projects/linux·blob master · refs

3097 lines · 91.1 KiB

1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the TCP module.
8 *
9 * Version: @(#)tcp.h 1.0.5 05/23/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <[email protected]>
13 */
14#ifndef _TCP_H
15#define _TCP_H
16
17#define FASTRETRANS_DEBUG 1
18
19#include <linux/list.h>
20#include <linux/tcp.h>
21#include <linux/bug.h>
22#include <linux/slab.h>
23#include <linux/cache.h>
24#include <linux/percpu.h>
25#include <linux/skbuff.h>
26#include <linux/kref.h>
27#include <linux/ktime.h>
28#include <linux/indirect_call_wrapper.h>
29#include <linux/bits.h>
30
31#include <net/inet_connection_sock.h>
32#include <net/inet_timewait_sock.h>
33#include <net/inet_hashtables.h>
34#include <net/checksum.h>
35#include <net/request_sock.h>
36#include <net/sock_reuseport.h>
37#include <net/sock.h>
38#include <net/snmp.h>
39#include <net/ip.h>
40#include <net/tcp_states.h>
41#include <net/tcp_ao.h>
42#include <net/inet_ecn.h>
43#include <net/dst.h>
44#include <net/mptcp.h>
45#include <net/xfrm.h>
46#include <net/secure_seq.h>
47
48#include <linux/seq_file.h>
49#include <linux/memcontrol.h>
50#include <linux/bpf-cgroup.h>
51#include <linux/siphash.h>
52
53extern struct inet_hashinfo tcp_hashinfo;
54
55DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
56int tcp_orphan_count_sum(void);
57
58static inline void tcp_orphan_count_inc(void)
59{
60 this_cpu_inc(tcp_orphan_count);
61}
62
63static inline void tcp_orphan_count_dec(void)
64{
65 this_cpu_dec(tcp_orphan_count);
66}
67
68void tcp_time_wait(struct sock *sk, int state, int timeo);
69
70#define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
71#define MAX_TCP_OPTION_SPACE 40
72#define TCP_MIN_SND_MSS 48
73#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
74
75/*
76 * Never offer a window over 32767 without using window scaling. Some
77 * poor stacks do signed 16bit maths!
78 */
79#define MAX_TCP_WINDOW 32767U
80
81/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
82#define TCP_MIN_MSS 88U
83
84/* The initial MTU to use for probing */
85#define TCP_BASE_MSS 1024
86
87/* probing interval, default to 10 minutes as per RFC4821 */
88#define TCP_PROBE_INTERVAL 600
89
90/* Specify interval when tcp mtu probing will stop */
91#define TCP_PROBE_THRESHOLD 8
92
93/* After receiving this amount of duplicate ACKs fast retransmit starts. */
94#define TCP_FASTRETRANS_THRESH 3
95
96/* Maximal number of ACKs sent quickly to accelerate slow-start. */
97#define TCP_MAX_QUICKACKS 16U
98
99/* Maximal number of window scale according to RFC1323 */
100#define TCP_MAX_WSCALE 14U
101
102/* Default sending frequency of accurate ECN option per RTT */
103#define TCP_ACCECN_OPTION_BEACON 3
104
105/* urg_data states */
106#define TCP_URG_VALID 0x0100
107#define TCP_URG_NOTYET 0x0200
108#define TCP_URG_READ 0x0400
109
110#define TCP_RETR1 3 /*
111 * This is how many retries it does before it
112 * tries to figure out if the gateway is
113 * down. Minimal RFC value is 3; it corresponds
114 * to ~3sec-8min depending on RTO.
115 */
116
117#define TCP_RETR2 15 /*
118 * This should take at least
119 * 90 minutes to time out.
120 * RFC1122 says that the limit is 100 sec.
121 * 15 is ~13-30min depending on RTO.
122 */
123
124#define TCP_SYN_RETRIES 6 /* This is how many retries are done
125 * when active opening a connection.
126 * RFC1122 says the minimum retry MUST
127 * be at least 180secs. Nevertheless
128 * this value is corresponding to
129 * 63secs of retransmission with the
130 * current initial RTO.
131 */
132
133#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
134 * when passive opening a connection.
135 * This is corresponding to 31secs of
136 * retransmission with the current
137 * initial RTO.
138 */
139
140#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
141 * state, about 60 seconds */
142#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
143 /* BSD style FIN_WAIT2 deadlock breaker.
144 * It used to be 3min, new value is 60sec,
145 * to combine FIN-WAIT-2 timeout with
146 * TIME-WAIT timer.
147 */
148#define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
149
150#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
151static_assert((1 << ATO_BITS) > TCP_DELACK_MAX);
152
153#if HZ >= 100
154#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
155#define TCP_ATO_MIN ((unsigned)(HZ/25))
156#else
157#define TCP_DELACK_MIN 4U
158#define TCP_ATO_MIN 4U
159#endif
160#define TCP_RTO_MAX_SEC 120
161#define TCP_RTO_MAX ((unsigned)(TCP_RTO_MAX_SEC * HZ))
162#define TCP_RTO_MIN ((unsigned)(HZ / 5))
163#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
164
165#define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
166
167#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
168#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
169 * used as a fallback RTO for the
170 * initial data transmission if no
171 * valid RTT sample has been acquired,
172 * most likely due to retrans in 3WHS.
173 */
174
175#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
176 * for local resources.
177 */
178#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
179#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
180#define TCP_KEEPALIVE_INTVL (75*HZ)
181
182#define MAX_TCP_KEEPIDLE 32767
183#define MAX_TCP_KEEPINTVL 32767
184#define MAX_TCP_KEEPCNT 127
185#define MAX_TCP_SYNCNT 127
186
187/* Ensure that TCP PAWS checks are relaxed after ~2147 seconds
188 * to avoid overflows. This assumes a clock smaller than 1 Mhz.
189 * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz.
190 */
191#define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC)
192
193#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
194 * after this time. It should be equal
195 * (or greater than) TCP_TIMEWAIT_LEN
196 * to provide reliability equal to one
197 * provided by timewait state.
198 */
199#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
200 * timestamps. It must be less than
201 * minimal timewait lifetime.
202 */
203/*
204 * TCP option
205 */
206
207#define TCPOPT_NOP 1 /* Padding */
208#define TCPOPT_EOL 0 /* End of options */
209#define TCPOPT_MSS 2 /* Segment size negotiating */
210#define TCPOPT_WINDOW 3 /* Window scaling */
211#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
212#define TCPOPT_SACK 5 /* SACK Block */
213#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
214#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
215#define TCPOPT_AO 29 /* Authentication Option (RFC5925) */
216#define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
217#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
218#define TCPOPT_ACCECN0 172 /* 0xAC: Accurate ECN Order 0 */
219#define TCPOPT_ACCECN1 174 /* 0xAE: Accurate ECN Order 1 */
220#define TCPOPT_EXP 254 /* Experimental */
221/* Magic number to be after the option value for sharing TCP
222 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
223 */
224#define TCPOPT_FASTOPEN_MAGIC 0xF989
225#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
226
227/*
228 * TCP option lengths
229 */
230
231#define TCPOLEN_MSS 4
232#define TCPOLEN_WINDOW 3
233#define TCPOLEN_SACK_PERM 2
234#define TCPOLEN_TIMESTAMP 10
235#define TCPOLEN_MD5SIG 18
236#define TCPOLEN_FASTOPEN_BASE 2
237#define TCPOLEN_ACCECN_BASE 2
238#define TCPOLEN_EXP_FASTOPEN_BASE 4
239#define TCPOLEN_EXP_SMC_BASE 6
240
241/* But this is what stacks really send out. */
242#define TCPOLEN_TSTAMP_ALIGNED 12
243#define TCPOLEN_WSCALE_ALIGNED 4
244#define TCPOLEN_SACKPERM_ALIGNED 4
245#define TCPOLEN_SACK_BASE 2
246#define TCPOLEN_SACK_BASE_ALIGNED 4
247#define TCPOLEN_SACK_PERBLOCK 8
248#define TCPOLEN_MD5SIG_ALIGNED 20
249#define TCPOLEN_MSS_ALIGNED 4
250#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
251#define TCPOLEN_ACCECN_PERFIELD 3
252
253/* Maximum number of byte counters in AccECN option + size */
254#define TCP_ACCECN_NUMFIELDS 3
255#define TCP_ACCECN_MAXSIZE (TCPOLEN_ACCECN_BASE + \
256 TCPOLEN_ACCECN_PERFIELD * \
257 TCP_ACCECN_NUMFIELDS)
258#define TCP_ACCECN_SAFETY_SHIFT 1 /* SAFETY_FACTOR in accecn draft */
259
260/* Flags in tp->nonagle */
261#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
262#define TCP_NAGLE_CORK 2 /* Socket is corked */
263#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
264
265/* TCP thin-stream limits */
266#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
267
268/* TCP initial congestion window as per rfc6928 */
269#define TCP_INIT_CWND 10
270
271/* Bit Flags for sysctl_tcp_fastopen */
272#define TFO_CLIENT_ENABLE 1
273#define TFO_SERVER_ENABLE 2
274#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
275
276/* Accept SYN data w/o any cookie option */
277#define TFO_SERVER_COOKIE_NOT_REQD 0x200
278
279/* Force enable TFO on all listeners, i.e., not requiring the
280 * TCP_FASTOPEN socket option.
281 */
282#define TFO_SERVER_WO_SOCKOPT1 0x400
283
284
285/* sysctl variables for tcp */
286extern int sysctl_tcp_max_orphans;
287extern long sysctl_tcp_mem[3];
288
289#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
290#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
291#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
292
293DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
294
295extern struct percpu_counter tcp_sockets_allocated;
296extern unsigned long tcp_memory_pressure;
297
298/* optimized version of sk_under_memory_pressure() for TCP sockets */
299static inline bool tcp_under_memory_pressure(const struct sock *sk)
300{
301 if (mem_cgroup_sk_enabled(sk) &&
302 mem_cgroup_sk_under_memory_pressure(sk))
303 return true;
304
305 if (sk->sk_bypass_prot_mem)
306 return false;
307
308 return READ_ONCE(tcp_memory_pressure);
309}
310/*
311 * The next routines deal with comparing 32 bit unsigned ints
312 * and worry about wraparound (automatic with unsigned arithmetic).
313 */
314
315static inline bool before(__u32 seq1, __u32 seq2)
316{
317 return (__s32)(seq1-seq2) < 0;
318}
319#define after(seq2, seq1) before(seq1, seq2)
320
321/* is s2<=s1<=s3 ? */
322static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
323{
324 return seq3 - seq2 >= seq1 - seq2;
325}
326
327static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
328{
329 sk_wmem_queued_add(sk, -skb->truesize);
330 if (!skb_zcopy_pure(skb))
331 sk_mem_uncharge(sk, skb->truesize);
332 else
333 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
334 __kfree_skb(skb);
335}
336
337void sk_forced_mem_schedule(struct sock *sk, int size);
338
339bool tcp_check_oom(const struct sock *sk, int shift);
340
341
342extern struct proto tcp_prot;
343
344#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
345#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
346#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
347#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
348
349/*
350 * TCP splice context
351 */
352struct tcp_splice_state {
353 struct pipe_inode_info *pipe;
354 size_t len;
355 unsigned int flags;
356};
357
358void tcp_tsq_work_init(void);
359
360int tcp_v4_err(struct sk_buff *skb, u32);
361
362void tcp_shutdown(struct sock *sk, int how);
363
364int tcp_v4_rcv(struct sk_buff *skb);
365
366void tcp_remove_empty_skb(struct sock *sk);
367int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
368int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
369int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
370 size_t size, struct ubuf_info *uarg);
371void tcp_splice_eof(struct socket *sock);
372int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
373int tcp_wmem_schedule(struct sock *sk, int copy);
374void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
375 int size_goal);
376
377void tcp_release_cb(struct sock *sk);
378
379static inline bool tcp_release_cb_cond(struct sock *sk)
380{
381#ifdef CONFIG_INET
382 if (likely(sk->sk_prot->release_cb == tcp_release_cb)) {
383 if (unlikely(smp_load_acquire(&sk->sk_tsq_flags) & TCP_DEFERRED_ALL))
384 tcp_release_cb(sk);
385 return true;
386 }
387#endif
388 return false;
389}
390
391void tcp_write_timer_handler(struct sock *sk);
392void tcp_delack_timer_handler(struct sock *sk);
393int tcp_ioctl(struct sock *sk, int cmd, int *karg);
394enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
395void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
396void tcp_rcvbuf_grow(struct sock *sk, u32 newval);
397void tcp_rcv_space_adjust(struct sock *sk);
398int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
399void tcp_twsk_destructor(struct sock *sk);
400void tcp_twsk_purge(struct list_head *net_exit_list);
401int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
402 unsigned int offset, size_t len);
403ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
404 struct pipe_inode_info *pipe, size_t len,
405 unsigned int flags);
406struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp,
407 bool force_schedule);
408
409static inline void tcp_dec_quickack_mode(struct sock *sk)
410{
411 struct inet_connection_sock *icsk = inet_csk(sk);
412
413 if (icsk->icsk_ack.quick) {
414 /* How many ACKs S/ACKing new data have we sent? */
415 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
416
417 if (pkts >= icsk->icsk_ack.quick) {
418 icsk->icsk_ack.quick = 0;
419 /* Leaving quickack mode we deflate ATO. */
420 icsk->icsk_ack.ato = TCP_ATO_MIN;
421 } else
422 icsk->icsk_ack.quick -= pkts;
423 }
424}
425
426#define TCP_ECN_MODE_RFC3168 BIT(0)
427#define TCP_ECN_QUEUE_CWR BIT(1)
428#define TCP_ECN_DEMAND_CWR BIT(2)
429#define TCP_ECN_SEEN BIT(3)
430#define TCP_ECN_MODE_ACCECN BIT(4)
431
432#define TCP_ECN_DISABLED 0
433#define TCP_ECN_MODE_PENDING (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN)
434#define TCP_ECN_MODE_ANY (TCP_ECN_MODE_RFC3168 | TCP_ECN_MODE_ACCECN)
435
436static inline bool tcp_ecn_mode_any(const struct tcp_sock *tp)
437{
438 return tp->ecn_flags & TCP_ECN_MODE_ANY;
439}
440
441static inline bool tcp_ecn_mode_rfc3168(const struct tcp_sock *tp)
442{
443 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_RFC3168;
444}
445
446static inline bool tcp_ecn_mode_accecn(const struct tcp_sock *tp)
447{
448 return (tp->ecn_flags & TCP_ECN_MODE_ANY) == TCP_ECN_MODE_ACCECN;
449}
450
451static inline bool tcp_ecn_disabled(const struct tcp_sock *tp)
452{
453 return !tcp_ecn_mode_any(tp);
454}
455
456static inline bool tcp_ecn_mode_pending(const struct tcp_sock *tp)
457{
458 return (tp->ecn_flags & TCP_ECN_MODE_PENDING) == TCP_ECN_MODE_PENDING;
459}
460
461static inline void tcp_ecn_mode_set(struct tcp_sock *tp, u8 mode)
462{
463 tp->ecn_flags &= ~TCP_ECN_MODE_ANY;
464 tp->ecn_flags |= mode;
465}
466
467enum tcp_tw_status {
468 TCP_TW_SUCCESS = 0,
469 TCP_TW_RST = 1,
470 TCP_TW_ACK = 2,
471 TCP_TW_SYN = 3,
472 TCP_TW_ACK_OOW = 4
473};
474
475
476enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
477 struct sk_buff *skb,
478 const struct tcphdr *th,
479 u32 *tw_isn,
480 enum skb_drop_reason *drop_reason);
481struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
482 struct request_sock *req, bool fastopen,
483 bool *lost_race, enum skb_drop_reason *drop_reason);
484enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
485 struct sk_buff *skb);
486void tcp_enter_loss(struct sock *sk);
487void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
488void tcp_clear_retrans(struct tcp_sock *tp);
489void tcp_update_pacing_rate(struct sock *sk);
490void tcp_set_rto(struct sock *sk);
491void tcp_update_metrics(struct sock *sk);
492void tcp_init_metrics(struct sock *sk);
493void tcp_metrics_init(void);
494bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
495void __tcp_close(struct sock *sk, long timeout);
496void tcp_close(struct sock *sk, long timeout);
497void tcp_init_sock(struct sock *sk);
498void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
499__poll_t tcp_poll(struct file *file, struct socket *sock,
500 struct poll_table_struct *wait);
501int do_tcp_getsockopt(struct sock *sk, int level,
502 int optname, sockptr_t optval, sockptr_t optlen);
503int tcp_getsockopt(struct sock *sk, int level, int optname,
504 char __user *optval, int __user *optlen);
505bool tcp_bpf_bypass_getsockopt(int level, int optname);
506int do_tcp_setsockopt(struct sock *sk, int level, int optname,
507 sockptr_t optval, unsigned int optlen);
508int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
509 unsigned int optlen);
510void tcp_reset_keepalive_timer(struct sock *sk, unsigned long timeout);
511void tcp_set_keepalive(struct sock *sk, int val);
512void tcp_syn_ack_timeout(const struct request_sock *req);
513int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
514 int flags);
515int tcp_set_rcvlowat(struct sock *sk, int val);
516void tcp_set_rcvbuf(struct sock *sk, int val);
517int tcp_set_window_clamp(struct sock *sk, int val);
518
519static inline void
520tcp_update_recv_tstamps(struct sk_buff *skb,
521 struct scm_timestamping_internal *tss)
522{
523 tss->ts[0] = skb->tstamp;
524 tss->ts[2] = skb_hwtstamps(skb)->hwtstamp;
525}
526
527void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
528 struct scm_timestamping_internal *tss);
529void tcp_data_ready(struct sock *sk);
530#ifdef CONFIG_MMU
531int tcp_mmap(struct file *file, struct socket *sock,
532 struct vm_area_struct *vma);
533#endif
534void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
535 struct tcp_options_received *opt_rx,
536 int estab, struct tcp_fastopen_cookie *foc);
537
538/*
539 * BPF SKB-less helpers
540 */
541u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
542 struct tcphdr *th, u32 *cookie);
543u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
544 struct tcphdr *th, u32 *cookie);
545u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
546u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
547 const struct tcp_request_sock_ops *af_ops,
548 struct sock *sk, struct tcphdr *th);
549/*
550 * TCP v4 functions exported for the inet6 API
551 */
552
553void tcp_v4_mtu_reduced(struct sock *sk);
554void tcp_req_err(struct sock *sk, u32 seq, bool abort);
555void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
556int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
557struct sock *tcp_create_openreq_child(const struct sock *sk,
558 struct request_sock *req,
559 struct sk_buff *skb);
560void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
561struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
562 struct request_sock *req,
563 struct dst_entry *dst,
564 struct request_sock *req_unhash,
565 bool *own_req,
566 void (*opt_child_init)(struct sock *newsk,
567 const struct sock *sk));
568int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
569int tcp_v4_connect(struct sock *sk, struct sockaddr_unsized *uaddr, int addr_len);
570int tcp_connect(struct sock *sk);
571enum tcp_synack_type {
572 TCP_SYNACK_NORMAL,
573 TCP_SYNACK_FASTOPEN,
574 TCP_SYNACK_COOKIE,
575 TCP_SYNACK_RETRANS,
576};
577struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
578 struct request_sock *req,
579 struct tcp_fastopen_cookie *foc,
580 enum tcp_synack_type synack_type,
581 struct sk_buff *syn_skb);
582int tcp_disconnect(struct sock *sk, int flags);
583
584void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
585int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
586void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
587
588/* From syncookies.c */
589struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
590 struct request_sock *req,
591 struct dst_entry *dst);
592int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th);
593struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
594struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
595 struct sock *sk, struct sk_buff *skb,
596 struct tcp_options_received *tcp_opt,
597 int mss, u32 tsoff);
598
599#if IS_ENABLED(CONFIG_BPF)
600struct bpf_tcp_req_attrs {
601 u32 rcv_tsval;
602 u32 rcv_tsecr;
603 u16 mss;
604 u8 rcv_wscale;
605 u8 snd_wscale;
606 u8 ecn_ok;
607 u8 wscale_ok;
608 u8 sack_ok;
609 u8 tstamp_ok;
610 u8 usec_ts_ok;
611 u8 reserved[3];
612};
613#endif
614
615#ifdef CONFIG_SYN_COOKIES
616
617/* Syncookies use a monotonic timer which increments every 60 seconds.
618 * This counter is used both as a hash input and partially encoded into
619 * the cookie value. A cookie is only validated further if the delta
620 * between the current counter value and the encoded one is less than this,
621 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
622 * the counter advances immediately after a cookie is generated).
623 */
624#define MAX_SYNCOOKIE_AGE 2
625#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
626#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
627
628/* syncookies: remember time of last synqueue overflow
629 * But do not dirty this field too often (once per second is enough)
630 * It is racy as we do not hold a lock, but race is very minor.
631 */
632static inline void tcp_synq_overflow(const struct sock *sk)
633{
634 unsigned int last_overflow;
635 unsigned int now = jiffies;
636
637 if (sk->sk_reuseport) {
638 struct sock_reuseport *reuse;
639
640 reuse = rcu_dereference(sk->sk_reuseport_cb);
641 if (likely(reuse)) {
642 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
643 if (!time_between32(now, last_overflow,
644 last_overflow + HZ))
645 WRITE_ONCE(reuse->synq_overflow_ts, now);
646 return;
647 }
648 }
649
650 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
651 if (!time_between32(now, last_overflow, last_overflow + HZ))
652 WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now);
653}
654
655/* syncookies: no recent synqueue overflow on this listening socket? */
656static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
657{
658 unsigned int last_overflow;
659 unsigned int now = jiffies;
660
661 if (sk->sk_reuseport) {
662 struct sock_reuseport *reuse;
663
664 reuse = rcu_dereference(sk->sk_reuseport_cb);
665 if (likely(reuse)) {
666 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
667 return !time_between32(now, last_overflow - HZ,
668 last_overflow +
669 TCP_SYNCOOKIE_VALID);
670 }
671 }
672
673 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
674
675 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
676 * then we're under synflood. However, we have to use
677 * 'last_overflow - HZ' as lower bound. That's because a concurrent
678 * tcp_synq_overflow() could update .ts_recent_stamp after we read
679 * jiffies but before we store .ts_recent_stamp into last_overflow,
680 * which could lead to rejecting a valid syncookie.
681 */
682 return !time_between32(now, last_overflow - HZ,
683 last_overflow + TCP_SYNCOOKIE_VALID);
684}
685
686static inline u32 tcp_cookie_time(void)
687{
688 u64 val = get_jiffies_64();
689
690 do_div(val, TCP_SYNCOOKIE_PERIOD);
691 return val;
692}
693
694/* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
695static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val)
696{
697 if (usec_ts)
698 return div_u64(val, NSEC_PER_USEC);
699
700 return div_u64(val, NSEC_PER_MSEC);
701}
702
703u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
704 u16 *mssp);
705__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
706u64 cookie_init_timestamp(struct request_sock *req, u64 now);
707bool cookie_timestamp_decode(const struct net *net,
708 struct tcp_options_received *opt);
709
710static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst)
711{
712 return READ_ONCE(net->ipv4.sysctl_tcp_ecn) ||
713 dst_feature(dst, RTAX_FEATURE_ECN);
714}
715
716#if IS_ENABLED(CONFIG_BPF)
717static inline bool cookie_bpf_ok(struct sk_buff *skb)
718{
719 return skb->sk;
720}
721
722struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb);
723#else
724static inline bool cookie_bpf_ok(struct sk_buff *skb)
725{
726 return false;
727}
728
729static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk,
730 struct sk_buff *skb)
731{
732 return NULL;
733}
734#endif
735
736/* From net/ipv6/syncookies.c */
737int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th);
738struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
739
740u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
741 const struct tcphdr *th, u16 *mssp);
742__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
743#endif
744/* tcp_output.c */
745
746void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
747void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
748void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
749 int nonagle);
750int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
751int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
752void tcp_retransmit_timer(struct sock *sk);
753void tcp_xmit_retransmit_queue(struct sock *);
754void tcp_simple_retransmit(struct sock *);
755void tcp_enter_recovery(struct sock *sk, bool ece_ack);
756int tcp_trim_head(struct sock *, struct sk_buff *, u32);
757enum tcp_queue {
758 TCP_FRAG_IN_WRITE_QUEUE,
759 TCP_FRAG_IN_RTX_QUEUE,
760};
761int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
762 struct sk_buff *skb, u32 len,
763 unsigned int mss_now, gfp_t gfp);
764
765void tcp_send_probe0(struct sock *);
766int tcp_write_wakeup(struct sock *, int mib);
767void tcp_send_fin(struct sock *sk);
768void tcp_send_active_reset(struct sock *sk, enum sk_rst_reason reason);
769int tcp_send_synack(struct sock *);
770void tcp_push_one(struct sock *, unsigned int mss_now);
771void __tcp_send_ack(struct sock *sk, u32 rcv_nxt, u16 flags);
772void tcp_send_ack(struct sock *sk);
773void tcp_send_delayed_ack(struct sock *sk);
774void tcp_send_loss_probe(struct sock *sk);
775bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
776void tcp_skb_collapse_tstamp(struct sk_buff *skb,
777 const struct sk_buff *next_skb);
778
779/* tcp_input.c */
780void tcp_rearm_rto(struct sock *sk);
781void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
782void tcp_done_with_error(struct sock *sk, int err);
783void tcp_reset(struct sock *sk, struct sk_buff *skb);
784void tcp_fin(struct sock *sk);
785void __tcp_check_space(struct sock *sk);
786static inline void tcp_check_space(struct sock *sk)
787{
788 /* pairs with tcp_poll() */
789 smp_mb();
790
791 if (sk->sk_socket && test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))
792 __tcp_check_space(sk);
793}
794void tcp_sack_compress_send_ack(struct sock *sk);
795
796static inline void tcp_cleanup_skb(struct sk_buff *skb)
797{
798 skb_dst_drop(skb);
799 secpath_reset(skb);
800}
801
802static inline void tcp_add_receive_queue(struct sock *sk, struct sk_buff *skb)
803{
804 DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
805 DEBUG_NET_WARN_ON_ONCE(secpath_exists(skb));
806 __skb_queue_tail(&sk->sk_receive_queue, skb);
807}
808
809/* tcp_timer.c */
810void tcp_init_xmit_timers(struct sock *);
811static inline void tcp_clear_xmit_timers(struct sock *sk)
812{
813 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
814 __sock_put(sk);
815
816 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
817 __sock_put(sk);
818
819 inet_csk_clear_xmit_timers(sk);
820}
821
822unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
823unsigned int tcp_current_mss(struct sock *sk);
824u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
825
826/* Bound MSS / TSO packet size with the half of the window */
827static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
828{
829 int cutoff;
830
831 /* When peer uses tiny windows, there is no use in packetizing
832 * to sub-MSS pieces for the sake of SWS or making sure there
833 * are enough packets in the pipe for fast recovery.
834 *
835 * On the other hand, for extremely large MSS devices, handling
836 * smaller than MSS windows in this way does make sense.
837 */
838 if (tp->max_window > TCP_MSS_DEFAULT)
839 cutoff = (tp->max_window >> 1);
840 else
841 cutoff = tp->max_window;
842
843 if (cutoff && pktsize > cutoff)
844 return max_t(int, cutoff, 68U - tp->tcp_header_len);
845 else
846 return pktsize;
847}
848
849/* tcp.c */
850void tcp_get_info(struct sock *, struct tcp_info *);
851void tcp_rate_check_app_limited(struct sock *sk);
852
853/* Read 'sendfile()'-style from a TCP socket */
854int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
855 sk_read_actor_t recv_actor);
856int tcp_read_sock_noack(struct sock *sk, read_descriptor_t *desc,
857 sk_read_actor_t recv_actor, bool noack,
858 u32 *copied_seq);
859int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
860struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
861void tcp_read_done(struct sock *sk, size_t len);
862
863void tcp_initialize_rcv_mss(struct sock *sk);
864
865int tcp_mtu_to_mss(struct sock *sk, int pmtu);
866int tcp_mss_to_mtu(struct sock *sk, int mss);
867void tcp_mtup_init(struct sock *sk);
868
869static inline unsigned int tcp_rto_max(const struct sock *sk)
870{
871 return READ_ONCE(inet_csk(sk)->icsk_rto_max);
872}
873
874static inline void tcp_bound_rto(struct sock *sk)
875{
876 inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk));
877}
878
879static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
880{
881 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
882}
883
884static inline unsigned long tcp_reqsk_timeout(struct request_sock *req)
885{
886 u64 timeout = (u64)req->timeout << req->num_timeout;
887
888 return (unsigned long)min_t(u64, timeout,
889 tcp_rto_max(req->rsk_listener));
890}
891
892u32 tcp_delack_max(const struct sock *sk);
893
894/* Compute the actual rto_min value */
895static inline u32 tcp_rto_min(const struct sock *sk)
896{
897 const struct dst_entry *dst = __sk_dst_get(sk);
898 u32 rto_min = READ_ONCE(inet_csk(sk)->icsk_rto_min);
899
900 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
901 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
902 return rto_min;
903}
904
905static inline u32 tcp_rto_min_us(const struct sock *sk)
906{
907 return jiffies_to_usecs(tcp_rto_min(sk));
908}
909
910static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
911{
912 return dst_metric_locked(dst, RTAX_CC_ALGO);
913}
914
915/* Minimum RTT in usec. ~0 means not available. */
916static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
917{
918 return minmax_get(&tp->rtt_min);
919}
920
921/* Compute the actual receive window we are currently advertising.
922 * Rcv_nxt can be after the window if our peer push more data
923 * than the offered window.
924 */
925static inline u32 tcp_receive_window(const struct tcp_sock *tp)
926{
927 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
928
929 if (win < 0)
930 win = 0;
931 return (u32) win;
932}
933
934/* Compute the maximum receive window we ever advertised.
935 * Rcv_nxt can be after the window if our peer push more data
936 * than the offered window.
937 */
938static inline u32 tcp_max_receive_window(const struct tcp_sock *tp)
939{
940 s32 win = tp->rcv_mwnd_seq - tp->rcv_nxt;
941
942 if (win < 0)
943 win = 0;
944 return (u32) win;
945}
946
947/* Check if we need to update the maximum receive window sequence number */
948static inline void tcp_update_max_rcv_wnd_seq(struct tcp_sock *tp)
949{
950 u32 wre = tp->rcv_wup + tp->rcv_wnd;
951
952 if (after(wre, tp->rcv_mwnd_seq))
953 tp->rcv_mwnd_seq = wre;
954}
955
956/* Choose a new window, without checks for shrinking, and without
957 * scaling applied to the result. The caller does these things
958 * if necessary. This is a "raw" window selection.
959 */
960u32 __tcp_select_window(struct sock *sk);
961
962void tcp_send_window_probe(struct sock *sk);
963
964/* TCP uses 32bit jiffies to save some space.
965 * Note that this is different from tcp_time_stamp, which
966 * historically has been the same until linux-4.13.
967 */
968#define tcp_jiffies32 ((u32)jiffies)
969
970/*
971 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
972 * It is no longer tied to jiffies, but to 1 ms clock.
973 * Note: double check if you want to use tcp_jiffies32 instead of this.
974 */
975#define TCP_TS_HZ 1000
976
977static inline u64 tcp_clock_ns(void)
978{
979 return ktime_get_ns();
980}
981
982static inline u64 tcp_clock_us(void)
983{
984 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
985}
986
987static inline u64 tcp_clock_ms(void)
988{
989 return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
990}
991
992/* TCP Timestamp included in TS option (RFC 1323) can either use ms
993 * or usec resolution. Each socket carries a flag to select one or other
994 * resolution, as the route attribute could change anytime.
995 * Each flow must stick to initial resolution.
996 */
997static inline u32 tcp_clock_ts(bool usec_ts)
998{
999 return usec_ts ? tcp_clock_us() : tcp_clock_ms();
1000}
1001
1002static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
1003{
1004 return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
1005}
1006
1007static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
1008{
1009 if (tp->tcp_usec_ts)
1010 return tp->tcp_mstamp;
1011 return tcp_time_stamp_ms(tp);
1012}
1013
1014/* Refresh clocks of a TCP socket,
1015 * ensuring monotically increasing values.
1016 */
1017static inline void tcp_mstamp_refresh_inline(struct tcp_sock *tp)
1018{
1019 u64 val = tcp_clock_ns();
1020
1021 tp->tcp_clock_cache = val;
1022 tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC);
1023}
1024void tcp_mstamp_refresh(struct tcp_sock *tp);
1025
1026static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
1027{
1028 return max_t(s64, t1 - t0, 0);
1029}
1030
1031/* provide the departure time in us unit */
1032static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
1033{
1034 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
1035}
1036
1037/* Provide skb TSval in usec or ms unit */
1038static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
1039{
1040 if (usec_ts)
1041 return tcp_skb_timestamp_us(skb);
1042
1043 return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC);
1044}
1045
1046static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
1047{
1048 return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset;
1049}
1050
1051static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
1052{
1053 return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
1054}
1055
1056#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
1057
1058#define TCPHDR_FIN BIT(0)
1059#define TCPHDR_SYN BIT(1)
1060#define TCPHDR_RST BIT(2)
1061#define TCPHDR_PSH BIT(3)
1062#define TCPHDR_ACK BIT(4)
1063#define TCPHDR_URG BIT(5)
1064#define TCPHDR_ECE BIT(6)
1065#define TCPHDR_CWR BIT(7)
1066#define TCPHDR_AE BIT(8)
1067#define TCPHDR_FLAGS_MASK (TCPHDR_FIN | TCPHDR_SYN | TCPHDR_RST | \
1068 TCPHDR_PSH | TCPHDR_ACK | TCPHDR_URG | \
1069 TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE)
1070#define tcp_flags_ntohs(th) (ntohs(*(__be16 *)&tcp_flag_word(th)) & \
1071 TCPHDR_FLAGS_MASK)
1072
1073#define TCPHDR_ACE (TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE)
1074#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
1075#define TCPHDR_SYNACK_ACCECN (TCPHDR_SYN | TCPHDR_ACK | TCPHDR_CWR)
1076
1077#define TCP_ACCECN_CEP_ACE_MASK 0x7
1078#define TCP_ACCECN_ACE_MAX_DELTA 6
1079
1080/* To avoid/detect middlebox interference, not all counters start at 0.
1081 * See draft-ietf-tcpm-accurate-ecn for the latest values.
1082 */
1083#define TCP_ACCECN_CEP_INIT_OFFSET 5
1084#define TCP_ACCECN_E1B_INIT_OFFSET 1
1085#define TCP_ACCECN_E0B_INIT_OFFSET 1
1086#define TCP_ACCECN_CEB_INIT_OFFSET 0
1087
1088/* State flags for sacked in struct tcp_skb_cb */
1089enum tcp_skb_cb_sacked_flags {
1090 TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */
1091 TCPCB_SACKED_RETRANS = (1 << 1), /* SKB retransmitted */
1092 TCPCB_LOST = (1 << 2), /* SKB is lost */
1093 TCPCB_TAGBITS = (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS |
1094 TCPCB_LOST), /* All tag bits */
1095 TCPCB_REPAIRED = (1 << 4), /* SKB repaired (no skb_mstamp_ns) */
1096 TCPCB_EVER_RETRANS = (1 << 7), /* Ever retransmitted frame */
1097 TCPCB_RETRANS = (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS |
1098 TCPCB_REPAIRED),
1099};
1100
1101/* This is what the send packet queuing engine uses to pass
1102 * TCP per-packet control information to the transmission code.
1103 * We also store the host-order sequence numbers in here too.
1104 * This is 44 bytes if IPV6 is enabled.
1105 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
1106 */
1107struct tcp_skb_cb {
1108 __u32 seq; /* Starting sequence number */
1109 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
1110 union {
1111 /* Notes :
1112 * tcp_tw_isn is used in input path only
1113 * (isn chosen by tcp_timewait_state_process())
1114 * tcp_gso_segs/size are used in write queue only,
1115 * cf tcp_skb_pcount()/tcp_skb_mss()
1116 */
1117 u32 tcp_tw_isn;
1118 struct {
1119 u16 tcp_gso_segs;
1120 u16 tcp_gso_size;
1121 };
1122 };
1123 __u16 tcp_flags; /* TCP header flags (tcp[12-13])*/
1124
1125 __u8 sacked; /* State flags for SACK. */
1126 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
1127#define TSTAMP_ACK_SK 0x1
1128#define TSTAMP_ACK_BPF 0x2
1129 __u8 txstamp_ack:2, /* Record TX timestamp for ack? */
1130 eor:1, /* Is skb MSG_EOR marked? */
1131 has_rxtstamp:1, /* SKB has a RX timestamp */
1132 unused:4;
1133 __u32 ack_seq; /* Sequence number ACK'd */
1134 union {
1135 struct {
1136#define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
1137 /* There is space for up to 24 bytes */
1138 __u32 is_app_limited:1, /* cwnd not fully used? */
1139 delivered_ce:20,
1140 unused:11;
1141 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
1142 __u32 delivered;
1143 /* start of send pipeline phase */
1144 u64 first_tx_mstamp;
1145 /* when we reached the "delivered" count */
1146 u64 delivered_mstamp;
1147 } tx; /* only used for outgoing skbs */
1148 union {
1149 struct inet_skb_parm h4;
1150#if IS_ENABLED(CONFIG_IPV6)
1151 struct inet6_skb_parm h6;
1152#endif
1153 } header; /* For incoming skbs */
1154 };
1155};
1156
1157#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
1158
1159extern const struct inet_connection_sock_af_ops ipv4_specific;
1160
1161#if IS_ENABLED(CONFIG_IPV6)
1162/* This is the variant of inet6_iif() that must be used by TCP,
1163 * as TCP moves IP6CB into a different location in skb->cb[]
1164 */
1165static inline int tcp_v6_iif(const struct sk_buff *skb)
1166{
1167 return TCP_SKB_CB(skb)->header.h6.iif;
1168}
1169
1170static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
1171{
1172 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
1173
1174 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
1175}
1176
1177/* TCP_SKB_CB reference means this can not be used from early demux */
1178static inline int tcp_v6_sdif(const struct sk_buff *skb)
1179{
1180#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1181 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
1182 return TCP_SKB_CB(skb)->header.h6.iif;
1183#endif
1184 return 0;
1185}
1186
1187extern const struct inet_connection_sock_af_ops ipv6_specific;
1188
1189INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
1190
1191#endif
1192
1193/* TCP_SKB_CB reference means this can not be used from early demux */
1194static inline int tcp_v4_sdif(struct sk_buff *skb)
1195{
1196#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1197 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
1198 return TCP_SKB_CB(skb)->header.h4.iif;
1199#endif
1200 return 0;
1201}
1202
1203/* Due to TSO, an SKB can be composed of multiple actual
1204 * packets. To keep these tracked properly, we use this.
1205 */
1206static inline int tcp_skb_pcount(const struct sk_buff *skb)
1207{
1208 return TCP_SKB_CB(skb)->tcp_gso_segs;
1209}
1210
1211static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
1212{
1213 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
1214}
1215
1216static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
1217{
1218 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
1219}
1220
1221/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
1222static inline int tcp_skb_mss(const struct sk_buff *skb)
1223{
1224 return TCP_SKB_CB(skb)->tcp_gso_size;
1225}
1226
1227static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
1228{
1229 return likely(!TCP_SKB_CB(skb)->eor);
1230}
1231
1232static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
1233 const struct sk_buff *from)
1234{
1235 return likely(tcp_skb_can_collapse_to(to) &&
1236 mptcp_skb_can_collapse(to, from) &&
1237 !skb_cmp_decrypted(to, from) &&
1238 skb_pure_zcopy_same(to, from) &&
1239 skb_frags_readable(to) == skb_frags_readable(from));
1240}
1241
1242static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to,
1243 const struct sk_buff *from)
1244{
1245 return likely(mptcp_skb_can_collapse(to, from) &&
1246 !skb_cmp_decrypted(to, from));
1247}
1248
1249/* Events passed to congestion control interface */
1250enum tcp_ca_event {
1251 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1252 CA_EVENT_CWND_RESTART, /* congestion window restart */
1253 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1254 CA_EVENT_LOSS, /* loss timeout */
1255 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
1256 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1257};
1258
1259/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1260enum tcp_ca_ack_event_flags {
1261 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1262 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1263 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1264};
1265
1266/*
1267 * Interface for adding new TCP congestion control handlers
1268 */
1269#define TCP_CA_NAME_MAX 16
1270#define TCP_CA_MAX 128
1271#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1272
1273#define TCP_CA_UNSPEC 0
1274
1275/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1276#define TCP_CONG_NON_RESTRICTED BIT(0)
1277/* Requires ECN/ECT set on all packets */
1278#define TCP_CONG_NEEDS_ECN BIT(1)
1279/* Require successfully negotiated AccECN capability */
1280#define TCP_CONG_NEEDS_ACCECN BIT(2)
1281/* Use ECT(1) instead of ECT(0) while the CA is uninitialized */
1282#define TCP_CONG_ECT_1_NEGOTIATION BIT(3)
1283/* Cannot fallback to RFC3168 during AccECN negotiation */
1284#define TCP_CONG_NO_FALLBACK_RFC3168 BIT(4)
1285#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN | \
1286 TCP_CONG_NEEDS_ACCECN | TCP_CONG_ECT_1_NEGOTIATION | \
1287 TCP_CONG_NO_FALLBACK_RFC3168)
1288
1289union tcp_cc_info;
1290
1291struct ack_sample {
1292 u32 pkts_acked;
1293 s32 rtt_us;
1294 u32 in_flight;
1295};
1296
1297/* A rate sample measures the number of (original/retransmitted) data
1298 * packets delivered "delivered" over an interval of time "interval_us".
1299 * The tcp_rate.c code fills in the rate sample, and congestion
1300 * control modules that define a cong_control function to run at the end
1301 * of ACK processing can optionally chose to consult this sample when
1302 * setting cwnd and pacing rate.
1303 * A sample is invalid if "delivered" or "interval_us" is negative.
1304 */
1305struct rate_sample {
1306 u64 prior_mstamp; /* starting timestamp for interval */
1307 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1308 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1309 s32 delivered; /* number of packets delivered over interval */
1310 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1311 long interval_us; /* time for tp->delivered to incr "delivered" */
1312 u32 snd_interval_us; /* snd interval for delivered packets */
1313 u32 rcv_interval_us; /* rcv interval for delivered packets */
1314 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1315 int losses; /* number of packets marked lost upon ACK */
1316 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1317 u32 prior_in_flight; /* in flight before this ACK */
1318 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1319 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1320 bool is_retrans; /* is sample from retransmission? */
1321 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1322};
1323
1324struct tcp_congestion_ops {
1325/* fast path fields are put first to fill one cache line */
1326
1327 /* A congestion control (CC) must provide one of either:
1328 *
1329 * (a) a cong_avoid function, if the CC wants to use the core TCP
1330 * stack's default functionality to implement a "classic"
1331 * (Reno/CUBIC-style) response to packet loss, RFC3168 ECN,
1332 * idle periods, pacing rate computations, etc.
1333 *
1334 * (b) a cong_control function, if the CC wants custom behavior and
1335 * complete control of all congestion control behaviors.
1336 */
1337 /* (a) "classic" response: calculate new cwnd.
1338 */
1339 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1340 /* (b) "custom" response: call when packets are delivered to update
1341 * cwnd and pacing rate, after all the ca_state processing.
1342 */
1343 void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs);
1344
1345 /* return slow start threshold (required) */
1346 u32 (*ssthresh)(struct sock *sk);
1347
1348 /* call before changing ca_state (optional) */
1349 void (*set_state)(struct sock *sk, u8 new_state);
1350
1351 /* call when cwnd event occurs (optional) */
1352 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1353
1354 /* call when CA_EVENT_TX_START cwnd event occurs (optional) */
1355 void (*cwnd_event_tx_start)(struct sock *sk);
1356
1357 /* call when ack arrives (optional) */
1358 void (*in_ack_event)(struct sock *sk, u32 flags);
1359
1360 /* hook for packet ack accounting (optional) */
1361 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1362
1363 /* override sysctl_tcp_min_tso_segs (optional) */
1364 u32 (*min_tso_segs)(struct sock *sk);
1365
1366 /* new value of cwnd after loss (required) */
1367 u32 (*undo_cwnd)(struct sock *sk);
1368 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1369 u32 (*sndbuf_expand)(struct sock *sk);
1370
1371/* control/slow paths put last */
1372 /* get info for inet_diag (optional) */
1373 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1374 union tcp_cc_info *info);
1375
1376 char name[TCP_CA_NAME_MAX];
1377 struct module *owner;
1378 struct list_head list;
1379 u32 key;
1380 u32 flags;
1381
1382 /* initialize private data (optional) */
1383 void (*init)(struct sock *sk);
1384 /* cleanup private data (optional) */
1385 void (*release)(struct sock *sk);
1386} ____cacheline_aligned_in_smp;
1387
1388int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1389void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1390int tcp_update_congestion_control(struct tcp_congestion_ops *type,
1391 struct tcp_congestion_ops *old_type);
1392int tcp_validate_congestion_control(struct tcp_congestion_ops *ca);
1393
1394void tcp_assign_congestion_control(struct sock *sk);
1395void tcp_init_congestion_control(struct sock *sk);
1396void tcp_cleanup_congestion_control(struct sock *sk);
1397int tcp_set_default_congestion_control(struct net *net, const char *name);
1398void tcp_get_default_congestion_control(struct net *net, char *name);
1399void tcp_get_available_congestion_control(char *buf, size_t len);
1400void tcp_get_allowed_congestion_control(char *buf, size_t len);
1401int tcp_set_allowed_congestion_control(char *allowed);
1402int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1403 bool cap_net_admin);
1404u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1405void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1406
1407u32 tcp_reno_ssthresh(struct sock *sk);
1408u32 tcp_reno_undo_cwnd(struct sock *sk);
1409void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1410extern struct tcp_congestion_ops tcp_reno;
1411
1412struct tcp_congestion_ops *tcp_ca_find(const char *name);
1413struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1414u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1415#ifdef CONFIG_INET
1416char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1417#else
1418static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1419{
1420 return NULL;
1421}
1422#endif
1423
1424static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1425{
1426 const struct inet_connection_sock *icsk = inet_csk(sk);
1427
1428 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1429}
1430
1431static inline bool tcp_ca_needs_accecn(const struct sock *sk)
1432{
1433 const struct inet_connection_sock *icsk = inet_csk(sk);
1434
1435 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ACCECN;
1436}
1437
1438static inline bool tcp_ca_ect_1_negotiation(const struct sock *sk)
1439{
1440 const struct inet_connection_sock *icsk = inet_csk(sk);
1441
1442 return icsk->icsk_ca_ops->flags & TCP_CONG_ECT_1_NEGOTIATION;
1443}
1444
1445static inline bool tcp_ca_no_fallback_rfc3168(const struct sock *sk)
1446{
1447 const struct inet_connection_sock *icsk = inet_csk(sk);
1448
1449 return icsk->icsk_ca_ops->flags & TCP_CONG_NO_FALLBACK_RFC3168;
1450}
1451
1452static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1453{
1454 const struct inet_connection_sock *icsk = inet_csk(sk);
1455
1456 if (event == CA_EVENT_TX_START) {
1457 if (icsk->icsk_ca_ops->cwnd_event_tx_start)
1458 icsk->icsk_ca_ops->cwnd_event_tx_start(sk);
1459 return;
1460 }
1461 if (icsk->icsk_ca_ops->cwnd_event)
1462 icsk->icsk_ca_ops->cwnd_event(sk, event);
1463}
1464
1465/* From tcp_cong.c */
1466void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1467
1468
1469static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1470{
1471 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1472}
1473
1474/* These functions determine how the current flow behaves in respect of SACK
1475 * handling. SACK is negotiated with the peer, and therefore it can vary
1476 * between different flows.
1477 *
1478 * tcp_is_sack - SACK enabled
1479 * tcp_is_reno - No SACK
1480 */
1481static inline int tcp_is_sack(const struct tcp_sock *tp)
1482{
1483 return likely(tp->rx_opt.sack_ok);
1484}
1485
1486static inline bool tcp_is_reno(const struct tcp_sock *tp)
1487{
1488 return !tcp_is_sack(tp);
1489}
1490
1491static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1492{
1493 return tp->sacked_out + tp->lost_out;
1494}
1495
1496/* This determines how many packets are "in the network" to the best
1497 * of our knowledge. In many cases it is conservative, but where
1498 * detailed information is available from the receiver (via SACK
1499 * blocks etc.) we can make more aggressive calculations.
1500 *
1501 * Use this for decisions involving congestion control, use just
1502 * tp->packets_out to determine if the send queue is empty or not.
1503 *
1504 * Read this equation as:
1505 *
1506 * "Packets sent once on transmission queue" MINUS
1507 * "Packets left network, but not honestly ACKed yet" PLUS
1508 * "Packets fast retransmitted"
1509 */
1510static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1511{
1512 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1513}
1514
1515#define TCP_INFINITE_SSTHRESH 0x7fffffff
1516
1517static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1518{
1519 return tp->snd_cwnd;
1520}
1521
1522static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1523{
1524 WARN_ON_ONCE((int)val <= 0);
1525 WRITE_ONCE(tp->snd_cwnd, val);
1526}
1527
1528static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1529{
1530 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1531}
1532
1533static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1534{
1535 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1536}
1537
1538static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1539{
1540 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1541 (1 << inet_csk(sk)->icsk_ca_state);
1542}
1543
1544/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1545 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1546 * ssthresh.
1547 */
1548static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1549{
1550 const struct tcp_sock *tp = tcp_sk(sk);
1551
1552 if (tcp_in_cwnd_reduction(sk))
1553 return tp->snd_ssthresh;
1554 else
1555 return max(tp->snd_ssthresh,
1556 ((tcp_snd_cwnd(tp) >> 1) +
1557 (tcp_snd_cwnd(tp) >> 2)));
1558}
1559
1560/* Use define here intentionally to get WARN_ON location shown at the caller */
1561#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1562
1563void tcp_enter_cwr(struct sock *sk);
1564__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1565
1566/* The maximum number of MSS of available cwnd for which TSO defers
1567 * sending if not using sysctl_tcp_tso_win_divisor.
1568 */
1569static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1570{
1571 return 3;
1572}
1573
1574/* Returns end sequence number of the receiver's advertised window */
1575static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1576{
1577 return tp->snd_una + tp->snd_wnd;
1578}
1579
1580/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1581 * flexible approach. The RFC suggests cwnd should not be raised unless
1582 * it was fully used previously. And that's exactly what we do in
1583 * congestion avoidance mode. But in slow start we allow cwnd to grow
1584 * as long as the application has used half the cwnd.
1585 * Example :
1586 * cwnd is 10 (IW10), but application sends 9 frames.
1587 * We allow cwnd to reach 18 when all frames are ACKed.
1588 * This check is safe because it's as aggressive as slow start which already
1589 * risks 100% overshoot. The advantage is that we discourage application to
1590 * either send more filler packets or data to artificially blow up the cwnd
1591 * usage, and allow application-limited process to probe bw more aggressively.
1592 */
1593static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1594{
1595 const struct tcp_sock *tp = tcp_sk(sk);
1596
1597 if (tp->is_cwnd_limited)
1598 return true;
1599
1600 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1601 if (tcp_in_slow_start(tp))
1602 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1603
1604 return false;
1605}
1606
1607/* BBR congestion control needs pacing.
1608 * Same remark for SO_MAX_PACING_RATE.
1609 * sch_fq packet scheduler is efficiently handling pacing,
1610 * but is not always installed/used.
1611 * Return true if TCP stack should pace packets itself.
1612 */
1613static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1614{
1615 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1616}
1617
1618/* Estimates in how many jiffies next packet for this flow can be sent.
1619 * Scheduling a retransmit timer too early would be silly.
1620 */
1621static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1622{
1623 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1624
1625 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1626}
1627
1628static inline void tcp_reset_xmit_timer(struct sock *sk,
1629 const int what,
1630 unsigned long when,
1631 bool pace_delay)
1632{
1633 if (pace_delay)
1634 when += tcp_pacing_delay(sk);
1635 inet_csk_reset_xmit_timer(sk, what, when,
1636 tcp_rto_max(sk));
1637}
1638
1639/* Something is really bad, we could not queue an additional packet,
1640 * because qdisc is full or receiver sent a 0 window, or we are paced.
1641 * We do not want to add fuel to the fire, or abort too early,
1642 * so make sure the timer we arm now is at least 200ms in the future,
1643 * regardless of current icsk_rto value (as it could be ~2ms)
1644 */
1645static inline unsigned long tcp_probe0_base(const struct sock *sk)
1646{
1647 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1648}
1649
1650/* Variant of inet_csk_rto_backoff() used for zero window probes */
1651static inline unsigned long tcp_probe0_when(const struct sock *sk,
1652 unsigned long max_when)
1653{
1654 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1655 inet_csk(sk)->icsk_backoff);
1656 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1657
1658 return (unsigned long)min_t(u64, when, max_when);
1659}
1660
1661static inline void tcp_check_probe_timer(struct sock *sk)
1662{
1663 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1664 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1665 tcp_probe0_base(sk), true);
1666}
1667
1668static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1669{
1670 tp->snd_wl1 = seq;
1671}
1672
1673static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1674{
1675 tp->snd_wl1 = seq;
1676}
1677
1678/*
1679 * Calculate(/check) TCP checksum
1680 */
1681static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1682 __be32 daddr, __wsum base)
1683{
1684 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1685}
1686
1687static inline bool tcp_checksum_complete(struct sk_buff *skb)
1688{
1689 return !skb_csum_unnecessary(skb) &&
1690 __skb_checksum_complete(skb);
1691}
1692
1693enum skb_drop_reason tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1694
1695static inline enum skb_drop_reason
1696tcp_filter(struct sock *sk, struct sk_buff *skb)
1697{
1698 const struct tcphdr *th = (const struct tcphdr *)skb->data;
1699
1700 return sk_filter_trim_cap(sk, skb, __tcp_hdrlen(th));
1701}
1702
1703void tcp_set_state(struct sock *sk, int state);
1704void tcp_done(struct sock *sk);
1705int tcp_abort(struct sock *sk, int err);
1706
1707static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1708{
1709 rx_opt->dsack = 0;
1710 rx_opt->num_sacks = 0;
1711}
1712
1713void tcp_cwnd_restart(struct sock *sk, s32 delta);
1714
1715static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1716{
1717 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1718 struct tcp_sock *tp = tcp_sk(sk);
1719 s32 delta;
1720
1721 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1722 tp->packets_out || ca_ops->cong_control)
1723 return;
1724 delta = tcp_jiffies32 - tp->lsndtime;
1725 if (delta > inet_csk(sk)->icsk_rto)
1726 tcp_cwnd_restart(sk, delta);
1727}
1728
1729/* Determine a window scaling and initial window to offer. */
1730void tcp_select_initial_window(const struct sock *sk, int __space,
1731 __u32 mss, __u32 *rcv_wnd,
1732 __u32 *window_clamp, int wscale_ok,
1733 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1734
1735static inline int __tcp_win_from_space(u8 scaling_ratio, int space)
1736{
1737 s64 scaled_space = (s64)space * scaling_ratio;
1738
1739 return scaled_space >> TCP_RMEM_TO_WIN_SCALE;
1740}
1741
1742static inline int tcp_win_from_space(const struct sock *sk, int space)
1743{
1744 return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space);
1745}
1746
1747/* inverse of __tcp_win_from_space() */
1748static inline int __tcp_space_from_win(u8 scaling_ratio, int win)
1749{
1750 u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE;
1751
1752 do_div(val, scaling_ratio);
1753 return val;
1754}
1755
1756static inline int tcp_space_from_win(const struct sock *sk, int win)
1757{
1758 return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win);
1759}
1760
1761/* Assume a 50% default for skb->len/skb->truesize ratio.
1762 * This may be adjusted later in tcp_measure_rcv_mss().
1763 */
1764#define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1))
1765
1766static inline void tcp_scaling_ratio_init(struct sock *sk)
1767{
1768 tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
1769}
1770
1771/* Note: caller must be prepared to deal with negative returns */
1772static inline int tcp_space(const struct sock *sk)
1773{
1774 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1775 READ_ONCE(sk->sk_backlog.len) -
1776 atomic_read(&sk->sk_rmem_alloc));
1777}
1778
1779static inline int tcp_full_space(const struct sock *sk)
1780{
1781 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1782}
1783
1784static inline u32 tcp_dst_advmss(const struct dst_entry *dst)
1785{
1786 return max_t(u32, dst_metric_advmss(dst), TCP_MIN_MSS);
1787}
1788
1789static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh)
1790{
1791 int unused_mem = sk_unused_reserved_mem(sk);
1792 struct tcp_sock *tp = tcp_sk(sk);
1793
1794 tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh);
1795 if (unused_mem)
1796 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1797 tcp_win_from_space(sk, unused_mem));
1798}
1799
1800static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1801{
1802 __tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss);
1803}
1804
1805void tcp_cleanup_rbuf(struct sock *sk, int copied);
1806void __tcp_cleanup_rbuf(struct sock *sk, int copied);
1807
1808
1809/* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1810 * If 87.5 % (7/8) of the space has been consumed, we want to override
1811 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1812 * len/truesize ratio.
1813 */
1814static inline bool tcp_rmem_pressure(const struct sock *sk)
1815{
1816 int rcvbuf, threshold;
1817
1818 if (tcp_under_memory_pressure(sk))
1819 return true;
1820
1821 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1822 threshold = rcvbuf - (rcvbuf >> 3);
1823
1824 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1825}
1826
1827static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1828{
1829 const struct tcp_sock *tp = tcp_sk(sk);
1830 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1831
1832 if (avail <= 0)
1833 return false;
1834
1835 return (avail >= target) || tcp_rmem_pressure(sk) ||
1836 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1837}
1838
1839extern void tcp_openreq_init_rwin(struct request_sock *req,
1840 const struct sock *sk_listener,
1841 const struct dst_entry *dst);
1842
1843void tcp_enter_memory_pressure(struct sock *sk);
1844void tcp_leave_memory_pressure(struct sock *sk);
1845
1846static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1847{
1848 struct net *net = sock_net((struct sock *)tp);
1849 int val;
1850
1851 /* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl()
1852 * and do_tcp_setsockopt().
1853 */
1854 val = READ_ONCE(tp->keepalive_intvl);
1855
1856 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1857}
1858
1859static inline int keepalive_time_when(const struct tcp_sock *tp)
1860{
1861 struct net *net = sock_net((struct sock *)tp);
1862 int val;
1863
1864 /* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */
1865 val = READ_ONCE(tp->keepalive_time);
1866
1867 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1868}
1869
1870static inline int keepalive_probes(const struct tcp_sock *tp)
1871{
1872 struct net *net = sock_net((struct sock *)tp);
1873 int val;
1874
1875 /* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt()
1876 * and do_tcp_setsockopt().
1877 */
1878 val = READ_ONCE(tp->keepalive_probes);
1879
1880 return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1881}
1882
1883static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1884{
1885 const struct inet_connection_sock *icsk = &tp->inet_conn;
1886
1887 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1888 tcp_jiffies32 - tp->rcv_tstamp);
1889}
1890
1891static inline int tcp_fin_time(const struct sock *sk)
1892{
1893 int fin_timeout = tcp_sk(sk)->linger2 ? :
1894 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1895 const int rto = inet_csk(sk)->icsk_rto;
1896
1897 if (fin_timeout < (rto << 2) - (rto >> 1))
1898 fin_timeout = (rto << 2) - (rto >> 1);
1899
1900 return fin_timeout;
1901}
1902
1903static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1904 int paws_win)
1905{
1906 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1907 return true;
1908 if (unlikely(!time_before32(ktime_get_seconds(),
1909 rx_opt->ts_recent_stamp + TCP_PAWS_WRAP)))
1910 return true;
1911 /*
1912 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1913 * then following tcp messages have valid values. Ignore 0 value,
1914 * or else 'negative' tsval might forbid us to accept their packets.
1915 */
1916 if (!rx_opt->ts_recent)
1917 return true;
1918 return false;
1919}
1920
1921static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1922 int rst)
1923{
1924 if (tcp_paws_check(rx_opt, 0))
1925 return false;
1926
1927 /* RST segments are not recommended to carry timestamp,
1928 and, if they do, it is recommended to ignore PAWS because
1929 "their cleanup function should take precedence over timestamps."
1930 Certainly, it is mistake. It is necessary to understand the reasons
1931 of this constraint to relax it: if peer reboots, clock may go
1932 out-of-sync and half-open connections will not be reset.
1933 Actually, the problem would be not existing if all
1934 the implementations followed draft about maintaining clock
1935 via reboots. Linux-2.2 DOES NOT!
1936
1937 However, we can relax time bounds for RST segments to MSL.
1938 */
1939 if (rst && !time_before32(ktime_get_seconds(),
1940 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1941 return false;
1942 return true;
1943}
1944
1945static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1946{
1947 u32 ace;
1948
1949 /* mptcp hooks are only on the slow path */
1950 if (sk_is_mptcp((struct sock *)tp))
1951 return;
1952
1953 ace = tcp_ecn_mode_accecn(tp) ?
1954 ((tp->delivered_ce + TCP_ACCECN_CEP_INIT_OFFSET) &
1955 TCP_ACCECN_CEP_ACE_MASK) : 0;
1956
1957 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
1958 (ace << 22) |
1959 ntohl(TCP_FLAG_ACK) |
1960 snd_wnd);
1961}
1962
1963static inline void tcp_fast_path_on(struct tcp_sock *tp)
1964{
1965 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
1966}
1967
1968static inline void tcp_fast_path_check(struct sock *sk)
1969{
1970 struct tcp_sock *tp = tcp_sk(sk);
1971
1972 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
1973 tp->rcv_wnd &&
1974 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
1975 !tp->urg_data)
1976 tcp_fast_path_on(tp);
1977}
1978
1979bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1980 int mib_idx, u32 *last_oow_ack_time);
1981void tcp_reqsk_send_challenge_ack(struct sock *sk, struct sk_buff *skb,
1982 struct request_sock *req);
1983
1984static inline void tcp_mib_init(struct net *net)
1985{
1986 /* See RFC 2012 */
1987 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1988 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1989 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1990 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1991}
1992
1993/* from STCP */
1994static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1995{
1996 tp->retransmit_skb_hint = NULL;
1997}
1998
1999#define tcp_md5_addr tcp_ao_addr
2000
2001/* - key database */
2002struct tcp_md5sig_key {
2003 struct hlist_node node;
2004 u8 keylen;
2005 u8 family; /* AF_INET or AF_INET6 */
2006 u8 prefixlen;
2007 u8 flags;
2008 union tcp_md5_addr addr;
2009 int l3index; /* set if key added with L3 scope */
2010 u8 key[TCP_MD5SIG_MAXKEYLEN];
2011 struct rcu_head rcu;
2012};
2013
2014/* - sock block */
2015struct tcp_md5sig_info {
2016 struct hlist_head head;
2017 struct rcu_head rcu;
2018};
2019
2020/* - pseudo header */
2021struct tcp4_pseudohdr {
2022 __be32 saddr;
2023 __be32 daddr;
2024 __u8 pad;
2025 __u8 protocol;
2026 __be16 len;
2027};
2028
2029struct tcp6_pseudohdr {
2030 struct in6_addr saddr;
2031 struct in6_addr daddr;
2032 __be32 len;
2033 __be32 protocol; /* including padding */
2034};
2035
2036void tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
2037 const struct sock *sk, const struct sk_buff *skb);
2038int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
2039 int family, u8 prefixlen, int l3index, u8 flags,
2040 const u8 *newkey, u8 newkeylen);
2041int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
2042 int family, u8 prefixlen, int l3index,
2043 struct tcp_md5sig_key *key);
2044
2045int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
2046 int family, u8 prefixlen, int l3index, u8 flags);
2047void tcp_clear_md5_list(struct sock *sk);
2048struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
2049 const struct sock *addr_sk);
2050
2051#ifdef CONFIG_TCP_MD5SIG
2052struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
2053 const union tcp_md5_addr *addr,
2054 int family, bool any_l3index);
2055static inline struct tcp_md5sig_key *
2056tcp_md5_do_lookup(const struct sock *sk, int l3index,
2057 const union tcp_md5_addr *addr, int family)
2058{
2059 if (!static_branch_unlikely(&tcp_md5_needed.key))
2060 return NULL;
2061 return __tcp_md5_do_lookup(sk, l3index, addr, family, false);
2062}
2063
2064static inline struct tcp_md5sig_key *
2065tcp_md5_do_lookup_any_l3index(const struct sock *sk,
2066 const union tcp_md5_addr *addr, int family)
2067{
2068 if (!static_branch_unlikely(&tcp_md5_needed.key))
2069 return NULL;
2070 return __tcp_md5_do_lookup(sk, 0, addr, family, true);
2071}
2072
2073#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
2074void tcp_md5_destruct_sock(struct sock *sk);
2075#else
2076static inline struct tcp_md5sig_key *
2077tcp_md5_do_lookup(const struct sock *sk, int l3index,
2078 const union tcp_md5_addr *addr, int family)
2079{
2080 return NULL;
2081}
2082
2083static inline struct tcp_md5sig_key *
2084tcp_md5_do_lookup_any_l3index(const struct sock *sk,
2085 const union tcp_md5_addr *addr, int family)
2086{
2087 return NULL;
2088}
2089
2090#define tcp_twsk_md5_key(twsk) NULL
2091static inline void tcp_md5_destruct_sock(struct sock *sk)
2092{
2093}
2094#endif
2095
2096struct md5_ctx;
2097void tcp_md5_hash_skb_data(struct md5_ctx *ctx, const struct sk_buff *skb,
2098 unsigned int header_len);
2099void tcp_md5_hash_key(struct md5_ctx *ctx, const struct tcp_md5sig_key *key);
2100
2101/* From tcp_fastopen.c */
2102void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
2103 struct tcp_fastopen_cookie *cookie);
2104void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
2105 struct tcp_fastopen_cookie *cookie, bool syn_lost,
2106 u16 try_exp);
2107struct tcp_fastopen_request {
2108 /* Fast Open cookie. Size 0 means a cookie request */
2109 struct tcp_fastopen_cookie cookie;
2110 struct msghdr *data; /* data in MSG_FASTOPEN */
2111 size_t size;
2112 int copied; /* queued in tcp_connect() */
2113 struct ubuf_info *uarg;
2114};
2115void tcp_free_fastopen_req(struct tcp_sock *tp);
2116void tcp_fastopen_destroy_cipher(struct sock *sk);
2117void tcp_fastopen_ctx_destroy(struct net *net);
2118int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
2119 void *primary_key, void *backup_key);
2120int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
2121 u64 *key);
2122void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
2123struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
2124 struct request_sock *req,
2125 struct tcp_fastopen_cookie *foc,
2126 const struct dst_entry *dst);
2127void tcp_fastopen_init_key_once(struct net *net);
2128bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
2129 struct tcp_fastopen_cookie *cookie);
2130bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
2131#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
2132#define TCP_FASTOPEN_KEY_MAX 2
2133#define TCP_FASTOPEN_KEY_BUF_LENGTH \
2134 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
2135
2136/* Fastopen key context */
2137struct tcp_fastopen_context {
2138 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
2139 int num;
2140 struct rcu_head rcu;
2141};
2142
2143void tcp_fastopen_active_disable(struct sock *sk);
2144bool tcp_fastopen_active_should_disable(struct sock *sk);
2145void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
2146void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
2147
2148/* Caller needs to wrap with rcu_read_(un)lock() */
2149static inline
2150struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
2151{
2152 struct tcp_fastopen_context *ctx;
2153
2154 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
2155 if (!ctx)
2156 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
2157 return ctx;
2158}
2159
2160static inline
2161bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
2162 const struct tcp_fastopen_cookie *orig)
2163{
2164 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
2165 orig->len == foc->len &&
2166 !memcmp(orig->val, foc->val, foc->len))
2167 return true;
2168 return false;
2169}
2170
2171static inline
2172int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
2173{
2174 return ctx->num;
2175}
2176
2177/* Latencies incurred by various limits for a sender. They are
2178 * chronograph-like stats that are mutually exclusive.
2179 */
2180enum tcp_chrono {
2181 TCP_CHRONO_UNSPEC,
2182 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
2183 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
2184 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
2185 __TCP_CHRONO_MAX,
2186};
2187
2188static inline void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new)
2189{
2190 const u32 now = tcp_jiffies32;
2191 enum tcp_chrono old = tp->chrono_type;
2192
2193 /* Following WRITE_ONCE()s pair with READ_ONCE()s in
2194 * tcp_get_info_chrono_stats().
2195 */
2196 if (old > TCP_CHRONO_UNSPEC)
2197 WRITE_ONCE(tp->chrono_stat[old - 1],
2198 tp->chrono_stat[old - 1] + now - tp->chrono_start);
2199 WRITE_ONCE(tp->chrono_start, now);
2200 WRITE_ONCE(tp->chrono_type, new);
2201}
2202
2203static inline void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type)
2204{
2205 struct tcp_sock *tp = tcp_sk(sk);
2206
2207 /* If there are multiple conditions worthy of tracking in a
2208 * chronograph then the highest priority enum takes precedence
2209 * over the other conditions. So that if something "more interesting"
2210 * starts happening, stop the previous chrono and start a new one.
2211 */
2212 if (type > tp->chrono_type)
2213 tcp_chrono_set(tp, type);
2214}
2215
2216void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
2217
2218/* This helper is needed, because skb->tcp_tsorted_anchor uses
2219 * the same memory storage than skb->destructor/_skb_refdst
2220 */
2221static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
2222{
2223 skb->destructor = NULL;
2224 skb->_skb_refdst = 0UL;
2225}
2226
2227#define tcp_skb_tsorted_save(skb) { \
2228 unsigned long _save = skb->_skb_refdst; \
2229 skb->_skb_refdst = 0UL;
2230
2231#define tcp_skb_tsorted_restore(skb) \
2232 skb->_skb_refdst = _save; \
2233}
2234
2235void tcp_write_queue_purge(struct sock *sk);
2236
2237static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
2238{
2239 return skb_rb_first(&sk->tcp_rtx_queue);
2240}
2241
2242static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
2243{
2244 return skb_rb_last(&sk->tcp_rtx_queue);
2245}
2246
2247static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
2248{
2249 return skb_peek_tail(&sk->sk_write_queue);
2250}
2251
2252#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
2253 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
2254
2255static inline struct sk_buff *tcp_send_head(const struct sock *sk)
2256{
2257 return skb_peek(&sk->sk_write_queue);
2258}
2259
2260static inline bool tcp_skb_is_last(const struct sock *sk,
2261 const struct sk_buff *skb)
2262{
2263 return skb_queue_is_last(&sk->sk_write_queue, skb);
2264}
2265
2266/**
2267 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
2268 * @sk: socket
2269 *
2270 * Since the write queue can have a temporary empty skb in it,
2271 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
2272 */
2273static inline bool tcp_write_queue_empty(const struct sock *sk)
2274{
2275 const struct tcp_sock *tp = tcp_sk(sk);
2276
2277 return tp->write_seq == tp->snd_nxt;
2278}
2279
2280static inline bool tcp_rtx_queue_empty(const struct sock *sk)
2281{
2282 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
2283}
2284
2285static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
2286{
2287 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
2288}
2289
2290static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
2291{
2292 __skb_queue_tail(&sk->sk_write_queue, skb);
2293
2294 /* Queue it, remembering where we must start sending. */
2295 if (sk->sk_write_queue.next == skb)
2296 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
2297}
2298
2299/* Insert new before skb on the write queue of sk. */
2300static inline void tcp_insert_write_queue_before(struct sk_buff *new,
2301 struct sk_buff *skb,
2302 struct sock *sk)
2303{
2304 __skb_queue_before(&sk->sk_write_queue, skb, new);
2305}
2306
2307static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
2308{
2309 tcp_skb_tsorted_anchor_cleanup(skb);
2310 __skb_unlink(skb, &sk->sk_write_queue);
2311}
2312
2313void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
2314
2315static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
2316{
2317 tcp_skb_tsorted_anchor_cleanup(skb);
2318 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
2319}
2320
2321static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
2322{
2323 list_del(&skb->tcp_tsorted_anchor);
2324 tcp_rtx_queue_unlink(skb, sk);
2325 tcp_wmem_free_skb(sk, skb);
2326}
2327
2328static inline void tcp_write_collapse_fence(struct sock *sk)
2329{
2330 struct sk_buff *skb = tcp_write_queue_tail(sk) ?: tcp_rtx_queue_tail(sk);
2331
2332 if (skb)
2333 TCP_SKB_CB(skb)->eor = 1;
2334}
2335
2336static inline void tcp_push_pending_frames(struct sock *sk)
2337{
2338 if (tcp_send_head(sk)) {
2339 struct tcp_sock *tp = tcp_sk(sk);
2340
2341 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
2342 }
2343}
2344
2345/* Start sequence of the skb just after the highest skb with SACKed
2346 * bit, valid only if sacked_out > 0 or when the caller has ensured
2347 * validity by itself.
2348 */
2349static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
2350{
2351 if (!tp->sacked_out)
2352 return tp->snd_una;
2353
2354 if (tp->highest_sack == NULL)
2355 return tp->snd_nxt;
2356
2357 return TCP_SKB_CB(tp->highest_sack)->seq;
2358}
2359
2360static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
2361{
2362 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
2363}
2364
2365static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
2366{
2367 return tcp_sk(sk)->highest_sack;
2368}
2369
2370static inline void tcp_highest_sack_reset(struct sock *sk)
2371{
2372 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
2373}
2374
2375/* Called when old skb is about to be deleted and replaced by new skb */
2376static inline void tcp_highest_sack_replace(struct sock *sk,
2377 struct sk_buff *old,
2378 struct sk_buff *new)
2379{
2380 if (old == tcp_highest_sack(sk))
2381 tcp_sk(sk)->highest_sack = new;
2382}
2383
2384/* This helper checks if socket has IP_TRANSPARENT set */
2385static inline bool inet_sk_transparent(const struct sock *sk)
2386{
2387 switch (sk->sk_state) {
2388 case TCP_TIME_WAIT:
2389 return inet_twsk(sk)->tw_transparent;
2390 case TCP_NEW_SYN_RECV:
2391 return inet_rsk(inet_reqsk(sk))->no_srccheck;
2392 }
2393 return inet_test_bit(TRANSPARENT, sk);
2394}
2395
2396/* Determines whether this is a thin stream (which may suffer from
2397 * increased latency). Used to trigger latency-reducing mechanisms.
2398 */
2399static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2400{
2401 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2402}
2403
2404/* /proc */
2405enum tcp_seq_states {
2406 TCP_SEQ_STATE_LISTENING,
2407 TCP_SEQ_STATE_ESTABLISHED,
2408};
2409
2410void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2411void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2412void tcp_seq_stop(struct seq_file *seq, void *v);
2413
2414struct tcp_seq_afinfo {
2415 sa_family_t family;
2416};
2417
2418struct tcp_iter_state {
2419 struct seq_net_private p;
2420 enum tcp_seq_states state;
2421 struct sock *syn_wait_sk;
2422 int bucket, offset, sbucket, num;
2423 loff_t last_pos;
2424};
2425
2426extern struct request_sock_ops tcp_request_sock_ops;
2427extern struct request_sock_ops tcp6_request_sock_ops;
2428
2429void tcp_v4_destroy_sock(struct sock *sk);
2430
2431struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2432 netdev_features_t features);
2433struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th);
2434struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb,
2435 struct tcphdr *th);
2436INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2437INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2438#ifdef CONFIG_INET
2439void tcp_gro_complete(struct sk_buff *skb);
2440#else
2441static inline void tcp_gro_complete(struct sk_buff *skb) { }
2442#endif
2443
2444static inline void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr,
2445 __be32 daddr)
2446{
2447 struct tcphdr *th = tcp_hdr(skb);
2448
2449 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
2450 skb->csum_start = skb_transport_header(skb) - skb->head;
2451 skb->csum_offset = offsetof(struct tcphdr, check);
2452}
2453
2454static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2455{
2456 struct net *net = sock_net((struct sock *)tp);
2457 u32 val;
2458
2459 val = READ_ONCE(tp->notsent_lowat);
2460
2461 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2462}
2463
2464bool tcp_stream_memory_free(const struct sock *sk, int wake);
2465
2466#ifdef CONFIG_PROC_FS
2467int tcp4_proc_init(void);
2468void tcp4_proc_exit(void);
2469#endif
2470
2471int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2472int tcp_conn_request(struct request_sock_ops *rsk_ops,
2473 const struct tcp_request_sock_ops *af_ops,
2474 struct sock *sk, struct sk_buff *skb);
2475
2476/* TCP af-specific functions */
2477struct tcp_sock_af_ops {
2478#ifdef CONFIG_TCP_MD5SIG
2479 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2480 const struct sock *addr_sk);
2481 void (*calc_md5_hash)(char *location,
2482 const struct tcp_md5sig_key *md5,
2483 const struct sock *sk,
2484 const struct sk_buff *skb);
2485 int (*md5_parse)(struct sock *sk,
2486 int optname,
2487 sockptr_t optval,
2488 int optlen);
2489#endif
2490#ifdef CONFIG_TCP_AO
2491 int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen);
2492 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2493 struct sock *addr_sk,
2494 int sndid, int rcvid);
2495 void (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key,
2496 const struct sock *sk,
2497 __be32 sisn, __be32 disn, bool send);
2498 int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao,
2499 const struct sock *sk, const struct sk_buff *skb,
2500 const u8 *tkey, int hash_offset, u32 sne);
2501#endif
2502};
2503
2504struct tcp_request_sock_ops {
2505 u16 mss_clamp;
2506#ifdef CONFIG_TCP_MD5SIG
2507 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2508 const struct sock *addr_sk);
2509 void (*calc_md5_hash) (char *location,
2510 const struct tcp_md5sig_key *md5,
2511 const struct sock *sk,
2512 const struct sk_buff *skb);
2513#endif
2514#ifdef CONFIG_TCP_AO
2515 struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2516 struct request_sock *req,
2517 int sndid, int rcvid);
2518 void (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk);
2519 int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt,
2520 struct request_sock *req, const struct sk_buff *skb,
2521 int hash_offset, u32 sne);
2522#endif
2523#ifdef CONFIG_SYN_COOKIES
2524 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2525 __u16 *mss);
2526#endif
2527 struct dst_entry *(*route_req)(const struct sock *sk,
2528 struct sk_buff *skb,
2529 struct flowi *fl,
2530 struct request_sock *req,
2531 u32 tw_isn);
2532 union tcp_seq_and_ts_off (*init_seq_and_ts_off)(
2533 const struct net *net,
2534 const struct sk_buff *skb);
2535 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2536 struct flowi *fl, struct request_sock *req,
2537 struct tcp_fastopen_cookie *foc,
2538 enum tcp_synack_type synack_type,
2539 struct sk_buff *syn_skb);
2540};
2541
2542extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2543#if IS_ENABLED(CONFIG_IPV6)
2544extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2545#endif
2546
2547#ifdef CONFIG_SYN_COOKIES
2548static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2549 struct sk_buff *skb, __u16 *mss)
2550{
2551 return ops->cookie_init_seq(skb, mss);
2552}
2553#else
2554static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2555 struct sk_buff *skb, __u16 *mss)
2556{
2557 return 0;
2558}
2559#endif
2560
2561#ifdef CONFIG_SYN_COOKIES
2562static inline void cookie_record_sent(const struct sock *sk)
2563{
2564 tcp_synq_overflow(sk);
2565 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2566}
2567#else
2568static inline void cookie_record_sent(const struct sock *sk)
2569{
2570}
2571#endif
2572
2573struct tcp_key {
2574 union {
2575 struct {
2576 struct tcp_ao_key *ao_key;
2577 char *traffic_key;
2578 u32 sne;
2579 u8 rcv_next;
2580 };
2581 struct tcp_md5sig_key *md5_key;
2582 };
2583 enum {
2584 TCP_KEY_NONE = 0,
2585 TCP_KEY_MD5,
2586 TCP_KEY_AO,
2587 } type;
2588};
2589
2590static inline void tcp_get_current_key(const struct sock *sk,
2591 struct tcp_key *out)
2592{
2593#if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG)
2594 const struct tcp_sock *tp = tcp_sk(sk);
2595#endif
2596
2597#ifdef CONFIG_TCP_AO
2598 if (static_branch_unlikely(&tcp_ao_needed.key)) {
2599 struct tcp_ao_info *ao;
2600
2601 ao = rcu_dereference_protected(tp->ao_info,
2602 lockdep_sock_is_held(sk));
2603 if (ao) {
2604 out->ao_key = READ_ONCE(ao->current_key);
2605 out->type = TCP_KEY_AO;
2606 return;
2607 }
2608 }
2609#endif
2610#ifdef CONFIG_TCP_MD5SIG
2611 if (static_branch_unlikely(&tcp_md5_needed.key) &&
2612 rcu_access_pointer(tp->md5sig_info)) {
2613 out->md5_key = tp->af_specific->md5_lookup(sk, sk);
2614 if (out->md5_key) {
2615 out->type = TCP_KEY_MD5;
2616 return;
2617 }
2618 }
2619#endif
2620 out->type = TCP_KEY_NONE;
2621}
2622
2623static inline bool tcp_key_is_md5(const struct tcp_key *key)
2624{
2625 if (static_branch_tcp_md5())
2626 return key->type == TCP_KEY_MD5;
2627 return false;
2628}
2629
2630static inline bool tcp_key_is_ao(const struct tcp_key *key)
2631{
2632 if (static_branch_tcp_ao())
2633 return key->type == TCP_KEY_AO;
2634 return false;
2635}
2636
2637int tcpv4_offload_init(void);
2638
2639void tcp_v4_init(void);
2640void tcp_init(void);
2641
2642/* tcp_recovery.c */
2643void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2644void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2645extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2646 u32 reo_wnd);
2647extern bool tcp_rack_mark_lost(struct sock *sk);
2648extern void tcp_rack_reo_timeout(struct sock *sk);
2649
2650/* tcp_plb.c */
2651
2652/*
2653 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2654 * expects cong_ratio which represents fraction of traffic that experienced
2655 * congestion over a single RTT. In order to avoid floating point operations,
2656 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2657 */
2658#define TCP_PLB_SCALE 8
2659
2660/* State for PLB (Protective Load Balancing) for a single TCP connection. */
2661struct tcp_plb_state {
2662 u8 consec_cong_rounds:5, /* consecutive congested rounds */
2663 unused:3;
2664 u32 pause_until; /* jiffies32 when PLB can resume rerouting */
2665};
2666
2667static inline void tcp_plb_init(const struct sock *sk,
2668 struct tcp_plb_state *plb)
2669{
2670 plb->consec_cong_rounds = 0;
2671 plb->pause_until = 0;
2672}
2673void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2674 const int cong_ratio);
2675void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2676void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2677
2678static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str)
2679{
2680 WARN_ONCE(cond,
2681 "%scwn:%u out:%u sacked:%u lost:%u retrans:%u tlp_high_seq:%u sk_state:%u ca_state:%u advmss:%u mss_cache:%u pmtu:%u\n",
2682 str,
2683 tcp_snd_cwnd(tcp_sk(sk)),
2684 tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out,
2685 tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out,
2686 tcp_sk(sk)->tlp_high_seq, sk->sk_state,
2687 inet_csk(sk)->icsk_ca_state,
2688 tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache,
2689 inet_csk(sk)->icsk_pmtu_cookie);
2690}
2691
2692/* At how many usecs into the future should the RTO fire? */
2693static inline s64 tcp_rto_delta_us(const struct sock *sk)
2694{
2695 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2696 u32 rto = inet_csk(sk)->icsk_rto;
2697
2698 if (likely(skb)) {
2699 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2700
2701 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2702 } else {
2703 tcp_warn_once(sk, 1, "rtx queue empty: ");
2704 return jiffies_to_usecs(rto);
2705 }
2706
2707}
2708
2709/*
2710 * Save and compile IPv4 options, return a pointer to it
2711 */
2712static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2713 struct sk_buff *skb)
2714{
2715 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2716 struct ip_options_rcu *dopt = NULL;
2717
2718 if (opt->optlen) {
2719 int opt_size = sizeof(*dopt) + opt->optlen;
2720
2721 dopt = kmalloc(opt_size, GFP_ATOMIC);
2722 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2723 kfree(dopt);
2724 dopt = NULL;
2725 }
2726 }
2727 return dopt;
2728}
2729
2730/* locally generated TCP pure ACKs have skb->truesize == 2
2731 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2732 * This is much faster than dissecting the packet to find out.
2733 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2734 */
2735static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2736{
2737 return skb->truesize == 2;
2738}
2739
2740static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2741{
2742 skb->truesize = 2;
2743}
2744
2745static inline int tcp_inq(struct sock *sk)
2746{
2747 struct tcp_sock *tp = tcp_sk(sk);
2748 int answ;
2749
2750 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2751 answ = 0;
2752 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2753 !tp->urg_data ||
2754 before(tp->urg_seq, tp->copied_seq) ||
2755 !before(tp->urg_seq, tp->rcv_nxt)) {
2756
2757 answ = tp->rcv_nxt - tp->copied_seq;
2758
2759 /* Subtract 1, if FIN was received */
2760 if (answ && sock_flag(sk, SOCK_DONE))
2761 answ--;
2762 } else {
2763 answ = tp->urg_seq - tp->copied_seq;
2764 }
2765
2766 return answ;
2767}
2768
2769int tcp_peek_len(struct socket *sock);
2770
2771static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2772{
2773 u16 segs_in;
2774
2775 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2776
2777 /* We update these fields while other threads might
2778 * read them from tcp_get_info()
2779 */
2780 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2781 if (skb->len > tcp_hdrlen(skb))
2782 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2783}
2784
2785/*
2786 * TCP listen path runs lockless.
2787 * We forced "struct sock" to be const qualified to make sure
2788 * we don't modify one of its field by mistake.
2789 * Here, we increment sk_drops which is an atomic_t, so we can safely
2790 * make sock writable again.
2791 */
2792static inline void tcp_listendrop(const struct sock *sk)
2793{
2794 sk_drops_inc((struct sock *)sk);
2795 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2796}
2797
2798enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2799
2800/*
2801 * Interface for adding Upper Level Protocols over TCP
2802 */
2803
2804#define TCP_ULP_NAME_MAX 16
2805#define TCP_ULP_MAX 128
2806#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2807
2808struct tcp_ulp_ops {
2809 struct list_head list;
2810
2811 /* initialize ulp */
2812 int (*init)(struct sock *sk);
2813 /* update ulp */
2814 void (*update)(struct sock *sk, struct proto *p,
2815 void (*write_space)(struct sock *sk));
2816 /* cleanup ulp */
2817 void (*release)(struct sock *sk);
2818 /* diagnostic */
2819 int (*get_info)(struct sock *sk, struct sk_buff *skb, bool net_admin);
2820 size_t (*get_info_size)(const struct sock *sk, bool net_admin);
2821 /* clone ulp */
2822 void (*clone)(const struct request_sock *req, struct sock *newsk,
2823 const gfp_t priority);
2824
2825 char name[TCP_ULP_NAME_MAX];
2826 struct module *owner;
2827};
2828int tcp_register_ulp(struct tcp_ulp_ops *type);
2829void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2830int tcp_set_ulp(struct sock *sk, const char *name);
2831void tcp_get_available_ulp(char *buf, size_t len);
2832void tcp_cleanup_ulp(struct sock *sk);
2833void tcp_update_ulp(struct sock *sk, struct proto *p,
2834 void (*write_space)(struct sock *sk));
2835
2836#define MODULE_ALIAS_TCP_ULP(name) \
2837 MODULE_INFO(alias, name); \
2838 MODULE_INFO(alias, "tcp-ulp-" name)
2839
2840#ifdef CONFIG_NET_SOCK_MSG
2841struct sk_msg;
2842struct sk_psock;
2843
2844#ifdef CONFIG_BPF_SYSCALL
2845int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2846void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2847#ifdef CONFIG_BPF_STREAM_PARSER
2848struct strparser;
2849int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc,
2850 sk_read_actor_t recv_actor);
2851#endif /* CONFIG_BPF_STREAM_PARSER */
2852#endif /* CONFIG_BPF_SYSCALL */
2853
2854#ifdef CONFIG_INET
2855void tcp_eat_skb(struct sock *sk, struct sk_buff *skb);
2856#else
2857static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
2858{
2859}
2860#endif
2861
2862int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2863 struct sk_msg *msg, u32 bytes, int flags);
2864#endif /* CONFIG_NET_SOCK_MSG */
2865
2866#if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2867static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2868{
2869}
2870#endif
2871
2872#ifdef CONFIG_CGROUP_BPF
2873static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2874 struct sk_buff *skb,
2875 unsigned int end_offset)
2876{
2877 skops->skb = skb;
2878 skops->skb_data_end = skb->data + end_offset;
2879}
2880#else
2881static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2882 struct sk_buff *skb,
2883 unsigned int end_offset)
2884{
2885}
2886#endif
2887
2888/* Call BPF_SOCK_OPS program that returns an int. If the return value
2889 * is < 0, then the BPF op failed (for example if the loaded BPF
2890 * program does not support the chosen operation or there is no BPF
2891 * program loaded).
2892 */
2893#ifdef CONFIG_BPF
2894static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2895{
2896 struct bpf_sock_ops_kern sock_ops;
2897 int ret;
2898
2899 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2900 if (sk_fullsock(sk)) {
2901 sock_ops.is_fullsock = 1;
2902 sock_ops.is_locked_tcp_sock = 1;
2903 sock_owned_by_me(sk);
2904 }
2905
2906 sock_ops.sk = sk;
2907 sock_ops.op = op;
2908 if (nargs > 0)
2909 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2910
2911 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2912 if (ret == 0)
2913 ret = sock_ops.reply;
2914 else
2915 ret = -1;
2916 return ret;
2917}
2918
2919static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2920{
2921 u32 args[2] = {arg1, arg2};
2922
2923 return tcp_call_bpf(sk, op, 2, args);
2924}
2925
2926static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2927 u32 arg3)
2928{
2929 u32 args[3] = {arg1, arg2, arg3};
2930
2931 return tcp_call_bpf(sk, op, 3, args);
2932}
2933
2934static inline void tcp_clear_sock_ops_cb_flags(struct sock *sk)
2935{
2936 tcp_sk(sk)->bpf_sock_ops_cb_flags = 0;
2937}
2938
2939#else
2940static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2941{
2942 return -EPERM;
2943}
2944
2945static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2946{
2947 return -EPERM;
2948}
2949
2950static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2951 u32 arg3)
2952{
2953 return -EPERM;
2954}
2955
2956static inline void tcp_clear_sock_ops_cb_flags(struct sock *sk)
2957{
2958}
2959
2960#endif
2961
2962static inline u32 tcp_timeout_init(struct sock *sk)
2963{
2964 int timeout;
2965
2966 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2967
2968 if (timeout <= 0)
2969 timeout = TCP_TIMEOUT_INIT;
2970 return min_t(int, timeout, TCP_RTO_MAX);
2971}
2972
2973static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2974{
2975 int rwnd;
2976
2977 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2978
2979 if (rwnd < 0)
2980 rwnd = 0;
2981 return rwnd;
2982}
2983
2984static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2985{
2986 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2987}
2988
2989static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt)
2990{
2991 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2992 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt);
2993}
2994
2995#if IS_ENABLED(CONFIG_SMC)
2996extern struct static_key_false tcp_have_smc;
2997#endif
2998
2999#if IS_ENABLED(CONFIG_TLS_DEVICE)
3000void clean_acked_data_enable(struct tcp_sock *tp,
3001 void (*cad)(struct sock *sk, u32 ack_seq));
3002void clean_acked_data_disable(struct tcp_sock *tp);
3003void clean_acked_data_flush(void);
3004#endif
3005
3006DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
3007static inline void tcp_add_tx_delay(struct sk_buff *skb,
3008 const struct tcp_sock *tp)
3009{
3010 if (static_branch_unlikely(&tcp_tx_delay_enabled))
3011 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
3012}
3013
3014/* Compute Earliest Departure Time for some control packets
3015 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
3016 */
3017static inline u64 tcp_transmit_time(const struct sock *sk)
3018{
3019 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
3020 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
3021 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
3022
3023 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
3024 }
3025 return 0;
3026}
3027
3028static inline int tcp_parse_auth_options(const struct tcphdr *th,
3029 const u8 **md5_hash, const struct tcp_ao_hdr **aoh)
3030{
3031 const u8 *md5_tmp, *ao_tmp;
3032 int ret;
3033
3034 ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp);
3035 if (ret)
3036 return ret;
3037
3038 if (md5_hash)
3039 *md5_hash = md5_tmp;
3040
3041 if (aoh) {
3042 if (!ao_tmp)
3043 *aoh = NULL;
3044 else
3045 *aoh = (struct tcp_ao_hdr *)(ao_tmp - 2);
3046 }
3047
3048 return 0;
3049}
3050
3051static inline bool tcp_ao_required(struct sock *sk, const void *saddr,
3052 int family, int l3index, bool stat_inc)
3053{
3054#ifdef CONFIG_TCP_AO
3055 struct tcp_ao_info *ao_info;
3056 struct tcp_ao_key *ao_key;
3057
3058 if (!static_branch_unlikely(&tcp_ao_needed.key))
3059 return false;
3060
3061 ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info,
3062 lockdep_sock_is_held(sk));
3063 if (!ao_info)
3064 return false;
3065
3066 ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1);
3067 if (ao_info->ao_required || ao_key) {
3068 if (stat_inc) {
3069 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED);
3070 atomic64_inc(&ao_info->counters.ao_required);
3071 }
3072 return true;
3073 }
3074#endif
3075 return false;
3076}
3077
3078enum skb_drop_reason tcp_inbound_hash(struct sock *sk,
3079 const struct request_sock *req, const struct sk_buff *skb,
3080 const void *saddr, const void *daddr,
3081 int family, int dif, int sdif);
3082
3083static inline int tcp_recv_should_stop(struct sock *sk)
3084{
3085 return sk->sk_err ||
3086 sk->sk_state == TCP_CLOSE ||
3087 (sk->sk_shutdown & RCV_SHUTDOWN) ||
3088 signal_pending(current);
3089}
3090
3091INDIRECT_CALLABLE_DECLARE(union tcp_seq_and_ts_off
3092 tcp_v4_init_seq_and_ts_off(const struct net *net,
3093 const struct sk_buff *skb));
3094INDIRECT_CALLABLE_DECLARE(union tcp_seq_and_ts_off
3095 tcp_v6_init_seq_and_ts_off(const struct net *net,
3096 const struct sk_buff *skb));
3097#endif /* _TCP_H */