(struct request_sock
*req
,
1059 struct tcp_options_received
*rx_opt
,
1060 struct sk_buff
*skb
)
1062 struct inet_request_sock
*ireq
= inet_rsk(req
);
1064 req
->rcv_wnd
= 0; /* So that tcp_send_synack() knows! */
1066 tcp_rsk(req
)->rcv_isn
= TCP_SKB_CB(skb
)->seq
;
1067 req
->mss
= rx_opt
->mss_clamp
;
1068 req
->ts_recent
= rx_opt
->saw_tstamp
? rx_opt
->rcv_tsval
: 0;
1069 ireq
->tstamp_ok
= rx_opt
->tstamp_ok
;
1070 ireq
->sack_ok
= rx_opt
->sack_ok
;
1071 ireq
->snd_wscale
= rx_opt
->snd_wscale
;
1072 ireq
->wscale_ok
= rx_opt
->wscale_ok
;
1075 ireq
->rmt_port
= tcp_hdr(skb
)->source
;
1076 ireq
->loc_port
= tcp_hdr(skb
)->dest
;
1079 extern void tcp_enter_memory_pressure(struct sock
*sk
);
1081 static inline int keepalive_intvl_when(const struct tcp_sock
*tp
)
1083 return tp
->keepalive_intvl
? : sysctl_tcp_keepalive_intvl
;
1086 static inline int keepalive_time_when(const struct tcp_sock
*tp
)
1088 return tp
->keepalive_time
? : sysctl_tcp_keepalive_time
;
1091 static inline int keepalive_probes(const struct tcp_sock
*tp
)
1093 return tp
->keepalive_probes
? : sysctl_tcp_keepalive_probes
;
1096 static inline u32
keepalive_time_elapsed(const struct tcp_sock
*tp
)
1098 const struct inet_connection_sock
*icsk
= &tp
->inet_conn
;
1100 return min_t(u32
, tcp_time_stamp
- icsk
->icsk_ack
.lrcvtime
,
1101 tcp_time_stamp
- tp
->rcv_tstamp
);
1104 static inline int tcp_fin_time(const struct sock
*sk
)
1106 int fin_timeout
= tcp_sk(sk
)->linger2
? : sysctl_tcp_fin_timeout
;
1107 const int rto
= inet_csk(sk
)->icsk_rto
;
1109 if (fin_timeout
< (rto
<< 2) - (rto
>> 1))
1110 fin_timeout
= (rto
<< 2) - (rto
>> 1);
1115 static inline bool tcp_paws_check(const struct tcp_options_received
*rx_opt
,
1118 if ((s32
)(rx_opt
->ts_recent
- rx_opt
->rcv_tsval
) <= paws_win
)
1120 if (unlikely(get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_24DAYS
))
1123 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1124 * then following tcp messages have valid values. Ignore 0 value,
1125 * or else 'negative' tsval might forbid us to accept their packets.
1127 if (!rx_opt
->ts_recent
)
1132 static inline bool tcp_paws_reject(const struct tcp_options_received
*rx_opt
,
1135 if (tcp_paws_check(rx_opt
, 0))
1138 /* RST segments are not recommended to carry timestamp,
1139 and, if they do, it is recommended to ignore PAWS because
1140 "their cleanup function should take precedence over timestamps."
1141 Certainly, it is mistake. It is necessary to understand the reasons
1142 of this constraint to relax it: if peer reboots, clock may go
1143 out-of-sync and half-open connections will not be reset.
1144 Actually, the problem would be not existing if all
1145 the implementations followed draft about maintaining clock
1146 via reboots. Linux-2.2 DOES NOT!
1148 However, we can relax time bounds for RST segments to MSL.
1150 if (rst
&& get_seconds() >= rx_opt
->ts_recent_stamp
+ TCP_PAWS_MSL
)
1155 static inline void tcp_mib_init(struct net
*net
)
1158 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOALGORITHM
, 1);
1159 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMIN
, TCP_RTO_MIN
*1000/HZ
);
1160 TCP_ADD_STATS_USER(net
, TCP_MIB_RTOMAX
, TCP_RTO_MAX
*1000/HZ
);
1161 TCP_ADD_STATS_USER(net
, TCP_MIB_MAXCONN
, -1);
1165 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock
*tp
)
1167 tp
->lost_skb_hint
= NULL
;
1168 tp
->scoreboard_skb_hint
= NULL
;
1171 static inline void tcp_clear_all_retrans_hints(struct tcp_sock
*tp
)
1173 tcp_clear_retrans_hints_partial(tp
);
1174 tp
->retransmit_skb_hint
= NULL
;
1180 union tcp_md5_addr
{
1182 #if IS_ENABLED(CONFIG_IPV6)
1187 /* - key database */
1188 struct tcp_md5sig_key
{
1189 struct hlist_node node
;
1191 u8 family
; /* AF_INET or AF_INET6 */
1192 union tcp_md5_addr addr
;
1193 u8 key
[TCP_MD5SIG_MAXKEYLEN
];
1194 struct rcu_head rcu
;
1198 struct tcp_md5sig_info
{
1199 struct hlist_head head
;
1200 struct rcu_head rcu
;
1203 /* - pseudo header */
1204 struct tcp4_pseudohdr
{
1212 struct tcp6_pseudohdr
{
1213 struct in6_addr saddr
;
1214 struct in6_addr daddr
;
1216 __be32 protocol
; /* including padding */
1219 union tcp_md5sum_block
{
1220 struct tcp4_pseudohdr ip4
;
1221 #if IS_ENABLED(CONFIG_IPV6)
1222 struct tcp6_pseudohdr ip6
;
1226 /* - pool: digest algorithm, hash description and scratch buffer */
1227 struct tcp_md5sig_pool
{
1228 struct hash_desc md5_desc
;
1229 union tcp_md5sum_block md5_blk
;
1233 extern int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1234 const struct sock
*sk
,
1235 const struct request_sock
*req
,
1236 const struct sk_buff
*skb
);
1237 extern int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1238 int family
, const u8
*newkey
,
1239 u8 newkeylen
, gfp_t gfp
);
1240 extern int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1242 extern struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
1243 struct sock
*addr_sk
);
1245 #ifdef CONFIG_TCP_MD5SIG
1246 extern struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1247 const union tcp_md5_addr
*addr
, int family
);
1248 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1250 static inline struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
1251 const union tcp_md5_addr
*addr
,
1256 #define tcp_twsk_md5_key(twsk) NULL
1259 extern struct tcp_md5sig_pool __percpu
*tcp_alloc_md5sig_pool(struct sock
*);
1260 extern void tcp_free_md5sig_pool(void);
1262 extern struct tcp_md5sig_pool
*tcp_get_md5sig_pool(void);
1263 extern void tcp_put_md5sig_pool(void);
1265 extern int tcp_md5_hash_header(struct tcp_md5sig_pool
*, const struct tcphdr
*);
1266 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool
*, const struct sk_buff
*,
1267 unsigned int header_len
);
1268 extern int tcp_md5_hash_key(struct tcp_md5sig_pool
*hp
,
1269 const struct tcp_md5sig_key
*key
);
1271 /* write queue abstraction */
1272 static inline void tcp_write_queue_purge(struct sock
*sk
)
1274 struct sk_buff
*skb
;
1276 while ((skb
= __skb_dequeue(&sk
->sk_write_queue
)) != NULL
)
1277 sk_wmem_free_skb(sk
, skb
);
1279 tcp_clear_all_retrans_hints(tcp_sk(sk
));
1282 static inline struct sk_buff
*tcp_write_queue_head(const struct sock
*sk
)
1284 return skb_peek(&sk
->sk_write_queue
);
1287 static inline struct sk_buff
*tcp_write_queue_tail(const struct sock
*sk
)
1289 return skb_peek_tail(&sk
->sk_write_queue
);
1292 static inline struct sk_buff
*tcp_write_queue_next(const struct sock
*sk
,
1293 const struct sk_buff
*skb
)
1295 return skb_queue_next(&sk
->sk_write_queue
, skb
);
1298 static inline struct sk_buff
*tcp_write_queue_prev(const struct sock
*sk
,
1299 const struct sk_buff
*skb
)
1301 return skb_queue_prev(&sk
->sk_write_queue
, skb
);
1304 #define tcp_for_write_queue(skb, sk) \
1305 skb_queue_walk(&(sk)->sk_write_queue, skb)
1307 #define tcp_for_write_queue_from(skb, sk) \
1308 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1310 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1311 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1313 static inline struct sk_buff
*tcp_send_head(const struct sock
*sk
)
1315 return sk
->sk_send_head
;
1318 static inline bool tcp_skb_is_last(const struct sock
*sk
,
1319 const struct sk_buff
*skb
)
1321 return skb_queue_is_last(&sk
->sk_write_queue
, skb
);
1324 static inline void tcp_advance_send_head(struct sock
*sk
, const struct sk_buff
*skb
)
1326 if (tcp_skb_is_last(sk
, skb
))
1327 sk
->sk_send_head
= NULL
;
1329 sk
->sk_send_head
= tcp_write_queue_next(sk
, skb
);
1332 static inline void tcp_check_send_head(struct sock
*sk
, struct sk_buff
*skb_unlinked
)
1334 if (sk
->sk_send_head
== skb_unlinked
)
1335 sk
->sk_send_head
= NULL
;
1338 static inline void tcp_init_send_head(struct sock
*sk
)
1340 sk
->sk_send_head
= NULL
;
1343 static inline void __tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1345 __skb_queue_tail(&sk
->sk_write_queue
, skb
);
1348 static inline void tcp_add_write_queue_tail(struct sock
*sk
, struct sk_buff
*skb
)
1350 __tcp_add_write_queue_tail(sk
, skb
);
1352 /* Queue it, remembering where we must start sending. */
1353 if (sk
->sk_send_head
== NULL
) {
1354 sk
->sk_send_head
= skb
;
1356 if (tcp_sk(sk
)->highest_sack
== NULL
)
1357 tcp_sk(sk
)->highest_sack
= skb
;
1361 static inline void __tcp_add_write_queue_head(struct sock
*sk
, struct sk_buff
*skb
)
1363 __skb_queue_head(&sk
->sk_write_queue
, skb
);
1366 /* Insert buff after skb on the write queue of sk. */
1367 static inline void tcp_insert_write_queue_after(struct sk_buff
*skb
,
1368 struct sk_buff
*buff
,
1371 __skb_queue_after(&sk
->sk_write_queue
, skb
, buff
);
1374 /* Insert new before skb on the write queue of sk. */
1375 static inline void tcp_insert_write_queue_before(struct sk_buff
*new,
1376 struct sk_buff
*skb
,
1379 __skb_queue_before(&sk
->sk_write_queue
, skb
, new);
1381 if (sk
->sk_send_head
== skb
)
1382 sk
->sk_send_head
= new;
1385 static inline void tcp_unlink_write_queue(struct sk_buff
*skb
, struct sock
*sk
)
1387 __skb_unlink(skb
, &sk
->sk_write_queue
);
1390 static inline bool tcp_write_queue_empty(struct sock
*sk
)
1392 return skb_queue_empty(&sk
->sk_write_queue
);
1395 static inline void tcp_push_pending_frames(struct sock
*sk
)
1397 if (tcp_send_head(sk
)) {
1398 struct tcp_sock
*tp
= tcp_sk(sk
);
1400 __tcp_push_pending_frames(sk
, tcp_current_mss(sk
), tp
->nonagle
);
1404 /* Start sequence of the skb just after the highest skb with SACKed
1405 * bit, valid only if sacked_out > 0 or when the caller has ensured
1406 * validity by itself.
1408 static inline u32
tcp_highest_sack_seq(struct tcp_sock
*tp
)
1410 if (!tp
->sacked_out
)
1413 if (tp
->highest_sack
== NULL
)
1416 return TCP_SKB_CB(tp
->highest_sack
)->seq
;
1419 static inline void tcp_advance_highest_sack(struct sock
*sk
, struct sk_buff
*skb
)
1421 tcp_sk(sk
)->highest_sack
= tcp_skb_is_last(sk
, skb
) ? NULL
:
1422 tcp_write_queue_next(sk
, skb
);
1425 static inline struct sk_buff
*tcp_highest_sack(struct sock
*sk
)
1427 return tcp_sk(sk
)->highest_sack
;
1430 static inline void tcp_highest_sack_reset(struct sock
*sk
)
1432 tcp_sk(sk
)->highest_sack
= tcp_write_queue_head(sk
);
1435 /* Called when old skb is about to be deleted (to be combined with new skb) */
1436 static inline void tcp_highest_sack_combine(struct sock
*sk
,
1437 struct sk_buff
*old
,
1438 struct sk_buff
*new)
1440 if (tcp_sk(sk
)->sacked_out
&& (old
== tcp_sk(sk
)->highest_sack
))
1441 tcp_sk(sk
)->highest_sack
= new;
1444 /* Determines whether this is a thin stream (which may suffer from
1445 * increased latency). Used to trigger latency-reducing mechanisms.
1447 static inline bool tcp_stream_is_thin(struct tcp_sock
*tp
)
1449 return tp
->packets_out
< 4 && !tcp_in_initial_slowstart(tp
);
1453 enum tcp_seq_states
{
1454 TCP_SEQ_STATE_LISTENING
,
1455 TCP_SEQ_STATE_OPENREQ
,
1456 TCP_SEQ_STATE_ESTABLISHED
,
1457 TCP_SEQ_STATE_TIME_WAIT
,
1460 int tcp_seq_open(struct inode
*inode
, struct file
*file
);
1462 struct tcp_seq_afinfo
{
1465 const struct file_operations
*seq_fops
;
1466 struct seq_operations seq_ops
;
1469 struct tcp_iter_state
{
1470 struct seq_net_private p
;
1472 enum tcp_seq_states state
;
1473 struct sock
*syn_wait_sk
;
1474 int bucket
, offset
, sbucket
, num
, uid
;
1478 extern int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1479 extern void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
);
1481 extern struct request_sock_ops tcp_request_sock_ops
;
1482 extern struct request_sock_ops tcp6_request_sock_ops
;
1484 extern void tcp_v4_destroy_sock(struct sock
*sk
);
1486 extern int tcp_v4_gso_send_check(struct sk_buff
*skb
);
1487 extern struct sk_buff
*tcp_tso_segment(struct sk_buff
*skb
,
1488 netdev_features_t features
);
1489 extern struct sk_buff
**tcp_gro_receive(struct sk_buff
**head
,
1490 struct sk_buff
*skb
);
1491 extern struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
,
1492 struct sk_buff
*skb
);
1493 extern int tcp_gro_complete(struct sk_buff
*skb
);
1494 extern int tcp4_gro_complete(struct sk_buff
*skb
);
1496 #ifdef CONFIG_PROC_FS
1497 extern int tcp4_proc_init(void);
1498 extern void tcp4_proc_exit(void);
1501 /* TCP af-specific functions */
1502 struct tcp_sock_af_ops
{
1503 #ifdef CONFIG_TCP_MD5SIG
1504 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1505 struct sock
*addr_sk
);
1506 int (*calc_md5_hash
) (char *location
,
1507 struct tcp_md5sig_key
*md5
,
1508 const struct sock
*sk
,
1509 const struct request_sock
*req
,
1510 const struct sk_buff
*skb
);
1511 int (*md5_parse
) (struct sock
*sk
,
1512 char __user
*optval
,
1517 struct tcp_request_sock_ops
{
1518 #ifdef CONFIG_TCP_MD5SIG
1519 struct tcp_md5sig_key
*(*md5_lookup
) (struct sock
*sk
,
1520 struct request_sock
*req
);
1521 int (*calc_md5_hash
) (char *location
,
1522 struct tcp_md5sig_key
*md5
,
1523 const struct sock
*sk
,
1524 const struct request_sock
*req
,
1525 const struct sk_buff
*skb
);
1529 /* Using SHA1 for now, define some constants.
1531 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1532 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1533 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1535 extern int tcp_cookie_generator(u32
*bakery
);
1538 * struct tcp_cookie_values - each socket needs extra space for the
1539 * cookies, together with (optional) space for any SYN data.
1541 * A tcp_sock contains a pointer to the current value, and this is
1542 * cloned to the tcp_timewait_sock.
1544 * @cookie_pair: variable data from the option exchange.
1546 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1547 * indicates default (sysctl_tcp_cookie_size).
1548 * After cookie sent, remembers size of cookie.
1549 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1551 * @s_data_desired: user specified tcpct_s_data_desired. When the
1552 * constant payload is specified (@s_data_constant),
1553 * holds its length instead.
1554 * Range 0 to TCP_MSS_DESIRED.
1556 * @s_data_payload: constant data that is to be included in the
1557 * payload of SYN or SYNACK segments when the
1558 * cookie option is present.
1560 struct tcp_cookie_values
{
1562 u8 cookie_pair
[TCP_COOKIE_PAIR_SIZE
];
1563 u8 cookie_pair_size
;
1565 u16 s_data_desired
:11,
1570 u8 s_data_payload
[0];
1573 static inline void tcp_cookie_values_release(struct kref
*kref
)
1575 kfree(container_of(kref
, struct tcp_cookie_values
, kref
));
1578 /* The length of constant payload data. Note that s_data_desired is
1579 * overloaded, depending on s_data_constant: either the length of constant
1580 * data (returned here) or the limit on variable data.
1582 static inline int tcp_s_data_size(const struct tcp_sock
*tp
)
1584 return (tp
->cookie_values
!= NULL
&& tp
->cookie_values
->s_data_constant
)
1585 ? tp
->cookie_values
->s_data_desired
1590 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1592 * As tcp_request_sock has already been extended in other places, the
1593 * only remaining method is to pass stack values along as function
1594 * parameters. These parameters are not needed after sending SYNACK.
1596 * @cookie_bakery: cryptographic secret and message workspace.
1598 * @cookie_plus: bytes in authenticator/cookie option, copied from
1599 * struct tcp_options_received (above).
1601 struct tcp_extend_values
{
1602 struct request_values rv
;
1603 u32 cookie_bakery
[COOKIE_WORKSPACE_WORDS
];
1609 static inline struct tcp_extend_values
*tcp_xv(struct request_values
*rvp
)
1611 return (struct tcp_extend_values
*)rvp
;
1614 extern void tcp_v4_init(void);
1615 extern void tcp_init(void);