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);
584 void tcp_init_xmit_timers(struct sock *);
585 static inline void tcp_clear_xmit_timers(struct sock *sk)
587 inet_csk_clear_xmit_timers(sk);
590 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
591 unsigned int tcp_current_mss(struct sock *sk);
593 /* Bound MSS / TSO packet size with the half of the window */
594 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
598 /* When peer uses tiny windows, there is no use in packetizing
599 * to sub-MSS pieces for the sake of SWS or making sure there
600 * are enough packets in the pipe for fast recovery.
602 * On the other hand, for extremely large MSS devices, handling
603 * smaller than MSS windows in this way does make sense.
605 if (tp->max_window > TCP_MSS_DEFAULT)
606 cutoff = (tp->max_window >> 1);
608 cutoff = tp->max_window;
610 if (cutoff && pktsize > cutoff)
611 return max_t(int, cutoff, 68U - tp->tcp_header_len);
617 void tcp_get_info(struct sock *, struct tcp_info *);
619 /* Read 'sendfile()'-style from a TCP socket */
620 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
621 sk_read_actor_t recv_actor);
623 void tcp_initialize_rcv_mss(struct sock *sk);
625 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
626 int tcp_mss_to_mtu(struct sock *sk, int mss);
627 void tcp_mtup_init(struct sock *sk);
628 void tcp_init_buffer_space(struct sock *sk);
630 static inline void tcp_bound_rto(const struct sock *sk)
632 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
633 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
636 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
638 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
641 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
643 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
644 ntohl(TCP_FLAG_ACK) |
648 static inline void tcp_fast_path_on(struct tcp_sock *tp)
650 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
653 static inline void tcp_fast_path_check(struct sock *sk)
655 struct tcp_sock *tp = tcp_sk(sk);
657 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
659 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
661 tcp_fast_path_on(tp);
664 /* Compute the actual rto_min value */
665 static inline u32 tcp_rto_min(struct sock *sk)
667 const struct dst_entry *dst = __sk_dst_get(sk);
668 u32 rto_min = TCP_RTO_MIN;
670 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
671 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
675 static inline u32 tcp_rto_min_us(struct sock *sk)
677 return jiffies_to_usecs(tcp_rto_min(sk));
680 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
682 return dst_metric_locked(dst, RTAX_CC_ALGO);
685 /* Minimum RTT in usec. ~0 means not available. */
686 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
688 return minmax_get(&tp->rtt_min);
691 /* Compute the actual receive window we are currently advertising.
692 * Rcv_nxt can be after the window if our peer push more data
693 * than the offered window.
695 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
697 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
704 /* Choose a new window, without checks for shrinking, and without
705 * scaling applied to the result. The caller does these things
706 * if necessary. This is a "raw" window selection.
708 u32 __tcp_select_window(struct sock *sk);
710 void tcp_send_window_probe(struct sock *sk);
712 /* TCP timestamps are only 32-bits, this causes a slight
713 * complication on 64-bit systems since we store a snapshot
714 * of jiffies in the buffer control blocks below. We decided
715 * to use only the low 32-bits of jiffies and hide the ugly
716 * casts with the following macro.
718 #define tcp_time_stamp ((__u32)(jiffies))
720 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
722 return skb->skb_mstamp.stamp_jiffies;
726 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
728 #define TCPHDR_FIN 0x01
729 #define TCPHDR_SYN 0x02
730 #define TCPHDR_RST 0x04
731 #define TCPHDR_PSH 0x08
732 #define TCPHDR_ACK 0x10
733 #define TCPHDR_URG 0x20
734 #define TCPHDR_ECE 0x40
735 #define TCPHDR_CWR 0x80
737 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
739 /* This is what the send packet queuing engine uses to pass
740 * TCP per-packet control information to the transmission code.
741 * We also store the host-order sequence numbers in here too.
742 * This is 44 bytes if IPV6 is enabled.
743 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
746 __u32 seq; /* Starting sequence number */
747 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
749 /* Note : tcp_tw_isn is used in input path only
750 * (isn chosen by tcp_timewait_state_process())
752 * tcp_gso_segs/size are used in write queue only,
753 * cf tcp_skb_pcount()/tcp_skb_mss()
761 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
763 __u8 sacked; /* State flags for SACK/FACK. */
764 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
765 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
766 #define TCPCB_LOST 0x04 /* SKB is lost */
767 #define TCPCB_TAGBITS 0x07 /* All tag bits */
768 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
769 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
770 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
773 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
774 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
775 eor:1, /* Is skb MSG_EOR marked? */
777 __u32 ack_seq; /* Sequence number ACK'd */
780 /* There is space for up to 24 bytes */
781 __u32 in_flight:30,/* Bytes in flight at transmit */
782 is_app_limited:1, /* cwnd not fully used? */
784 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
786 /* start of send pipeline phase */
787 struct skb_mstamp first_tx_mstamp;
788 /* when we reached the "delivered" count */
789 struct skb_mstamp delivered_mstamp;
790 } tx; /* only used for outgoing skbs */
792 struct inet_skb_parm h4;
793 #if IS_ENABLED(CONFIG_IPV6)
794 struct inet6_skb_parm h6;
796 } header; /* For incoming skbs */
800 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
803 #if IS_ENABLED(CONFIG_IPV6)
804 /* This is the variant of inet6_iif() that must be used by TCP,
805 * as TCP moves IP6CB into a different location in skb->cb[]
807 static inline int tcp_v6_iif(const struct sk_buff *skb)
809 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
811 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
815 /* TCP_SKB_CB reference means this can not be used from early demux */
816 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
818 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
819 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
820 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
826 /* Due to TSO, an SKB can be composed of multiple actual
827 * packets. To keep these tracked properly, we use this.
829 static inline int tcp_skb_pcount(const struct sk_buff *skb)
831 return TCP_SKB_CB(skb)->tcp_gso_segs;
834 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
836 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
839 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
841 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
844 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
845 static inline int tcp_skb_mss(const struct sk_buff *skb)
847 return TCP_SKB_CB(skb)->tcp_gso_size;
850 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
852 return likely(!TCP_SKB_CB(skb)->eor);
855 /* Events passed to congestion control interface */
857 CA_EVENT_TX_START, /* first transmit when no packets in flight */
858 CA_EVENT_CWND_RESTART, /* congestion window restart */
859 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
860 CA_EVENT_LOSS, /* loss timeout */
861 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
862 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
865 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
866 enum tcp_ca_ack_event_flags {
867 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
868 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
869 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
873 * Interface for adding new TCP congestion control handlers
875 #define TCP_CA_NAME_MAX 16
876 #define TCP_CA_MAX 128
877 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
879 #define TCP_CA_UNSPEC 0
881 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
882 #define TCP_CONG_NON_RESTRICTED 0x1
883 /* Requires ECN/ECT set on all packets */
884 #define TCP_CONG_NEEDS_ECN 0x2
894 /* A rate sample measures the number of (original/retransmitted) data
895 * packets delivered "delivered" over an interval of time "interval_us".
896 * The tcp_rate.c code fills in the rate sample, and congestion
897 * control modules that define a cong_control function to run at the end
898 * of ACK processing can optionally chose to consult this sample when
899 * setting cwnd and pacing rate.
900 * A sample is invalid if "delivered" or "interval_us" is negative.
903 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
904 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
905 s32 delivered; /* number of packets delivered over interval */
906 long interval_us; /* time for tp->delivered to incr "delivered" */
907 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
908 int losses; /* number of packets marked lost upon ACK */
909 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
910 u32 prior_in_flight; /* in flight before this ACK */
911 bool is_app_limited; /* is sample from packet with bubble in pipe? */
912 bool is_retrans; /* is sample from retransmission? */
915 struct tcp_congestion_ops {
916 struct list_head list;
920 /* initialize private data (optional) */
921 void (*init)(struct sock *sk);
922 /* cleanup private data (optional) */
923 void (*release)(struct sock *sk);
925 /* return slow start threshold (required) */
926 u32 (*ssthresh)(struct sock *sk);
927 /* do new cwnd calculation (required) */
928 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
929 /* call before changing ca_state (optional) */
930 void (*set_state)(struct sock *sk, u8 new_state);
931 /* call when cwnd event occurs (optional) */
932 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
933 /* call when ack arrives (optional) */
934 void (*in_ack_event)(struct sock *sk, u32 flags);
935 /* new value of cwnd after loss (optional) */
936 u32 (*undo_cwnd)(struct sock *sk);
937 /* hook for packet ack accounting (optional) */
938 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
939 /* suggest number of segments for each skb to transmit (optional) */
940 u32 (*tso_segs_goal)(struct sock *sk);
941 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
942 u32 (*sndbuf_expand)(struct sock *sk);
943 /* call when packets are delivered to update cwnd and pacing rate,
944 * after all the ca_state processing. (optional)
946 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
947 /* get info for inet_diag (optional) */
948 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
949 union tcp_cc_info *info);
951 char name[TCP_CA_NAME_MAX];
952 struct module *owner;
955 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
956 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
958 void tcp_assign_congestion_control(struct sock *sk);
959 void tcp_init_congestion_control(struct sock *sk);
960 void tcp_cleanup_congestion_control(struct sock *sk);
961 int tcp_set_default_congestion_control(const char *name);
962 void tcp_get_default_congestion_control(char *name);
963 void tcp_get_available_congestion_control(char *buf, size_t len);
964 void tcp_get_allowed_congestion_control(char *buf, size_t len);
965 int tcp_set_allowed_congestion_control(char *allowed);
966 int tcp_set_congestion_control(struct sock *sk, const char *name);
967 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
968 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
970 u32 tcp_reno_ssthresh(struct sock *sk);
971 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
972 extern struct tcp_congestion_ops tcp_reno;
974 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
975 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
977 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
979 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
985 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
987 const struct inet_connection_sock *icsk = inet_csk(sk);
989 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
992 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
994 struct inet_connection_sock *icsk = inet_csk(sk);
996 if (icsk->icsk_ca_ops->set_state)
997 icsk->icsk_ca_ops->set_state(sk, ca_state);
998 icsk->icsk_ca_state = ca_state;
1001 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1003 const struct inet_connection_sock *icsk = inet_csk(sk);
1005 if (icsk->icsk_ca_ops->cwnd_event)
1006 icsk->icsk_ca_ops->cwnd_event(sk, event);
1009 /* From tcp_rate.c */
1010 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1011 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1012 struct rate_sample *rs);
1013 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1014 bool is_sack_reneg, struct skb_mstamp *now, struct rate_sample *rs);
1015 void tcp_rate_check_app_limited(struct sock *sk);
1017 /* These functions determine how the current flow behaves in respect of SACK
1018 * handling. SACK is negotiated with the peer, and therefore it can vary
1019 * between different flows.
1021 * tcp_is_sack - SACK enabled
1022 * tcp_is_reno - No SACK
1023 * tcp_is_fack - FACK enabled, implies SACK enabled
1025 static inline int tcp_is_sack(const struct tcp_sock *tp)
1027 return tp->rx_opt.sack_ok;
1030 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1032 return !tcp_is_sack(tp);
1035 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1037 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1040 static inline void tcp_enable_fack(struct tcp_sock *tp)
1042 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1045 /* TCP early-retransmit (ER) is similar to but more conservative than
1046 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1048 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1050 struct net *net = sock_net((struct sock *)tp);
1052 tp->do_early_retrans = sysctl_tcp_early_retrans &&
1053 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1054 net->ipv4.sysctl_tcp_reordering == 3;
1057 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1059 tp->do_early_retrans = 0;
1062 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1064 return tp->sacked_out + tp->lost_out;
1067 /* This determines how many packets are "in the network" to the best
1068 * of our knowledge. In many cases it is conservative, but where
1069 * detailed information is available from the receiver (via SACK
1070 * blocks etc.) we can make more aggressive calculations.
1072 * Use this for decisions involving congestion control, use just
1073 * tp->packets_out to determine if the send queue is empty or not.
1075 * Read this equation as:
1077 * "Packets sent once on transmission queue" MINUS
1078 * "Packets left network, but not honestly ACKed yet" PLUS
1079 * "Packets fast retransmitted"
1081 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1083 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1086 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1088 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1090 return tp->snd_cwnd < tp->snd_ssthresh;
1093 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1095 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1098 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1100 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1101 (1 << inet_csk(sk)->icsk_ca_state);
1104 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1105 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1108 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1110 const struct tcp_sock *tp = tcp_sk(sk);
1112 if (tcp_in_cwnd_reduction(sk))
1113 return tp->snd_ssthresh;
1115 return max(tp->snd_ssthresh,
1116 ((tp->snd_cwnd >> 1) +
1117 (tp->snd_cwnd >> 2)));
1120 /* Use define here intentionally to get WARN_ON location shown at the caller */
1121 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1123 void tcp_enter_cwr(struct sock *sk);
1124 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1126 /* The maximum number of MSS of available cwnd for which TSO defers
1127 * sending if not using sysctl_tcp_tso_win_divisor.
1129 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1134 /* Returns end sequence number of the receiver's advertised window */
1135 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1137 return tp->snd_una + tp->snd_wnd;
1140 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1141 * flexible approach. The RFC suggests cwnd should not be raised unless
1142 * it was fully used previously. And that's exactly what we do in
1143 * congestion avoidance mode. But in slow start we allow cwnd to grow
1144 * as long as the application has used half the cwnd.
1146 * cwnd is 10 (IW10), but application sends 9 frames.
1147 * We allow cwnd to reach 18 when all frames are ACKed.
1148 * This check is safe because it's as aggressive as slow start which already
1149 * risks 100% overshoot. The advantage is that we discourage application to
1150 * either send more filler packets or data to artificially blow up the cwnd
1151 * usage, and allow application-limited process to probe bw more aggressively.
1153 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1155 const struct tcp_sock *tp = tcp_sk(sk);
1157 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1158 if (tcp_in_slow_start(tp))
1159 return tp->snd_cwnd < 2 * tp->max_packets_out;
1161 return tp->is_cwnd_limited;
1164 /* Something is really bad, we could not queue an additional packet,
1165 * because qdisc is full or receiver sent a 0 window.
1166 * We do not want to add fuel to the fire, or abort too early,
1167 * so make sure the timer we arm now is at least 200ms in the future,
1168 * regardless of current icsk_rto value (as it could be ~2ms)
1170 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1172 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1175 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1176 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1177 unsigned long max_when)
1179 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1181 return (unsigned long)min_t(u64, when, max_when);
1184 static inline void tcp_check_probe_timer(struct sock *sk)
1186 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1187 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1188 tcp_probe0_base(sk), TCP_RTO_MAX);
1191 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1196 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1202 * Calculate(/check) TCP checksum
1204 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1205 __be32 daddr, __wsum base)
1207 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1210 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1212 return __skb_checksum_complete(skb);
1215 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1217 return !skb_csum_unnecessary(skb) &&
1218 __tcp_checksum_complete(skb);
1221 /* Prequeue for VJ style copy to user, combined with checksumming. */
1223 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1225 tp->ucopy.task = NULL;
1227 tp->ucopy.memory = 0;
1228 skb_queue_head_init(&tp->ucopy.prequeue);
1231 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1232 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1233 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1238 static const char *statename[]={
1239 "Unused","Established","Syn Sent","Syn Recv",
1240 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1241 "Close Wait","Last ACK","Listen","Closing"
1244 void tcp_set_state(struct sock *sk, int state);
1246 void tcp_done(struct sock *sk);
1248 int tcp_abort(struct sock *sk, int err);
1250 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1253 rx_opt->num_sacks = 0;
1256 u32 tcp_default_init_rwnd(u32 mss);
1257 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1259 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1261 struct tcp_sock *tp = tcp_sk(sk);
1264 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1266 delta = tcp_time_stamp - tp->lsndtime;
1267 if (delta > inet_csk(sk)->icsk_rto)
1268 tcp_cwnd_restart(sk, delta);
1271 /* Determine a window scaling and initial window to offer. */
1272 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1273 __u32 *window_clamp, int wscale_ok,
1274 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1276 static inline int tcp_win_from_space(int space)
1278 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1280 return tcp_adv_win_scale <= 0 ?
1281 (space>>(-tcp_adv_win_scale)) :
1282 space - (space>>tcp_adv_win_scale);
1285 /* Note: caller must be prepared to deal with negative returns */
1286 static inline int tcp_space(const struct sock *sk)
1288 return tcp_win_from_space(sk->sk_rcvbuf -
1289 atomic_read(&sk->sk_rmem_alloc));
1292 static inline int tcp_full_space(const struct sock *sk)
1294 return tcp_win_from_space(sk->sk_rcvbuf);
1297 extern void tcp_openreq_init_rwin(struct request_sock *req,
1298 const struct sock *sk_listener,
1299 const struct dst_entry *dst);
1301 void tcp_enter_memory_pressure(struct sock *sk);
1303 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1305 struct net *net = sock_net((struct sock *)tp);
1307 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1310 static inline int keepalive_time_when(const struct tcp_sock *tp)
1312 struct net *net = sock_net((struct sock *)tp);
1314 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1317 static inline int keepalive_probes(const struct tcp_sock *tp)
1319 struct net *net = sock_net((struct sock *)tp);
1321 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1324 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1326 const struct inet_connection_sock *icsk = &tp->inet_conn;
1328 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1329 tcp_time_stamp - tp->rcv_tstamp);
1332 static inline int tcp_fin_time(const struct sock *sk)
1334 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1335 const int rto = inet_csk(sk)->icsk_rto;
1337 if (fin_timeout < (rto << 2) - (rto >> 1))
1338 fin_timeout = (rto << 2) - (rto >> 1);
1343 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1346 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1348 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1351 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1352 * then following tcp messages have valid values. Ignore 0 value,
1353 * or else 'negative' tsval might forbid us to accept their packets.
1355 if (!rx_opt->ts_recent)
1360 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1363 if (tcp_paws_check(rx_opt, 0))
1366 /* RST segments are not recommended to carry timestamp,
1367 and, if they do, it is recommended to ignore PAWS because
1368 "their cleanup function should take precedence over timestamps."
1369 Certainly, it is mistake. It is necessary to understand the reasons
1370 of this constraint to relax it: if peer reboots, clock may go
1371 out-of-sync and half-open connections will not be reset.
1372 Actually, the problem would be not existing if all
1373 the implementations followed draft about maintaining clock
1374 via reboots. Linux-2.2 DOES NOT!
1376 However, we can relax time bounds for RST segments to MSL.
1378 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1383 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1384 int mib_idx, u32 *last_oow_ack_time);
1386 static inline void tcp_mib_init(struct net *net)
1389 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1390 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1391 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1392 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1396 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1398 tp->lost_skb_hint = NULL;
1401 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1403 tcp_clear_retrans_hints_partial(tp);
1404 tp->retransmit_skb_hint = NULL;
1407 union tcp_md5_addr {
1409 #if IS_ENABLED(CONFIG_IPV6)
1414 /* - key database */
1415 struct tcp_md5sig_key {
1416 struct hlist_node node;
1418 u8 family; /* AF_INET or AF_INET6 */
1419 union tcp_md5_addr addr;
1420 u8 key[TCP_MD5SIG_MAXKEYLEN];
1421 struct rcu_head rcu;
1425 struct tcp_md5sig_info {
1426 struct hlist_head head;
1427 struct rcu_head rcu;
1430 /* - pseudo header */
1431 struct tcp4_pseudohdr {
1439 struct tcp6_pseudohdr {
1440 struct in6_addr saddr;
1441 struct in6_addr daddr;
1443 __be32 protocol; /* including padding */
1446 union tcp_md5sum_block {
1447 struct tcp4_pseudohdr ip4;
1448 #if IS_ENABLED(CONFIG_IPV6)
1449 struct tcp6_pseudohdr ip6;
1453 /* - pool: digest algorithm, hash description and scratch buffer */
1454 struct tcp_md5sig_pool {
1455 struct ahash_request *md5_req;
1460 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1461 const struct sock *sk, const struct sk_buff *skb);
1462 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1463 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1464 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1466 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1467 const struct sock *addr_sk);
1469 #ifdef CONFIG_TCP_MD5SIG
1470 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1471 const union tcp_md5_addr *addr,
1473 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1475 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1476 const union tcp_md5_addr *addr,
1481 #define tcp_twsk_md5_key(twsk) NULL
1484 bool tcp_alloc_md5sig_pool(void);
1486 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1487 static inline void tcp_put_md5sig_pool(void)
1492 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1493 unsigned int header_len);
1494 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1495 const struct tcp_md5sig_key *key);
1497 /* From tcp_fastopen.c */
1498 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1499 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1500 unsigned long *last_syn_loss);
1501 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1502 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1504 struct tcp_fastopen_request {
1505 /* Fast Open cookie. Size 0 means a cookie request */
1506 struct tcp_fastopen_cookie cookie;
1507 struct msghdr *data; /* data in MSG_FASTOPEN */
1509 int copied; /* queued in tcp_connect() */
1511 void tcp_free_fastopen_req(struct tcp_sock *tp);
1513 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1514 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1515 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1516 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1517 struct request_sock *req,
1518 struct tcp_fastopen_cookie *foc,
1519 struct dst_entry *dst);
1520 void tcp_fastopen_init_key_once(bool publish);
1521 #define TCP_FASTOPEN_KEY_LENGTH 16
1523 static inline void tcp_init_send_head(struct sock *sk)
1525 sk->sk_send_head = NULL;
1528 /* Fastopen key context */
1529 struct tcp_fastopen_context {
1530 struct crypto_cipher *tfm;
1531 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1532 struct rcu_head rcu;
1535 /* write queue abstraction */
1536 static inline void tcp_write_queue_purge(struct sock *sk)
1538 struct sk_buff *skb;
1540 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1541 sk_wmem_free_skb(sk, skb);
1543 tcp_clear_all_retrans_hints(tcp_sk(sk));
1544 tcp_init_send_head(sk);
1545 inet_csk(sk)->icsk_backoff = 0;
1548 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1550 return skb_peek(&sk->sk_write_queue);
1553 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1555 return skb_peek_tail(&sk->sk_write_queue);
1558 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1559 const struct sk_buff *skb)
1561 return skb_queue_next(&sk->sk_write_queue, skb);
1564 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1565 const struct sk_buff *skb)
1567 return skb_queue_prev(&sk->sk_write_queue, skb);
1570 #define tcp_for_write_queue(skb, sk) \
1571 skb_queue_walk(&(sk)->sk_write_queue, skb)
1573 #define tcp_for_write_queue_from(skb, sk) \
1574 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1576 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1577 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1579 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1581 return sk->sk_send_head;
1584 static inline bool tcp_skb_is_last(const struct sock *sk,
1585 const struct sk_buff *skb)
1587 return skb_queue_is_last(&sk->sk_write_queue, skb);
1590 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1592 if (tcp_skb_is_last(sk, skb))
1593 sk->sk_send_head = NULL;
1595 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1598 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1600 if (sk->sk_send_head == skb_unlinked)
1601 sk->sk_send_head = NULL;
1602 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1603 tcp_sk(sk)->highest_sack = NULL;
1606 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1608 struct sk_buff *skb = tcp_write_queue_head(sk);
1610 if (skb == tcp_send_head(sk))
1616 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1618 struct sk_buff *skb = tcp_send_head(sk);
1620 /* empty retransmit queue, for example due to zero window */
1621 if (skb == tcp_write_queue_head(sk))
1624 return skb ? tcp_write_queue_prev(sk, skb) : tcp_write_queue_tail(sk);
1627 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1629 __skb_queue_tail(&sk->sk_write_queue, skb);
1632 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1634 __tcp_add_write_queue_tail(sk, skb);
1636 /* Queue it, remembering where we must start sending. */
1637 if (sk->sk_send_head == NULL) {
1638 sk->sk_send_head = skb;
1640 if (tcp_sk(sk)->highest_sack == NULL)
1641 tcp_sk(sk)->highest_sack = skb;
1645 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1647 __skb_queue_head(&sk->sk_write_queue, skb);
1650 /* Insert buff after skb on the write queue of sk. */
1651 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1652 struct sk_buff *buff,
1655 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1658 /* Insert new before skb on the write queue of sk. */
1659 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1660 struct sk_buff *skb,
1663 __skb_queue_before(&sk->sk_write_queue, skb, new);
1665 if (sk->sk_send_head == skb)
1666 sk->sk_send_head = new;
1669 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1671 __skb_unlink(skb, &sk->sk_write_queue);
1674 static inline bool tcp_write_queue_empty(struct sock *sk)
1676 return skb_queue_empty(&sk->sk_write_queue);
1679 static inline void tcp_push_pending_frames(struct sock *sk)
1681 if (tcp_send_head(sk)) {
1682 struct tcp_sock *tp = tcp_sk(sk);
1684 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1688 /* Start sequence of the skb just after the highest skb with SACKed
1689 * bit, valid only if sacked_out > 0 or when the caller has ensured
1690 * validity by itself.
1692 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1694 if (!tp->sacked_out)
1697 if (tp->highest_sack == NULL)
1700 return TCP_SKB_CB(tp->highest_sack)->seq;
1703 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1705 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1706 tcp_write_queue_next(sk, skb);
1709 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1711 return tcp_sk(sk)->highest_sack;
1714 static inline void tcp_highest_sack_reset(struct sock *sk)
1716 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1719 /* Called when old skb is about to be deleted and replaced by new skb */
1720 static inline void tcp_highest_sack_replace(struct sock *sk,
1721 struct sk_buff *old,
1722 struct sk_buff *new)
1724 if (old == tcp_highest_sack(sk))
1725 tcp_sk(sk)->highest_sack = new;
1728 /* This helper checks if socket has IP_TRANSPARENT set */
1729 static inline bool inet_sk_transparent(const struct sock *sk)
1731 switch (sk->sk_state) {
1733 return inet_twsk(sk)->tw_transparent;
1734 case TCP_NEW_SYN_RECV:
1735 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1737 return inet_sk(sk)->transparent;
1740 /* Determines whether this is a thin stream (which may suffer from
1741 * increased latency). Used to trigger latency-reducing mechanisms.
1743 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1745 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1749 enum tcp_seq_states {
1750 TCP_SEQ_STATE_LISTENING,
1751 TCP_SEQ_STATE_ESTABLISHED,
1754 int tcp_seq_open(struct inode *inode, struct file *file);
1756 struct tcp_seq_afinfo {
1759 const struct file_operations *seq_fops;
1760 struct seq_operations seq_ops;
1763 struct tcp_iter_state {
1764 struct seq_net_private p;
1766 enum tcp_seq_states state;
1767 struct sock *syn_wait_sk;
1768 int bucket, offset, sbucket, num;
1772 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1773 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1775 extern struct request_sock_ops tcp_request_sock_ops;
1776 extern struct request_sock_ops tcp6_request_sock_ops;
1778 void tcp_v4_destroy_sock(struct sock *sk);
1780 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1781 netdev_features_t features);
1782 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1783 int tcp_gro_complete(struct sk_buff *skb);
1785 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1787 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1789 struct net *net = sock_net((struct sock *)tp);
1790 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1793 static inline bool tcp_stream_memory_free(const struct sock *sk)
1795 const struct tcp_sock *tp = tcp_sk(sk);
1796 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1798 return notsent_bytes < tcp_notsent_lowat(tp);
1801 #ifdef CONFIG_PROC_FS
1802 int tcp4_proc_init(void);
1803 void tcp4_proc_exit(void);
1806 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1807 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1808 const struct tcp_request_sock_ops *af_ops,
1809 struct sock *sk, struct sk_buff *skb);
1811 /* TCP af-specific functions */
1812 struct tcp_sock_af_ops {
1813 #ifdef CONFIG_TCP_MD5SIG
1814 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1815 const struct sock *addr_sk);
1816 int (*calc_md5_hash)(char *location,
1817 const struct tcp_md5sig_key *md5,
1818 const struct sock *sk,
1819 const struct sk_buff *skb);
1820 int (*md5_parse)(struct sock *sk,
1821 char __user *optval,
1826 struct tcp_request_sock_ops {
1828 #ifdef CONFIG_TCP_MD5SIG
1829 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1830 const struct sock *addr_sk);
1831 int (*calc_md5_hash) (char *location,
1832 const struct tcp_md5sig_key *md5,
1833 const struct sock *sk,
1834 const struct sk_buff *skb);
1836 void (*init_req)(struct request_sock *req,
1837 const struct sock *sk_listener,
1838 struct sk_buff *skb);
1839 #ifdef CONFIG_SYN_COOKIES
1840 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1843 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1844 const struct request_sock *req,
1846 __u32 (*init_seq)(const struct sk_buff *skb);
1847 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1848 struct flowi *fl, struct request_sock *req,
1849 struct tcp_fastopen_cookie *foc,
1850 enum tcp_synack_type synack_type);
1853 #ifdef CONFIG_SYN_COOKIES
1854 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1855 const struct sock *sk, struct sk_buff *skb,
1858 tcp_synq_overflow(sk);
1859 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1860 return ops->cookie_init_seq(skb, mss);
1863 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1864 const struct sock *sk, struct sk_buff *skb,
1871 int tcpv4_offload_init(void);
1873 void tcp_v4_init(void);
1874 void tcp_init(void);
1876 /* tcp_recovery.c */
1878 /* Flags to enable various loss recovery features. See below */
1879 extern int sysctl_tcp_recovery;
1881 /* Use TCP RACK to detect (some) tail and retransmit losses */
1882 #define TCP_RACK_LOST_RETRANS 0x1
1884 extern int tcp_rack_mark_lost(struct sock *sk);
1886 extern void tcp_rack_advance(struct tcp_sock *tp,
1887 const struct skb_mstamp *xmit_time, u8 sacked);
1890 * Save and compile IPv4 options, return a pointer to it
1892 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1894 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1895 struct ip_options_rcu *dopt = NULL;
1898 int opt_size = sizeof(*dopt) + opt->optlen;
1900 dopt = kmalloc(opt_size, GFP_ATOMIC);
1901 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1909 /* locally generated TCP pure ACKs have skb->truesize == 2
1910 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1911 * This is much faster than dissecting the packet to find out.
1912 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1914 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1916 return skb->truesize == 2;
1919 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1924 static inline int tcp_inq(struct sock *sk)
1926 struct tcp_sock *tp = tcp_sk(sk);
1929 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1931 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1933 before(tp->urg_seq, tp->copied_seq) ||
1934 !before(tp->urg_seq, tp->rcv_nxt)) {
1936 answ = tp->rcv_nxt - tp->copied_seq;
1938 /* Subtract 1, if FIN was received */
1939 if (answ && sock_flag(sk, SOCK_DONE))
1942 answ = tp->urg_seq - tp->copied_seq;
1948 int tcp_peek_len(struct socket *sock);
1950 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1954 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1955 tp->segs_in += segs_in;
1956 if (skb->len > tcp_hdrlen(skb))
1957 tp->data_segs_in += segs_in;
1961 * TCP listen path runs lockless.
1962 * We forced "struct sock" to be const qualified to make sure
1963 * we don't modify one of its field by mistake.
1964 * Here, we increment sk_drops which is an atomic_t, so we can safely
1965 * make sock writable again.
1967 static inline void tcp_listendrop(const struct sock *sk)
1969 atomic_inc(&((struct sock *)sk)->sk_drops);
1970 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);