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/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 extern struct percpu_counter tcp_orphan_count;
52 void tcp_time_wait(struct sock *sk, int state, int timeo);
54 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
56 #define TCP_MIN_SND_MSS 48
57 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
63 #define MAX_TCP_WINDOW 32767U
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS 88U
68 /* The least MTU to use for probing */
69 #define TCP_BASE_MSS 1024
71 /* probing interval, default to 10 minutes as per RFC4821 */
72 #define TCP_PROBE_INTERVAL 600
74 /* Specify interval when tcp mtu probing will stop */
75 #define TCP_PROBE_THRESHOLD 8
77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
78 #define TCP_FASTRETRANS_THRESH 3
80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
81 #define TCP_MAX_QUICKACKS 16U
84 #define TCP_URG_VALID 0x0100
85 #define TCP_URG_NOTYET 0x0200
86 #define TCP_URG_READ 0x0400
88 #define TCP_RETR1 3 /*
89 * This is how many retries it does before it
90 * tries to figure out if the gateway is
91 * down. Minimal RFC value is 3; it corresponds
92 * to ~3sec-8min depending on RTO.
95 #define TCP_RETR2 15 /*
96 * This should take at least
97 * 90 minutes to time out.
98 * RFC1122 says that the limit is 100 sec.
99 * 15 is ~13-30min depending on RTO.
102 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
103 * when active opening a connection.
104 * RFC1122 says the minimum retry MUST
105 * be at least 180secs. Nevertheless
106 * this value is corresponding to
107 * 63secs of retransmission with the
108 * current initial RTO.
111 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
112 * when passive opening a connection.
113 * This is corresponding to 31secs of
114 * retransmission with the current
118 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
119 * state, about 60 seconds */
120 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
121 /* BSD style FIN_WAIT2 deadlock breaker.
122 * It used to be 3min, new value is 60sec,
123 * to combine FIN-WAIT-2 timeout with
127 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
129 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
130 #define TCP_ATO_MIN ((unsigned)(HZ/25))
132 #define TCP_DELACK_MIN 4U
133 #define TCP_ATO_MIN 4U
135 #define TCP_RTO_MAX ((unsigned)(120*HZ))
136 #define TCP_RTO_MIN ((unsigned)(HZ/5))
137 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
138 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
139 * used as a fallback RTO for the
140 * initial data transmission if no
141 * valid RTT sample has been acquired,
142 * most likely due to retrans in 3WHS.
145 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
146 * for local resources.
149 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
150 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
151 #define TCP_KEEPALIVE_INTVL (75*HZ)
153 #define MAX_TCP_KEEPIDLE 32767
154 #define MAX_TCP_KEEPINTVL 32767
155 #define MAX_TCP_KEEPCNT 127
156 #define MAX_TCP_SYNCNT 127
158 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
160 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
161 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
162 * after this time. It should be equal
163 * (or greater than) TCP_TIMEWAIT_LEN
164 * to provide reliability equal to one
165 * provided by timewait state.
167 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
168 * timestamps. It must be less than
169 * minimal timewait lifetime.
175 #define TCPOPT_NOP 1 /* Padding */
176 #define TCPOPT_EOL 0 /* End of options */
177 #define TCPOPT_MSS 2 /* Segment size negotiating */
178 #define TCPOPT_WINDOW 3 /* Window scaling */
179 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
180 #define TCPOPT_SACK 5 /* SACK Block */
181 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
182 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
183 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
184 #define TCPOPT_EXP 254 /* Experimental */
185 /* Magic number to be after the option value for sharing TCP
186 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
188 #define TCPOPT_FASTOPEN_MAGIC 0xF989
194 #define TCPOLEN_MSS 4
195 #define TCPOLEN_WINDOW 3
196 #define TCPOLEN_SACK_PERM 2
197 #define TCPOLEN_TIMESTAMP 10
198 #define TCPOLEN_MD5SIG 18
199 #define TCPOLEN_FASTOPEN_BASE 2
200 #define TCPOLEN_EXP_FASTOPEN_BASE 4
202 /* But this is what stacks really send out. */
203 #define TCPOLEN_TSTAMP_ALIGNED 12
204 #define TCPOLEN_WSCALE_ALIGNED 4
205 #define TCPOLEN_SACKPERM_ALIGNED 4
206 #define TCPOLEN_SACK_BASE 2
207 #define TCPOLEN_SACK_BASE_ALIGNED 4
208 #define TCPOLEN_SACK_PERBLOCK 8
209 #define TCPOLEN_MD5SIG_ALIGNED 20
210 #define TCPOLEN_MSS_ALIGNED 4
212 /* Flags in tp->nonagle */
213 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
214 #define TCP_NAGLE_CORK 2 /* Socket is corked */
215 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
217 /* TCP thin-stream limits */
218 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
220 /* TCP initial congestion window as per rfc6928 */
221 #define TCP_INIT_CWND 10
223 /* Bit Flags for sysctl_tcp_fastopen */
224 #define TFO_CLIENT_ENABLE 1
225 #define TFO_SERVER_ENABLE 2
226 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
228 /* Accept SYN data w/o any cookie option */
229 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
231 /* Force enable TFO on all listeners, i.e., not requiring the
232 * TCP_FASTOPEN socket option.
234 #define TFO_SERVER_WO_SOCKOPT1 0x400
236 extern struct inet_timewait_death_row tcp_death_row;
238 /* sysctl variables for tcp */
239 extern int sysctl_tcp_timestamps;
240 extern int sysctl_tcp_window_scaling;
241 extern int sysctl_tcp_sack;
242 extern int sysctl_tcp_fastopen;
243 extern int sysctl_tcp_retrans_collapse;
244 extern int sysctl_tcp_stdurg;
245 extern int sysctl_tcp_rfc1337;
246 extern int sysctl_tcp_abort_on_overflow;
247 extern int sysctl_tcp_max_orphans;
248 extern int sysctl_tcp_fack;
249 extern int sysctl_tcp_reordering;
250 extern int sysctl_tcp_max_reordering;
251 extern int sysctl_tcp_dsack;
252 extern long sysctl_tcp_mem[3];
253 extern int sysctl_tcp_wmem[3];
254 extern int sysctl_tcp_rmem[3];
255 extern int sysctl_tcp_app_win;
256 extern int sysctl_tcp_adv_win_scale;
257 extern int sysctl_tcp_tw_reuse;
258 extern int sysctl_tcp_frto;
259 extern int sysctl_tcp_low_latency;
260 extern int sysctl_tcp_nometrics_save;
261 extern int sysctl_tcp_moderate_rcvbuf;
262 extern int sysctl_tcp_tso_win_divisor;
263 extern int sysctl_tcp_workaround_signed_windows;
264 extern int sysctl_tcp_slow_start_after_idle;
265 extern int sysctl_tcp_thin_linear_timeouts;
266 extern int sysctl_tcp_thin_dupack;
267 extern int sysctl_tcp_early_retrans;
268 extern int sysctl_tcp_limit_output_bytes;
269 extern int sysctl_tcp_challenge_ack_limit;
270 extern int sysctl_tcp_min_tso_segs;
271 extern int sysctl_tcp_min_rtt_wlen;
272 extern int sysctl_tcp_autocorking;
273 extern int sysctl_tcp_invalid_ratelimit;
274 extern int sysctl_tcp_pacing_ss_ratio;
275 extern int sysctl_tcp_pacing_ca_ratio;
277 extern atomic_long_t tcp_memory_allocated;
278 extern struct percpu_counter tcp_sockets_allocated;
279 extern int tcp_memory_pressure;
281 /* optimized version of sk_under_memory_pressure() for TCP sockets */
282 static inline bool tcp_under_memory_pressure(const struct sock *sk)
284 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
285 mem_cgroup_under_socket_pressure(sk->sk_memcg))
288 return tcp_memory_pressure;
291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
295 static inline bool before(__u32 seq1, __u32 seq2)
297 return (__s32)(seq1-seq2) < 0;
299 #define after(seq2, seq1) before(seq1, seq2)
301 /* is s2<=s1<=s3 ? */
302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
304 return seq3 - seq2 >= seq1 - seq2;
307 static inline bool tcp_out_of_memory(struct sock *sk)
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
315 void sk_forced_mem_schedule(struct sock *sk, int size);
317 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
319 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
320 int orphans = percpu_counter_read_positive(ocp);
322 if (orphans << shift > sysctl_tcp_max_orphans) {
323 orphans = percpu_counter_sum_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans)
330 bool tcp_check_oom(struct sock *sk, int shift);
333 extern struct proto tcp_prot;
335 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
336 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
337 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
338 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
340 void tcp_tasklet_init(void);
342 void tcp_v4_err(struct sk_buff *skb, u32);
344 void tcp_shutdown(struct sock *sk, int how);
346 void tcp_v4_early_demux(struct sk_buff *skb);
347 int tcp_v4_rcv(struct sk_buff *skb);
349 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
350 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
351 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
353 void tcp_release_cb(struct sock *sk);
354 void tcp_wfree(struct sk_buff *skb);
355 void tcp_write_timer_handler(struct sock *sk);
356 void tcp_delack_timer_handler(struct sock *sk);
357 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
358 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
359 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
360 const struct tcphdr *th, unsigned int len);
361 void tcp_rcv_space_adjust(struct sock *sk);
362 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
363 void tcp_twsk_destructor(struct sock *sk);
364 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
365 struct pipe_inode_info *pipe, size_t len,
368 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
369 static inline void tcp_dec_quickack_mode(struct sock *sk,
370 const unsigned int pkts)
372 struct inet_connection_sock *icsk = inet_csk(sk);
374 if (icsk->icsk_ack.quick) {
375 if (pkts >= icsk->icsk_ack.quick) {
376 icsk->icsk_ack.quick = 0;
377 /* Leaving quickack mode we deflate ATO. */
378 icsk->icsk_ack.ato = TCP_ATO_MIN;
380 icsk->icsk_ack.quick -= pkts;
385 #define TCP_ECN_QUEUE_CWR 2
386 #define TCP_ECN_DEMAND_CWR 4
387 #define TCP_ECN_SEEN 8
397 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
399 const struct tcphdr *th);
400 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
401 struct request_sock *req, bool fastopen);
402 int tcp_child_process(struct sock *parent, struct sock *child,
403 struct sk_buff *skb);
404 void tcp_enter_loss(struct sock *sk);
405 void tcp_clear_retrans(struct tcp_sock *tp);
406 void tcp_update_metrics(struct sock *sk);
407 void tcp_init_metrics(struct sock *sk);
408 void tcp_metrics_init(void);
409 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
410 bool paws_check, bool timestamps);
411 bool tcp_remember_stamp(struct sock *sk);
412 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
413 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
414 void tcp_disable_fack(struct tcp_sock *tp);
415 void tcp_close(struct sock *sk, long timeout);
416 void tcp_init_sock(struct sock *sk);
417 unsigned int tcp_poll(struct file *file, struct socket *sock,
418 struct poll_table_struct *wait);
419 int tcp_getsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, int __user *optlen);
421 int tcp_setsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, unsigned int optlen);
423 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
424 char __user *optval, int __user *optlen);
425 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
426 char __user *optval, unsigned int optlen);
427 void tcp_set_keepalive(struct sock *sk, int val);
428 void tcp_syn_ack_timeout(const struct request_sock *req);
429 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
430 int flags, int *addr_len);
431 void tcp_parse_options(const struct sk_buff *skb,
432 struct tcp_options_received *opt_rx,
433 int estab, struct tcp_fastopen_cookie *foc);
434 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
437 * TCP v4 functions exported for the inet6 API
440 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
441 void tcp_v4_mtu_reduced(struct sock *sk);
442 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
443 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
444 struct sock *tcp_create_openreq_child(const struct sock *sk,
445 struct request_sock *req,
446 struct sk_buff *skb);
447 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
448 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
449 struct request_sock *req,
450 struct dst_entry *dst,
451 struct request_sock *req_unhash,
453 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455 int tcp_connect(struct sock *sk);
456 enum tcp_synack_type {
461 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
462 struct request_sock *req,
463 struct tcp_fastopen_cookie *foc,
464 enum tcp_synack_type synack_type);
465 int tcp_disconnect(struct sock *sk, int flags);
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
471 /* From syncookies.c */
472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
478 #ifdef CONFIG_SYN_COOKIES
480 /* Syncookies use a monotonic timer which increments every 60 seconds.
481 * This counter is used both as a hash input and partially encoded into
482 * the cookie value. A cookie is only validated further if the delta
483 * between the current counter value and the encoded one is less than this,
484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
485 * the counter advances immediately after a cookie is generated).
487 #define MAX_SYNCOOKIE_AGE 2
488 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
489 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
491 /* syncookies: remember time of last synqueue overflow
492 * But do not dirty this field too often (once per second is enough)
493 * It is racy as we do not hold a lock, but race is very minor.
495 static inline void tcp_synq_overflow(const struct sock *sk)
497 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
498 unsigned long now = jiffies;
500 if (!time_between32(now, last_overflow, last_overflow + HZ))
501 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
504 /* syncookies: no recent synqueue overflow on this listening socket? */
505 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
507 unsigned long last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
509 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
510 * then we're under synflood. However, we have to use
511 * 'last_overflow - HZ' as lower bound. That's because a concurrent
512 * tcp_synq_overflow() could update .ts_recent_stamp after we read
513 * jiffies but before we store .ts_recent_stamp into last_overflow,
514 * which could lead to rejecting a valid syncookie.
516 return !time_between32(jiffies, last_overflow - HZ,
517 last_overflow + TCP_SYNCOOKIE_VALID);
520 static inline u32 tcp_cookie_time(void)
522 u64 val = get_jiffies_64();
524 do_div(val, TCP_SYNCOOKIE_PERIOD);
528 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
530 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
531 __u32 cookie_init_timestamp(struct request_sock *req);
532 bool cookie_timestamp_decode(struct tcp_options_received *opt);
533 bool cookie_ecn_ok(const struct tcp_options_received *opt,
534 const struct net *net, const struct dst_entry *dst);
536 /* From net/ipv6/syncookies.c */
537 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
539 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
541 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
542 const struct tcphdr *th, u16 *mssp);
543 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
547 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
549 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
551 bool tcp_may_send_now(struct sock *sk);
552 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
553 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
554 void tcp_retransmit_timer(struct sock *sk);
555 void tcp_xmit_retransmit_queue(struct sock *);
556 void tcp_simple_retransmit(struct sock *);
557 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
558 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
560 void tcp_send_probe0(struct sock *);
561 void tcp_send_partial(struct sock *);
562 int tcp_write_wakeup(struct sock *, int mib);
563 void tcp_send_fin(struct sock *sk);
564 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
565 int tcp_send_synack(struct sock *);
566 void tcp_push_one(struct sock *, unsigned int mss_now);
567 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
568 void tcp_send_ack(struct sock *sk);
569 void tcp_send_delayed_ack(struct sock *sk);
570 void tcp_send_loss_probe(struct sock *sk);
571 bool tcp_schedule_loss_probe(struct sock *sk);
572 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
573 const struct sk_buff *next_skb);
576 void tcp_resume_early_retransmit(struct sock *sk);
577 void tcp_rearm_rto(struct sock *sk);
578 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
579 void tcp_reset(struct sock *sk);
580 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
581 void tcp_fin(struct sock *sk);
582 void tcp_check_space(struct sock *sk);
585 void tcp_init_xmit_timers(struct sock *);
586 static inline void tcp_clear_xmit_timers(struct sock *sk)
588 inet_csk_clear_xmit_timers(sk);
591 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
592 unsigned int tcp_current_mss(struct sock *sk);
594 /* Bound MSS / TSO packet size with the half of the window */
595 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
599 /* When peer uses tiny windows, there is no use in packetizing
600 * to sub-MSS pieces for the sake of SWS or making sure there
601 * are enough packets in the pipe for fast recovery.
603 * On the other hand, for extremely large MSS devices, handling
604 * smaller than MSS windows in this way does make sense.
606 if (tp->max_window > TCP_MSS_DEFAULT)
607 cutoff = (tp->max_window >> 1);
609 cutoff = tp->max_window;
611 if (cutoff && pktsize > cutoff)
612 return max_t(int, cutoff, 68U - tp->tcp_header_len);
618 void tcp_get_info(struct sock *, struct tcp_info *);
620 /* Read 'sendfile()'-style from a TCP socket */
621 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
622 sk_read_actor_t recv_actor);
624 void tcp_initialize_rcv_mss(struct sock *sk);
626 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
627 int tcp_mss_to_mtu(struct sock *sk, int mss);
628 void tcp_mtup_init(struct sock *sk);
629 void tcp_init_buffer_space(struct sock *sk);
631 static inline void tcp_bound_rto(const struct sock *sk)
633 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
634 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
637 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
639 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
642 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
644 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
645 ntohl(TCP_FLAG_ACK) |
649 static inline void tcp_fast_path_on(struct tcp_sock *tp)
651 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
654 static inline void tcp_fast_path_check(struct sock *sk)
656 struct tcp_sock *tp = tcp_sk(sk);
658 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
660 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
662 tcp_fast_path_on(tp);
665 /* Compute the actual rto_min value */
666 static inline u32 tcp_rto_min(struct sock *sk)
668 const struct dst_entry *dst = __sk_dst_get(sk);
669 u32 rto_min = TCP_RTO_MIN;
671 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
672 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
676 static inline u32 tcp_rto_min_us(struct sock *sk)
678 return jiffies_to_usecs(tcp_rto_min(sk));
681 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
683 return dst_metric_locked(dst, RTAX_CC_ALGO);
686 /* Minimum RTT in usec. ~0 means not available. */
687 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
689 return minmax_get(&tp->rtt_min);
692 /* Compute the actual receive window we are currently advertising.
693 * Rcv_nxt can be after the window if our peer push more data
694 * than the offered window.
696 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
698 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
705 /* Choose a new window, without checks for shrinking, and without
706 * scaling applied to the result. The caller does these things
707 * if necessary. This is a "raw" window selection.
709 u32 __tcp_select_window(struct sock *sk);
711 void tcp_send_window_probe(struct sock *sk);
713 /* TCP timestamps are only 32-bits, this causes a slight
714 * complication on 64-bit systems since we store a snapshot
715 * of jiffies in the buffer control blocks below. We decided
716 * to use only the low 32-bits of jiffies and hide the ugly
717 * casts with the following macro.
719 #define tcp_time_stamp ((__u32)(jiffies))
721 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
723 return skb->skb_mstamp.stamp_jiffies;
727 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
729 #define TCPHDR_FIN 0x01
730 #define TCPHDR_SYN 0x02
731 #define TCPHDR_RST 0x04
732 #define TCPHDR_PSH 0x08
733 #define TCPHDR_ACK 0x10
734 #define TCPHDR_URG 0x20
735 #define TCPHDR_ECE 0x40
736 #define TCPHDR_CWR 0x80
738 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
740 /* This is what the send packet queuing engine uses to pass
741 * TCP per-packet control information to the transmission code.
742 * We also store the host-order sequence numbers in here too.
743 * This is 44 bytes if IPV6 is enabled.
744 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
747 __u32 seq; /* Starting sequence number */
748 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
750 /* Note : tcp_tw_isn is used in input path only
751 * (isn chosen by tcp_timewait_state_process())
753 * tcp_gso_segs/size are used in write queue only,
754 * cf tcp_skb_pcount()/tcp_skb_mss()
762 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
764 __u8 sacked; /* State flags for SACK/FACK. */
765 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
766 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
767 #define TCPCB_LOST 0x04 /* SKB is lost */
768 #define TCPCB_TAGBITS 0x07 /* All tag bits */
769 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
770 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
771 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
774 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
775 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
776 eor:1, /* Is skb MSG_EOR marked? */
778 __u32 ack_seq; /* Sequence number ACK'd */
781 /* There is space for up to 24 bytes */
782 __u32 in_flight:30,/* Bytes in flight at transmit */
783 is_app_limited:1, /* cwnd not fully used? */
785 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
787 /* start of send pipeline phase */
788 struct skb_mstamp first_tx_mstamp;
789 /* when we reached the "delivered" count */
790 struct skb_mstamp delivered_mstamp;
791 } tx; /* only used for outgoing skbs */
793 struct inet_skb_parm h4;
794 #if IS_ENABLED(CONFIG_IPV6)
795 struct inet6_skb_parm h6;
797 } header; /* For incoming skbs */
801 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
804 #if IS_ENABLED(CONFIG_IPV6)
805 /* This is the variant of inet6_iif() that must be used by TCP,
806 * as TCP moves IP6CB into a different location in skb->cb[]
808 static inline int tcp_v6_iif(const struct sk_buff *skb)
810 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
812 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
816 /* TCP_SKB_CB reference means this can not be used from early demux */
817 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
819 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
820 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
821 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
827 /* Due to TSO, an SKB can be composed of multiple actual
828 * packets. To keep these tracked properly, we use this.
830 static inline int tcp_skb_pcount(const struct sk_buff *skb)
832 return TCP_SKB_CB(skb)->tcp_gso_segs;
835 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
837 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
840 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
842 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
845 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
846 static inline int tcp_skb_mss(const struct sk_buff *skb)
848 return TCP_SKB_CB(skb)->tcp_gso_size;
851 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
853 return likely(!TCP_SKB_CB(skb)->eor);
856 /* Events passed to congestion control interface */
858 CA_EVENT_TX_START, /* first transmit when no packets in flight */
859 CA_EVENT_CWND_RESTART, /* congestion window restart */
860 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
861 CA_EVENT_LOSS, /* loss timeout */
862 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
863 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
866 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
867 enum tcp_ca_ack_event_flags {
868 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
869 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
870 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
874 * Interface for adding new TCP congestion control handlers
876 #define TCP_CA_NAME_MAX 16
877 #define TCP_CA_MAX 128
878 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
880 #define TCP_CA_UNSPEC 0
882 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
883 #define TCP_CONG_NON_RESTRICTED 0x1
884 /* Requires ECN/ECT set on all packets */
885 #define TCP_CONG_NEEDS_ECN 0x2
895 /* A rate sample measures the number of (original/retransmitted) data
896 * packets delivered "delivered" over an interval of time "interval_us".
897 * The tcp_rate.c code fills in the rate sample, and congestion
898 * control modules that define a cong_control function to run at the end
899 * of ACK processing can optionally chose to consult this sample when
900 * setting cwnd and pacing rate.
901 * A sample is invalid if "delivered" or "interval_us" is negative.
904 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
905 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
906 s32 delivered; /* number of packets delivered over interval */
907 long interval_us; /* time for tp->delivered to incr "delivered" */
908 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
909 int losses; /* number of packets marked lost upon ACK */
910 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
911 u32 prior_in_flight; /* in flight before this ACK */
912 bool is_app_limited; /* is sample from packet with bubble in pipe? */
913 bool is_retrans; /* is sample from retransmission? */
916 struct tcp_congestion_ops {
917 struct list_head list;
921 /* initialize private data (optional) */
922 void (*init)(struct sock *sk);
923 /* cleanup private data (optional) */
924 void (*release)(struct sock *sk);
926 /* return slow start threshold (required) */
927 u32 (*ssthresh)(struct sock *sk);
928 /* do new cwnd calculation (required) */
929 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
930 /* call before changing ca_state (optional) */
931 void (*set_state)(struct sock *sk, u8 new_state);
932 /* call when cwnd event occurs (optional) */
933 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
934 /* call when ack arrives (optional) */
935 void (*in_ack_event)(struct sock *sk, u32 flags);
936 /* new value of cwnd after loss (optional) */
937 u32 (*undo_cwnd)(struct sock *sk);
938 /* hook for packet ack accounting (optional) */
939 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
940 /* suggest number of segments for each skb to transmit (optional) */
941 u32 (*tso_segs_goal)(struct sock *sk);
942 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
943 u32 (*sndbuf_expand)(struct sock *sk);
944 /* call when packets are delivered to update cwnd and pacing rate,
945 * after all the ca_state processing. (optional)
947 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
948 /* get info for inet_diag (optional) */
949 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
950 union tcp_cc_info *info);
952 char name[TCP_CA_NAME_MAX];
953 struct module *owner;
956 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
957 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
959 void tcp_assign_congestion_control(struct sock *sk);
960 void tcp_init_congestion_control(struct sock *sk);
961 void tcp_cleanup_congestion_control(struct sock *sk);
962 int tcp_set_default_congestion_control(const char *name);
963 void tcp_get_default_congestion_control(char *name);
964 void tcp_get_available_congestion_control(char *buf, size_t len);
965 void tcp_get_allowed_congestion_control(char *buf, size_t len);
966 int tcp_set_allowed_congestion_control(char *allowed);
967 int tcp_set_congestion_control(struct sock *sk, const char *name);
968 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
969 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
971 u32 tcp_reno_ssthresh(struct sock *sk);
972 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
973 extern struct tcp_congestion_ops tcp_reno;
975 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
976 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
978 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
980 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
986 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
988 const struct inet_connection_sock *icsk = inet_csk(sk);
990 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
993 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
995 struct inet_connection_sock *icsk = inet_csk(sk);
997 if (icsk->icsk_ca_ops->set_state)
998 icsk->icsk_ca_ops->set_state(sk, ca_state);
999 icsk->icsk_ca_state = ca_state;
1002 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1004 const struct inet_connection_sock *icsk = inet_csk(sk);
1006 if (icsk->icsk_ca_ops->cwnd_event)
1007 icsk->icsk_ca_ops->cwnd_event(sk, event);
1010 /* From tcp_rate.c */
1011 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1012 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1013 struct rate_sample *rs);
1014 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1015 bool is_sack_reneg, struct skb_mstamp *now, struct rate_sample *rs);
1016 void tcp_rate_check_app_limited(struct sock *sk);
1018 /* These functions determine how the current flow behaves in respect of SACK
1019 * handling. SACK is negotiated with the peer, and therefore it can vary
1020 * between different flows.
1022 * tcp_is_sack - SACK enabled
1023 * tcp_is_reno - No SACK
1024 * tcp_is_fack - FACK enabled, implies SACK enabled
1026 static inline int tcp_is_sack(const struct tcp_sock *tp)
1028 return tp->rx_opt.sack_ok;
1031 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1033 return !tcp_is_sack(tp);
1036 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1038 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1041 static inline void tcp_enable_fack(struct tcp_sock *tp)
1043 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1046 /* TCP early-retransmit (ER) is similar to but more conservative than
1047 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1049 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1051 struct net *net = sock_net((struct sock *)tp);
1053 tp->do_early_retrans = sysctl_tcp_early_retrans &&
1054 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1055 net->ipv4.sysctl_tcp_reordering == 3;
1058 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1060 tp->do_early_retrans = 0;
1063 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1065 return tp->sacked_out + tp->lost_out;
1068 /* This determines how many packets are "in the network" to the best
1069 * of our knowledge. In many cases it is conservative, but where
1070 * detailed information is available from the receiver (via SACK
1071 * blocks etc.) we can make more aggressive calculations.
1073 * Use this for decisions involving congestion control, use just
1074 * tp->packets_out to determine if the send queue is empty or not.
1076 * Read this equation as:
1078 * "Packets sent once on transmission queue" MINUS
1079 * "Packets left network, but not honestly ACKed yet" PLUS
1080 * "Packets fast retransmitted"
1082 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1084 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1087 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1089 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1091 return tp->snd_cwnd < tp->snd_ssthresh;
1094 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1096 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1099 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1101 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1102 (1 << inet_csk(sk)->icsk_ca_state);
1105 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1106 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1109 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1111 const struct tcp_sock *tp = tcp_sk(sk);
1113 if (tcp_in_cwnd_reduction(sk))
1114 return tp->snd_ssthresh;
1116 return max(tp->snd_ssthresh,
1117 ((tp->snd_cwnd >> 1) +
1118 (tp->snd_cwnd >> 2)));
1121 /* Use define here intentionally to get WARN_ON location shown at the caller */
1122 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1124 void tcp_enter_cwr(struct sock *sk);
1125 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1127 /* The maximum number of MSS of available cwnd for which TSO defers
1128 * sending if not using sysctl_tcp_tso_win_divisor.
1130 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1135 /* Returns end sequence number of the receiver's advertised window */
1136 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1138 return tp->snd_una + tp->snd_wnd;
1141 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1142 * flexible approach. The RFC suggests cwnd should not be raised unless
1143 * it was fully used previously. And that's exactly what we do in
1144 * congestion avoidance mode. But in slow start we allow cwnd to grow
1145 * as long as the application has used half the cwnd.
1147 * cwnd is 10 (IW10), but application sends 9 frames.
1148 * We allow cwnd to reach 18 when all frames are ACKed.
1149 * This check is safe because it's as aggressive as slow start which already
1150 * risks 100% overshoot. The advantage is that we discourage application to
1151 * either send more filler packets or data to artificially blow up the cwnd
1152 * usage, and allow application-limited process to probe bw more aggressively.
1154 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1156 const struct tcp_sock *tp = tcp_sk(sk);
1158 if (tp->is_cwnd_limited)
1161 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1162 if (tcp_in_slow_start(tp))
1163 return tp->snd_cwnd < 2 * tp->max_packets_out;
1168 /* Something is really bad, we could not queue an additional packet,
1169 * because qdisc is full or receiver sent a 0 window.
1170 * We do not want to add fuel to the fire, or abort too early,
1171 * so make sure the timer we arm now is at least 200ms in the future,
1172 * regardless of current icsk_rto value (as it could be ~2ms)
1174 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1176 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1179 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1180 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1181 unsigned long max_when)
1183 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1185 return (unsigned long)min_t(u64, when, max_when);
1188 static inline void tcp_check_probe_timer(struct sock *sk)
1190 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1191 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1192 tcp_probe0_base(sk), TCP_RTO_MAX);
1195 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1200 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1206 * Calculate(/check) TCP checksum
1208 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1209 __be32 daddr, __wsum base)
1211 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1214 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1216 return __skb_checksum_complete(skb);
1219 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1221 return !skb_csum_unnecessary(skb) &&
1222 __tcp_checksum_complete(skb);
1225 /* Prequeue for VJ style copy to user, combined with checksumming. */
1227 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1229 tp->ucopy.task = NULL;
1231 tp->ucopy.memory = 0;
1232 skb_queue_head_init(&tp->ucopy.prequeue);
1235 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1236 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1237 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1242 static const char *statename[]={
1243 "Unused","Established","Syn Sent","Syn Recv",
1244 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1245 "Close Wait","Last ACK","Listen","Closing"
1248 void tcp_set_state(struct sock *sk, int state);
1250 void tcp_done(struct sock *sk);
1252 int tcp_abort(struct sock *sk, int err);
1254 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1257 rx_opt->num_sacks = 0;
1260 u32 tcp_default_init_rwnd(u32 mss);
1261 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1263 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1265 struct tcp_sock *tp = tcp_sk(sk);
1268 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1270 delta = tcp_time_stamp - tp->lsndtime;
1271 if (delta > inet_csk(sk)->icsk_rto)
1272 tcp_cwnd_restart(sk, delta);
1275 /* Determine a window scaling and initial window to offer. */
1276 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1277 __u32 *window_clamp, int wscale_ok,
1278 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1280 static inline int tcp_win_from_space(int space)
1282 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1284 return tcp_adv_win_scale <= 0 ?
1285 (space>>(-tcp_adv_win_scale)) :
1286 space - (space>>tcp_adv_win_scale);
1289 /* Note: caller must be prepared to deal with negative returns */
1290 static inline int tcp_space(const struct sock *sk)
1292 return tcp_win_from_space(sk->sk_rcvbuf -
1293 atomic_read(&sk->sk_rmem_alloc));
1296 static inline int tcp_full_space(const struct sock *sk)
1298 return tcp_win_from_space(sk->sk_rcvbuf);
1301 extern void tcp_openreq_init_rwin(struct request_sock *req,
1302 const struct sock *sk_listener,
1303 const struct dst_entry *dst);
1305 void tcp_enter_memory_pressure(struct sock *sk);
1307 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1309 struct net *net = sock_net((struct sock *)tp);
1311 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1314 static inline int keepalive_time_when(const struct tcp_sock *tp)
1316 struct net *net = sock_net((struct sock *)tp);
1318 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1321 static inline int keepalive_probes(const struct tcp_sock *tp)
1323 struct net *net = sock_net((struct sock *)tp);
1325 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1328 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1330 const struct inet_connection_sock *icsk = &tp->inet_conn;
1332 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1333 tcp_time_stamp - tp->rcv_tstamp);
1336 static inline int tcp_fin_time(const struct sock *sk)
1338 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1339 const int rto = inet_csk(sk)->icsk_rto;
1341 if (fin_timeout < (rto << 2) - (rto >> 1))
1342 fin_timeout = (rto << 2) - (rto >> 1);
1347 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1350 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1352 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1355 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1356 * then following tcp messages have valid values. Ignore 0 value,
1357 * or else 'negative' tsval might forbid us to accept their packets.
1359 if (!rx_opt->ts_recent)
1364 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1367 if (tcp_paws_check(rx_opt, 0))
1370 /* RST segments are not recommended to carry timestamp,
1371 and, if they do, it is recommended to ignore PAWS because
1372 "their cleanup function should take precedence over timestamps."
1373 Certainly, it is mistake. It is necessary to understand the reasons
1374 of this constraint to relax it: if peer reboots, clock may go
1375 out-of-sync and half-open connections will not be reset.
1376 Actually, the problem would be not existing if all
1377 the implementations followed draft about maintaining clock
1378 via reboots. Linux-2.2 DOES NOT!
1380 However, we can relax time bounds for RST segments to MSL.
1382 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1387 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1388 int mib_idx, u32 *last_oow_ack_time);
1390 static inline void tcp_mib_init(struct net *net)
1393 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1394 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1395 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1396 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1400 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1402 tp->lost_skb_hint = NULL;
1405 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1407 tcp_clear_retrans_hints_partial(tp);
1408 tp->retransmit_skb_hint = NULL;
1411 union tcp_md5_addr {
1413 #if IS_ENABLED(CONFIG_IPV6)
1418 /* - key database */
1419 struct tcp_md5sig_key {
1420 struct hlist_node node;
1422 u8 family; /* AF_INET or AF_INET6 */
1423 union tcp_md5_addr addr;
1424 u8 key[TCP_MD5SIG_MAXKEYLEN];
1425 struct rcu_head rcu;
1429 struct tcp_md5sig_info {
1430 struct hlist_head head;
1431 struct rcu_head rcu;
1434 /* - pseudo header */
1435 struct tcp4_pseudohdr {
1443 struct tcp6_pseudohdr {
1444 struct in6_addr saddr;
1445 struct in6_addr daddr;
1447 __be32 protocol; /* including padding */
1450 union tcp_md5sum_block {
1451 struct tcp4_pseudohdr ip4;
1452 #if IS_ENABLED(CONFIG_IPV6)
1453 struct tcp6_pseudohdr ip6;
1457 /* - pool: digest algorithm, hash description and scratch buffer */
1458 struct tcp_md5sig_pool {
1459 struct ahash_request *md5_req;
1464 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1465 const struct sock *sk, const struct sk_buff *skb);
1466 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1467 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1468 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1470 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1471 const struct sock *addr_sk);
1473 #ifdef CONFIG_TCP_MD5SIG
1474 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1475 const union tcp_md5_addr *addr,
1477 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1479 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1480 const union tcp_md5_addr *addr,
1485 #define tcp_twsk_md5_key(twsk) NULL
1488 bool tcp_alloc_md5sig_pool(void);
1490 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1491 static inline void tcp_put_md5sig_pool(void)
1496 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1497 unsigned int header_len);
1498 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1499 const struct tcp_md5sig_key *key);
1501 /* From tcp_fastopen.c */
1502 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1503 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1504 unsigned long *last_syn_loss);
1505 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1506 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1508 struct tcp_fastopen_request {
1509 /* Fast Open cookie. Size 0 means a cookie request */
1510 struct tcp_fastopen_cookie cookie;
1511 struct msghdr *data; /* data in MSG_FASTOPEN */
1513 int copied; /* queued in tcp_connect() */
1515 void tcp_free_fastopen_req(struct tcp_sock *tp);
1517 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1518 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1519 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1520 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1521 struct request_sock *req,
1522 struct tcp_fastopen_cookie *foc,
1523 struct dst_entry *dst);
1524 void tcp_fastopen_init_key_once(bool publish);
1525 #define TCP_FASTOPEN_KEY_LENGTH 16
1527 static inline void tcp_init_send_head(struct sock *sk)
1529 sk->sk_send_head = NULL;
1532 /* Fastopen key context */
1533 struct tcp_fastopen_context {
1534 struct crypto_cipher *tfm;
1535 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1536 struct rcu_head rcu;
1539 /* write queue abstraction */
1540 static inline void tcp_write_queue_purge(struct sock *sk)
1542 struct sk_buff *skb;
1544 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1545 sk_wmem_free_skb(sk, skb);
1547 tcp_clear_all_retrans_hints(tcp_sk(sk));
1548 tcp_init_send_head(sk);
1549 inet_csk(sk)->icsk_backoff = 0;
1552 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1554 return skb_peek(&sk->sk_write_queue);
1557 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1559 return skb_peek_tail(&sk->sk_write_queue);
1562 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1563 const struct sk_buff *skb)
1565 return skb_queue_next(&sk->sk_write_queue, skb);
1568 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1569 const struct sk_buff *skb)
1571 return skb_queue_prev(&sk->sk_write_queue, skb);
1574 #define tcp_for_write_queue(skb, sk) \
1575 skb_queue_walk(&(sk)->sk_write_queue, skb)
1577 #define tcp_for_write_queue_from(skb, sk) \
1578 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1580 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1581 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1583 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1585 return sk->sk_send_head;
1588 static inline bool tcp_skb_is_last(const struct sock *sk,
1589 const struct sk_buff *skb)
1591 return skb_queue_is_last(&sk->sk_write_queue, skb);
1594 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1596 if (tcp_skb_is_last(sk, skb))
1597 sk->sk_send_head = NULL;
1599 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1602 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1604 if (sk->sk_send_head == skb_unlinked)
1605 sk->sk_send_head = NULL;
1606 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1607 tcp_sk(sk)->highest_sack = NULL;
1610 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1612 struct sk_buff *skb = tcp_write_queue_head(sk);
1614 if (skb == tcp_send_head(sk))
1620 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1622 struct sk_buff *skb = tcp_send_head(sk);
1624 /* empty retransmit queue, for example due to zero window */
1625 if (skb == tcp_write_queue_head(sk))
1628 return skb ? tcp_write_queue_prev(sk, skb) : tcp_write_queue_tail(sk);
1631 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1633 __skb_queue_tail(&sk->sk_write_queue, skb);
1636 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1638 __tcp_add_write_queue_tail(sk, skb);
1640 /* Queue it, remembering where we must start sending. */
1641 if (sk->sk_send_head == NULL) {
1642 sk->sk_send_head = skb;
1644 if (tcp_sk(sk)->highest_sack == NULL)
1645 tcp_sk(sk)->highest_sack = skb;
1649 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1651 __skb_queue_head(&sk->sk_write_queue, skb);
1654 /* Insert buff after skb on the write queue of sk. */
1655 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1656 struct sk_buff *buff,
1659 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1662 /* Insert new before skb on the write queue of sk. */
1663 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1664 struct sk_buff *skb,
1667 __skb_queue_before(&sk->sk_write_queue, skb, new);
1669 if (sk->sk_send_head == skb)
1670 sk->sk_send_head = new;
1673 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1675 __skb_unlink(skb, &sk->sk_write_queue);
1678 static inline bool tcp_write_queue_empty(struct sock *sk)
1680 return skb_queue_empty(&sk->sk_write_queue);
1683 static inline void tcp_push_pending_frames(struct sock *sk)
1685 if (tcp_send_head(sk)) {
1686 struct tcp_sock *tp = tcp_sk(sk);
1688 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1692 /* Start sequence of the skb just after the highest skb with SACKed
1693 * bit, valid only if sacked_out > 0 or when the caller has ensured
1694 * validity by itself.
1696 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1698 if (!tp->sacked_out)
1701 if (tp->highest_sack == NULL)
1704 return TCP_SKB_CB(tp->highest_sack)->seq;
1707 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1709 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1710 tcp_write_queue_next(sk, skb);
1713 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1715 return tcp_sk(sk)->highest_sack;
1718 static inline void tcp_highest_sack_reset(struct sock *sk)
1720 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1723 /* Called when old skb is about to be deleted and replaced by new skb */
1724 static inline void tcp_highest_sack_replace(struct sock *sk,
1725 struct sk_buff *old,
1726 struct sk_buff *new)
1728 if (old == tcp_highest_sack(sk))
1729 tcp_sk(sk)->highest_sack = new;
1732 /* This helper checks if socket has IP_TRANSPARENT set */
1733 static inline bool inet_sk_transparent(const struct sock *sk)
1735 switch (sk->sk_state) {
1737 return inet_twsk(sk)->tw_transparent;
1738 case TCP_NEW_SYN_RECV:
1739 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1741 return inet_sk(sk)->transparent;
1744 /* Determines whether this is a thin stream (which may suffer from
1745 * increased latency). Used to trigger latency-reducing mechanisms.
1747 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1749 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1753 enum tcp_seq_states {
1754 TCP_SEQ_STATE_LISTENING,
1755 TCP_SEQ_STATE_ESTABLISHED,
1758 int tcp_seq_open(struct inode *inode, struct file *file);
1760 struct tcp_seq_afinfo {
1763 const struct file_operations *seq_fops;
1764 struct seq_operations seq_ops;
1767 struct tcp_iter_state {
1768 struct seq_net_private p;
1770 enum tcp_seq_states state;
1771 struct sock *syn_wait_sk;
1772 int bucket, offset, sbucket, num;
1776 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1777 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1779 extern struct request_sock_ops tcp_request_sock_ops;
1780 extern struct request_sock_ops tcp6_request_sock_ops;
1782 void tcp_v4_destroy_sock(struct sock *sk);
1784 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1785 netdev_features_t features);
1786 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1787 int tcp_gro_complete(struct sk_buff *skb);
1789 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1791 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1793 struct net *net = sock_net((struct sock *)tp);
1794 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
1797 static inline bool tcp_stream_memory_free(const struct sock *sk)
1799 const struct tcp_sock *tp = tcp_sk(sk);
1800 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1802 return notsent_bytes < tcp_notsent_lowat(tp);
1805 #ifdef CONFIG_PROC_FS
1806 int tcp4_proc_init(void);
1807 void tcp4_proc_exit(void);
1810 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1811 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1812 const struct tcp_request_sock_ops *af_ops,
1813 struct sock *sk, struct sk_buff *skb);
1815 /* TCP af-specific functions */
1816 struct tcp_sock_af_ops {
1817 #ifdef CONFIG_TCP_MD5SIG
1818 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1819 const struct sock *addr_sk);
1820 int (*calc_md5_hash)(char *location,
1821 const struct tcp_md5sig_key *md5,
1822 const struct sock *sk,
1823 const struct sk_buff *skb);
1824 int (*md5_parse)(struct sock *sk,
1825 char __user *optval,
1830 struct tcp_request_sock_ops {
1832 #ifdef CONFIG_TCP_MD5SIG
1833 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1834 const struct sock *addr_sk);
1835 int (*calc_md5_hash) (char *location,
1836 const struct tcp_md5sig_key *md5,
1837 const struct sock *sk,
1838 const struct sk_buff *skb);
1840 void (*init_req)(struct request_sock *req,
1841 const struct sock *sk_listener,
1842 struct sk_buff *skb);
1843 #ifdef CONFIG_SYN_COOKIES
1844 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1847 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1848 const struct request_sock *req,
1850 __u32 (*init_seq)(const struct sk_buff *skb);
1851 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1852 struct flowi *fl, struct request_sock *req,
1853 struct tcp_fastopen_cookie *foc,
1854 enum tcp_synack_type synack_type);
1857 #ifdef CONFIG_SYN_COOKIES
1858 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1859 const struct sock *sk, struct sk_buff *skb,
1862 tcp_synq_overflow(sk);
1863 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1864 return ops->cookie_init_seq(skb, mss);
1867 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1868 const struct sock *sk, struct sk_buff *skb,
1875 int tcpv4_offload_init(void);
1877 void tcp_v4_init(void);
1878 void tcp_init(void);
1880 /* tcp_recovery.c */
1882 /* Flags to enable various loss recovery features. See below */
1883 extern int sysctl_tcp_recovery;
1885 /* Use TCP RACK to detect (some) tail and retransmit losses */
1886 #define TCP_RACK_LOST_RETRANS 0x1
1888 extern int tcp_rack_mark_lost(struct sock *sk);
1890 extern void tcp_rack_advance(struct tcp_sock *tp,
1891 const struct skb_mstamp *xmit_time, u8 sacked);
1894 * Save and compile IPv4 options, return a pointer to it
1896 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1898 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1899 struct ip_options_rcu *dopt = NULL;
1902 int opt_size = sizeof(*dopt) + opt->optlen;
1904 dopt = kmalloc(opt_size, GFP_ATOMIC);
1905 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1913 /* locally generated TCP pure ACKs have skb->truesize == 2
1914 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1915 * This is much faster than dissecting the packet to find out.
1916 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1918 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1920 return skb->truesize == 2;
1923 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1928 static inline int tcp_inq(struct sock *sk)
1930 struct tcp_sock *tp = tcp_sk(sk);
1933 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1935 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1937 before(tp->urg_seq, tp->copied_seq) ||
1938 !before(tp->urg_seq, tp->rcv_nxt)) {
1940 answ = tp->rcv_nxt - tp->copied_seq;
1942 /* Subtract 1, if FIN was received */
1943 if (answ && sock_flag(sk, SOCK_DONE))
1946 answ = tp->urg_seq - tp->copied_seq;
1952 int tcp_peek_len(struct socket *sock);
1954 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1958 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1959 tp->segs_in += segs_in;
1960 if (skb->len > tcp_hdrlen(skb))
1961 tp->data_segs_in += segs_in;
1965 * TCP listen path runs lockless.
1966 * We forced "struct sock" to be const qualified to make sure
1967 * we don't modify one of its field by mistake.
1968 * Here, we increment sk_drops which is an atomic_t, so we can safely
1969 * make sock writable again.
1971 static inline void tcp_listendrop(const struct sock *sk)
1973 atomic_inc(&((struct sock *)sk)->sk_drops);
1974 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);