root/net/ipv4/tcp_input.c

DEFINITIONS

1. clean_acked_data_enable
2. clean_acked_data_disable
3. clean_acked_data_flush
4. tcp_gro_dev_warn
5. tcp_measure_rcv_mss
6. tcp_incr_quickack
7. tcp_enter_quickack_mode
8. tcp_in_quickack_mode
9. tcp_ecn_queue_cwr
10. tcp_ecn_accept_cwr
11. tcp_ecn_withdraw_cwr
12. __tcp_ecn_check_ce
13. tcp_ecn_check_ce
14. tcp_ecn_rcv_synack
15. tcp_ecn_rcv_syn
16. tcp_ecn_rcv_ecn_echo
17. tcp_sndbuf_expand
18. __tcp_grow_window
19. tcp_grow_window
20. tcp_init_buffer_space
21. tcp_clamp_window
22. tcp_initialize_rcv_mss
23. tcp_rcv_rtt_update
24. tcp_rcv_rtt_measure
25. tcp_rcv_rtt_measure_ts
26. tcp_rcv_space_adjust
27. tcp_event_data_recv
28. tcp_rtt_estimator
29. tcp_update_pacing_rate
30. tcp_set_rto
31. tcp_init_cwnd
32. tcp_dsack_seen
33. tcp_check_sack_reordering
34. tcp_verify_retransmit_hint
35. tcp_sum_lost
36. tcp_skb_mark_lost
37. tcp_skb_mark_lost_uncond_verify
38. tcp_is_sackblock_valid
39. tcp_check_dsack
40. tcp_match_skb_to_sack
41. tcp_sacktag_one
42. tcp_shifted_skb
43. tcp_skb_seglen
44. skb_can_shift
45. tcp_skb_shift
46. tcp_shift_skb_data
47. tcp_sacktag_walk
48. tcp_sacktag_bsearch
49. tcp_sacktag_skip
50. tcp_maybe_skipping_dsack
51. tcp_sack_cache_ok
52. tcp_sacktag_write_queue
53. tcp_limit_reno_sacked
54. tcp_check_reno_reordering
55. tcp_add_reno_sack
56. tcp_remove_reno_sacks
57. tcp_reset_reno_sack
58. tcp_clear_retrans
59. tcp_init_undo
60. tcp_is_rack
61. tcp_timeout_mark_lost
62. tcp_enter_loss
63. tcp_check_sack_reneging
64. tcp_dupack_heuristics
65. tcp_time_to_recover
66. tcp_mark_head_lost
67. tcp_update_scoreboard
68. tcp_tsopt_ecr_before
69. tcp_skb_spurious_retrans
70. tcp_packet_delayed
71. tcp_any_retrans_done
72. DBGUNDO
73. tcp_undo_cwnd_reduction
74. tcp_may_undo
75. tcp_try_undo_recovery
76. tcp_try_undo_dsack
77. tcp_try_undo_loss
78. tcp_init_cwnd_reduction
79. tcp_cwnd_reduction
80. tcp_end_cwnd_reduction
81. tcp_enter_cwr
82. tcp_try_keep_open
83. tcp_try_to_open
84. tcp_mtup_probe_failed
85. tcp_mtup_probe_success
86. tcp_simple_retransmit
87. tcp_enter_recovery
88. tcp_process_loss
89. tcp_try_undo_partial
90. tcp_identify_packet_loss
91. tcp_force_fast_retransmit
92. tcp_fastretrans_alert
93. tcp_update_rtt_min
94. tcp_ack_update_rtt
95. tcp_synack_rtt_meas
96. tcp_cong_avoid
97. tcp_rearm_rto
98. tcp_set_xmit_timer
99. tcp_tso_acked
100. tcp_ack_tstamp
101. tcp_clean_rtx_queue
102. tcp_ack_probe
103. tcp_ack_is_dubious
104. tcp_may_raise_cwnd
105. tcp_cong_control
106. tcp_may_update_window
107. tcp_snd_una_update
108. tcp_rcv_nxt_update
109. tcp_ack_update_window
110. __tcp_oow_rate_limited
111. tcp_oow_rate_limited
112. tcp_send_challenge_ack
113. tcp_store_ts_recent
114. tcp_replace_ts_recent
115. tcp_process_tlp_ack
116. tcp_in_ack_event
117. tcp_xmit_recovery
118. tcp_newly_delivered
119. tcp_ack
120. tcp_parse_fastopen_option
121. smc_parse_options
122. tcp_parse_mss_option
123. tcp_parse_options
124. tcp_parse_aligned_timestamp
125. tcp_fast_parse_options
126. tcp_parse_md5sig_option
127. tcp_disordered_ack
128. tcp_paws_discard
129. tcp_sequence
130. tcp_reset
131. tcp_fin
132. tcp_sack_extend
133. tcp_dsack_set
134. tcp_dsack_extend
135. tcp_rcv_spurious_retrans
136. tcp_send_dupack
137. tcp_sack_maybe_coalesce
138. tcp_sack_new_ofo_skb
139. tcp_sack_remove
140. tcp_try_coalesce
141. tcp_ooo_try_coalesce
142. tcp_drop
143. tcp_ofo_queue
144. tcp_try_rmem_schedule
145. tcp_data_queue_ofo
146. tcp_queue_rcv
147. tcp_send_rcvq
148. tcp_data_ready
149. tcp_data_queue
150. tcp_skb_next
151. tcp_collapse_one
152. tcp_rbtree_insert
153. tcp_collapse
154. tcp_collapse_ofo_queue
155. tcp_prune_ofo_queue
156. tcp_prune_queue
157. tcp_should_expand_sndbuf
158. tcp_new_space
159. tcp_check_space
160. tcp_data_snd_check
161. __tcp_ack_snd_check
162. tcp_ack_snd_check
163. tcp_check_urg
164. tcp_urg
165. tcp_reset_check
166. tcp_validate_incoming
167. tcp_rcv_established
168. tcp_init_transfer
169. tcp_finish_connect
170. tcp_rcv_fastopen_synack
171. smc_check_reset_syn
172. tcp_try_undo_spurious_syn
173. tcp_rcv_synsent_state_process
174. tcp_rcv_synrecv_state_fastopen
175. tcp_rcv_state_process
176. pr_drop_req
177. tcp_ecn_create_request
178. tcp_openreq_init
179. inet_reqsk_alloc
180. tcp_syn_flood_action
181. tcp_reqsk_record_syn
182. tcp_get_syncookie_mss
183. tcp_conn_request

   1 // SPDX-License-Identifier: GPL-2.0
   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  *              Implementation of the Transmission Control Protocol(TCP).
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Mark Evans, <evansmp@uhura.aston.ac.uk>
  12  *              Corey Minyard <wf-rch!minyard@relay.EU.net>
  13  *              Florian La Roche, <flla@stud.uni-sb.de>
  14  *              Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
  15  *              Linus Torvalds, <torvalds@cs.helsinki.fi>
  16  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  17  *              Matthew Dillon, <dillon@apollo.west.oic.com>
  18  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  19  *              Jorge Cwik, <jorge@laser.satlink.net>
  20  */
  21 
  22 /*
  23  * Changes:
  24  *              Pedro Roque     :       Fast Retransmit/Recovery.
  25  *                                      Two receive queues.
  26  *                                      Retransmit queue handled by TCP.
  27  *                                      Better retransmit timer handling.
  28  *                                      New congestion avoidance.
  29  *                                      Header prediction.
  30  *                                      Variable renaming.
  31  *
  32  *              Eric            :       Fast Retransmit.
  33  *              Randy Scott     :       MSS option defines.
  34  *              Eric Schenk     :       Fixes to slow start algorithm.
  35  *              Eric Schenk     :       Yet another double ACK bug.
  36  *              Eric Schenk     :       Delayed ACK bug fixes.
  37  *              Eric Schenk     :       Floyd style fast retrans war avoidance.
  38  *              David S. Miller :       Don't allow zero congestion window.
  39  *              Eric Schenk     :       Fix retransmitter so that it sends
  40  *                                      next packet on ack of previous packet.
  41  *              Andi Kleen      :       Moved open_request checking here
  42  *                                      and process RSTs for open_requests.
  43  *              Andi Kleen      :       Better prune_queue, and other fixes.
  44  *              Andrey Savochkin:       Fix RTT measurements in the presence of
  45  *                                      timestamps.
  46  *              Andrey Savochkin:       Check sequence numbers correctly when
  47  *                                      removing SACKs due to in sequence incoming
  48  *                                      data segments.
  49  *              Andi Kleen:             Make sure we never ack data there is not
  50  *                                      enough room for. Also make this condition
  51  *                                      a fatal error if it might still happen.
  52  *              Andi Kleen:             Add tcp_measure_rcv_mss to make
  53  *                                      connections with MSS<min(MTU,ann. MSS)
  54  *                                      work without delayed acks.
  55  *              Andi Kleen:             Process packets with PSH set in the
  56  *                                      fast path.
  57  *              J Hadi Salim:           ECN support
  58  *              Andrei Gurtov,
  59  *              Pasi Sarolahti,
  60  *              Panu Kuhlberg:          Experimental audit of TCP (re)transmission
  61  *                                      engine. Lots of bugs are found.
  62  *              Pasi Sarolahti:         F-RTO for dealing with spurious RTOs
  63  */
  64 
  65 #define pr_fmt(fmt) "TCP: " fmt
  66 
  67 #include <linux/mm.h>
  68 #include <linux/slab.h>
  69 #include <linux/module.h>
  70 #include <linux/sysctl.h>
  71 #include <linux/kernel.h>
  72 #include <linux/prefetch.h>
  73 #include <net/dst.h>
  74 #include <net/tcp.h>
  75 #include <net/inet_common.h>
  76 #include <linux/ipsec.h>
  77 #include <asm/unaligned.h>
  78 #include <linux/errqueue.h>
  79 #include <trace/events/tcp.h>
  80 #include <linux/jump_label_ratelimit.h>
  81 #include <net/busy_poll.h>
  82 
  83 int sysctl_tcp_max_orphans __read_mostly = NR_FILE;
  84 
  85 #define FLAG_DATA               0x01 /* Incoming frame contained data.          */
  86 #define FLAG_WIN_UPDATE         0x02 /* Incoming ACK was a window update.       */
  87 #define FLAG_DATA_ACKED         0x04 /* This ACK acknowledged new data.         */
  88 #define FLAG_RETRANS_DATA_ACKED 0x08 /* "" "" some of which was retransmitted.  */
  89 #define FLAG_SYN_ACKED          0x10 /* This ACK acknowledged SYN.              */
  90 #define FLAG_DATA_SACKED        0x20 /* New SACK.                               */
  91 #define FLAG_ECE                0x40 /* ECE in this ACK                         */
  92 #define FLAG_LOST_RETRANS       0x80 /* This ACK marks some retransmission lost */
  93 #define FLAG_SLOWPATH           0x100 /* Do not skip RFC checks for window update.*/
  94 #define FLAG_ORIG_SACK_ACKED    0x200 /* Never retransmitted data are (s)acked  */
  95 #define FLAG_SND_UNA_ADVANCED   0x400 /* Snd_una was changed (!= FLAG_DATA_ACKED) */
  96 #define FLAG_DSACKING_ACK       0x800 /* SACK blocks contained D-SACK info */
  97 #define FLAG_SET_XMIT_TIMER     0x1000 /* Set TLP or RTO timer */
  98 #define FLAG_SACK_RENEGING      0x2000 /* snd_una advanced to a sacked seq */
  99 #define FLAG_UPDATE_TS_RECENT   0x4000 /* tcp_replace_ts_recent() */
 100 #define FLAG_NO_CHALLENGE_ACK   0x8000 /* do not call tcp_send_challenge_ack()  */
 101 #define FLAG_ACK_MAYBE_DELAYED  0x10000 /* Likely a delayed ACK */
 102 
 103 #define FLAG_ACKED              (FLAG_DATA_ACKED|FLAG_SYN_ACKED)
 104 #define FLAG_NOT_DUP            (FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED)
 105 #define FLAG_CA_ALERT           (FLAG_DATA_SACKED|FLAG_ECE|FLAG_DSACKING_ACK)
 106 #define FLAG_FORWARD_PROGRESS   (FLAG_ACKED|FLAG_DATA_SACKED)
 107 
 108 #define TCP_REMNANT (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN|TCP_FLAG_PSH)
 109 #define TCP_HP_BITS (~(TCP_RESERVED_BITS|TCP_FLAG_PSH))
 110 
 111 #define REXMIT_NONE     0 /* no loss recovery to do */
 112 #define REXMIT_LOST     1 /* retransmit packets marked lost */
 113 #define REXMIT_NEW      2 /* FRTO-style transmit of unsent/new packets */
 114 
 115 #if IS_ENABLED(CONFIG_TLS_DEVICE)
 116 static DEFINE_STATIC_KEY_DEFERRED_FALSE(clean_acked_data_enabled, HZ);
 117 
 118 void clean_acked_data_enable(struct inet_connection_sock *icsk,
 119                              void (*cad)(struct sock *sk, u32 ack_seq))
 120 {
 121         icsk->icsk_clean_acked = cad;
 122         static_branch_deferred_inc(&clean_acked_data_enabled);
 123 }
 124 EXPORT_SYMBOL_GPL(clean_acked_data_enable);
 125 
 126 void clean_acked_data_disable(struct inet_connection_sock *icsk)
 127 {
 128         static_branch_slow_dec_deferred(&clean_acked_data_enabled);
 129         icsk->icsk_clean_acked = NULL;
 130 }
 131 EXPORT_SYMBOL_GPL(clean_acked_data_disable);
 132 
 133 void clean_acked_data_flush(void)
 134 {
 135         static_key_deferred_flush(&clean_acked_data_enabled);
 136 }
 137 EXPORT_SYMBOL_GPL(clean_acked_data_flush);
 138 #endif
 139 
 140 static void tcp_gro_dev_warn(struct sock *sk, const struct sk_buff *skb,
 141                              unsigned int len)
 142 {
 143         static bool __once __read_mostly;
 144 
 145         if (!__once) {
 146                 struct net_device *dev;
 147 
 148                 __once = true;
 149 
 150                 rcu_read_lock();
 151                 dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif);
 152                 if (!dev || len >= dev->mtu)
 153                         pr_warn("%s: Driver has suspect GRO implementation, TCP performance may be compromised.\n",
 154                                 dev ? dev->name : "Unknown driver");
 155                 rcu_read_unlock();
 156         }
 157 }
 158 
 159 /* Adapt the MSS value used to make delayed ack decision to the
 160  * real world.
 161  */
 162 static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
 163 {
 164         struct inet_connection_sock *icsk = inet_csk(sk);
 165         const unsigned int lss = icsk->icsk_ack.last_seg_size;
 166         unsigned int len;
 167 
 168         icsk->icsk_ack.last_seg_size = 0;
 169 
 170         /* skb->len may jitter because of SACKs, even if peer
 171          * sends good full-sized frames.
 172          */
 173         len = skb_shinfo(skb)->gso_size ? : skb->len;
 174         if (len >= icsk->icsk_ack.rcv_mss) {
 175                 icsk->icsk_ack.rcv_mss = min_t(unsigned int, len,
 176                                                tcp_sk(sk)->advmss);
 177                 /* Account for possibly-removed options */
 178                 if (unlikely(len > icsk->icsk_ack.rcv_mss +
 179                                    MAX_TCP_OPTION_SPACE))
 180                         tcp_gro_dev_warn(sk, skb, len);
 181         } else {
 182                 /* Otherwise, we make more careful check taking into account,
 183                  * that SACKs block is variable.
 184                  *
 185                  * "len" is invariant segment length, including TCP header.
 186                  */
 187                 len += skb->data - skb_transport_header(skb);
 188                 if (len >= TCP_MSS_DEFAULT + sizeof(struct tcphdr) ||
 189                     /* If PSH is not set, packet should be
 190                      * full sized, provided peer TCP is not badly broken.
 191                      * This observation (if it is correct 8)) allows
 192                      * to handle super-low mtu links fairly.
 193                      */
 194                     (len >= TCP_MIN_MSS + sizeof(struct tcphdr) &&
 195                      !(tcp_flag_word(tcp_hdr(skb)) & TCP_REMNANT))) {
 196                         /* Subtract also invariant (if peer is RFC compliant),
 197                          * tcp header plus fixed timestamp option length.
 198                          * Resulting "len" is MSS free of SACK jitter.
 199                          */
 200                         len -= tcp_sk(sk)->tcp_header_len;
 201                         icsk->icsk_ack.last_seg_size = len;
 202                         if (len == lss) {
 203                                 icsk->icsk_ack.rcv_mss = len;
 204                                 return;
 205                         }
 206                 }
 207                 if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)
 208                         icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2;
 209                 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
 210         }
 211 }
 212 
 213 static void tcp_incr_quickack(struct sock *sk, unsigned int max_quickacks)
 214 {
 215         struct inet_connection_sock *icsk = inet_csk(sk);
 216         unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss);
 217 
 218         if (quickacks == 0)
 219                 quickacks = 2;
 220         quickacks = min(quickacks, max_quickacks);
 221         if (quickacks > icsk->icsk_ack.quick)
 222                 icsk->icsk_ack.quick = quickacks;
 223 }
 224 
 225 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks)
 226 {
 227         struct inet_connection_sock *icsk = inet_csk(sk);
 228 
 229         tcp_incr_quickack(sk, max_quickacks);
 230         inet_csk_exit_pingpong_mode(sk);
 231         icsk->icsk_ack.ato = TCP_ATO_MIN;
 232 }
 233 EXPORT_SYMBOL(tcp_enter_quickack_mode);
 234 
 235 /* Send ACKs quickly, if "quick" count is not exhausted
 236  * and the session is not interactive.
 237  */
 238 
 239 static bool tcp_in_quickack_mode(struct sock *sk)
 240 {
 241         const struct inet_connection_sock *icsk = inet_csk(sk);
 242         const struct dst_entry *dst = __sk_dst_get(sk);
 243 
 244         return (dst && dst_metric(dst, RTAX_QUICKACK)) ||
 245                 (icsk->icsk_ack.quick && !inet_csk_in_pingpong_mode(sk));
 246 }
 247 
 248 static void tcp_ecn_queue_cwr(struct tcp_sock *tp)
 249 {
 250         if (tp->ecn_flags & TCP_ECN_OK)
 251                 tp->ecn_flags |= TCP_ECN_QUEUE_CWR;
 252 }
 253 
 254 static void tcp_ecn_accept_cwr(struct sock *sk, const struct sk_buff *skb)
 255 {
 256         if (tcp_hdr(skb)->cwr) {
 257                 tcp_sk(sk)->ecn_flags &= ~TCP_ECN_DEMAND_CWR;
 258 
 259                 /* If the sender is telling us it has entered CWR, then its
 260                  * cwnd may be very low (even just 1 packet), so we should ACK
 261                  * immediately.
 262                  */
 263                 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
 264         }
 265 }
 266 
 267 static void tcp_ecn_withdraw_cwr(struct tcp_sock *tp)
 268 {
 269         tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
 270 }
 271 
 272 static void __tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb)
 273 {
 274         struct tcp_sock *tp = tcp_sk(sk);
 275 
 276         switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) {
 277         case INET_ECN_NOT_ECT:
 278                 /* Funny extension: if ECT is not set on a segment,
 279                  * and we already seen ECT on a previous segment,
 280                  * it is probably a retransmit.
 281                  */
 282                 if (tp->ecn_flags & TCP_ECN_SEEN)
 283                         tcp_enter_quickack_mode(sk, 2);
 284                 break;
 285         case INET_ECN_CE:
 286                 if (tcp_ca_needs_ecn(sk))
 287                         tcp_ca_event(sk, CA_EVENT_ECN_IS_CE);
 288 
 289                 if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) {
 290                         /* Better not delay acks, sender can have a very low cwnd */
 291                         tcp_enter_quickack_mode(sk, 2);
 292                         tp->ecn_flags |= TCP_ECN_DEMAND_CWR;
 293                 }
 294                 tp->ecn_flags |= TCP_ECN_SEEN;
 295                 break;
 296         default:
 297                 if (tcp_ca_needs_ecn(sk))
 298                         tcp_ca_event(sk, CA_EVENT_ECN_NO_CE);
 299                 tp->ecn_flags |= TCP_ECN_SEEN;
 300                 break;
 301         }
 302 }
 303 
 304 static void tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb)
 305 {
 306         if (tcp_sk(sk)->ecn_flags & TCP_ECN_OK)
 307                 __tcp_ecn_check_ce(sk, skb);
 308 }
 309 
 310 static void tcp_ecn_rcv_synack(struct tcp_sock *tp, const struct tcphdr *th)
 311 {
 312         if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr))
 313                 tp->ecn_flags &= ~TCP_ECN_OK;
 314 }
 315 
 316 static void tcp_ecn_rcv_syn(struct tcp_sock *tp, const struct tcphdr *th)
 317 {
 318         if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr))
 319                 tp->ecn_flags &= ~TCP_ECN_OK;
 320 }
 321 
 322 static bool tcp_ecn_rcv_ecn_echo(const struct tcp_sock *tp, const struct tcphdr *th)
 323 {
 324         if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK))
 325                 return true;
 326         return false;
 327 }
 328 
 329 /* Buffer size and advertised window tuning.
 330  *
 331  * 1. Tuning sk->sk_sndbuf, when connection enters established state.
 332  */
 333 
 334 static void tcp_sndbuf_expand(struct sock *sk)
 335 {
 336         const struct tcp_sock *tp = tcp_sk(sk);
 337         const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
 338         int sndmem, per_mss;
 339         u32 nr_segs;
 340 
 341         /* Worst case is non GSO/TSO : each frame consumes one skb
 342          * and skb->head is kmalloced using power of two area of memory
 343          */
 344         per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) +
 345                   MAX_TCP_HEADER +
 346                   SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 347 
 348         per_mss = roundup_pow_of_two(per_mss) +
 349                   SKB_DATA_ALIGN(sizeof(struct sk_buff));
 350 
 351         nr_segs = max_t(u32, TCP_INIT_CWND, tp->snd_cwnd);
 352         nr_segs = max_t(u32, nr_segs, tp->reordering + 1);
 353 
 354         /* Fast Recovery (RFC 5681 3.2) :
 355          * Cubic needs 1.7 factor, rounded to 2 to include
 356          * extra cushion (application might react slowly to EPOLLOUT)
 357          */
 358         sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2;
 359         sndmem *= nr_segs * per_mss;
 360 
 361         if (sk->sk_sndbuf < sndmem)
 362                 WRITE_ONCE(sk->sk_sndbuf,
 363                            min(sndmem, sock_net(sk)->ipv4.sysctl_tcp_wmem[2]));
 364 }
 365 
 366 /* 2. Tuning advertised window (window_clamp, rcv_ssthresh)
 367  *
 368  * All tcp_full_space() is split to two parts: "network" buffer, allocated
 369  * forward and advertised in receiver window (tp->rcv_wnd) and
 370  * "application buffer", required to isolate scheduling/application
 371  * latencies from network.
 372  * window_clamp is maximal advertised window. It can be less than
 373  * tcp_full_space(), in this case tcp_full_space() - window_clamp
 374  * is reserved for "application" buffer. The less window_clamp is
 375  * the smoother our behaviour from viewpoint of network, but the lower
 376  * throughput and the higher sensitivity of the connection to losses. 8)
 377  *
 378  * rcv_ssthresh is more strict window_clamp used at "slow start"
 379  * phase to predict further behaviour of this connection.
 380  * It is used for two goals:
 381  * - to enforce header prediction at sender, even when application
 382  *   requires some significant "application buffer". It is check #1.
 383  * - to prevent pruning of receive queue because of misprediction
 384  *   of receiver window. Check #2.
 385  *
 386  * The scheme does not work when sender sends good segments opening
 387  * window and then starts to feed us spaghetti. But it should work
 388  * in common situations. Otherwise, we have to rely on queue collapsing.
 389  */
 390 
 391 /* Slow part of check#2. */
 392 static int __tcp_grow_window(const struct sock *sk, const struct sk_buff *skb)
 393 {
 394         struct tcp_sock *tp = tcp_sk(sk);
 395         /* Optimize this! */
 396         int truesize = tcp_win_from_space(sk, skb->truesize) >> 1;
 397         int window = tcp_win_from_space(sk, sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1;
 398 
 399         while (tp->rcv_ssthresh <= window) {
 400                 if (truesize <= skb->len)
 401                         return 2 * inet_csk(sk)->icsk_ack.rcv_mss;
 402 
 403                 truesize >>= 1;
 404                 window >>= 1;
 405         }
 406         return 0;
 407 }
 408 
 409 static void tcp_grow_window(struct sock *sk, const struct sk_buff *skb)
 410 {
 411         struct tcp_sock *tp = tcp_sk(sk);
 412         int room;
 413 
 414         room = min_t(int, tp->window_clamp, tcp_space(sk)) - tp->rcv_ssthresh;
 415 
 416         /* Check #1 */
 417         if (room > 0 && !tcp_under_memory_pressure(sk)) {
 418                 int incr;
 419 
 420                 /* Check #2. Increase window, if skb with such overhead
 421                  * will fit to rcvbuf in future.
 422                  */
 423                 if (tcp_win_from_space(sk, skb->truesize) <= skb->len)
 424                         incr = 2 * tp->advmss;
 425                 else
 426                         incr = __tcp_grow_window(sk, skb);
 427 
 428                 if (incr) {
 429                         incr = max_t(int, incr, 2 * skb->len);
 430                         tp->rcv_ssthresh += min(room, incr);
 431                         inet_csk(sk)->icsk_ack.quick |= 1;
 432                 }
 433         }
 434 }
 435 
 436 /* 3. Try to fixup all. It is made immediately after connection enters
 437  *    established state.
 438  */
 439 void tcp_init_buffer_space(struct sock *sk)
 440 {
 441         int tcp_app_win = sock_net(sk)->ipv4.sysctl_tcp_app_win;
 442         struct tcp_sock *tp = tcp_sk(sk);
 443         int maxwin;
 444 
 445         if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK))
 446                 tcp_sndbuf_expand(sk);
 447 
 448         tp->rcvq_space.space = min_t(u32, tp->rcv_wnd, TCP_INIT_CWND * tp->advmss);
 449         tcp_mstamp_refresh(tp);
 450         tp->rcvq_space.time = tp->tcp_mstamp;
 451         tp->rcvq_space.seq = tp->copied_seq;
 452 
 453         maxwin = tcp_full_space(sk);
 454 
 455         if (tp->window_clamp >= maxwin) {
 456                 tp->window_clamp = maxwin;
 457 
 458                 if (tcp_app_win && maxwin > 4 * tp->advmss)
 459                         tp->window_clamp = max(maxwin -
 460                                                (maxwin >> tcp_app_win),
 461                                                4 * tp->advmss);
 462         }
 463 
 464         /* Force reservation of one segment. */
 465         if (tcp_app_win &&
 466             tp->window_clamp > 2 * tp->advmss &&
 467             tp->window_clamp + tp->advmss > maxwin)
 468                 tp->window_clamp = max(2 * tp->advmss, maxwin - tp->advmss);
 469 
 470         tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp);
 471         tp->snd_cwnd_stamp = tcp_jiffies32;
 472 }
 473 
 474 /* 4. Recalculate window clamp after socket hit its memory bounds. */
 475 static void tcp_clamp_window(struct sock *sk)
 476 {
 477         struct tcp_sock *tp = tcp_sk(sk);
 478         struct inet_connection_sock *icsk = inet_csk(sk);
 479         struct net *net = sock_net(sk);
 480 
 481         icsk->icsk_ack.quick = 0;
 482 
 483         if (sk->sk_rcvbuf < net->ipv4.sysctl_tcp_rmem[2] &&
 484             !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) &&
 485             !tcp_under_memory_pressure(sk) &&
 486             sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)) {
 487                 WRITE_ONCE(sk->sk_rcvbuf,
 488                            min(atomic_read(&sk->sk_rmem_alloc),
 489                                net->ipv4.sysctl_tcp_rmem[2]));
 490         }
 491         if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)
 492                 tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);
 493 }
 494 
 495 /* Initialize RCV_MSS value.
 496  * RCV_MSS is an our guess about MSS used by the peer.
 497  * We haven't any direct information about the MSS.
 498  * It's better to underestimate the RCV_MSS rather than overestimate.
 499  * Overestimations make us ACKing less frequently than needed.
 500  * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
 501  */
 502 void tcp_initialize_rcv_mss(struct sock *sk)
 503 {
 504         const struct tcp_sock *tp = tcp_sk(sk);
 505         unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
 506 
 507         hint = min(hint, tp->rcv_wnd / 2);
 508         hint = min(hint, TCP_MSS_DEFAULT);
 509         hint = max(hint, TCP_MIN_MSS);
 510 
 511         inet_csk(sk)->icsk_ack.rcv_mss = hint;
 512 }
 513 EXPORT_SYMBOL(tcp_initialize_rcv_mss);
 514 
 515 /* Receiver "autotuning" code.
 516  *
 517  * The algorithm for RTT estimation w/o timestamps is based on
 518  * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL.
 519  * <http://public.lanl.gov/radiant/pubs.html#DRS>
 520  *
 521  * More detail on this code can be found at
 522  * <http://staff.psc.edu/jheffner/>,
 523  * though this reference is out of date.  A new paper
 524  * is pending.
 525  */
 526 static void tcp_rcv_rtt_update(struct tcp_sock *tp, u32 sample, int win_dep)
 527 {
 528         u32 new_sample = tp->rcv_rtt_est.rtt_us;
 529         long m = sample;
 530 
 531         if (new_sample != 0) {
 532                 /* If we sample in larger samples in the non-timestamp
 533                  * case, we could grossly overestimate the RTT especially
 534                  * with chatty applications or bulk transfer apps which
 535                  * are stalled on filesystem I/O.
 536                  *
 537                  * Also, since we are only going for a minimum in the
 538                  * non-timestamp case, we do not smooth things out
 539                  * else with timestamps disabled convergence takes too
 540                  * long.
 541                  */
 542                 if (!win_dep) {
 543                         m -= (new_sample >> 3);
 544                         new_sample += m;
 545                 } else {
 546                         m <<= 3;
 547                         if (m < new_sample)
 548                                 new_sample = m;
 549                 }
 550         } else {
 551                 /* No previous measure. */
 552                 new_sample = m << 3;
 553         }
 554 
 555         tp->rcv_rtt_est.rtt_us = new_sample;
 556 }
 557 
 558 static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp)
 559 {
 560         u32 delta_us;
 561 
 562         if (tp->rcv_rtt_est.time == 0)
 563                 goto new_measure;
 564         if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq))
 565                 return;
 566         delta_us = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcv_rtt_est.time);
 567         if (!delta_us)
 568                 delta_us = 1;
 569         tcp_rcv_rtt_update(tp, delta_us, 1);
 570 
 571 new_measure:
 572         tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd;
 573         tp->rcv_rtt_est.time = tp->tcp_mstamp;
 574 }
 575 
 576 static inline void tcp_rcv_rtt_measure_ts(struct sock *sk,
 577                                           const struct sk_buff *skb)
 578 {
 579         struct tcp_sock *tp = tcp_sk(sk);
 580 
 581         if (tp->rx_opt.rcv_tsecr == tp->rcv_rtt_last_tsecr)
 582                 return;
 583         tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;
 584 
 585         if (TCP_SKB_CB(skb)->end_seq -
 586             TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) {
 587                 u32 delta = tcp_time_stamp(tp) - tp->rx_opt.rcv_tsecr;
 588                 u32 delta_us;
 589 
 590                 if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
 591                         if (!delta)
 592                                 delta = 1;
 593                         delta_us = delta * (USEC_PER_SEC / TCP_TS_HZ);
 594                         tcp_rcv_rtt_update(tp, delta_us, 0);
 595                 }
 596         }
 597 }
 598 
 599 /*
 600  * This function should be called every time data is copied to user space.
 601  * It calculates the appropriate TCP receive buffer space.
 602  */
 603 void tcp_rcv_space_adjust(struct sock *sk)
 604 {
 605         struct tcp_sock *tp = tcp_sk(sk);
 606         u32 copied;
 607         int time;
 608 
 609         trace_tcp_rcv_space_adjust(sk);
 610 
 611         tcp_mstamp_refresh(tp);
 612         time = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcvq_space.time);
 613         if (time < (tp->rcv_rtt_est.rtt_us >> 3) || tp->rcv_rtt_est.rtt_us == 0)
 614                 return;
 615 
 616         /* Number of bytes copied to user in last RTT */
 617         copied = tp->copied_seq - tp->rcvq_space.seq;
 618         if (copied <= tp->rcvq_space.space)
 619                 goto new_measure;
 620 
 621         /* A bit of theory :
 622          * copied = bytes received in previous RTT, our base window
 623          * To cope with packet losses, we need a 2x factor
 624          * To cope with slow start, and sender growing its cwin by 100 %
 625          * every RTT, we need a 4x factor, because the ACK we are sending
 626          * now is for the next RTT, not the current one :
 627          * <prev RTT . ><current RTT .. ><next RTT .... >
 628          */
 629 
 630         if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
 631             !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
 632                 int rcvmem, rcvbuf;
 633                 u64 rcvwin, grow;
 634 
 635                 /* minimal window to cope with packet losses, assuming
 636                  * steady state. Add some cushion because of small variations.
 637                  */
 638                 rcvwin = ((u64)copied << 1) + 16 * tp->advmss;
 639 
 640                 /* Accommodate for sender rate increase (eg. slow start) */
 641                 grow = rcvwin * (copied - tp->rcvq_space.space);
 642                 do_div(grow, tp->rcvq_space.space);
 643                 rcvwin += (grow << 1);
 644 
 645                 rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER);
 646                 while (tcp_win_from_space(sk, rcvmem) < tp->advmss)
 647                         rcvmem += 128;
 648 
 649                 do_div(rcvwin, tp->advmss);
 650                 rcvbuf = min_t(u64, rcvwin * rcvmem,
 651                                sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
 652                 if (rcvbuf > sk->sk_rcvbuf) {
 653                         WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
 654 
 655                         /* Make the window clamp follow along.  */
 656                         tp->window_clamp = tcp_win_from_space(sk, rcvbuf);
 657                 }
 658         }
 659         tp->rcvq_space.space = copied;
 660 
 661 new_measure:
 662         tp->rcvq_space.seq = tp->copied_seq;
 663         tp->rcvq_space.time = tp->tcp_mstamp;
 664 }
 665 
 666 /* There is something which you must keep in mind when you analyze the
 667  * behavior of the tp->ato delayed ack timeout interval.  When a
 668  * connection starts up, we want to ack as quickly as possible.  The
 669  * problem is that "good" TCP's do slow start at the beginning of data
 670  * transmission.  The means that until we send the first few ACK's the
 671  * sender will sit on his end and only queue most of his data, because
 672  * he can only send snd_cwnd unacked packets at any given time.  For
 673  * each ACK we send, he increments snd_cwnd and transmits more of his
 674  * queue.  -DaveM
 675  */
 676 static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb)
 677 {
 678         struct tcp_sock *tp = tcp_sk(sk);
 679         struct inet_connection_sock *icsk = inet_csk(sk);
 680         u32 now;
 681 
 682         inet_csk_schedule_ack(sk);
 683 
 684         tcp_measure_rcv_mss(sk, skb);
 685 
 686         tcp_rcv_rtt_measure(tp);
 687 
 688         now = tcp_jiffies32;
 689 
 690         if (!icsk->icsk_ack.ato) {
 691                 /* The _first_ data packet received, initialize
 692                  * delayed ACK engine.
 693                  */
 694                 tcp_incr_quickack(sk, TCP_MAX_QUICKACKS);
 695                 icsk->icsk_ack.ato = TCP_ATO_MIN;
 696         } else {
 697                 int m = now - icsk->icsk_ack.lrcvtime;
 698 
 699                 if (m <= TCP_ATO_MIN / 2) {
 700                         /* The fastest case is the first. */
 701                         icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2;
 702                 } else if (m < icsk->icsk_ack.ato) {
 703                         icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m;
 704                         if (icsk->icsk_ack.ato > icsk->icsk_rto)
 705                                 icsk->icsk_ack.ato = icsk->icsk_rto;
 706                 } else if (m > icsk->icsk_rto) {
 707                         /* Too long gap. Apparently sender failed to
 708                          * restart window, so that we send ACKs quickly.
 709                          */
 710                         tcp_incr_quickack(sk, TCP_MAX_QUICKACKS);
 711                         sk_mem_reclaim(sk);
 712                 }
 713         }
 714         icsk->icsk_ack.lrcvtime = now;
 715 
 716         tcp_ecn_check_ce(sk, skb);
 717 
 718         if (skb->len >= 128)
 719                 tcp_grow_window(sk, skb);
 720 }
 721 
 722 /* Called to compute a smoothed rtt estimate. The data fed to this
 723  * routine either comes from timestamps, or from segments that were
 724  * known _not_ to have been retransmitted [see Karn/Partridge
 725  * Proceedings SIGCOMM 87]. The algorithm is from the SIGCOMM 88
 726  * piece by Van Jacobson.
 727  * NOTE: the next three routines used to be one big routine.
 728  * To save cycles in the RFC 1323 implementation it was better to break
 729  * it up into three procedures. -- erics
 730  */
 731 static void tcp_rtt_estimator(struct sock *sk, long mrtt_us)
 732 {
 733         struct tcp_sock *tp = tcp_sk(sk);
 734         long m = mrtt_us; /* RTT */
 735         u32 srtt = tp->srtt_us;
 736 
 737         /*      The following amusing code comes from Jacobson's
 738          *      article in SIGCOMM '88.  Note that rtt and mdev
 739          *      are scaled versions of rtt and mean deviation.
 740          *      This is designed to be as fast as possible
 741          *      m stands for "measurement".
 742          *
 743          *      On a 1990 paper the rto value is changed to:
 744          *      RTO = rtt + 4 * mdev
 745          *
 746          * Funny. This algorithm seems to be very broken.
 747          * These formulae increase RTO, when it should be decreased, increase
 748          * too slowly, when it should be increased quickly, decrease too quickly
 749          * etc. I guess in BSD RTO takes ONE value, so that it is absolutely
 750          * does not matter how to _calculate_ it. Seems, it was trap
 751          * that VJ failed to avoid. 8)
 752          */
 753         if (srtt != 0) {
 754                 m -= (srtt >> 3);       /* m is now error in rtt est */
 755                 srtt += m;              /* rtt = 7/8 rtt + 1/8 new */
 756                 if (m < 0) {
 757                         m = -m;         /* m is now abs(error) */
 758                         m -= (tp->mdev_us >> 2);   /* similar update on mdev */
 759                         /* This is similar to one of Eifel findings.
 760                          * Eifel blocks mdev updates when rtt decreases.
 761                          * This solution is a bit different: we use finer gain
 762                          * for mdev in this case (alpha*beta).
 763                          * Like Eifel it also prevents growth of rto,
 764                          * but also it limits too fast rto decreases,
 765                          * happening in pure Eifel.
 766                          */
 767                         if (m > 0)
 768                                 m >>= 3;
 769                 } else {
 770                         m -= (tp->mdev_us >> 2);   /* similar update on mdev */
 771                 }
 772                 tp->mdev_us += m;               /* mdev = 3/4 mdev + 1/4 new */
 773                 if (tp->mdev_us > tp->mdev_max_us) {
 774                         tp->mdev_max_us = tp->mdev_us;
 775                         if (tp->mdev_max_us > tp->rttvar_us)
 776                                 tp->rttvar_us = tp->mdev_max_us;
 777                 }
 778                 if (after(tp->snd_una, tp->rtt_seq)) {
 779                         if (tp->mdev_max_us < tp->rttvar_us)
 780                                 tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2;
 781                         tp->rtt_seq = tp->snd_nxt;
 782                         tp->mdev_max_us = tcp_rto_min_us(sk);
 783 
 784                         tcp_bpf_rtt(sk);
 785                 }
 786         } else {
 787                 /* no previous measure. */
 788                 srtt = m << 3;          /* take the measured time to be rtt */
 789                 tp->mdev_us = m << 1;   /* make sure rto = 3*rtt */
 790                 tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk));
 791                 tp->mdev_max_us = tp->rttvar_us;
 792                 tp->rtt_seq = tp->snd_nxt;
 793 
 794                 tcp_bpf_rtt(sk);
 795         }
 796         tp->srtt_us = max(1U, srtt);
 797 }
 798 
 799 static void tcp_update_pacing_rate(struct sock *sk)
 800 {
 801         const struct tcp_sock *tp = tcp_sk(sk);
 802         u64 rate;
 803 
 804         /* set sk_pacing_rate to 200 % of current rate (mss * cwnd / srtt) */
 805         rate = (u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3);
 806 
 807         /* current rate is (cwnd * mss) / srtt
 808          * In Slow Start [1], set sk_pacing_rate to 200 % the current rate.
 809          * In Congestion Avoidance phase, set it to 120 % the current rate.
 810          *
 811          * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh)
 812          *       If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching
 813          *       end of slow start and should slow down.
 814          */
 815         if (tp->snd_cwnd < tp->snd_ssthresh / 2)
 816                 rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio;
 817         else
 818                 rate *= sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio;
 819 
 820         rate *= max(tp->snd_cwnd, tp->packets_out);
 821 
 822         if (likely(tp->srtt_us))
 823                 do_div(rate, tp->srtt_us);
 824 
 825         /* WRITE_ONCE() is needed because sch_fq fetches sk_pacing_rate
 826          * without any lock. We want to make sure compiler wont store
 827          * intermediate values in this location.
 828          */
 829         WRITE_ONCE(sk->sk_pacing_rate, min_t(u64, rate,
 830                                              sk->sk_max_pacing_rate));
 831 }
 832 
 833 /* Calculate rto without backoff.  This is the second half of Van Jacobson's
 834  * routine referred to above.
 835  */
 836 static void tcp_set_rto(struct sock *sk)
 837 {
 838         const struct tcp_sock *tp = tcp_sk(sk);
 839         /* Old crap is replaced with new one. 8)
 840          *
 841          * More seriously:
 842          * 1. If rtt variance happened to be less 50msec, it is hallucination.
 843          *    It cannot be less due to utterly erratic ACK generation made
 844          *    at least by solaris and freebsd. "Erratic ACKs" has _nothing_
 845          *    to do with delayed acks, because at cwnd>2 true delack timeout
 846          *    is invisible. Actually, Linux-2.4 also generates erratic
 847          *    ACKs in some circumstances.
 848          */
 849         inet_csk(sk)->icsk_rto = __tcp_set_rto(tp);
 850 
 851         /* 2. Fixups made earlier cannot be right.
 852          *    If we do not estimate RTO correctly without them,
 853          *    all the algo is pure shit and should be replaced
 854          *    with correct one. It is exactly, which we pretend to do.
 855          */
 856 
 857         /* NOTE: clamping at TCP_RTO_MIN is not required, current algo
 858          * guarantees that rto is higher.
 859          */
 860         tcp_bound_rto(sk);
 861 }
 862 
 863 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst)
 864 {
 865         __u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
 866 
 867         if (!cwnd)
 868                 cwnd = TCP_INIT_CWND;
 869         return min_t(__u32, cwnd, tp->snd_cwnd_clamp);
 870 }
 871 
 872 /* Take a notice that peer is sending D-SACKs */
 873 static void tcp_dsack_seen(struct tcp_sock *tp)
 874 {
 875         tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;
 876         tp->rack.dsack_seen = 1;
 877         tp->dsack_dups++;
 878 }
 879 
 880 /* It's reordering when higher sequence was delivered (i.e. sacked) before
 881  * some lower never-retransmitted sequence ("low_seq"). The maximum reordering
 882  * distance is approximated in full-mss packet distance ("reordering").
 883  */
 884 static void tcp_check_sack_reordering(struct sock *sk, const u32 low_seq,
 885                                       const int ts)
 886 {
 887         struct tcp_sock *tp = tcp_sk(sk);
 888         const u32 mss = tp->mss_cache;
 889         u32 fack, metric;
 890 
 891         fack = tcp_highest_sack_seq(tp);
 892         if (!before(low_seq, fack))
 893                 return;
 894 
 895         metric = fack - low_seq;
 896         if ((metric > tp->reordering * mss) && mss) {
 897 #if FASTRETRANS_DEBUG > 1
 898                 pr_debug("Disorder%d %d %u f%u s%u rr%d\n",
 899                          tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state,
 900                          tp->reordering,
 901                          0,
 902                          tp->sacked_out,
 903                          tp->undo_marker ? tp->undo_retrans : 0);
 904 #endif
 905                 tp->reordering = min_t(u32, (metric + mss - 1) / mss,
 906                                        sock_net(sk)->ipv4.sysctl_tcp_max_reordering);
 907         }
 908 
 909         /* This exciting event is worth to be remembered. 8) */
 910         tp->reord_seen++;
 911         NET_INC_STATS(sock_net(sk),
 912                       ts ? LINUX_MIB_TCPTSREORDER : LINUX_MIB_TCPSACKREORDER);
 913 }
 914 
 915 /* This must be called before lost_out is incremented */
 916 static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
 917 {
 918         if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||
 919             (tp->retransmit_skb_hint &&
 920              before(TCP_SKB_CB(skb)->seq,
 921                     TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))
 922                 tp->retransmit_skb_hint = skb;
 923 }
 924 
 925 /* Sum the number of packets on the wire we have marked as lost.
 926  * There are two cases we care about here:
 927  * a) Packet hasn't been marked lost (nor retransmitted),
 928  *    and this is the first loss.
 929  * b) Packet has been marked both lost and retransmitted,
 930  *    and this means we think it was lost again.
 931  */
 932 static void tcp_sum_lost(struct tcp_sock *tp, struct sk_buff *skb)
 933 {
 934         __u8 sacked = TCP_SKB_CB(skb)->sacked;
 935 
 936         if (!(sacked & TCPCB_LOST) ||
 937             ((sacked & TCPCB_LOST) && (sacked & TCPCB_SACKED_RETRANS)))
 938                 tp->lost += tcp_skb_pcount(skb);
 939 }
 940 
 941 static void tcp_skb_mark_lost(struct tcp_sock *tp, struct sk_buff *skb)
 942 {
 943         if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) {
 944                 tcp_verify_retransmit_hint(tp, skb);
 945 
 946                 tp->lost_out += tcp_skb_pcount(skb);
 947                 tcp_sum_lost(tp, skb);
 948                 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
 949         }
 950 }
 951 
 952 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb)
 953 {
 954         tcp_verify_retransmit_hint(tp, skb);
 955 
 956         tcp_sum_lost(tp, skb);
 957         if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_ACKED))) {
 958                 tp->lost_out += tcp_skb_pcount(skb);
 959                 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
 960         }
 961 }
 962 
 963 /* This procedure tags the retransmission queue when SACKs arrive.
 964  *
 965  * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L).
 966  * Packets in queue with these bits set are counted in variables
 967  * sacked_out, retrans_out and lost_out, correspondingly.
 968  *
 969  * Valid combinations are:
 970  * Tag  InFlight        Description
 971  * 0    1               - orig segment is in flight.
 972  * S    0               - nothing flies, orig reached receiver.
 973  * L    0               - nothing flies, orig lost by net.
 974  * R    2               - both orig and retransmit are in flight.
 975  * L|R  1               - orig is lost, retransmit is in flight.
 976  * S|R  1               - orig reached receiver, retrans is still in flight.
 977  * (L|S|R is logically valid, it could occur when L|R is sacked,
 978  *  but it is equivalent to plain S and code short-curcuits it to S.
 979  *  L|S is logically invalid, it would mean -1 packet in flight 8))
 980  *
 981  * These 6 states form finite state machine, controlled by the following events:
 982  * 1. New ACK (+SACK) arrives. (tcp_sacktag_write_queue())
 983  * 2. Retransmission. (tcp_retransmit_skb(), tcp_xmit_retransmit_queue())
 984  * 3. Loss detection event of two flavors:
 985  *      A. Scoreboard estimator decided the packet is lost.
 986  *         A'. Reno "three dupacks" marks head of queue lost.
 987  *      B. SACK arrives sacking SND.NXT at the moment, when the
 988  *         segment was retransmitted.
 989  * 4. D-SACK added new rule: D-SACK changes any tag to S.
 990  *
 991  * It is pleasant to note, that state diagram turns out to be commutative,
 992  * so that we are allowed not to be bothered by order of our actions,
 993  * when multiple events arrive simultaneously. (see the function below).
 994  *
 995  * Reordering detection.
 996  * --------------------
 997  * Reordering metric is maximal distance, which a packet can be displaced
 998  * in packet stream. With SACKs we can estimate it:
 999  *
1000  * 1. SACK fills old hole and the corresponding segment was not
1001  *    ever retransmitted -> reordering. Alas, we cannot use it
1002  *    when segment was retransmitted.
1003  * 2. The last flaw is solved with D-SACK. D-SACK arrives
1004  *    for retransmitted and already SACKed segment -> reordering..
1005  * Both of these heuristics are not used in Loss state, when we cannot
1006  * account for retransmits accurately.
1007  *
1008  * SACK block validation.
1009  * ----------------------
1010  *
1011  * SACK block range validation checks that the received SACK block fits to
1012  * the expected sequence limits, i.e., it is between SND.UNA and SND.NXT.
1013  * Note that SND.UNA is not included to the range though being valid because
1014  * it means that the receiver is rather inconsistent with itself reporting
1015  * SACK reneging when it should advance SND.UNA. Such SACK block this is
1016  * perfectly valid, however, in light of RFC2018 which explicitly states
1017  * that "SACK block MUST reflect the newest segment.  Even if the newest
1018  * segment is going to be discarded ...", not that it looks very clever
1019  * in case of head skb. Due to potentional receiver driven attacks, we
1020  * choose to avoid immediate execution of a walk in write queue due to
1021  * reneging and defer head skb's loss recovery to standard loss recovery
1022  * procedure that will eventually trigger (nothing forbids us doing this).
1023  *
1024  * Implements also blockage to start_seq wrap-around. Problem lies in the
1025  * fact that though start_seq (s) is before end_seq (i.e., not reversed),
1026  * there's no guarantee that it will be before snd_nxt (n). The problem
1027  * happens when start_seq resides between end_seq wrap (e_w) and snd_nxt
1028  * wrap (s_w):
1029  *
1030  *         <- outs wnd ->                          <- wrapzone ->
1031  *         u     e      n                         u_w   e_w  s n_w
1032  *         |     |      |                          |     |   |  |
1033  * |<------------+------+----- TCP seqno space --------------+---------->|
1034  * ...-- <2^31 ->|                                           |<--------...
1035  * ...---- >2^31 ------>|                                    |<--------...
1036  *
1037  * Current code wouldn't be vulnerable but it's better still to discard such
1038  * crazy SACK blocks. Doing this check for start_seq alone closes somewhat
1039  * similar case (end_seq after snd_nxt wrap) as earlier reversed check in
1040  * snd_nxt wrap -> snd_una region will then become "well defined", i.e.,
1041  * equal to the ideal case (infinite seqno space without wrap caused issues).
1042  *
1043  * With D-SACK the lower bound is extended to cover sequence space below
1044  * SND.UNA down to undo_marker, which is the last point of interest. Yet
1045  * again, D-SACK block must not to go across snd_una (for the same reason as
1046  * for the normal SACK blocks, explained above). But there all simplicity
1047  * ends, TCP might receive valid D-SACKs below that. As long as they reside
1048  * fully below undo_marker they do not affect behavior in anyway and can
1049  * therefore be safely ignored. In rare cases (which are more or less
1050  * theoretical ones), the D-SACK will nicely cross that boundary due to skb
1051  * fragmentation and packet reordering past skb's retransmission. To consider
1052  * them correctly, the acceptable range must be extended even more though
1053  * the exact amount is rather hard to quantify. However, tp->max_window can
1054  * be used as an exaggerated estimate.
1055  */
1056 static bool tcp_is_sackblock_valid(struct tcp_sock *tp, bool is_dsack,
1057                                    u32 start_seq, u32 end_seq)
1058 {
1059         /* Too far in future, or reversed (interpretation is ambiguous) */
1060         if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq))
1061                 return false;
1062 
1063         /* Nasty start_seq wrap-around check (see comments above) */
1064         if (!before(start_seq, tp->snd_nxt))
1065                 return false;
1066 
1067         /* In outstanding window? ...This is valid exit for D-SACKs too.
1068          * start_seq == snd_una is non-sensical (see comments above)
1069          */
1070         if (after(start_seq, tp->snd_una))
1071                 return true;
1072 
1073         if (!is_dsack || !tp->undo_marker)
1074                 return false;
1075 
1076         /* ...Then it's D-SACK, and must reside below snd_una completely */
1077         if (after(end_seq, tp->snd_una))
1078                 return false;
1079 
1080         if (!before(start_seq, tp->undo_marker))
1081                 return true;
1082 
1083         /* Too old */
1084         if (!after(end_seq, tp->undo_marker))
1085                 return false;
1086 
1087         /* Undo_marker boundary crossing (overestimates a lot). Known already:
1088          *   start_seq < undo_marker and end_seq >= undo_marker.
1089          */
1090         return !before(start_seq, end_seq - tp->max_window);
1091 }
1092 
1093 static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb,
1094                             struct tcp_sack_block_wire *sp, int num_sacks,
1095                             u32 prior_snd_una)
1096 {
1097         struct tcp_sock *tp = tcp_sk(sk);
1098         u32 start_seq_0 = get_unaligned_be32(&sp[0].start_seq);
1099         u32 end_seq_0 = get_unaligned_be32(&sp[0].end_seq);
1100         bool dup_sack = false;
1101 
1102         if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) {
1103                 dup_sack = true;
1104                 tcp_dsack_seen(tp);
1105                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDSACKRECV);
1106         } else if (num_sacks > 1) {
1107                 u32 end_seq_1 = get_unaligned_be32(&sp[1].end_seq);
1108                 u32 start_seq_1 = get_unaligned_be32(&sp[1].start_seq);
1109 
1110                 if (!after(end_seq_0, end_seq_1) &&
1111                     !before(start_seq_0, start_seq_1)) {
1112                         dup_sack = true;
1113                         tcp_dsack_seen(tp);
1114                         NET_INC_STATS(sock_net(sk),
1115                                         LINUX_MIB_TCPDSACKOFORECV);
1116                 }
1117         }
1118 
1119         /* D-SACK for already forgotten data... Do dumb counting. */
1120         if (dup_sack && tp->undo_marker && tp->undo_retrans > 0 &&
1121             !after(end_seq_0, prior_snd_una) &&
1122             after(end_seq_0, tp->undo_marker))
1123                 tp->undo_retrans--;
1124 
1125         return dup_sack;
1126 }
1127 
1128 struct tcp_sacktag_state {
1129         u32     reord;
1130         /* Timestamps for earliest and latest never-retransmitted segment
1131          * that was SACKed. RTO needs the earliest RTT to stay conservative,
1132          * but congestion control should still get an accurate delay signal.
1133          */
1134         u64     first_sackt;
1135         u64     last_sackt;
1136         struct rate_sample *rate;
1137         int     flag;
1138         unsigned int mss_now;
1139 };
1140 
1141 /* Check if skb is fully within the SACK block. In presence of GSO skbs,
1142  * the incoming SACK may not exactly match but we can find smaller MSS
1143  * aligned portion of it that matches. Therefore we might need to fragment
1144  * which may fail and creates some hassle (caller must handle error case
1145  * returns).
1146  *
1147  * FIXME: this could be merged to shift decision code
1148  */
1149 static int tcp_match_skb_to_sack(struct sock *sk, struct sk_buff *skb,
1150                                   u32 start_seq, u32 end_seq)
1151 {
1152         int err;
1153         bool in_sack;
1154         unsigned int pkt_len;
1155         unsigned int mss;
1156 
1157         in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
1158                   !before(end_seq, TCP_SKB_CB(skb)->end_seq);
1159 
1160         if (tcp_skb_pcount(skb) > 1 && !in_sack &&
1161             after(TCP_SKB_CB(skb)->end_seq, start_seq)) {
1162                 mss = tcp_skb_mss(skb);
1163                 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);
1164 
1165                 if (!in_sack) {
1166                         pkt_len = start_seq - TCP_SKB_CB(skb)->seq;
1167                         if (pkt_len < mss)
1168                                 pkt_len = mss;
1169                 } else {
1170                         pkt_len = end_seq - TCP_SKB_CB(skb)->seq;
1171                         if (pkt_len < mss)
1172                                 return -EINVAL;
1173                 }
1174 
1175                 /* Round if necessary so that SACKs cover only full MSSes
1176                  * and/or the remaining small portion (if present)
1177                  */
1178                 if (pkt_len > mss) {
1179                         unsigned int new_len = (pkt_len / mss) * mss;
1180                         if (!in_sack && new_len < pkt_len)
1181                                 new_len += mss;
1182                         pkt_len = new_len;
1183                 }
1184 
1185                 if (pkt_len >= skb->len && !in_sack)
1186                         return 0;
1187 
1188                 err = tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb,
1189                                    pkt_len, mss, GFP_ATOMIC);
1190                 if (err < 0)
1191                         return err;
1192         }
1193 
1194         return in_sack;
1195 }
1196 
1197 /* Mark the given newly-SACKed range as such, adjusting counters and hints. */
1198 static u8 tcp_sacktag_one(struct sock *sk,
1199                           struct tcp_sacktag_state *state, u8 sacked,
1200                           u32 start_seq, u32 end_seq,
1201                           int dup_sack, int pcount,
1202                           u64 xmit_time)
1203 {
1204         struct tcp_sock *tp = tcp_sk(sk);
1205 
1206         /* Account D-SACK for retransmitted packet. */
1207         if (dup_sack && (sacked & TCPCB_RETRANS)) {
1208                 if (tp->undo_marker && tp->undo_retrans > 0 &&
1209                     after(end_seq, tp->undo_marker))
1210                         tp->undo_retrans--;
1211                 if ((sacked & TCPCB_SACKED_ACKED) &&
1212                     before(start_seq, state->reord))
1213                                 state->reord = start_seq;
1214         }
1215 
1216         /* Nothing to do; acked frame is about to be dropped (was ACKed). */
1217         if (!after(end_seq, tp->snd_una))
1218                 return sacked;
1219 
1220         if (!(sacked & TCPCB_SACKED_ACKED)) {
1221                 tcp_rack_advance(tp, sacked, end_seq, xmit_time);
1222 
1223                 if (sacked & TCPCB_SACKED_RETRANS) {
1224                         /* If the segment is not tagged as lost,
1225                          * we do not clear RETRANS, believing
1226                          * that retransmission is still in flight.
1227                          */
1228                         if (sacked & TCPCB_LOST) {
1229                                 sacked &= ~(TCPCB_LOST|TCPCB_SACKED_RETRANS);
1230                                 tp->lost_out -= pcount;
1231                                 tp->retrans_out -= pcount;
1232                         }
1233                 } else {
1234                         if (!(sacked & TCPCB_RETRANS)) {
1235                                 /* New sack for not retransmitted frame,
1236                                  * which was in hole. It is reordering.
1237                                  */
1238                                 if (before(start_seq,
1239                                            tcp_highest_sack_seq(tp)) &&
1240                                     before(start_seq, state->reord))
1241                                         state->reord = start_seq;
1242 
1243                                 if (!after(end_seq, tp->high_seq))
1244                                         state->flag |= FLAG_ORIG_SACK_ACKED;
1245                                 if (state->first_sackt == 0)
1246                                         state->first_sackt = xmit_time;
1247                                 state->last_sackt = xmit_time;
1248                         }
1249 
1250                         if (sacked & TCPCB_LOST) {
1251                                 sacked &= ~TCPCB_LOST;
1252                                 tp->lost_out -= pcount;
1253                         }
1254                 }
1255 
1256                 sacked |= TCPCB_SACKED_ACKED;
1257                 state->flag |= FLAG_DATA_SACKED;
1258                 tp->sacked_out += pcount;
1259                 tp->delivered += pcount;  /* Out-of-order packets delivered */
1260 
1261                 /* Lost marker hint past SACKed? Tweak RFC3517 cnt */
1262                 if (tp->lost_skb_hint &&
1263                     before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq))
1264                         tp->lost_cnt_hint += pcount;
1265         }
1266 
1267         /* D-SACK. We can detect redundant retransmission in S|R and plain R
1268          * frames and clear it. undo_retrans is decreased above, L|R frames
1269          * are accounted above as well.
1270          */
1271         if (dup_sack && (sacked & TCPCB_SACKED_RETRANS)) {
1272                 sacked &= ~TCPCB_SACKED_RETRANS;
1273                 tp->retrans_out -= pcount;
1274         }
1275 
1276         return sacked;
1277 }
1278 
1279 /* Shift newly-SACKed bytes from this skb to the immediately previous
1280  * already-SACKed sk_buff. Mark the newly-SACKed bytes as such.
1281  */
1282 static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *prev,
1283                             struct sk_buff *skb,
1284                             struct tcp_sacktag_state *state,
1285                             unsigned int pcount, int shifted, int mss,
1286                             bool dup_sack)
1287 {
1288         struct tcp_sock *tp = tcp_sk(sk);
1289         u32 start_seq = TCP_SKB_CB(skb)->seq;   /* start of newly-SACKed */
1290         u32 end_seq = start_seq + shifted;      /* end of newly-SACKed */
1291 
1292         BUG_ON(!pcount);
1293 
1294         /* Adjust counters and hints for the newly sacked sequence
1295          * range but discard the return value since prev is already
1296          * marked. We must tag the range first because the seq
1297          * advancement below implicitly advances
1298          * tcp_highest_sack_seq() when skb is highest_sack.
1299          */
1300         tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,
1301                         start_seq, end_seq, dup_sack, pcount,
1302                         tcp_skb_timestamp_us(skb));
1303         tcp_rate_skb_delivered(sk, skb, state->rate);
1304 
1305         if (skb == tp->lost_skb_hint)
1306                 tp->lost_cnt_hint += pcount;
1307 
1308         TCP_SKB_CB(prev)->end_seq += shifted;
1309         TCP_SKB_CB(skb)->seq += shifted;
1310 
1311         tcp_skb_pcount_add(prev, pcount);
1312         WARN_ON_ONCE(tcp_skb_pcount(skb) < pcount);
1313         tcp_skb_pcount_add(skb, -pcount);
1314 
1315         /* When we're adding to gso_segs == 1, gso_size will be zero,
1316          * in theory this shouldn't be necessary but as long as DSACK
1317          * code can come after this skb later on it's better to keep
1318          * setting gso_size to something.
1319          */
1320         if (!TCP_SKB_CB(prev)->tcp_gso_size)
1321                 TCP_SKB_CB(prev)->tcp_gso_size = mss;
1322 
1323         /* CHECKME: To clear or not to clear? Mimics normal skb currently */
1324         if (tcp_skb_pcount(skb) <= 1)
1325                 TCP_SKB_CB(skb)->tcp_gso_size = 0;
1326 
1327         /* Difference in this won't matter, both ACKed by the same cumul. ACK */
1328         TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS);
1329 
1330         if (skb->len > 0) {
1331                 BUG_ON(!tcp_skb_pcount(skb));
1332                 NET_INC_STATS(sock_net(sk), LINUX_MIB_SACKSHIFTED);
1333                 return false;
1334         }
1335 
1336         /* Whole SKB was eaten :-) */
1337 
1338         if (skb == tp->retransmit_skb_hint)
1339                 tp->retransmit_skb_hint = prev;
1340         if (skb == tp->lost_skb_hint) {
1341                 tp->lost_skb_hint = prev;
1342                 tp->lost_cnt_hint -= tcp_skb_pcount(prev);
1343         }
1344 
1345         TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1346         TCP_SKB_CB(prev)->eor = TCP_SKB_CB(skb)->eor;
1347         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1348                 TCP_SKB_CB(prev)->end_seq++;
1349 
1350         if (skb == tcp_highest_sack(sk))
1351                 tcp_advance_highest_sack(sk, skb);
1352 
1353         tcp_skb_collapse_tstamp(prev, skb);
1354         if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp))
1355                 TCP_SKB_CB(prev)->tx.delivered_mstamp = 0;
1356 
1357         tcp_rtx_queue_unlink_and_free(skb, sk);
1358 
1359         NET_INC_STATS(sock_net(sk), LINUX_MIB_SACKMERGED);
1360 
1361         return true;
1362 }
1363 
1364 /* I wish gso_size would have a bit more sane initialization than
1365  * something-or-zero which complicates things
1366  */
1367 static int tcp_skb_seglen(const struct sk_buff *skb)
1368 {
1369         return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb);
1370 }
1371 
1372 /* Shifting pages past head area doesn't work */
1373 static int skb_can_shift(const struct sk_buff *skb)
1374 {
1375         return !skb_headlen(skb) && skb_is_nonlinear(skb);
1376 }
1377 
1378 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from,
1379                   int pcount, int shiftlen)
1380 {
1381         /* TCP min gso_size is 8 bytes (TCP_MIN_GSO_SIZE)
1382          * Since TCP_SKB_CB(skb)->tcp_gso_segs is 16 bits, we need
1383          * to make sure not storing more than 65535 * 8 bytes per skb,
1384          * even if current MSS is bigger.
1385          */
1386         if (unlikely(to->len + shiftlen >= 65535 * TCP_MIN_GSO_SIZE))
1387                 return 0;
1388         if (unlikely(tcp_skb_pcount(to) + pcount > 65535))
1389                 return 0;
1390         return skb_shift(to, from, shiftlen);
1391 }
1392 
1393 /* Try collapsing SACK blocks spanning across multiple skbs to a single
1394  * skb.
1395  */
1396 static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb,
1397                                           struct tcp_sacktag_state *state,
1398                                           u32 start_seq, u32 end_seq,
1399                                           bool dup_sack)
1400 {
1401         struct tcp_sock *tp = tcp_sk(sk);
1402         struct sk_buff *prev;
1403         int mss;
1404         int pcount = 0;
1405         int len;
1406         int in_sack;
1407 
1408         /* Normally R but no L won't result in plain S */
1409         if (!dup_sack &&
1410             (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS)
1411                 goto fallback;
1412         if (!skb_can_shift(skb))
1413                 goto fallback;
1414         /* This frame is about to be dropped (was ACKed). */
1415         if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
1416                 goto fallback;
1417 
1418         /* Can only happen with delayed DSACK + discard craziness */
1419         prev = skb_rb_prev(skb);
1420         if (!prev)
1421                 goto fallback;
1422 
1423         if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED)
1424                 goto fallback;
1425 
1426         if (!tcp_skb_can_collapse_to(prev))
1427                 goto fallback;
1428 
1429         in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
1430                   !before(end_seq, TCP_SKB_CB(skb)->end_seq);
1431 
1432         if (in_sack) {
1433                 len = skb->len;
1434                 pcount = tcp_skb_pcount(skb);
1435                 mss = tcp_skb_seglen(skb);
1436 
1437                 /* TODO: Fix DSACKs to not fragment already SACKed and we can
1438                  * drop this restriction as unnecessary
1439                  */
1440                 if (mss != tcp_skb_seglen(prev))
1441                         goto fallback;
1442         } else {
1443                 if (!after(TCP_SKB_CB(skb)->end_seq, start_seq))
1444                         goto noop;
1445                 /* CHECKME: This is non-MSS split case only?, this will
1446                  * cause skipped skbs due to advancing loop btw, original
1447                  * has that feature too
1448                  */
1449                 if (tcp_skb_pcount(skb) <= 1)
1450                         goto noop;
1451 
1452                 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);
1453                 if (!in_sack) {
1454                         /* TODO: head merge to next could be attempted here
1455                          * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)),
1456                          * though it might not be worth of the additional hassle
1457                          *
1458                          * ...we can probably just fallback to what was done
1459                          * previously. We could try merging non-SACKed ones
1460                          * as well but it probably isn't going to buy off
1461                          * because later SACKs might again split them, and
1462                          * it would make skb timestamp tracking considerably
1463                          * harder problem.
1464                          */
1465                         goto fallback;
1466                 }
1467 
1468                 len = end_seq - TCP_SKB_CB(skb)->seq;
1469                 BUG_ON(len < 0);
1470                 BUG_ON(len > skb->len);
1471 
1472                 /* MSS boundaries should be honoured or else pcount will
1473                  * severely break even though it makes things bit trickier.
1474                  * Optimize common case to avoid most of the divides
1475                  */
1476                 mss = tcp_skb_mss(skb);
1477 
1478                 /* TODO: Fix DSACKs to not fragment already SACKed and we can
1479                  * drop this restriction as unnecessary
1480                  */
1481                 if (mss != tcp_skb_seglen(prev))
1482                         goto fallback;
1483 
1484                 if (len == mss) {
1485                         pcount = 1;
1486                 } else if (len < mss) {
1487                         goto noop;
1488                 } else {
1489                         pcount = len / mss;
1490                         len = pcount * mss;
1491                 }
1492         }
1493 
1494         /* tcp_sacktag_one() won't SACK-tag ranges below snd_una */
1495         if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una))
1496                 goto fallback;
1497 
1498         if (!tcp_skb_shift(prev, skb, pcount, len))
1499                 goto fallback;
1500         if (!tcp_shifted_skb(sk, prev, skb, state, pcount, len, mss, dup_sack))
1501                 goto out;
1502 
1503         /* Hole filled allows collapsing with the next as well, this is very
1504          * useful when hole on every nth skb pattern happens
1505          */
1506         skb = skb_rb_next(prev);
1507         if (!skb)
1508                 goto out;
1509 
1510         if (!skb_can_shift(skb) ||
1511             ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) ||
1512             (mss != tcp_skb_seglen(skb)))
1513                 goto out;
1514 
1515         len = skb->len;
1516         pcount = tcp_skb_pcount(skb);
1517         if (tcp_skb_shift(prev, skb, pcount, len))
1518                 tcp_shifted_skb(sk, prev, skb, state, pcount,
1519                                 len, mss, 0);
1520 
1521 out:
1522         return prev;
1523 
1524 noop:
1525         return skb;
1526 
1527 fallback:
1528         NET_INC_STATS(sock_net(sk), LINUX_MIB_SACKSHIFTFALLBACK);
1529         return NULL;
1530 }
1531 
1532 static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk,
1533                                         struct tcp_sack_block *next_dup,
1534                                         struct tcp_sacktag_state *state,
1535                                         u32 start_seq, u32 end_seq,
1536                                         bool dup_sack_in)
1537 {
1538         struct tcp_sock *tp = tcp_sk(sk);
1539         struct sk_buff *tmp;
1540 
1541         skb_rbtree_walk_from(skb) {
1542                 int in_sack = 0;
1543                 bool dup_sack = dup_sack_in;
1544 
1545                 /* queue is in-order => we can short-circuit the walk early */
1546                 if (!before(TCP_SKB_CB(skb)->seq, end_seq))
1547                         break;
1548 
1549                 if (next_dup  &&
1550                     before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) {
1551                         in_sack = tcp_match_skb_to_sack(sk, skb,
1552                                                         next_dup->start_seq,
1553                                                         next_dup->end_seq);
1554                         if (in_sack > 0)
1555                                 dup_sack = true;
1556                 }
1557 
1558                 /* skb reference here is a bit tricky to get right, since
1559                  * shifting can eat and free both this skb and the next,
1560                  * so not even _safe variant of the loop is enough.
1561                  */
1562                 if (in_sack <= 0) {
1563                         tmp = tcp_shift_skb_data(sk, skb, state,
1564                                                  start_seq, end_seq, dup_sack);
1565                         if (tmp) {
1566                                 if (tmp != skb) {
1567                                         skb = tmp;
1568                                         continue;
1569                                 }
1570 
1571                                 in_sack = 0;
1572                         } else {
1573                                 in_sack = tcp_match_skb_to_sack(sk, skb,
1574                                                                 start_seq,
1575                                                                 end_seq);
1576                         }
1577                 }
1578 
1579                 if (unlikely(in_sack < 0))
1580                         break;
1581 
1582                 if (in_sack) {
1583                         TCP_SKB_CB(skb)->sacked =
1584                                 tcp_sacktag_one(sk,
1585                                                 state,
1586                                                 TCP_SKB_CB(skb)->sacked,
1587                                                 TCP_SKB_CB(skb)->seq,
1588                                                 TCP_SKB_CB(skb)->end_seq,
1589                                                 dup_sack,
1590                                                 tcp_skb_pcount(skb),
1591                                                 tcp_skb_timestamp_us(skb));
1592                         tcp_rate_skb_delivered(sk, skb, state->rate);
1593                         if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
1594                                 list_del_init(&skb->tcp_tsorted_anchor);
1595 
1596                         if (!before(TCP_SKB_CB(skb)->seq,
1597                                     tcp_highest_sack_seq(tp)))
1598                                 tcp_advance_highest_sack(sk, skb);
1599                 }
1600         }
1601         return skb;
1602 }
1603 
1604 static struct sk_buff *tcp_sacktag_bsearch(struct sock *sk, u32 seq)
1605 {
1606         struct rb_node *parent, **p = &sk->tcp_rtx_queue.rb_node;
1607         struct sk_buff *skb;
1608 
1609         while (*p) {
1610                 parent = *p;
1611                 skb = rb_to_skb(parent);
1612                 if (before(seq, TCP_SKB_CB(skb)->seq)) {
1613                         p = &parent->rb_left;
1614                         continue;
1615                 }
1616                 if (!before(seq, TCP_SKB_CB(skb)->end_seq)) {
1617                         p = &parent->rb_right;
1618                         continue;
1619                 }
1620                 return skb;
1621         }
1622         return NULL;
1623 }
1624 
1625 static struct sk_buff *tcp_sacktag_skip(struct sk_buff *skb, struct sock *sk,
1626                                         u32 skip_to_seq)
1627 {
1628         if (skb && after(TCP_SKB_CB(skb)->seq, skip_to_seq))
1629                 return skb;
1630 
1631         return tcp_sacktag_bsearch(sk, skip_to_seq);
1632 }
1633 
1634 static struct sk_buff *tcp_maybe_skipping_dsack(struct sk_buff *skb,
1635                                                 struct sock *sk,
1636                                                 struct tcp_sack_block *next_dup,
1637                                                 struct tcp_sacktag_state *state,
1638                                                 u32 skip_to_seq)
1639 {
1640         if (!next_dup)
1641                 return skb;
1642 
1643         if (before(next_dup->start_seq, skip_to_seq)) {
1644                 skb = tcp_sacktag_skip(skb, sk, next_dup->start_seq);
1645                 skb = tcp_sacktag_walk(skb, sk, NULL, state,
1646                                        next_dup->start_seq, next_dup->end_seq,
1647                                        1);
1648         }
1649 
1650         return skb;
1651 }
1652 
1653 static int tcp_sack_cache_ok(const struct tcp_sock *tp, const struct tcp_sack_block *cache)
1654 {
1655         return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);
1656 }
1657 
1658 static int
1659 tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
1660                         u32 prior_snd_una, struct tcp_sacktag_state *state)
1661 {
1662         struct tcp_sock *tp = tcp_sk(sk);
1663         const unsigned char *ptr = (skb_transport_header(ack_skb) +
1664                                     TCP_SKB_CB(ack_skb)->sacked);
1665         struct tcp_sack_block_wire *sp_wire = (struct tcp_sack_block_wire *)(ptr+2);
1666         struct tcp_sack_block sp[TCP_NUM_SACKS];
1667         struct tcp_sack_block *cache;
1668         struct sk_buff *skb;
1669         int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3);
1670         int used_sacks;
1671         bool found_dup_sack = false;
1672         int i, j;
1673         int first_sack_index;
1674 
1675         state->flag = 0;
1676         state->reord = tp->snd_nxt;
1677 
1678         if (!tp->sacked_out)
1679                 tcp_highest_sack_reset(sk);
1680 
1681         found_dup_sack = tcp_check_dsack(sk, ack_skb, sp_wire,
1682                                          num_sacks, prior_snd_una);
1683         if (found_dup_sack) {
1684                 state->flag |= FLAG_DSACKING_ACK;
1685                 tp->delivered++; /* A spurious retransmission is delivered */
1686         }
1687 
1688         /* Eliminate too old ACKs, but take into
1689          * account more or less fresh ones, they can
1690          * contain valid SACK info.
1691          */
1692         if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window))
1693                 return 0;
1694 
1695         if (!tp->packets_out)
1696                 goto out;
1697 
1698         used_sacks = 0;
1699         first_sack_index = 0;
1700         for (i = 0; i < num_sacks; i++) {
1701                 bool dup_sack = !i && found_dup_sack;
1702 
1703                 sp[used_sacks].start_seq = get_unaligned_be32(&sp_wire[i].start_seq);
1704                 sp[used_sacks].end_seq = get_unaligned_be32(&sp_wire[i].end_seq);
1705 
1706                 if (!tcp_is_sackblock_valid(tp, dup_sack,
1707                                             sp[used_sacks].start_seq,
1708                                             sp[used_sacks].end_seq)) {
1709                         int mib_idx;
1710 
1711                         if (dup_sack) {
1712                                 if (!tp->undo_marker)
1713                                         mib_idx = LINUX_MIB_TCPDSACKIGNOREDNOUNDO;
1714                                 else
1715                                         mib_idx = LINUX_MIB_TCPDSACKIGNOREDOLD;
1716                         } else {
1717                                 /* Don't count olds caused by ACK reordering */
1718                                 if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) &&
1719                                     !after(sp[used_sacks].end_seq, tp->snd_una))
1720                                         continue;
1721                                 mib_idx = LINUX_MIB_TCPSACKDISCARD;
1722                         }
1723 
1724                         NET_INC_STATS(sock_net(sk), mib_idx);
1725                         if (i == 0)
1726                                 first_sack_index = -1;
1727                         continue;
1728                 }
1729 
1730                 /* Ignore very old stuff early */
1731                 if (!after(sp[used_sacks].end_seq, prior_snd_una)) {
1732                         if (i == 0)
1733                                 first_sack_index = -1;
1734                         continue;
1735                 }
1736 
1737                 used_sacks++;
1738         }
1739 
1740         /* order SACK blocks to allow in order walk of the retrans queue */
1741         for (i = used_sacks - 1; i > 0; i--) {
1742                 for (j = 0; j < i; j++) {
1743                         if (after(sp[j].start_seq, sp[j + 1].start_seq)) {
1744                                 swap(sp[j], sp[j + 1]);
1745 
1746                                 /* Track where the first SACK block goes to */
1747                                 if (j == first_sack_index)
1748                                         first_sack_index = j + 1;
1749                         }
1750                 }
1751         }
1752 
1753         state->mss_now = tcp_current_mss(sk);
1754         skb = NULL;
1755         i = 0;
1756 
1757         if (!tp->sacked_out) {
1758                 /* It's already past, so skip checking against it */
1759                 cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);
1760         } else {
1761                 cache = tp->recv_sack_cache;
1762                 /* Skip empty blocks in at head of the cache */
1763                 while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq &&
1764                        !cache->end_seq)
1765                         cache++;
1766         }
1767 
1768         while (i < used_sacks) {
1769                 u32 start_seq = sp[i].start_seq;
1770                 u32 end_seq = sp[i].end_seq;
1771                 bool dup_sack = (found_dup_sack && (i == first_sack_index));
1772                 struct tcp_sack_block *next_dup = NULL;
1773 
1774                 if (found_dup_sack && ((i + 1) == first_sack_index))
1775                         next_dup = &sp[i + 1];
1776 
1777                 /* Skip too early cached blocks */
1778                 while (tcp_sack_cache_ok(tp, cache) &&
1779                        !before(start_seq, cache->end_seq))
1780                         cache++;
1781 
1782                 /* Can skip some work by looking recv_sack_cache? */
1783                 if (tcp_sack_cache_ok(tp, cache) && !dup_sack &&
1784                     after(end_seq, cache->start_seq)) {
1785 
1786                         /* Head todo? */
1787                         if (before(start_seq, cache->start_seq)) {
1788                                 skb = tcp_sacktag_skip(skb, sk, start_seq);
1789                                 skb = tcp_sacktag_walk(skb, sk, next_dup,
1790                                                        state,
1791                                                        start_seq,
1792                                                        cache->start_seq,
1793                                                        dup_sack);
1794                         }
1795 
1796                         /* Rest of the block already fully processed? */
1797                         if (!after(end_seq, cache->end_seq))
1798                                 goto advance_sp;
1799 
1800                         skb = tcp_maybe_skipping_dsack(skb, sk, next_dup,
1801                                                        state,
1802                                                        cache->end_seq);
1803 
1804                         /* ...tail remains todo... */
1805                         if (tcp_highest_sack_seq(tp) == cache->end_seq) {
1806                                 /* ...but better entrypoint exists! */
1807                                 skb = tcp_highest_sack(sk);
1808                                 if (!skb)
1809                                         break;
1810                                 cache++;
1811                                 goto walk;
1812                         }
1813 
1814                         skb = tcp_sacktag_skip(skb, sk, cache->end_seq);
1815                         /* Check overlap against next cached too (past this one already) */
1816                         cache++;
1817                         continue;
1818                 }
1819 
1820                 if (!before(start_seq, tcp_highest_sack_seq(tp))) {
1821                         skb = tcp_highest_sack(sk);
1822                         if (!skb)
1823                                 break;
1824                 }
1825                 skb = tcp_sacktag_skip(skb, sk, start_seq);
1826 
1827 walk:
1828                 skb = tcp_sacktag_walk(skb, sk, next_dup, state,
1829                                        start_seq, end_seq, dup_sack);
1830 
1831 advance_sp:
1832                 i++;
1833         }
1834 
1835         /* Clear the head of the cache sack blocks so we can skip it next time */
1836         for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) {
1837                 tp->recv_sack_cache[i].start_seq = 0;
1838                 tp->recv_sack_cache[i].end_seq = 0;
1839         }
1840         for (j = 0; j < used_sacks; j++)
1841                 tp->recv_sack_cache[i++] = sp[j];
1842 
1843         if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss || tp->undo_marker)
1844                 tcp_check_sack_reordering(sk, state->reord, 0);
1845 
1846         tcp_verify_left_out(tp);
1847 out:
1848 
1849 #if FASTRETRANS_DEBUG > 0
1850         WARN_ON((int)tp->sacked_out < 0);
1851         WARN_ON((int)tp->lost_out < 0);
1852         WARN_ON((int)tp->retrans_out < 0);
1853         WARN_ON((int)tcp_packets_in_flight(tp) < 0);
1854 #endif
1855         return state->flag;
1856 }
1857 
1858 /* Limits sacked_out so that sum with lost_out isn't ever larger than
1859  * packets_out. Returns false if sacked_out adjustement wasn't necessary.
1860  */
1861 static bool tcp_limit_reno_sacked(struct tcp_sock *tp)
1862 {
1863         u32 holes;
1864 
1865         holes = max(tp->lost_out, 1U);
1866         holes = min(holes, tp->packets_out);
1867 
1868         if ((tp->sacked_out + holes) > tp->packets_out) {
1869                 tp->sacked_out = tp->packets_out - holes;
1870                 return true;
1871         }
1872         return false;
1873 }
1874 
1875 /* If we receive more dupacks than we expected counting segments
1876  * in assumption of absent reordering, interpret this as reordering.
1877  * The only another reason could be bug in receiver TCP.
1878  */
1879 static void tcp_check_reno_reordering(struct sock *sk, const int addend)
1880 {
1881         struct tcp_sock *tp = tcp_sk(sk);
1882 
1883         if (!tcp_limit_reno_sacked(tp))
1884                 return;
1885 
1886         tp->reordering = min_t(u32, tp->packets_out + addend,
1887                                sock_net(sk)->ipv4.sysctl_tcp_max_reordering);
1888         tp->reord_seen++;
1889         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRENOREORDER);
1890 }
1891 
1892 /* Emulate SACKs for SACKless connection: account for a new dupack. */
1893 
1894 static void tcp_add_reno_sack(struct sock *sk, int num_dupack)
1895 {
1896         if (num_dupack) {
1897                 struct tcp_sock *tp = tcp_sk(sk);
1898                 u32 prior_sacked = tp->sacked_out;
1899                 s32 delivered;
1900 
1901                 tp->sacked_out += num_dupack;
1902                 tcp_check_reno_reordering(sk, 0);
1903                 delivered = tp->sacked_out - prior_sacked;
1904                 if (delivered > 0)
1905                         tp->delivered += delivered;
1906                 tcp_verify_left_out(tp);
1907         }
1908 }
1909 
1910 /* Account for ACK, ACKing some data in Reno Recovery phase. */
1911 
1912 static void tcp_remove_reno_sacks(struct sock *sk, int acked)
1913 {
1914         struct tcp_sock *tp = tcp_sk(sk);
1915 
1916         if (acked > 0) {
1917                 /* One ACK acked hole. The rest eat duplicate ACKs. */
1918                 tp->delivered += max_t(int, acked - tp->sacked_out, 1);
1919                 if (acked - 1 >= tp->sacked_out)
1920                         tp->sacked_out = 0;
1921                 else
1922                         tp->sacked_out -= acked - 1;
1923         }
1924         tcp_check_reno_reordering(sk, acked);
1925         tcp_verify_left_out(tp);
1926 }
1927 
1928 static inline void tcp_reset_reno_sack(struct tcp_sock *tp)
1929 {
1930         tp->sacked_out = 0;
1931 }
1932 
1933 void tcp_clear_retrans(struct tcp_sock *tp)
1934 {
1935         tp->retrans_out = 0;
1936         tp->lost_out = 0;
1937         tp->undo_marker = 0;
1938         tp->undo_retrans = -1;
1939         tp->sacked_out = 0;
1940 }
1941 
1942 static inline void tcp_init_undo(struct tcp_sock *tp)
1943 {
1944         tp->undo_marker = tp->snd_una;
1945         /* Retransmission still in flight may cause DSACKs later. */
1946         tp->undo_retrans = tp->retrans_out ? : -1;
1947 }
1948 
1949 static bool tcp_is_rack(const struct sock *sk)
1950 {
1951         return sock_net(sk)->ipv4.sysctl_tcp_recovery & TCP_RACK_LOSS_DETECTION;
1952 }
1953 
1954 /* If we detect SACK reneging, forget all SACK information
1955  * and reset tags completely, otherwise preserve SACKs. If receiver
1956  * dropped its ofo queue, we will know this due to reneging detection.
1957  */
1958 static void tcp_timeout_mark_lost(struct sock *sk)
1959 {
1960         struct tcp_sock *tp = tcp_sk(sk);
1961         struct sk_buff *skb, *head;
1962         bool is_reneg;                  /* is receiver reneging on SACKs? */
1963 
1964         head = tcp_rtx_queue_head(sk);
1965         is_reneg = head && (TCP_SKB_CB(head)->sacked & TCPCB_SACKED_ACKED);
1966         if (is_reneg) {
1967                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSACKRENEGING);
1968                 tp->sacked_out = 0;
1969                 /* Mark SACK reneging until we recover from this loss event. */
1970                 tp->is_sack_reneg = 1;
1971         } else if (tcp_is_reno(tp)) {
1972                 tcp_reset_reno_sack(tp);
1973         }
1974 
1975         skb = head;
1976         skb_rbtree_walk_from(skb) {
1977                 if (is_reneg)
1978                         TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED;
1979                 else if (tcp_is_rack(sk) && skb != head &&
1980                          tcp_rack_skb_timeout(tp, skb, 0) > 0)
1981                         continue; /* Don't mark recently sent ones lost yet */
1982                 tcp_mark_skb_lost(sk, skb);
1983         }
1984         tcp_verify_left_out(tp);
1985         tcp_clear_all_retrans_hints(tp);
1986 }
1987 
1988 /* Enter Loss state. */
1989 void tcp_enter_loss(struct sock *sk)
1990 {
1991         const struct inet_connection_sock *icsk = inet_csk(sk);
1992         struct tcp_sock *tp = tcp_sk(sk);
1993         struct net *net = sock_net(sk);
1994         bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery;
1995 
1996         tcp_timeout_mark_lost(sk);
1997 
1998         /* Reduce ssthresh if it has not yet been made inside this window. */
1999         if (icsk->icsk_ca_state <= TCP_CA_Disorder ||
2000             !after(tp->high_seq, tp->snd_una) ||
2001             (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) {
2002                 tp->prior_ssthresh = tcp_current_ssthresh(sk);
2003                 tp->prior_cwnd = tp->snd_cwnd;
2004                 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
2005                 tcp_ca_event(sk, CA_EVENT_LOSS);
2006                 tcp_init_undo(tp);
2007         }
2008         tp->snd_cwnd       = tcp_packets_in_flight(tp) + 1;
2009         tp->snd_cwnd_cnt   = 0;
2010         tp->snd_cwnd_stamp = tcp_jiffies32;
2011 
2012         /* Timeout in disordered state after receiving substantial DUPACKs
2013          * suggests that the degree of reordering is over-estimated.
2014          */
2015         if (icsk->icsk_ca_state <= TCP_CA_Disorder &&
2016             tp->sacked_out >= net->ipv4.sysctl_tcp_reordering)
2017                 tp->reordering = min_t(unsigned int, tp->reordering,
2018                                        net->ipv4.sysctl_tcp_reordering);
2019         tcp_set_ca_state(sk, TCP_CA_Loss);
2020         tp->high_seq = tp->snd_nxt;
2021         tcp_ecn_queue_cwr(tp);
2022 
2023         /* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous
2024          * loss recovery is underway except recurring timeout(s) on
2025          * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing
2026          */
2027         tp->frto = net->ipv4.sysctl_tcp_frto &&
2028                    (new_recovery || icsk->icsk_retransmits) &&
2029                    !inet_csk(sk)->icsk_mtup.probe_size;
2030 }
2031 
2032 /* If ACK arrived pointing to a remembered SACK, it means that our
2033  * remembered SACKs do not reflect real state of receiver i.e.
2034  * receiver _host_ is heavily congested (or buggy).
2035  *
2036  * To avoid big spurious retransmission bursts due to transient SACK
2037  * scoreboard oddities that look like reneging, we give the receiver a
2038  * little time (max(RTT/2, 10ms)) to send us some more ACKs that will
2039  * restore sanity to the SACK scoreboard. If the apparent reneging
2040  * persists until this RTO then we'll clear the SACK scoreboard.
2041  */
2042 static bool tcp_check_sack_reneging(struct sock *sk, int flag)
2043 {
2044         if (flag & FLAG_SACK_RENEGING) {
2045                 struct tcp_sock *tp = tcp_sk(sk);
2046                 unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4),
2047                                           msecs_to_jiffies(10));
2048 
2049                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
2050                                           delay, TCP_RTO_MAX);
2051                 return true;
2052         }
2053         return false;
2054 }
2055 
2056 /* Heurestics to calculate number of duplicate ACKs. There's no dupACKs
2057  * counter when SACK is enabled (without SACK, sacked_out is used for
2058  * that purpose).
2059  *
2060  * With reordering, holes may still be in flight, so RFC3517 recovery
2061  * uses pure sacked_out (total number of SACKed segments) even though
2062  * it violates the RFC that uses duplicate ACKs, often these are equal
2063  * but when e.g. out-of-window ACKs or packet duplication occurs,
2064  * they differ. Since neither occurs due to loss, TCP should really
2065  * ignore them.
2066  */
2067 static inline int tcp_dupack_heuristics(const struct tcp_sock *tp)
2068 {
2069         return tp->sacked_out + 1;
2070 }
2071 
2072 /* Linux NewReno/SACK/ECN state machine.
2073  * --------------------------------------
2074  *
2075  * "Open"       Normal state, no dubious events, fast path.
2076  * "Disorder"   In all the respects it is "Open",
2077  *              but requires a bit more attention. It is entered when
2078  *              we see some SACKs or dupacks. It is split of "Open"
2079  *              mainly to move some processing from fast path to slow one.
2080  * "CWR"        CWND was reduced due to some Congestion Notification event.
2081  *              It can be ECN, ICMP source quench, local device congestion.
2082  * "Recovery"   CWND was reduced, we are fast-retransmitting.
2083  * "Loss"       CWND was reduced due to RTO timeout or SACK reneging.
2084  *
2085  * tcp_fastretrans_alert() is entered:
2086  * - each incoming ACK, if state is not "Open"
2087  * - when arrived ACK is unusual, namely:
2088  *      * SACK
2089  *      * Duplicate ACK.
2090  *      * ECN ECE.
2091  *
2092  * Counting packets in flight is pretty simple.
2093  *
2094  *      in_flight = packets_out - left_out + retrans_out
2095  *
2096  *      packets_out is SND.NXT-SND.UNA counted in packets.
2097  *
2098  *      retrans_out is number of retransmitted segments.
2099  *
2100  *      left_out is number of segments left network, but not ACKed yet.
2101  *
2102  *              left_out = sacked_out + lost_out
2103  *
2104  *     sacked_out: Packets, which arrived to receiver out of order
2105  *                 and hence not ACKed. With SACKs this number is simply
2106  *                 amount of SACKed data. Even without SACKs
2107  *                 it is easy to give pretty reliable estimate of this number,
2108  *                 counting duplicate ACKs.
2109  *
2110  *       lost_out: Packets lost by network. TCP has no explicit
2111  *                 "loss notification" feedback from network (for now).
2112  *                 It means that this number can be only _guessed_.
2113  *                 Actually, it is the heuristics to predict lossage that
2114  *                 distinguishes different algorithms.
2115  *
2116  *      F.e. after RTO, when all the queue is considered as lost,
2117  *      lost_out = packets_out and in_flight = retrans_out.
2118  *
2119  *              Essentially, we have now a few algorithms detecting
2120  *              lost packets.
2121  *
2122  *              If the receiver supports SACK:
2123  *
2124  *              RFC6675/3517: It is the conventional algorithm. A packet is
2125  *              considered lost if the number of higher sequence packets
2126  *              SACKed is greater than or equal the DUPACK thoreshold
2127  *              (reordering). This is implemented in tcp_mark_head_lost and
2128  *              tcp_update_scoreboard.
2129  *
2130  *              RACK (draft-ietf-tcpm-rack-01): it is a newer algorithm
2131  *              (2017-) that checks timing instead of counting DUPACKs.
2132  *              Essentially a packet is considered lost if it's not S/ACKed
2133  *              after RTT + reordering_window, where both metrics are
2134  *              dynamically measured and adjusted. This is implemented in
2135  *              tcp_rack_mark_lost.
2136  *
2137  *              If the receiver does not support SACK:
2138  *
2139  *              NewReno (RFC6582): in Recovery we assume that one segment
2140  *              is lost (classic Reno). While we are in Recovery and
2141  *              a partial ACK arrives, we assume that one more packet
2142  *              is lost (NewReno). This heuristics are the same in NewReno
2143  *              and SACK.
2144  *
2145  * Really tricky (and requiring careful tuning) part of algorithm
2146  * is hidden in functions tcp_time_to_recover() and tcp_xmit_retransmit_queue().
2147  * The first determines the moment _when_ we should reduce CWND and,
2148  * hence, slow down forward transmission. In fact, it determines the moment
2149  * when we decide that hole is caused by loss, rather than by a reorder.
2150  *
2151  * tcp_xmit_retransmit_queue() decides, _what_ we should retransmit to fill
2152  * holes, caused by lost packets.
2153  *
2154  * And the most logically complicated part of algorithm is undo
2155  * heuristics. We detect false retransmits due to both too early
2156  * fast retransmit (reordering) and underestimated RTO, analyzing
2157  * timestamps and D-SACKs. When we detect that some segments were
2158  * retransmitted by mistake and CWND reduction was wrong, we undo
2159  * window reduction and abort recovery phase. This logic is hidden
2160  * inside several functions named tcp_try_undo_<something>.
2161  */
2162 
2163 /* This function decides, when we should leave Disordered state
2164  * and enter Recovery phase, reducing congestion window.
2165  *
2166  * Main question: may we further continue forward transmission
2167  * with the same cwnd?
2168  */
2169 static bool tcp_time_to_recover(struct sock *sk, int flag)
2170 {
2171         struct tcp_sock *tp = tcp_sk(sk);
2172 
2173         /* Trick#1: The loss is proven. */
2174         if (tp->lost_out)
2175                 return true;
2176 
2177         /* Not-A-Trick#2 : Classic rule... */
2178         if (!tcp_is_rack(sk) && tcp_dupack_heuristics(tp) > tp->reordering)
2179                 return true;
2180 
2181         return false;
2182 }
2183 
2184 /* Detect loss in event "A" above by marking head of queue up as lost.
2185  * For non-SACK(Reno) senders, the first "packets" number of segments
2186  * are considered lost. For RFC3517 SACK, a segment is considered lost if it
2187  * has at least tp->reordering SACKed seqments above it; "packets" refers to
2188  * the maximum SACKed segments to pass before reaching this limit.
2189  */
2190 static void tcp_mark_head_lost(struct sock *sk, int packets, int mark_head)
2191 {
2192         struct tcp_sock *tp = tcp_sk(sk);
2193         struct sk_buff *skb;
2194         int cnt, oldcnt, lost;
2195         unsigned int mss;
2196         /* Use SACK to deduce losses of new sequences sent during recovery */
2197         const u32 loss_high = tcp_is_sack(tp) ?  tp->snd_nxt : tp->high_seq;
2198 
2199         WARN_ON(packets > tp->packets_out);
2200         skb = tp->lost_skb_hint;
2201         if (skb) {
2202                 /* Head already handled? */
2203                 if (mark_head && after(TCP_SKB_CB(skb)->seq, tp->snd_una))
2204                         return;
2205                 cnt = tp->lost_cnt_hint;
2206         } else {
2207                 skb = tcp_rtx_queue_head(sk);
2208                 cnt = 0;
2209         }
2210 
2211         skb_rbtree_walk_from(skb) {
2212                 /* TODO: do this better */
2213                 /* this is not the most efficient way to do this... */
2214                 tp->lost_skb_hint = skb;
2215                 tp->lost_cnt_hint = cnt;
2216 
2217                 if (after(TCP_SKB_CB(skb)->end_seq, loss_high))
2218                         break;
2219 
2220                 oldcnt = cnt;
2221                 if (tcp_is_reno(tp) ||
2222                     (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
2223                         cnt += tcp_skb_pcount(skb);
2224 
2225                 if (cnt > packets) {
2226                         if (tcp_is_sack(tp) ||
2227                             (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) ||
2228                             (oldcnt >= packets))
2229                                 break;
2230 
2231                         mss = tcp_skb_mss(skb);
2232                         /* If needed, chop off the prefix to mark as lost. */
2233                         lost = (packets - oldcnt) * mss;
2234                         if (lost < skb->len &&
2235                             tcp_fragment(sk, TCP_FRAG_IN_RTX_QUEUE, skb,
2236                                          lost, mss, GFP_ATOMIC) < 0)
2237                                 break;
2238                         cnt = packets;
2239                 }
2240 
2241                 tcp_skb_mark_lost(tp, skb);
2242 
2243                 if (mark_head)
2244                         break;
2245         }
2246         tcp_verify_left_out(tp);
2247 }
2248 
2249 /* Account newly detected lost packet(s) */
2250 
2251 static void tcp_update_scoreboard(struct sock *sk, int fast_rexmit)
2252 {
2253         struct tcp_sock *tp = tcp_sk(sk);
2254 
2255         if (tcp_is_sack(tp)) {
2256                 int sacked_upto = tp->sacked_out - tp->reordering;
2257                 if (sacked_upto >= 0)
2258                         tcp_mark_head_lost(sk, sacked_upto, 0);
2259                 else if (fast_rexmit)
2260                         tcp_mark_head_lost(sk, 1, 1);
2261         }
2262 }
2263 
2264 static bool tcp_tsopt_ecr_before(const struct tcp_sock *tp, u32 when)
2265 {
2266         return tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
2267                before(tp->rx_opt.rcv_tsecr, when);
2268 }
2269 
2270 /* skb is spurious retransmitted if the returned timestamp echo
2271  * reply is prior to the skb transmission time
2272  */
2273 static bool tcp_skb_spurious_retrans(const struct tcp_sock *tp,
2274                                      const struct sk_buff *skb)
2275 {
2276         return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) &&
2277                tcp_tsopt_ecr_before(tp, tcp_skb_timestamp(skb));
2278 }
2279 
2280 /* Nothing was retransmitted or returned timestamp is less
2281  * than timestamp of the first retransmission.
2282  */
2283 static inline bool tcp_packet_delayed(const struct tcp_sock *tp)
2284 {
2285         return tp->retrans_stamp &&
2286                tcp_tsopt_ecr_before(tp, tp->retrans_stamp);
2287 }
2288 
2289 /* Undo procedures. */
2290 
2291 /* We can clear retrans_stamp when there are no retransmissions in the
2292  * window. It would seem that it is trivially available for us in
2293  * tp->retrans_out, however, that kind of assumptions doesn't consider
2294  * what will happen if errors occur when sending retransmission for the
2295  * second time. ...It could the that such segment has only
2296  * TCPCB_EVER_RETRANS set at the present time. It seems that checking
2297  * the head skb is enough except for some reneging corner cases that
2298  * are not worth the effort.
2299  *
2300  * Main reason for all this complexity is the fact that connection dying
2301  * time now depends on the validity of the retrans_stamp, in particular,
2302  * that successive retransmissions of a segment must not advance
2303  * retrans_stamp under any conditions.
2304  */
2305 static bool tcp_any_retrans_done(const struct sock *sk)
2306 {
2307         const struct tcp_sock *tp = tcp_sk(sk);
2308         struct sk_buff *skb;
2309 
2310         if (tp->retrans_out)
2311                 return true;
2312 
2313         skb = tcp_rtx_queue_head(sk);
2314         if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS))
2315                 return true;
2316 
2317         return false;
2318 }
2319 
2320 static void DBGUNDO(struct sock *sk, const char *msg)
2321 {
2322 #if FASTRETRANS_DEBUG > 1
2323         struct tcp_sock *tp = tcp_sk(sk);
2324         struct inet_sock *inet = inet_sk(sk);
2325 
2326         if (sk->sk_family == AF_INET) {
2327                 pr_debug("Undo %s %pI4/%u c%u l%u ss%u/%u p%u\n",
2328                          msg,
2329                          &inet->inet_daddr, ntohs(inet->inet_dport),
2330                          tp->snd_cwnd, tcp_left_out(tp),
2331                          tp->snd_ssthresh, tp->prior_ssthresh,
2332                          tp->packets_out);
2333         }
2334 #if IS_ENABLED(CONFIG_IPV6)
2335         else if (sk->sk_family == AF_INET6) {
2336                 pr_debug("Undo %s %pI6/%u c%u l%u ss%u/%u p%u\n",
2337                          msg,
2338                          &sk->sk_v6_daddr, ntohs(inet->inet_dport),
2339                          tp->snd_cwnd, tcp_left_out(tp),
2340                          tp->snd_ssthresh, tp->prior_ssthresh,
2341                          tp->packets_out);
2342         }
2343 #endif
2344 #endif
2345 }
2346 
2347 static void tcp_undo_cwnd_reduction(struct sock *sk, bool unmark_loss)
2348 {
2349         struct tcp_sock *tp = tcp_sk(sk);
2350 
2351         if (unmark_loss) {
2352                 struct sk_buff *skb;
2353 
2354                 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
2355                         TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST;
2356                 }
2357                 tp->lost_out = 0;
2358                 tcp_clear_all_retrans_hints(tp);
2359         }
2360 
2361         if (tp->prior_ssthresh) {
2362                 const struct inet_connection_sock *icsk = inet_csk(sk);
2363 
2364                 tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk);
2365 
2366                 if (tp->prior_ssthresh > tp->snd_ssthresh) {
2367                         tp->snd_ssthresh = tp->prior_ssthresh;
2368                         tcp_ecn_withdraw_cwr(tp);
2369                 }
2370         }
2371         tp->snd_cwnd_stamp = tcp_jiffies32;
2372         tp->undo_marker = 0;
2373         tp->rack.advanced = 1; /* Force RACK to re-exam losses */
2374 }
2375 
2376 static inline bool tcp_may_undo(const struct tcp_sock *tp)
2377 {
2378         return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp));
2379 }
2380 
2381 /* People celebrate: "We love our President!" */
2382 static bool tcp_try_undo_recovery(struct sock *sk)
2383 {
2384         struct tcp_sock *tp = tcp_sk(sk);
2385 
2386         if (tcp_may_undo(tp)) {
2387                 int mib_idx;
2388 
2389                 /* Happy end! We did not retransmit anything
2390                  * or our original transmission succeeded.
2391                  */
2392                 DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans");
2393                 tcp_undo_cwnd_reduction(sk, false);
2394                 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss)
2395                         mib_idx = LINUX_MIB_TCPLOSSUNDO;
2396                 else
2397                         mib_idx = LINUX_MIB_TCPFULLUNDO;
2398 
2399                 NET_INC_STATS(sock_net(sk), mib_idx);
2400         } else if (tp->rack.reo_wnd_persist) {
2401                 tp->rack.reo_wnd_persist--;
2402         }
2403         if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) {
2404                 /* Hold old state until something *above* high_seq
2405                  * is ACKed. For Reno it is MUST to prevent false
2406                  * fast retransmits (RFC2582). SACK TCP is safe. */
2407                 if (!tcp_any_retrans_done(sk))
2408                         tp->retrans_stamp = 0;
2409                 return true;
2410         }
2411         tcp_set_ca_state(sk, TCP_CA_Open);
2412         tp->is_sack_reneg = 0;
2413         return false;
2414 }
2415 
2416 /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
2417 static bool tcp_try_undo_dsack(struct sock *sk)
2418 {
2419         struct tcp_sock *tp = tcp_sk(sk);
2420 
2421         if (tp->undo_marker && !tp->undo_retrans) {
2422                 tp->rack.reo_wnd_persist = min(TCP_RACK_RECOVERY_THRESH,
2423                                                tp->rack.reo_wnd_persist + 1);
2424                 DBGUNDO(sk, "D-SACK");
2425                 tcp_undo_cwnd_reduction(sk, false);
2426                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDSACKUNDO);
2427                 return true;
2428         }
2429         return false;
2430 }
2431 
2432 /* Undo during loss recovery after partial ACK or using F-RTO. */
2433 static bool tcp_try_undo_loss(struct sock *sk, bool frto_undo)
2434 {
2435         struct tcp_sock *tp = tcp_sk(sk);
2436 
2437         if (frto_undo || tcp_may_undo(tp)) {
2438                 tcp_undo_cwnd_reduction(sk, true);
2439 
2440                 DBGUNDO(sk, "partial loss");
2441                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSUNDO);
2442                 if (frto_undo)
2443                         NET_INC_STATS(sock_net(sk),
2444                                         LINUX_MIB_TCPSPURIOUSRTOS);
2445                 inet_csk(sk)->icsk_retransmits = 0;
2446                 if (frto_undo || tcp_is_sack(tp)) {
2447                         tcp_set_ca_state(sk, TCP_CA_Open);
2448                         tp->is_sack_reneg = 0;
2449                 }
2450                 return true;
2451         }
2452         return false;
2453 }
2454 
2455 /* The cwnd reduction in CWR and Recovery uses the PRR algorithm in RFC 6937.
2456  * It computes the number of packets to send (sndcnt) based on packets newly
2457  * delivered:
2458  *   1) If the packets in flight is larger than ssthresh, PRR spreads the
2459  *      cwnd reductions across a full RTT.
2460  *   2) Otherwise PRR uses packet conservation to send as much as delivered.
2461  *      But when the retransmits are acked without further losses, PRR
2462  *      slow starts cwnd up to ssthresh to speed up the recovery.
2463  */
2464 static void tcp_init_cwnd_reduction(struct sock *sk)
2465 {
2466         struct tcp_sock *tp = tcp_sk(sk);
2467 
2468         tp->high_seq = tp->snd_nxt;
2469         tp->tlp_high_seq = 0;
2470         tp->snd_cwnd_cnt = 0;
2471         tp->prior_cwnd = tp->snd_cwnd;
2472         tp->prr_delivered = 0;
2473         tp->prr_out = 0;
2474         tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk);
2475         tcp_ecn_queue_cwr(tp);
2476 }
2477 
2478 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag)
2479 {
2480         struct tcp_sock *tp = tcp_sk(sk);
2481         int sndcnt = 0;
2482         int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp);
2483 
2484         if (newly_acked_sacked <= 0 || WARN_ON_ONCE(!tp->prior_cwnd))
2485                 return;
2486 
2487         tp->prr_delivered += newly_acked_sacked;
2488         if (delta < 0) {
2489                 u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered +
2490                                tp->prior_cwnd - 1;
2491                 sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out;
2492         } else if ((flag & (FLAG_RETRANS_DATA_ACKED | FLAG_LOST_RETRANS)) ==
2493                    FLAG_RETRANS_DATA_ACKED) {
2494                 sndcnt = min_t(int, delta,
2495                                max_t(int, tp->prr_delivered - tp->prr_out,
2496                                      newly_acked_sacked) + 1);
2497         } else {
2498                 sndcnt = min(delta, newly_acked_sacked);
2499         }
2500         /* Force a fast retransmit upon entering fast recovery */
2501         sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1));
2502         tp->snd_cwnd = tcp_packets_in_flight(tp) + sndcnt;
2503 }
2504 
2505 static inline void tcp_end_cwnd_reduction(struct sock *sk)
2506 {
2507         struct tcp_sock *tp = tcp_sk(sk);
2508 
2509         if (inet_csk(sk)->icsk_ca_ops->cong_control)
2510                 return;
2511 
2512         /* Reset cwnd to ssthresh in CWR or Recovery (unless it's undone) */
2513         if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH &&
2514             (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) {
2515                 tp->snd_cwnd = tp->snd_ssthresh;
2516                 tp->snd_cwnd_stamp = tcp_jiffies32;
2517         }
2518         tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR);
2519 }
2520 
2521 /* Enter CWR state. Disable cwnd undo since congestion is proven with ECN */
2522 void tcp_enter_cwr(struct sock *sk)
2523 {
2524         struct tcp_sock *tp = tcp_sk(sk);
2525 
2526         tp->prior_ssthresh = 0;
2527         if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {
2528                 tp->undo_marker = 0;
2529                 tcp_init_cwnd_reduction(sk);
2530                 tcp_set_ca_state(sk, TCP_CA_CWR);
2531         }
2532 }
2533 EXPORT_SYMBOL(tcp_enter_cwr);
2534 
2535 static void tcp_try_keep_open(struct sock *sk)
2536 {
2537         struct tcp_sock *tp = tcp_sk(sk);
2538         int state = TCP_CA_Open;
2539 
2540         if (tcp_left_out(tp) || tcp_any_retrans_done(sk))
2541                 state = TCP_CA_Disorder;
2542 
2543         if (inet_csk(sk)->icsk_ca_state != state) {
2544                 tcp_set_ca_state(sk, state);
2545                 tp->high_seq = tp->snd_nxt;
2546         }
2547 }
2548 
2549 static void tcp_try_to_open(struct sock *sk, int flag)
2550 {
2551         struct tcp_sock *tp = tcp_sk(sk);
2552 
2553         tcp_verify_left_out(tp);
2554 
2555         if (!tcp_any_retrans_done(sk))
2556                 tp->retrans_stamp = 0;
2557 
2558         if (flag & FLAG_ECE)
2559                 tcp_enter_cwr(sk);
2560 
2561         if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {
2562                 tcp_try_keep_open(sk);
2563         }
2564 }
2565 
2566 static void tcp_mtup_probe_failed(struct sock *sk)
2567 {
2568         struct inet_connection_sock *icsk = inet_csk(sk);
2569 
2570         icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1;
2571         icsk->icsk_mtup.probe_size = 0;
2572         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMTUPFAIL);
2573 }
2574 
2575 static void tcp_mtup_probe_success(struct sock *sk)
2576 {
2577         struct tcp_sock *tp = tcp_sk(sk);
2578         struct inet_connection_sock *icsk = inet_csk(sk);
2579 
2580         /* FIXME: breaks with very large cwnd */
2581         tp->prior_ssthresh = tcp_current_ssthresh(sk);
2582         tp->snd_cwnd = tp->snd_cwnd *
2583                        tcp_mss_to_mtu(sk, tp->mss_cache) /
2584                        icsk->icsk_mtup.probe_size;
2585         tp->snd_cwnd_cnt = 0;
2586         tp->snd_cwnd_stamp = tcp_jiffies32;
2587         tp->snd_ssthresh = tcp_current_ssthresh(sk);
2588 
2589         icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size;
2590         icsk->icsk_mtup.probe_size = 0;
2591         tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
2592         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMTUPSUCCESS);
2593 }
2594 
2595 /* Do a simple retransmit without using the backoff mechanisms in
2596  * tcp_timer. This is used for path mtu discovery.
2597  * The socket is already locked here.
2598  */
2599 void tcp_simple_retransmit(struct sock *sk)
2600 {
2601         const struct inet_connection_sock *icsk = inet_csk(sk);
2602         struct tcp_sock *tp = tcp_sk(sk);
2603         struct sk_buff *skb;
2604         unsigned int mss = tcp_current_mss(sk);
2605 
2606         skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
2607                 if (tcp_skb_seglen(skb) > mss &&
2608                     !(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) {
2609                         if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
2610                                 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;
2611                                 tp->retrans_out -= tcp_skb_pcount(skb);
2612                         }
2613                         tcp_skb_mark_lost_uncond_verify(tp, skb);
2614                 }
2615         }
2616 
2617         tcp_clear_retrans_hints_partial(tp);
2618 
2619         if (!tp->lost_out)
2620                 return;
2621 
2622         if (tcp_is_reno(tp))
2623                 tcp_limit_reno_sacked(tp);
2624 
2625         tcp_verify_left_out(tp);
2626 
2627         /* Don't muck with the congestion window here.
2628          * Reason is that we do not increase amount of _data_
2629          * in network, but units changed and effective
2630          * cwnd/ssthresh really reduced now.
2631          */
2632         if (icsk->icsk_ca_state != TCP_CA_Loss) {
2633                 tp->high_seq = tp->snd_nxt;
2634                 tp->snd_ssthresh = tcp_current_ssthresh(sk);
2635                 tp->prior_ssthresh = 0;
2636                 tp->undo_marker = 0;
2637                 tcp_set_ca_state(sk, TCP_CA_Loss);
2638         }
2639         tcp_xmit_retransmit_queue(sk);
2640 }
2641 EXPORT_SYMBOL(tcp_simple_retransmit);
2642 
2643 void tcp_enter_recovery(struct sock *sk, bool ece_ack)
2644 {
2645         struct tcp_sock *tp = tcp_sk(sk);
2646         int mib_idx;
2647 
2648         if (tcp_is_reno(tp))
2649                 mib_idx = LINUX_MIB_TCPRENORECOVERY;
2650         else
2651                 mib_idx = LINUX_MIB_TCPSACKRECOVERY;
2652 
2653         NET_INC_STATS(sock_net(sk), mib_idx);
2654 
2655         tp->prior_ssthresh = 0;
2656         tcp_init_undo(tp);
2657 
2658         if (!tcp_in_cwnd_reduction(sk)) {
2659                 if (!ece_ack)
2660                         tp->prior_ssthresh = tcp_current_ssthresh(sk);
2661                 tcp_init_cwnd_reduction(sk);
2662         }
2663         tcp_set_ca_state(sk, TCP_CA_Recovery);
2664 }
2665 
2666 /* Process an ACK in CA_Loss state. Move to CA_Open if lost data are
2667  * recovered or spurious. Otherwise retransmits more on partial ACKs.
2668  */
2669 static void tcp_process_loss(struct sock *sk, int flag, int num_dupack,
2670                              int *rexmit)
2671 {
2672         struct tcp_sock *tp = tcp_sk(sk);
2673         bool recovered = !before(tp->snd_una, tp->high_seq);
2674 
2675         if ((flag & FLAG_SND_UNA_ADVANCED || rcu_access_pointer(tp->fastopen_rsk)) &&
2676             tcp_try_undo_loss(sk, false))
2677                 return;
2678 
2679         if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */
2680                 /* Step 3.b. A timeout is spurious if not all data are
2681                  * lost, i.e., never-retransmitted data are (s)acked.
2682                  */
2683                 if ((flag & FLAG_ORIG_SACK_ACKED) &&
2684                     tcp_try_undo_loss(sk, true))
2685                         return;
2686 
2687                 if (after(tp->snd_nxt, tp->high_seq)) {
2688                         if (flag & FLAG_DATA_SACKED || num_dupack)
2689                                 tp->frto = 0; /* Step 3.a. loss was real */
2690                 } else if (flag & FLAG_SND_UNA_ADVANCED && !recovered) {
2691                         tp->high_seq = tp->snd_nxt;
2692                         /* Step 2.b. Try send new data (but deferred until cwnd
2693                          * is updated in tcp_ack()). Otherwise fall back to
2694                          * the conventional recovery.
2695                          */
2696                         if (!tcp_write_queue_empty(sk) &&
2697                             after(tcp_wnd_end(tp), tp->snd_nxt)) {
2698                                 *rexmit = REXMIT_NEW;
2699                                 return;
2700                         }
2701                         tp->frto = 0;
2702                 }
2703         }
2704 
2705         if (recovered) {
2706                 /* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */
2707                 tcp_try_undo_recovery(sk);
2708                 return;
2709         }
2710         if (tcp_is_reno(tp)) {
2711                 /* A Reno DUPACK means new data in F-RTO step 2.b above are
2712                  * delivered. Lower inflight to clock out (re)tranmissions.
2713                  */
2714                 if (after(tp->snd_nxt, tp->high_seq) && num_dupack)
2715                         tcp_add_reno_sack(sk, num_dupack);
2716                 else if (flag & FLAG_SND_UNA_ADVANCED)
2717                         tcp_reset_reno_sack(tp);
2718         }
2719         *rexmit = REXMIT_LOST;
2720 }
2721 
2722 /* Undo during fast recovery after partial ACK. */
2723 static bool tcp_try_undo_partial(struct sock *sk, u32 prior_snd_una)
2724 {
2725         struct tcp_sock *tp = tcp_sk(sk);
2726 
2727         if (tp->undo_marker && tcp_packet_delayed(tp)) {
2728                 /* Plain luck! Hole if filled with delayed
2729                  * packet, rather than with a retransmit. Check reordering.
2730                  */
2731                 tcp_check_sack_reordering(sk, prior_snd_una, 1);
2732 
2733                 /* We are getting evidence that the reordering degree is higher
2734                  * than we realized. If there are no retransmits out then we
2735                  * can undo. Otherwise we clock out new packets but do not
2736                  * mark more packets lost or retransmit more.
2737                  */
2738                 if (tp->retrans_out)
2739                         return true;
2740 
2741                 if (!tcp_any_retrans_done(sk))
2742                         tp->retrans_stamp = 0;
2743 
2744                 DBGUNDO(sk, "partial recovery");
2745                 tcp_undo_cwnd_reduction(sk, true);
2746                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPPARTIALUNDO);
2747                 tcp_try_keep_open(sk);
2748                 return true;
2749         }
2750         return false;
2751 }
2752 
2753 static void tcp_identify_packet_loss(struct sock *sk, int *ack_flag)
2754 {
2755         struct tcp_sock *tp = tcp_sk(sk);
2756 
2757         if (tcp_rtx_queue_empty(sk))
2758                 return;
2759 
2760         if (unlikely(tcp_is_reno(tp))) {
2761                 tcp_newreno_mark_lost(sk, *ack_flag & FLAG_SND_UNA_ADVANCED);
2762         } else if (tcp_is_rack(sk)) {
2763                 u32 prior_retrans = tp->retrans_out;
2764 
2765                 tcp_rack_mark_lost(sk);
2766                 if (prior_retrans > tp->retrans_out)
2767                         *ack_flag |= FLAG_LOST_RETRANS;
2768         }
2769 }
2770 
2771 static bool tcp_force_fast_retransmit(struct sock *sk)
2772 {
2773         struct tcp_sock *tp = tcp_sk(sk);
2774 
2775         return after(tcp_highest_sack_seq(tp),
2776                      tp->snd_una + tp->reordering * tp->mss_cache);
2777 }
2778 
2779 /* Process an event, which can update packets-in-flight not trivially.
2780  * Main goal of this function is to calculate new estimate for left_out,
2781  * taking into account both packets sitting in receiver's buffer and
2782  * packets lost by network.
2783  *
2784  * Besides that it updates the congestion state when packet loss or ECN
2785  * is detected. But it does not reduce the cwnd, it is done by the
2786  * congestion control later.
2787  *
2788  * It does _not_ decide what to send, it is made in function
2789  * tcp_xmit_retransmit_queue().
2790  */
2791 static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
2792                                   int num_dupack, int *ack_flag, int *rexmit)
2793 {
2794         struct inet_connection_sock *icsk = inet_csk(sk);
2795         struct tcp_sock *tp = tcp_sk(sk);
2796         int fast_rexmit = 0, flag = *ack_flag;
2797         bool do_lost = num_dupack || ((flag & FLAG_DATA_SACKED) &&
2798                                       tcp_force_fast_retransmit(sk));
2799 
2800         if (!tp->packets_out && tp->sacked_out)
2801                 tp->sacked_out = 0;
2802 
2803         /* Now state machine starts.
2804          * A. ECE, hence prohibit cwnd undoing, the reduction is required. */
2805         if (flag & FLAG_ECE)
2806                 tp->prior_ssthresh = 0;
2807 
2808         /* B. In all the states check for reneging SACKs. */
2809         if (tcp_check_sack_reneging(sk, flag))
2810                 return;
2811 
2812         /* C. Check consistency of the current state. */
2813         tcp_verify_left_out(tp);
2814 
2815         /* D. Check state exit conditions. State can be terminated
2816          *    when high_seq is ACKed. */
2817         if (icsk->icsk_ca_state == TCP_CA_Open) {
2818                 WARN_ON(tp->retrans_out != 0);
2819                 tp->retrans_stamp = 0;
2820         } else if (!before(tp->snd_una, tp->high_seq)) {
2821                 switch (icsk->icsk_ca_state) {
2822                 case TCP_CA_CWR:
2823                         /* CWR is to be held something *above* high_seq
2824                          * is ACKed for CWR bit to reach receiver. */
2825                         if (tp->snd_una != tp->high_seq) {
2826                                 tcp_end_cwnd_reduction(sk);
2827                                 tcp_set_ca_state(sk, TCP_CA_Open);
2828                         }
2829                         break;
2830 
2831                 case TCP_CA_Recovery:
2832                         if (tcp_is_reno(tp))
2833                                 tcp_reset_reno_sack(tp);
2834                         if (tcp_try_undo_recovery(sk))
2835                                 return;
2836                         tcp_end_cwnd_reduction(sk);
2837                         break;
2838                 }
2839         }
2840 
2841         /* E. Process state. */
2842         switch (icsk->icsk_ca_state) {
2843         case TCP_CA_Recovery:
2844                 if (!(flag & FLAG_SND_UNA_ADVANCED)) {
2845                         if (tcp_is_reno(tp))
2846                                 tcp_add_reno_sack(sk, num_dupack);
2847                 } else {
2848                         if (tcp_try_undo_partial(sk, prior_snd_una))
2849                                 return;
2850                         /* Partial ACK arrived. Force fast retransmit. */
2851                         do_lost = tcp_is_reno(tp) ||
2852                                   tcp_force_fast_retransmit(sk);
2853                 }
2854                 if (tcp_try_undo_dsack(sk)) {
2855                         tcp_try_keep_open(sk);
2856                         return;
2857                 }
2858                 tcp_identify_packet_loss(sk, ack_flag);
2859                 break;
2860         case TCP_CA_Loss:
2861                 tcp_process_loss(sk, flag, num_dupack, rexmit);
2862                 tcp_identify_packet_loss(sk, ack_flag);
2863                 if (!(icsk->icsk_ca_state == TCP_CA_Open ||
2864                       (*ack_flag & FLAG_LOST_RETRANS)))
2865                         return;
2866                 /* Change state if cwnd is undone or retransmits are lost */
2867                 /* fall through */
2868         default:
2869                 if (tcp_is_reno(tp)) {
2870                         if (flag & FLAG_SND_UNA_ADVANCED)
2871                                 tcp_reset_reno_sack(tp);
2872                         tcp_add_reno_sack(sk, num_dupack);
2873                 }
2874 
2875                 if (icsk->icsk_ca_state <= TCP_CA_Disorder)
2876                         tcp_try_undo_dsack(sk);
2877 
2878                 tcp_identify_packet_loss(sk, ack_flag);
2879                 if (!tcp_time_to_recover(sk, flag)) {
2880                         tcp_try_to_open(sk, flag);
2881                         return;
2882                 }
2883 
2884                 /* MTU probe failure: don't reduce cwnd */
2885                 if (icsk->icsk_ca_state < TCP_CA_CWR &&
2886                     icsk->icsk_mtup.probe_size &&
2887                     tp->snd_una == tp->mtu_probe.probe_seq_start) {
2888                         tcp_mtup_probe_failed(sk);
2889                         /* Restores the reduction we did in tcp_mtup_probe() */
2890                         tp->snd_cwnd++;
2891                         tcp_simple_retransmit(sk);
2892                         return;
2893                 }
2894 
2895                 /* Otherwise enter Recovery state */
2896                 tcp_enter_recovery(sk, (flag & FLAG_ECE));
2897                 fast_rexmit = 1;
2898         }
2899 
2900         if (!tcp_is_rack(sk) && do_lost)
2901                 tcp_update_scoreboard(sk, fast_rexmit);
2902         *rexmit = REXMIT_LOST;
2903 }
2904 
2905 static void tcp_update_rtt_min(struct sock *sk, u32 rtt_us, const int flag)
2906 {
2907         u32 wlen = sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen * HZ;
2908         struct tcp_sock *tp = tcp_sk(sk);
2909 
2910         if ((flag & FLAG_ACK_MAYBE_DELAYED) && rtt_us > tcp_min_rtt(tp)) {
2911                 /* If the remote keeps returning delayed ACKs, eventually
2912                  * the min filter would pick it up and overestimate the
2913                  * prop. delay when it expires. Skip suspected delayed ACKs.
2914                  */
2915                 return;
2916         }
2917         minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32,
2918                            rtt_us ? : jiffies_to_usecs(1));
2919 }
2920 
2921 static bool tcp_ack_update_rtt(struct sock *sk, const int flag,
2922                                long seq_rtt_us, long sack_rtt_us,
2923                                long ca_rtt_us, struct rate_sample *rs)
2924 {
2925         const struct tcp_sock *tp = tcp_sk(sk);
2926 
2927         /* Prefer RTT measured from ACK's timing to TS-ECR. This is because
2928          * broken middle-boxes or peers may corrupt TS-ECR fields. But
2929          * Karn's algorithm forbids taking RTT if some retransmitted data
2930          * is acked (RFC6298).
2931          */
2932         if (seq_rtt_us < 0)
2933                 seq_rtt_us = sack_rtt_us;
2934 
2935         /* RTTM Rule: A TSecr value received in a segment is used to
2936          * update the averaged RTT measurement only if the segment
2937          * acknowledges some new data, i.e., only if it advances the
2938          * left edge of the send window.
2939          * See draft-ietf-tcplw-high-performance-00, section 3.3.
2940          */
2941         if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
2942             flag & FLAG_ACKED) {
2943                 u32 delta = tcp_time_stamp(tp) - tp->rx_opt.rcv_tsecr;
2944 
2945                 if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) {
2946                         seq_rtt_us = delta * (USEC_PER_SEC / TCP_TS_HZ);
2947                         ca_rtt_us = seq_rtt_us;
2948                 }
2949         }
2950         rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */
2951         if (seq_rtt_us < 0)
2952                 return false;
2953 
2954         /* ca_rtt_us >= 0 is counting on the invariant that ca_rtt_us is
2955          * always taken together with ACK, SACK, or TS-opts. Any negative
2956          * values will be skipped with the seq_rtt_us < 0 check above.
2957          */
2958         tcp_update_rtt_min(sk, ca_rtt_us, flag);
2959         tcp_rtt_estimator(sk, seq_rtt_us);
2960         tcp_set_rto(sk);
2961 
2962         /* RFC6298: only reset backoff on valid RTT measurement. */
2963         inet_csk(sk)->icsk_backoff = 0;
2964         return true;
2965 }
2966 
2967 /* Compute time elapsed between (last) SYNACK and the ACK completing 3WHS. */
2968 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req)
2969 {
2970         struct rate_sample rs;
2971         long rtt_us = -1L;
2972 
2973         if (req && !req->num_retrans && tcp_rsk(req)->snt_synack)
2974                 rtt_us = tcp_stamp_us_delta(tcp_clock_us(), tcp_rsk(req)->snt_synack);
2975 
2976         tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, rtt_us, -1L, rtt_us, &rs);
2977 }
2978 
2979 
2980 static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
2981 {
2982         const struct inet_connection_sock *icsk = inet_csk(sk);
2983 
2984         icsk->icsk_ca_ops->cong_avoid(sk, ack, acked);
2985         tcp_sk(sk)->snd_cwnd_stamp = tcp_jiffies32;
2986 }
2987 
2988 /* Restart timer after forward progress on connection.
2989  * RFC2988 recommends to restart timer to now+rto.
2990  */
2991 void tcp_rearm_rto(struct sock *sk)
2992 {
2993         const struct inet_connection_sock *icsk = inet_csk(sk);
2994         struct tcp_sock *tp = tcp_sk(sk);
2995 
2996         /* If the retrans timer is currently being used by Fast Open
2997          * for SYN-ACK retrans purpose, stay put.
2998          */
2999         if (rcu_access_pointer(tp->fastopen_rsk))
3000                 return;
3001 
3002         if (!tp->packets_out) {
3003                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
3004         } else {
3005                 u32 rto = inet_csk(sk)->icsk_rto;
3006                 /* Offset the time elapsed after installing regular RTO */
3007                 if (icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
3008                     icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
3009                         s64 delta_us = tcp_rto_delta_us(sk);
3010                         /* delta_us may not be positive if the socket is locked
3011                          * when the retrans timer fires and is rescheduled.
3012                          */
3013                         rto = usecs_to_jiffies(max_t(int, delta_us, 1));
3014                 }
3015                 tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto,
3016                                      TCP_RTO_MAX, tcp_rtx_queue_head(sk));
3017         }
3018 }
3019 
3020 /* Try to schedule a loss probe; if that doesn't work, then schedule an RTO. */
3021 static void tcp_set_xmit_timer(struct sock *sk)
3022 {
3023         if (!tcp_schedule_loss_probe(sk, true))
3024                 tcp_rearm_rto(sk);
3025 }
3026 
3027 /* If we get here, the whole TSO packet has not been acked. */
3028 static u32 tcp_tso_acked(struct sock *sk, struct sk_buff *skb)
3029 {
3030         struct tcp_sock *tp = tcp_sk(sk);
3031         u32 packets_acked;
3032 
3033         BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una));
3034 
3035         packets_acked = tcp_skb_pcount(skb);
3036         if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
3037                 return 0;
3038         packets_acked -= tcp_skb_pcount(skb);
3039 
3040         if (packets_acked) {
3041                 BUG_ON(tcp_skb_pcount(skb) == 0);
3042                 BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq));
3043         }
3044 
3045         return packets_acked;
3046 }
3047 
3048 static void tcp_ack_tstamp(struct sock *sk, struct sk_buff *skb,
3049                            u32 prior_snd_una)
3050 {
3051         const struct skb_shared_info *shinfo;
3052 
3053         /* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */
3054         if (likely(!TCP_SKB_CB(skb)->txstamp_ack))
3055                 return;
3056 
3057         shinfo = skb_shinfo(skb);
3058         if (!before(shinfo->tskey, prior_snd_una) &&
3059             before(shinfo->tskey, tcp_sk(sk)->snd_una)) {
3060                 tcp_skb_tsorted_save(skb) {
3061                         __skb_tstamp_tx(skb, NULL, sk, SCM_TSTAMP_ACK);
3062                 } tcp_skb_tsorted_restore(skb);
3063         }
3064 }
3065 
3066 /* Remove acknowledged frames from the retransmission queue. If our packet
3067  * is before the ack sequence we can discard it as it's confirmed to have
3068  * arrived at the other end.
3069  */
3070 static int tcp_clean_rtx_queue(struct sock *sk, u32 prior_fack,
3071                                u32 prior_snd_una,
3072                                struct tcp_sacktag_state *sack)
3073 {
3074         const struct inet_connection_sock *icsk = inet_csk(sk);
3075         u64 first_ackt, last_ackt;
3076         struct tcp_sock *tp = tcp_sk(sk);
3077         u32 prior_sacked = tp->sacked_out;
3078         u32 reord = tp->snd_nxt; /* lowest acked un-retx un-sacked seq */
3079         struct sk_buff *skb, *next;
3080         bool fully_acked = true;
3081         long sack_rtt_us = -1L;
3082         long seq_rtt_us = -1L;
3083         long ca_rtt_us = -1L;
3084         u32 pkts_acked = 0;
3085         u32 last_in_flight = 0;
3086         bool rtt_update;
3087         int flag = 0;
3088 
3089         first_ackt = 0;
3090 
3091         for (skb = skb_rb_first(&sk->tcp_rtx_queue); skb; skb = next) {
3092                 struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
3093                 const u32 start_seq = scb->seq;
3094                 u8 sacked = scb->sacked;
3095                 u32 acked_pcount;
3096 
3097                 tcp_ack_tstamp(sk, skb, prior_snd_una);
3098 
3099                 /* Determine how many packets and what bytes were acked, tso and else */
3100                 if (after(scb->end_seq, tp->snd_una)) {
3101                         if (tcp_skb_pcount(skb) == 1 ||
3102                             !after(tp->snd_una, scb->seq))
3103                                 break;
3104 
3105                         acked_pcount = tcp_tso_acked(sk, skb);
3106                         if (!acked_pcount)
3107                                 break;
3108                         fully_acked = false;
3109                 } else {
3110                         acked_pcount = tcp_skb_pcount(skb);
3111                 }
3112 
3113                 if (unlikely(sacked & TCPCB_RETRANS)) {
3114                         if (sacked & TCPCB_SACKED_RETRANS)
3115                                 tp->retrans_out -= acked_pcount;
3116                         flag |= FLAG_RETRANS_DATA_ACKED;
3117                 } else if (!(sacked & TCPCB_SACKED_ACKED)) {
3118                         last_ackt = tcp_skb_timestamp_us(skb);
3119                         WARN_ON_ONCE(last_ackt == 0);
3120                         if (!first_ackt)
3121                                 first_ackt = last_ackt;
3122 
3123                         last_in_flight = TCP_SKB_CB(skb)->tx.in_flight;
3124                         if (before(start_seq, reord))
3125                                 reord = start_seq;
3126                         if (!after(scb->end_seq, tp->high_seq))
3127                                 flag |= FLAG_ORIG_SACK_ACKED;
3128                 }
3129 
3130                 if (sacked & TCPCB_SACKED_ACKED) {
3131                         tp->sacked_out -= acked_pcount;
3132                 } else if (tcp_is_sack(tp)) {
3133                         tp->delivered += acked_pcount;
3134                         if (!tcp_skb_spurious_retrans(tp, skb))
3135                                 tcp_rack_advance(tp, sacked, scb->end_seq,
3136                                                  tcp_skb_timestamp_us(skb));
3137                 }
3138                 if (sacked & TCPCB_LOST)
3139                         tp->lost_out -= acked_pcount;
3140 
3141                 tp->packets_out -= acked_pcount;
3142                 pkts_acked += acked_pcount;
3143                 tcp_rate_skb_delivered(sk, skb, sack->rate);
3144 
3145                 /* Initial outgoing SYN's get put onto the write_queue
3146                  * just like anything else we transmit.  It is not
3147                  * true data, and if we misinform our callers that
3148                  * this ACK acks real data, we will erroneously exit
3149                  * connection startup slow start one packet too
3150                  * quickly.  This is severely frowned upon behavior.
3151                  */
3152                 if (likely(!(scb->tcp_flags & TCPHDR_SYN))) {
3153                         flag |= FLAG_DATA_ACKED;
3154                 } else {
3155                         flag |= FLAG_SYN_ACKED;
3156                         tp->retrans_stamp = 0;
3157                 }
3158 
3159                 if (!fully_acked)
3160                         break;
3161 
3162                 next = skb_rb_next(skb);
3163                 if (unlikely(skb == tp->retransmit_skb_hint))
3164                         tp->retransmit_skb_hint = NULL;
3165                 if (unlikely(skb == tp->lost_skb_hint))
3166                         tp->lost_skb_hint = NULL;
3167                 tcp_highest_sack_replace(sk, skb, next);
3168                 tcp_rtx_queue_unlink_and_free(skb, sk);
3169         }
3170 
3171         if (!skb)
3172                 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
3173 
3174         if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una)))
3175                 tp->snd_up = tp->snd_una;
3176 
3177         if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
3178                 flag |= FLAG_SACK_RENEGING;
3179 
3180         if (likely(first_ackt) && !(flag & FLAG_RETRANS_DATA_ACKED)) {
3181                 seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt);
3182                 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt);
3183 
3184                 if (pkts_acked == 1 && last_in_flight < tp->mss_cache &&
3185                     last_in_flight && !prior_sacked && fully_acked &&
3186                     sack->rate->prior_delivered + 1 == tp->delivered &&
3187                     !(flag & (FLAG_CA_ALERT | FLAG_SYN_ACKED))) {
3188                         /* Conservatively mark a delayed ACK. It's typically
3189                          * from a lone runt packet over the round trip to
3190                          * a receiver w/o out-of-order or CE events.
3191                          */
3192                         flag |= FLAG_ACK_MAYBE_DELAYED;
3193                 }
3194         }
3195         if (sack->first_sackt) {
3196                 sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt);
3197                 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->last_sackt);
3198         }
3199         rtt_update = tcp_ack_update_rtt(sk, flag, seq_rtt_us, sack_rtt_us,
3200                                         ca_rtt_us, sack->rate);
3201 
3202         if (flag & FLAG_ACKED) {
3203                 flag |= FLAG_SET_XMIT_TIMER;  /* set TLP or RTO timer */
3204                 if (unlikely(icsk->icsk_mtup.probe_size &&
3205                              !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) {
3206                         tcp_mtup_probe_success(sk);
3207                 }
3208 
3209                 if (tcp_is_reno(tp)) {
3210                         tcp_remove_reno_sacks(sk, pkts_acked);
3211 
3212                         /* If any of the cumulatively ACKed segments was
3213                          * retransmitted, non-SACK case cannot confirm that
3214                          * progress was due to original transmission due to
3215                          * lack of TCPCB_SACKED_ACKED bits even if some of
3216                          * the packets may have been never retransmitted.
3217                          */
3218                         if (flag & FLAG_RETRANS_DATA_ACKED)
3219                                 flag &= ~FLAG_ORIG_SACK_ACKED;
3220                 } else {
3221                         int delta;
3222 
3223                         /* Non-retransmitted hole got filled? That's reordering */
3224                         if (before(reord, prior_fack))
3225                                 tcp_check_sack_reordering(sk, reord, 0);
3226 
3227                         delta = prior_sacked - tp->sacked_out;
3228                         tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta);
3229                 }
3230         } else if (skb && rtt_update && sack_rtt_us >= 0 &&
3231                    sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp,
3232                                                     tcp_skb_timestamp_us(skb))) {
3233                 /* Do not re-arm RTO if the sack RTT is measured from data sent
3234                  * after when the head was last (re)transmitted. Otherwise the
3235                  * timeout may continue to extend in loss recovery.
3236                  */
3237                 flag |= FLAG_SET_XMIT_TIMER;  /* set TLP or RTO timer */
3238         }
3239 
3240         if (icsk->icsk_ca_ops->pkts_acked) {
3241                 struct ack_sample sample = { .pkts_acked = pkts_acked,
3242                                              .rtt_us = sack->rate->rtt_us,
3243                                              .in_flight = last_in_flight };
3244 
3245                 icsk->icsk_ca_ops->pkts_acked(sk, &sample);
3246         }
3247 
3248 #if FASTRETRANS_DEBUG > 0
3249         WARN_ON((int)tp->sacked_out < 0);
3250         WARN_ON((int)tp->lost_out < 0);
3251         WARN_ON((int)tp->retrans_out < 0);
3252         if (!tp->packets_out && tcp_is_sack(tp)) {
3253                 icsk = inet_csk(sk);
3254                 if (tp->lost_out) {
3255                         pr_debug("Leak l=%u %d\n",
3256                                  tp->lost_out, icsk->icsk_ca_state);
3257                         tp->lost_out = 0;
3258                 }
3259                 if (tp->sacked_out) {
3260                         pr_debug("Leak s=%u %d\n",
3261                                  tp->sacked_out, icsk->icsk_ca_state);
3262                         tp->sacked_out = 0;
3263                 }
3264                 if (tp->retrans_out) {
3265                         pr_debug("Leak r=%u %d\n",
3266                                  tp->retrans_out, icsk->icsk_ca_state);
3267                         tp->retrans_out = 0;
3268                 }
3269         }
3270 #endif
3271         return flag;
3272 }
3273 
3274 static void tcp_ack_probe(struct sock *sk)
3275 {
3276         struct inet_connection_sock *icsk = inet_csk(sk);
3277         struct sk_buff *head = tcp_send_head(sk);
3278         const struct tcp_sock *tp = tcp_sk(sk);
3279 
3280         /* Was it a usable window open? */
3281         if (!head)
3282                 return;
3283         if (!after(TCP_SKB_CB(head)->end_seq, tcp_wnd_end(tp))) {
3284                 icsk->icsk_backoff = 0;
3285                 inet_csk_clear_xmit_timer(sk, ICSK_TIME_PROBE0);
3286                 /* Socket must be waked up by subsequent tcp_data_snd_check().
3287                  * This function is not for random using!
3288                  */
3289         } else {
3290                 unsigned long when = tcp_probe0_when(sk, TCP_RTO_MAX);
3291 
3292                 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
3293                                      when, TCP_RTO_MAX, NULL);
3294         }
3295 }
3296 
3297 static inline bool tcp_ack_is_dubious(const struct sock *sk, const int flag)
3298 {
3299         return !(flag & FLAG_NOT_DUP) || (flag & FLAG_CA_ALERT) ||
3300                 inet_csk(sk)->icsk_ca_state != TCP_CA_Open;
3301 }
3302 
3303 /* Decide wheather to run the increase function of congestion control. */
3304 static inline bool tcp_may_raise_cwnd(const struct sock *sk, const int flag)
3305 {
3306         /* If reordering is high then always grow cwnd whenever data is
3307          * delivered regardless of its ordering. Otherwise stay conservative
3308          * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/
3309          * new SACK or ECE mark may first advance cwnd here and later reduce
3310          * cwnd in tcp_fastretrans_alert() based on more states.
3311          */
3312         if (tcp_sk(sk)->reordering > sock_net(sk)->ipv4.sysctl_tcp_reordering)
3313                 return flag & FLAG_FORWARD_PROGRESS;
3314 
3315         return flag & FLAG_DATA_ACKED;
3316 }
3317 
3318 /* The "ultimate" congestion control function that aims to replace the rigid
3319  * cwnd increase and decrease control (tcp_cong_avoid,tcp_*cwnd_reduction).
3320  * It's called toward the end of processing an ACK with precise rate
3321  * information. All transmission or retransmission are delayed afterwards.
3322  */
3323 static void tcp_cong_control(struct sock *sk, u32 ack, u32 acked_sacked,
3324                              int flag, const struct rate_sample *rs)
3325 {
3326         const struct inet_connection_sock *icsk = inet_csk(sk);
3327 
3328         if (icsk->icsk_ca_ops->cong_control) {
3329                 icsk->icsk_ca_ops->cong_control(sk, rs);
3330                 return;
3331         }
3332 
3333         if (tcp_in_cwnd_reduction(sk)) {
3334                 /* Reduce cwnd if state mandates */
3335                 tcp_cwnd_reduction(sk, acked_sacked, flag);
3336         } else if (tcp_may_raise_cwnd(sk, flag)) {
3337                 /* Advance cwnd if state allows */
3338                 tcp_cong_avoid(sk, ack, acked_sacked);
3339         }
3340         tcp_update_pacing_rate(sk);
3341 }
3342 
3343 /* Check that window update is acceptable.
3344  * The function assumes that snd_una<=ack<=snd_next.
3345  */
3346 static inline bool tcp_may_update_window(const struct tcp_sock *tp,
3347                                         const u32 ack, const u32 ack_seq,
3348                                         const u32 nwin)
3349 {
3350         return  after(ack, tp->snd_una) ||
3351                 after(ack_seq, tp->snd_wl1) ||
3352                 (ack_seq == tp->snd_wl1 && nwin > tp->snd_wnd);
3353 }
3354 
3355 /* If we update tp->snd_una, also update tp->bytes_acked */
3356 static void tcp_snd_una_update(struct tcp_sock *tp, u32 ack)
3357 {
3358         u32 delta = ack - tp->snd_una;
3359 
3360         sock_owned_by_me((struct sock *)tp);
3361         tp->bytes_acked += delta;
3362         tp->snd_una = ack;
3363 }
3364 
3365 /* If we update tp->rcv_nxt, also update tp->bytes_received */
3366 static void tcp_rcv_nxt_update(struct tcp_sock *tp, u32 seq)
3367 {
3368         u32 delta = seq - tp->rcv_nxt;
3369 
3370         sock_owned_by_me((struct sock *)tp);
3371         tp->bytes_received += delta;
3372         WRITE_ONCE(tp->rcv_nxt, seq);
3373 }
3374 
3375 /* Update our send window.
3376  *
3377  * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
3378  * and in FreeBSD. NetBSD's one is even worse.) is wrong.
3379  */
3380 static int tcp_ack_update_window(struct sock *sk, const struct sk_buff *skb, u32 ack,
3381                                  u32 ack_seq)
3382 {
3383         struct tcp_sock *tp = tcp_sk(sk);
3384         int flag = 0;
3385         u32 nwin = ntohs(tcp_hdr(skb)->window);
3386 
3387         if (likely(!tcp_hdr(skb)->syn))
3388                 nwin <<= tp->rx_opt.snd_wscale;
3389 
3390         if (tcp_may_update_window(tp, ack, ack_seq, nwin)) {
3391                 flag |= FLAG_WIN_UPDATE;
3392                 tcp_update_wl(tp, ack_seq);
3393 
3394                 if (tp->snd_wnd != nwin) {
3395                         tp->snd_wnd = nwin;
3396 
3397                         /* Note, it is the only place, where
3398                          * fast path is recovered for sending TCP.
3399                          */
3400                         tp->pred_flags = 0;
3401                         tcp_fast_path_check(sk);
3402 
3403                         if (!tcp_write_queue_empty(sk))
3404                                 tcp_slow_start_after_idle_check(sk);
3405 
3406                         if (nwin > tp->max_window) {
3407                                 tp->max_window = nwin;
3408                                 tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie);
3409                         }
3410                 }
3411         }
3412 
3413         tcp_snd_una_update(tp, ack);
3414 
3415         return flag;
3416 }
3417 
3418 static bool __tcp_oow_rate_limited(struct net *net, int mib_idx,
3419                                    u32 *last_oow_ack_time)
3420 {
3421         if (*last_oow_ack_time) {
3422                 s32 elapsed = (s32)(tcp_jiffies32 - *last_oow_ack_time);
3423 
3424                 if (0 <= elapsed && elapsed < net->ipv4.sysctl_tcp_invalid_ratelimit) {
3425                         NET_INC_STATS(net, mib_idx);
3426                         return true;    /* rate-limited: don't send yet! */
3427                 }
3428         }
3429 
3430         *last_oow_ack_time = tcp_jiffies32;
3431 
3432         return false;   /* not rate-limited: go ahead, send dupack now! */
3433 }
3434 
3435 /* Return true if we're currently rate-limiting out-of-window ACKs and
3436  * thus shouldn't send a dupack right now. We rate-limit dupacks in
3437  * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
3438  * attacks that send repeated SYNs or ACKs for the same connection. To
3439  * do this, we do not send a duplicate SYNACK or ACK if the remote
3440  * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
3441  */
3442 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
3443                           int mib_idx, u32 *last_oow_ack_time)
3444 {
3445         /* Data packets without SYNs are not likely part of an ACK loop. */
3446         if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&
3447             !tcp_hdr(skb)->syn)
3448                 return false;
3449 
3450         return __tcp_oow_rate_limited(net, mib_idx, last_oow_ack_time);
3451 }
3452 
3453 /* RFC 5961 7 [ACK Throttling] */
3454 static void tcp_send_challenge_ack(struct sock *sk, const struct sk_buff *skb)
3455 {
3456         /* unprotected vars, we dont care of overwrites */
3457         static u32 challenge_timestamp;
3458         static unsigned int challenge_count;
3459         struct tcp_sock *tp = tcp_sk(sk);
3460         struct net *net = sock_net(sk);
3461         u32 count, now;
3462 
3463         /* First check our per-socket dupack rate limit. */
3464         if (__tcp_oow_rate_limited(net,
3465                                    LINUX_MIB_TCPACKSKIPPEDCHALLENGE,
3466                                    &tp->last_oow_ack_time))
3467                 return;
3468 
3469         /* Then check host-wide RFC 5961 rate limit. */
3470         now = jiffies / HZ;
3471         if (now != challenge_timestamp) {
3472                 u32 ack_limit = net->ipv4.sysctl_tcp_challenge_ack_limit;
3473                 u32 half = (ack_limit + 1) >> 1;
3474 
3475                 challenge_timestamp = now;
3476                 WRITE_ONCE(challenge_count, half + prandom_u32_max(ack_limit));
3477         }
3478         count = READ_ONCE(challenge_count);
3479         if (count > 0) {
3480                 WRITE_ONCE(challenge_count, count - 1);
3481                 NET_INC_STATS(net, LINUX_MIB_TCPCHALLENGEACK);
3482                 tcp_send_ack(sk);
3483         }
3484 }
3485 
3486 static void tcp_store_ts_recent(struct tcp_sock *tp)
3487 {
3488         tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval;
3489         tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
3490 }
3491 
3492 static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq)
3493 {
3494         if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) {
3495                 /* PAWS bug workaround wrt. ACK frames, the PAWS discard
3496                  * extra check below makes sure this can only happen
3497                  * for pure ACK frames.  -DaveM
3498                  *
3499                  * Not only, also it occurs for expired timestamps.
3500                  */
3501 
3502                 if (tcp_paws_check(&tp->rx_opt, 0))
3503                         tcp_store_ts_recent(tp);
3504         }
3505 }
3506 
3507 /* This routine deals with acks during a TLP episode.
3508  * We mark the end of a TLP episode on receiving TLP dupack or when
3509  * ack is after tlp_high_seq.
3510  * Ref: loss detection algorithm in draft-dukkipati-tcpm-tcp-loss-probe.
3511  */
3512 static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
3513 {
3514         struct tcp_sock *tp = tcp_sk(sk);
3515 
3516         if (before(ack, tp->tlp_high_seq))
3517                 return;
3518 
3519         if (flag & FLAG_DSACKING_ACK) {
3520                 /* This DSACK means original and TLP probe arrived; no loss */
3521                 tp->tlp_high_seq = 0;
3522         } else if (after(ack, tp->tlp_high_seq)) {
3523                 /* ACK advances: there was a loss, so reduce cwnd. Reset
3524                  * tlp_high_seq in tcp_init_cwnd_reduction()
3525                  */
3526                 tcp_init_cwnd_reduction(sk);
3527                 tcp_set_ca_state(sk, TCP_CA_CWR);
3528                 tcp_end_cwnd_reduction(sk);
3529                 tcp_try_keep_open(sk);
3530                 NET_INC_STATS(sock_net(sk),
3531                                 LINUX_MIB_TCPLOSSPROBERECOVERY);
3532         } else if (!(flag & (FLAG_SND_UNA_ADVANCED |
3533                              FLAG_NOT_DUP | FLAG_DATA_SACKED))) {
3534                 /* Pure dupack: original and TLP probe arrived; no loss */
3535                 tp->tlp_high_seq = 0;
3536         }
3537 }
3538 
3539 static inline void tcp_in_ack_event(struct sock *sk, u32 flags)
3540 {
3541         const struct inet_connection_sock *icsk = inet_csk(sk);
3542 
3543         if (icsk->icsk_ca_ops->in_ack_event)
3544                 icsk->icsk_ca_ops->in_ack_event(sk, flags);
3545 }
3546 
3547 /* Congestion control has updated the cwnd already. So if we're in
3548  * loss recovery then now we do any new sends (for FRTO) or
3549  * retransmits (for CA_Loss or CA_recovery) that make sense.
3550  */
3551 static void tcp_xmit_recovery(struct sock *sk, int rexmit)
3552 {
3553         struct tcp_sock *tp = tcp_sk(sk);
3554 
3555         if (rexmit == REXMIT_NONE || sk->sk_state == TCP_SYN_SENT)
3556                 return;
3557 
3558         if (unlikely(rexmit == 2)) {
3559                 __tcp_push_pending_frames(sk, tcp_current_mss(sk),
3560                                           TCP_NAGLE_OFF);
3561                 if (after(tp->snd_nxt, tp->high_seq))
3562                         return;
3563                 tp->frto = 0;
3564         }
3565         tcp_xmit_retransmit_queue(sk);
3566 }
3567 
3568 /* Returns the number of packets newly acked or sacked by the current ACK */
3569 static u32 tcp_newly_delivered(struct sock *sk, u32 prior_delivered, int flag)
3570 {
3571         const struct net *net = sock_net(sk);
3572         struct tcp_sock *tp = tcp_sk(sk);
3573         u32 delivered;
3574 
3575         delivered = tp->delivered - prior_delivered;
3576         NET_ADD_STATS(net, LINUX_MIB_TCPDELIVERED, delivered);
3577         if (flag & FLAG_ECE) {
3578                 tp->delivered_ce += delivered;
3579                 NET_ADD_STATS(net, LINUX_MIB_TCPDELIVEREDCE, delivered);
3580         }
3581         return delivered;
3582 }
3583 
3584 /* This routine deals with incoming acks, but not outgoing ones. */
3585 static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
3586 {
3587         struct inet_connection_sock *icsk = inet_csk(sk);
3588         struct tcp_sock *tp = tcp_sk(sk);
3589         struct tcp_sacktag_state sack_state;
3590         struct rate_sample rs = { .prior_delivered = 0 };
3591         u32 prior_snd_una = tp->snd_una;
3592         bool is_sack_reneg = tp->is_sack_reneg;
3593         u32 ack_seq = TCP_SKB_CB(skb)->seq;
3594         u32 ack = TCP_SKB_CB(skb)->ack_seq;
3595         int num_dupack = 0;
3596         int prior_packets = tp->packets_out;
3597         u32 delivered = tp->delivered;
3598         u32 lost = tp->lost;
3599         int rexmit = REXMIT_NONE; /* Flag to (re)transmit to recover losses */
3600         u32 prior_fack;
3601 
3602         sack_state.first_sackt = 0;
3603         sack_state.rate = &rs;
3604 
3605         /* We very likely will need to access rtx queue. */
3606         prefetch(sk->tcp_rtx_queue.rb_node);
3607 
3608         /* If the ack is older than previous acks
3609          * then we can probably ignore it.
3610          */
3611         if (before(ack, prior_snd_una)) {
3612                 /* RFC 5961 5.2 [Blind Data Injection Attack].[Mitigation] */
3613                 if (before(ack, prior_snd_una - tp->max_window)) {
3614                         if (!(flag & FLAG_NO_CHALLENGE_ACK))
3615                                 tcp_send_challenge_ack(sk, skb);
3616                         return -1;
3617                 }
3618                 goto old_ack;
3619         }
3620 
3621         /* If the ack includes data we haven't sent yet, discard
3622          * this segment (RFC793 Section 3.9).
3623          */
3624         if (after(ack, tp->snd_nxt))
3625                 return -1;
3626 
3627         if (after(ack, prior_snd_una)) {
3628                 flag |= FLAG_SND_UNA_ADVANCED;
3629                 icsk->icsk_retransmits = 0;
3630 
3631 #if IS_ENABLED(CONFIG_TLS_DEVICE)
3632                 if (static_branch_unlikely(&clean_acked_data_enabled.key))
3633                         if (icsk->icsk_clean_acked)
3634                                 icsk->icsk_clean_acked(sk, ack);
3635 #endif
3636         }
3637 
3638         prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una;
3639         rs.prior_in_flight = tcp_packets_in_flight(tp);
3640 
3641         /* ts_recent update must be made after we are sure that the packet
3642          * is in window.
3643          */
3644         if (flag & FLAG_UPDATE_TS_RECENT)
3645                 tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq);
3646 
3647         if ((flag & (FLAG_SLOWPATH | FLAG_SND_UNA_ADVANCED)) ==
3648             FLAG_SND_UNA_ADVANCED) {
3649                 /* Window is constant, pure forward advance.
3650                  * No more checks are required.
3651                  * Note, we use the fact that SND.UNA>=SND.WL2.
3652                  */
3653                 tcp_update_wl(tp, ack_seq);
3654                 tcp_snd_una_update(tp, ack);
3655                 flag |= FLAG_WIN_UPDATE;
3656 
3657                 tcp_in_ack_event(sk, CA_ACK_WIN_UPDATE);
3658 
3659                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPHPACKS);
3660         } else {
3661                 u32 ack_ev_flags = CA_ACK_SLOWPATH;
3662 
3663                 if (ack_seq != TCP_SKB_CB(skb)->end_seq)
3664                         flag |= FLAG_DATA;
3665                 else
3666                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPPUREACKS);
3667 
3668                 flag |= tcp_ack_update_window(sk, skb, ack, ack_seq);
3669 
3670                 if (TCP_SKB_CB(skb)->sacked)
3671                         flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
3672                                                         &sack_state);
3673 
3674                 if (tcp_ecn_rcv_ecn_echo(tp, tcp_hdr(skb))) {
3675                         flag |= FLAG_ECE;
3676                         ack_ev_flags |= CA_ACK_ECE;
3677                 }
3678 
3679                 if (flag & FLAG_WIN_UPDATE)
3680                         ack_ev_flags |= CA_ACK_WIN_UPDATE;
3681 
3682                 tcp_in_ack_event(sk, ack_ev_flags);
3683         }
3684 
3685         /* We passed data and got it acked, remove any soft error
3686          * log. Something worked...
3687          */
3688         sk->sk_err_soft = 0;
3689         icsk->icsk_probes_out = 0;
3690         tp->rcv_tstamp = tcp_jiffies32;
3691         if (!prior_packets)
3692                 goto no_queue;
3693 
3694         /* See if we can take anything off of the retransmit queue. */
3695         flag |= tcp_clean_rtx_queue(sk, prior_fack, prior_snd_una, &sack_state);
3696 
3697         tcp_rack_update_reo_wnd(sk, &rs);
3698 
3699         if (tp->tlp_high_seq)
3700                 tcp_process_tlp_ack(sk, ack, flag);
3701         /* If needed, reset TLP/RTO timer; RACK may later override this. */
3702         if (flag & FLAG_SET_XMIT_TIMER)
3703                 tcp_set_xmit_timer(sk);
3704 
3705         if (tcp_ack_is_dubious(sk, flag)) {
3706                 if (!(flag & (FLAG_SND_UNA_ADVANCED | FLAG_NOT_DUP))) {
3707                         num_dupack = 1;
3708                         /* Consider if pure acks were aggregated in tcp_add_backlog() */
3709                         if (!(flag & FLAG_DATA))
3710                                 num_dupack = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
3711                 }
3712                 tcp_fastretrans_alert(sk, prior_snd_una, num_dupack, &flag,
3713                                       &rexmit);
3714         }
3715 
3716         if ((flag & FLAG_FORWARD_PROGRESS) || !(flag & FLAG_NOT_DUP))
3717                 sk_dst_confirm(sk);
3718 
3719         delivered = tcp_newly_delivered(sk, delivered, flag);
3720         lost = tp->lost - lost;                 /* freshly marked lost */
3721         rs.is_ack_delayed = !!(flag & FLAG_ACK_MAYBE_DELAYED);
3722         tcp_rate_gen(sk, delivered, lost, is_sack_reneg, sack_state.rate);
3723         tcp_cong_control(sk, ack, delivered, flag, sack_state.rate);
3724         tcp_xmit_recovery(sk, rexmit);
3725         return 1;
3726 
3727 no_queue:
3728         /* If data was DSACKed, see if we can undo a cwnd reduction. */
3729         if (flag & FLAG_DSACKING_ACK) {
3730                 tcp_fastretrans_alert(sk, prior_snd_una, num_dupack, &flag,
3731                                       &rexmit);
3732                 tcp_newly_delivered(sk, delivered, flag);
3733         }
3734         /* If this ack opens up a zero window, clear backoff.  It was
3735          * being used to time the probes, and is probably far higher than
3736          * it needs to be for normal retransmission.
3737          */
3738         tcp_ack_probe(sk);
3739 
3740         if (tp->tlp_high_seq)
3741                 tcp_process_tlp_ack(sk, ack, flag);
3742         return 1;
3743 
3744 old_ack:
3745         /* If data was SACKed, tag it and see if we should send more data.
3746          * If data was DSACKed, see if we can undo a cwnd reduction.
3747          */
3748         if (TCP_SKB_CB(skb)->sacked) {
3749                 flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
3750                                                 &sack_state);
3751                 tcp_fastretrans_alert(sk, prior_snd_una, num_dupack, &flag,
3752                                       &rexmit);
3753                 tcp_newly_delivered(sk, delivered, flag);
3754                 tcp_xmit_recovery(sk, rexmit);
3755         }
3756 
3757         return 0;
3758 }
3759 
3760 static void tcp_parse_fastopen_option(int len, const unsigned char *cookie,
3761                                       bool syn, struct tcp_fastopen_cookie *foc,
3762                                       bool exp_opt)
3763 {
3764         /* Valid only in SYN or SYN-ACK with an even length.  */
3765         if (!foc || !syn || len < 0 || (len & 1))
3766                 return;
3767 
3768         if (len >= TCP_FASTOPEN_COOKIE_MIN &&
3769             len <= TCP_FASTOPEN_COOKIE_MAX)
3770                 memcpy(foc->val, cookie, len);
3771         else if (len != 0)
3772                 len = -1;
3773         foc->len = len;
3774         foc->exp = exp_opt;
3775 }
3776 
3777 static void smc_parse_options(const struct tcphdr *th,
3778                               struct tcp_options_received *opt_rx,
3779                               const unsigned char *ptr,
3780                               int opsize)
3781 {
3782 #if IS_ENABLED(CONFIG_SMC)
3783         if (static_branch_unlikely(&tcp_have_smc)) {
3784                 if (th->syn && !(opsize & 1) &&
3785                     opsize >= TCPOLEN_EXP_SMC_BASE &&
3786                     get_unaligned_be32(ptr) == TCPOPT_SMC_MAGIC)
3787                         opt_rx->smc_ok = 1;
3788         }
3789 #endif
3790 }
3791 
3792 /* Try to parse the MSS option from the TCP header. Return 0 on failure, clamped
3793  * value on success.
3794  */
3795 static u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss)
3796 {
3797         const unsigned char *ptr = (const unsigned char *)(th + 1);
3798         int length = (th->doff * 4) - sizeof(struct tcphdr);
3799         u16 mss = 0;
3800 
3801         while (length > 0) {
3802                 int opcode = *ptr++;
3803                 int opsize;
3804 
3805                 switch (opcode) {
3806                 case TCPOPT_EOL:
3807                         return mss;
3808                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
3809                         length--;
3810                         continue;
3811                 default:
3812                         if (length < 2)
3813                                 return mss;
3814                         opsize = *ptr++;
3815                         if (opsize < 2) /* "silly options" */
3816                                 return mss;
3817                         if (opsize > length)
3818                                 return mss;     /* fail on partial options */
3819                         if (opcode == TCPOPT_MSS && opsize == TCPOLEN_MSS) {
3820                                 u16 in_mss = get_unaligned_be16(ptr);
3821 
3822                                 if (in_mss) {
3823                                         if (user_mss && user_mss < in_mss)
3824                                                 in_mss = user_mss;
3825                                         mss = in_mss;
3826                                 }
3827                         }
3828                         ptr += opsize - 2;
3829                         length -= opsize;
3830                 }
3831         }
3832         return mss;
3833 }
3834 
3835 /* Look for tcp options. Normally only called on SYN and SYNACK packets.
3836  * But, this can also be called on packets in the established flow when
3837  * the fast version below fails.
3838  */
3839 void tcp_parse_options(const struct net *net,
3840                        const struct sk_buff *skb,
3841                        struct tcp_options_received *opt_rx, int estab,
3842                        struct tcp_fastopen_cookie *foc)
3843 {
3844         const unsigned char *ptr;
3845         const struct tcphdr *th = tcp_hdr(skb);
3846         int length = (th->doff * 4) - sizeof(struct tcphdr);
3847 
3848         ptr = (const unsigned char *)(th + 1);
3849         opt_rx->saw_tstamp = 0;
3850 
3851         while (length > 0) {
3852                 int opcode = *ptr++;
3853                 int opsize;
3854 
3855                 switch (opcode) {
3856                 case TCPOPT_EOL:
3857                         return;
3858                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
3859                         length--;
3860                         continue;
3861                 default:
3862                         if (length < 2)
3863                                 return;
3864                         opsize = *ptr++;
3865                         if (opsize < 2) /* "silly options" */
3866                                 return;
3867                         if (opsize > length)
3868                                 return; /* don't parse partial options */
3869                         switch (opcode) {
3870                         case TCPOPT_MSS:
3871                                 if (opsize == TCPOLEN_MSS && th->syn && !estab) {
3872                                         u16 in_mss = get_unaligned_be16(ptr);
3873                                         if (in_mss) {
3874                                                 if (opt_rx->user_mss &&
3875                                                     opt_rx->user_mss < in_mss)
3876                                                         in_mss = opt_rx->user_mss;
3877                                                 opt_rx->mss_clamp = in_mss;
3878                                         }
3879                                 }
3880                                 break;
3881                         case TCPOPT_WINDOW:
3882                                 if (opsize == TCPOLEN_WINDOW && th->syn &&
3883                                     !estab && net->ipv4.sysctl_tcp_window_scaling) {
3884                                         __u8 snd_wscale = *(__u8 *)ptr;
3885                                         opt_rx->wscale_ok = 1;
3886                                         if (snd_wscale > TCP_MAX_WSCALE) {
3887                                                 net_info_ratelimited("%s: Illegal window scaling value %d > %u received\n",
3888                                                                      __func__,
3889                                                                      snd_wscale,
3890                                                                      TCP_MAX_WSCALE);
3891                                                 snd_wscale = TCP_MAX_WSCALE;
3892                                         }
3893                                         opt_rx->snd_wscale = snd_wscale;
3894                                 }
3895                                 break;
3896                         case TCPOPT_TIMESTAMP:
3897                                 if ((opsize == TCPOLEN_TIMESTAMP) &&
3898                                     ((estab && opt_rx->tstamp_ok) ||
3899                                      (!estab && net->ipv4.sysctl_tcp_timestamps))) {
3900                                         opt_rx->saw_tstamp = 1;
3901                                         opt_rx->rcv_tsval = get_unaligned_be32(ptr);
3902                                         opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4);
3903                                 }
3904                                 break;
3905                         case TCPOPT_SACK_PERM:
3906                                 if (opsize == TCPOLEN_SACK_PERM && th->syn &&
3907                                     !estab && net->ipv4.sysctl_tcp_sack) {
3908                                         opt_rx->sack_ok = TCP_SACK_SEEN;
3909                                         tcp_sack_reset(opt_rx);
3910                                 }
3911                                 break;
3912 
3913                         case TCPOPT_SACK:
3914                                 if ((opsize >= (TCPOLEN_SACK_BASE + TCPOLEN_SACK_PERBLOCK)) &&
3915                                    !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) &&
3916                                    opt_rx->sack_ok) {
3917                                         TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th;
3918                                 }
3919                                 break;
3920 #ifdef CONFIG_TCP_MD5SIG
3921                         case TCPOPT_MD5SIG:
3922                                 /*
3923                                  * The MD5 Hash has already been
3924                                  * checked (see tcp_v{4,6}_do_rcv()).
3925                                  */
3926                                 break;
3927 #endif
3928                         case TCPOPT_FASTOPEN:
3929                                 tcp_parse_fastopen_option(
3930                                         opsize - TCPOLEN_FASTOPEN_BASE,
3931                                         ptr, th->syn, foc, false);
3932                                 break;
3933 
3934                         case TCPOPT_EXP:
3935                                 /* Fast Open option shares code 254 using a
3936                                  * 16 bits magic number.
3937                                  */
3938                                 if (opsize >= TCPOLEN_EXP_FASTOPEN_BASE &&
3939                                     get_unaligned_be16(ptr) ==
3940                                     TCPOPT_FASTOPEN_MAGIC)
3941                                         tcp_parse_fastopen_option(opsize -
3942                                                 TCPOLEN_EXP_FASTOPEN_BASE,
3943                                                 ptr + 2, th->syn, foc, true);
3944                                 else
3945                                         smc_parse_options(th, opt_rx, ptr,
3946                                                           opsize);
3947                                 break;
3948 
3949                         }
3950                         ptr += opsize-2;
3951                         length -= opsize;
3952                 }
3953         }
3954 }
3955 EXPORT_SYMBOL(tcp_parse_options);
3956 
3957 static bool tcp_parse_aligned_timestamp(struct tcp_sock *tp, const struct tcphdr *th)
3958 {
3959         const __be32 *ptr = (const __be32 *)(th + 1);
3960 
3961         if (*ptr == htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16)
3962                           | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)) {
3963                 tp->rx_opt.saw_tstamp = 1;
3964                 ++ptr;
3965                 tp->rx_opt.rcv_tsval = ntohl(*ptr);
3966                 ++ptr;
3967                 if (*ptr)
3968                         tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset;
3969                 else
3970                         tp->rx_opt.rcv_tsecr = 0;
3971                 return true;
3972         }
3973         return false;
3974 }
3975 
3976 /* Fast parse options. This hopes to only see timestamps.
3977  * If it is wrong it falls back on tcp_parse_options().
3978  */
3979 static bool tcp_fast_parse_options(const struct net *net,
3980                                    const struct sk_buff *skb,
3981                                    const struct tcphdr *th, struct tcp_sock *tp)
3982 {
3983         /* In the spirit of fast parsing, compare doff directly to constant
3984          * values.  Because equality is used, short doff can be ignored here.
3985          */
3986         if (th->doff == (sizeof(*th) / 4)) {
3987                 tp->rx_opt.saw_tstamp = 0;
3988                 return false;
3989         } else if (tp->rx_opt.tstamp_ok &&
3990                    th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) {
3991                 if (tcp_parse_aligned_timestamp(tp, th))
3992                         return true;
3993         }
3994 
3995         tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL);
3996         if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
3997                 tp->rx_opt.rcv_tsecr -= tp->tsoffset;
3998 
3999         return true;
4000 }
4001 
4002 #ifdef CONFIG_TCP_MD5SIG
4003 /*
4004  * Parse MD5 Signature option
4005  */
4006 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th)
4007 {
4008         int length = (th->doff << 2) - sizeof(*th);
4009         const u8 *ptr = (const u8 *)(th + 1);
4010 
4011         /* If not enough data remaining, we can short cut */
4012         while (length >= TCPOLEN_MD5SIG) {
4013                 int opcode = *ptr++;
4014                 int opsize;
4015 
4016                 switch (opcode) {
4017                 case TCPOPT_EOL:
4018                         return NULL;
4019                 case TCPOPT_NOP:
4020                         length--;
4021                         continue;
4022                 default:
4023                         opsize = *ptr++;
4024                         if (opsize < 2 || opsize > length)
4025                                 return NULL;
4026                         if (opcode == TCPOPT_MD5SIG)
4027                                 return opsize == TCPOLEN_MD5SIG ? ptr : NULL;
4028                 }
4029                 ptr += opsize - 2;
4030                 length -= opsize;
4031         }
4032         return NULL;
4033 }
4034 EXPORT_SYMBOL(tcp_parse_md5sig_option);
4035 #endif
4036 
4037 /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
4038  *
4039  * It is not fatal. If this ACK does _not_ change critical state (seqs, window)
4040  * it can pass through stack. So, the following predicate verifies that
4041  * this segment is not used for anything but congestion avoidance or
4042  * fast retransmit. Moreover, we even are able to eliminate most of such
4043  * second order effects, if we apply some small "replay" window (~RTO)
4044  * to timestamp space.
4045  *
4046  * All these measures still do not guarantee that we reject wrapped ACKs
4047  * on networks with high bandwidth, when sequence space is recycled fastly,
4048  * but it guarantees that such events will be very rare and do not affect
4049  * connection seriously. This doesn't look nice, but alas, PAWS is really
4050  * buggy extension.
4051  *
4052  * [ Later note. Even worse! It is buggy for segments _with_ data. RFC
4053  * states that events when retransmit arrives after original data are rare.
4054  * It is a blatant lie. VJ forgot about fast retransmit! 8)8) It is
4055  * the biggest problem on large power networks even with minor reordering.
4056  * OK, let's give it small replay window. If peer clock is even 1hz, it is safe
4057  * up to bandwidth of 18Gigabit/sec. 8) ]
4058  */
4059 
4060 static int tcp_disordered_ack(const struct sock *sk, const struct sk_buff *skb)
4061 {
4062         const struct tcp_sock *tp = tcp_sk(sk);
4063         const struct tcphdr *th = tcp_hdr(skb);
4064         u32 seq = TCP_SKB_CB(skb)->seq;
4065         u32 ack = TCP_SKB_CB(skb)->ack_seq;
4066 
4067         return (/* 1. Pure ACK with correct sequence number. */
4068                 (th->ack && seq == TCP_SKB_CB(skb)->end_seq && seq == tp->rcv_nxt) &&
4069 
4070                 /* 2. ... and duplicate ACK. */
4071                 ack == tp->snd_una &&
4072 
4073                 /* 3. ... and does not update window. */
4074                 !tcp_may_update_window(tp, ack, seq, ntohs(th->window) << tp->rx_opt.snd_wscale) &&
4075 
4076                 /* 4. ... and sits in replay window. */
4077                 (s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) <= (inet_csk(sk)->icsk_rto * 1024) / HZ);
4078 }
4079 
4080 static inline bool tcp_paws_discard(const struct sock *sk,
4081                                    const struct sk_buff *skb)
4082 {
4083         const struct tcp_sock *tp = tcp_sk(sk);
4084 
4085         return !tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW) &&
4086                !tcp_disordered_ack(sk, skb);
4087 }
4088 
4089 /* Check segment sequence number for validity.
4090  *
4091  * Segment controls are considered valid, if the segment
4092  * fits to the window after truncation to the window. Acceptability
4093  * of data (and SYN, FIN, of course) is checked separately.
4094  * See tcp_data_queue(), for example.
4095  *
4096  * Also, controls (RST is main one) are accepted using RCV.WUP instead
4097  * of RCV.NXT. Peer still did not advance his SND.UNA when we
4098  * delayed ACK, so that hisSND.UNA<=ourRCV.WUP.
4099  * (borrowed from freebsd)
4100  */
4101 
4102 static inline bool tcp_sequence(const struct tcp_sock *tp, u32 seq, u32 end_seq)
4103 {
4104         return  !before(end_seq, tp->rcv_wup) &&
4105                 !after(seq, tp->rcv_nxt + tcp_receive_window(tp));
4106 }
4107 
4108 /* When we get a reset we do this. */
4109 void tcp_reset(struct sock *sk)
4110 {
4111         trace_tcp_receive_reset(sk);
4112 
4113         /* We want the right error as BSD sees it (and indeed as we do). */
4114         switch (sk->sk_state) {
4115         case TCP_SYN_SENT:
4116                 sk->sk_err = ECONNREFUSED;
4117                 break;
4118         case TCP_CLOSE_WAIT:
4119                 sk->sk_err = EPIPE;
4120                 break;
4121         case TCP_CLOSE:
4122                 return;
4123         default:
4124                 sk->sk_err = ECONNRESET;
4125         }
4126         /* This barrier is coupled with smp_rmb() in tcp_poll() */
4127         smp_wmb();
4128 
4129         tcp_write_queue_purge(sk);
4130         tcp_done(sk);
4131 
4132         if (!sock_flag(sk, SOCK_DEAD))
4133                 sk->sk_error_report(sk);
4134 }
4135 
4136 /*
4137  *      Process the FIN bit. This now behaves as it is supposed to work
4138  *      and the FIN takes effect when it is validly part of sequence
4139  *      space. Not before when we get holes.
4140  *
4141  *      If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT
4142  *      (and thence onto LAST-ACK and finally, CLOSE, we never enter
4143  *      TIME-WAIT)
4144  *
4145  *      If we are in FINWAIT-1, a received FIN indicates simultaneous
4146  *      close and we go into CLOSING (and later onto TIME-WAIT)
4147  *
4148  *      If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
4149  */
4150 void tcp_fin(struct sock *sk)
4151 {
4152         struct tcp_sock *tp = tcp_sk(sk);
4153 
4154         inet_csk_schedule_ack(sk);
4155 
4156         sk->sk_shutdown |= RCV_SHUTDOWN;
4157         sock_set_flag(sk, SOCK_DONE);
4158 
4159         switch (sk->sk_state) {
4160         case TCP_SYN_RECV:
4161         case TCP_ESTABLISHED:
4162                 /* Move to CLOSE_WAIT */
4163                 tcp_set_state(sk, TCP_CLOSE_WAIT);
4164                 inet_csk_enter_pingpong_mode(sk);
4165                 break;
4166 
4167         case TCP_CLOSE_WAIT:
4168         case TCP_CLOSING:
4169                 /* Received a retransmission of the FIN, do
4170                  * nothing.
4171                  */
4172                 break;
4173         case TCP_LAST_ACK:
4174                 /* RFC793: Remain in the LAST-ACK state. */
4175                 break;
4176 
4177         case TCP_FIN_WAIT1:
4178                 /* This case occurs when a simultaneous close
4179                  * happens, we must ack the received FIN and
4180                  * enter the CLOSING state.
4181                  */
4182                 tcp_send_ack(sk);
4183                 tcp_set_state(sk, TCP_CLOSING);
4184                 break;
4185         case TCP_FIN_WAIT2:
4186                 /* Received a FIN -- send ACK and enter TIME_WAIT. */
4187                 tcp_send_ack(sk);
4188                 tcp_time_wait(sk, TCP_TIME_WAIT, 0);
4189                 break;
4190         default:
4191                 /* Only TCP_LISTEN and TCP_CLOSE are left, in these
4192                  * cases we should never reach this piece of code.
4193                  */
4194                 pr_err("%s: Impossible, sk->sk_state=%d\n",
4195                        __func__, sk->sk_state);
4196                 break;
4197         }
4198 
4199         /* It _is_ possible, that we have something out-of-order _after_ FIN.
4200          * Probably, we should reset in this case. For now drop them.
4201          */
4202         skb_rbtree_purge(&tp->out_of_order_queue);
4203         if (tcp_is_sack(tp))
4204                 tcp_sack_reset(&tp->rx_opt);
4205         sk_mem_reclaim(sk);
4206 
4207         if (!sock_flag(sk, SOCK_DEAD)) {
4208                 sk->sk_state_change(sk);
4209 
4210                 /* Do not send POLL_HUP for half duplex close. */
4211                 if (sk->sk_shutdown == SHUTDOWN_MASK ||
4212                     sk->sk_state == TCP_CLOSE)
4213                         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
4214                 else
4215                         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
4216         }
4217 }
4218 
4219 static inline bool tcp_sack_extend(struct tcp_sack_block *sp, u32 seq,
4220                                   u32 end_seq)
4221 {
4222         if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) {
4223                 if (before(seq, sp->start_seq))
4224                         sp->start_seq = seq;
4225                 if (after(end_seq, sp->end_seq))
4226                         sp->end_seq = end_seq;
4227                 return true;
4228         }
4229         return false;
4230 }
4231 
4232 static void tcp_dsack_set(struct sock *sk, u32 seq, u32 end_seq)
4233 {
4234         struct tcp_sock *tp = tcp_sk(sk);
4235 
4236         if (tcp_is_sack(tp) && sock_net(sk)->ipv4.sysctl_tcp_dsack) {
4237                 int mib_idx;
4238 
4239                 if (before(seq, tp->rcv_nxt))
4240                         mib_idx = LINUX_MIB_TCPDSACKOLDSENT;
4241                 else
4242                         mib_idx = LINUX_MIB_TCPDSACKOFOSENT;
4243 
4244                 NET_INC_STATS(sock_net(sk), mib_idx);
4245 
4246                 tp->rx_opt.dsack = 1;
4247                 tp->duplicate_sack[0].start_seq = seq;
4248                 tp->duplicate_sack[0].end_seq = end_seq;
4249         }
4250 }
4251 
4252 static void tcp_dsack_extend(struct sock *sk, u32 seq, u32 end_seq)
4253 {
4254         struct tcp_sock *tp = tcp_sk(sk);
4255 
4256         if (!tp->rx_opt.dsack)
4257                 tcp_dsack_set(sk, seq, end_seq);
4258         else
4259                 tcp_sack_extend(tp->duplicate_sack, seq, end_seq);
4260 }
4261 
4262 static void tcp_rcv_spurious_retrans(struct sock *sk, const struct sk_buff *skb)
4263 {
4264         /* When the ACK path fails or drops most ACKs, the sender would
4265          * timeout and spuriously retransmit the same segment repeatedly.
4266          * The receiver remembers and reflects via DSACKs. Leverage the
4267          * DSACK state and change the txhash to re-route speculatively.
4268          */
4269         if (TCP_SKB_CB(skb)->seq == tcp_sk(sk)->duplicate_sack[0].start_seq)
4270                 sk_rethink_txhash(sk);
4271 }
4272 
4273 static void tcp_send_dupack(struct sock *sk, const struct sk_buff *skb)
4274 {
4275         struct tcp_sock *tp = tcp_sk(sk);
4276 
4277         if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
4278             before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
4279                 NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOST);
4280                 tcp_enter_quickack_mode(sk, TCP_MAX_QUICKACKS);
4281 
4282                 if (tcp_is_sack(tp) && sock_net(sk)->ipv4.sysctl_tcp_dsack) {
4283                         u32 end_seq = TCP_SKB_CB(skb)->end_seq;
4284 
4285                         tcp_rcv_spurious_retrans(sk, skb);
4286                         if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))
4287                                 end_seq = tp->rcv_nxt;
4288                         tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq);
4289                 }
4290         }
4291 
4292         tcp_send_ack(sk);
4293 }
4294 
4295 /* These routines update the SACK block as out-of-order packets arrive or
4296  * in-order packets close up the sequence space.
4297  */
4298 static void tcp_sack_maybe_coalesce(struct tcp_sock *tp)
4299 {
4300         int this_sack;
4301         struct tcp_sack_block *sp = &tp->selective_acks[0];
4302         struct tcp_sack_block *swalk = sp + 1;
4303 
4304         /* See if the recent change to the first SACK eats into
4305          * or hits the sequence space of other SACK blocks, if so coalesce.
4306          */
4307         for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) {
4308                 if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) {
4309                         int i;
4310 
4311                         /* Zap SWALK, by moving every further SACK up by one slot.
4312                          * Decrease num_sacks.
4313                          */
4314                         tp->rx_opt.num_sacks--;
4315                         for (i = this_sack; i < tp->rx_opt.num_sacks; i++)
4316                                 sp[i] = sp[i + 1];
4317                         continue;
4318                 }
4319                 this_sack++, swalk++;
4320         }
4321 }
4322 
4323 static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq)
4324 {
4325         struct tcp_sock *tp = tcp_sk(sk);
4326         struct tcp_sack_block *sp = &tp->selective_acks[0];
4327         int cur_sacks = tp->rx_opt.num_sacks;
4328         int this_sack;
4329 
4330         if (!cur_sacks)
4331                 goto new_sack;
4332 
4333         for (this_sack = 0; this_sack < cur_sacks; this_sack++, sp++) {
4334                 if (tcp_sack_extend(sp, seq, end_seq)) {
4335                         /* Rotate this_sack to the first one. */
4336                         for (; this_sack > 0; this_sack--, sp--)
4337                                 swap(*sp, *(sp - 1));
4338                         if (cur_sacks > 1)
4339                                 tcp_sack_maybe_coalesce(tp);
4340                         return;
4341                 }
4342         }
4343 
4344         /* Could not find an adjacent existing SACK, build a new one,
4345          * put it at the front, and shift everyone else down.  We
4346          * always know there is at least one SACK present already here.
4347          *
4348          * If the sack array is full, forget about the last one.
4349          */
4350         if (this_sack >= TCP_NUM_SACKS) {
4351                 if (tp->compressed_ack > TCP_FASTRETRANS_THRESH)
4352                         tcp_send_ack(sk);
4353                 this_sack--;
4354                 tp->rx_opt.num_sacks--;
4355                 sp--;
4356         }
4357         for (; this_sack > 0; this_sack--, sp--)
4358                 *sp = *(sp - 1);
4359 
4360 new_sack:
4361         /* Build the new head SACK, and we're done. */
4362         sp->start_seq = seq;
4363         sp->end_seq = end_seq;
4364         tp->rx_opt.num_sacks++;
4365 }
4366 
4367 /* RCV.NXT advances, some SACKs should be eaten. */
4368 
4369 static void tcp_sack_remove(struct tcp_sock *tp)
4370 {
4371         struct tcp_sack_block *sp = &tp->selective_acks[0];
4372         int num_sacks = tp->rx_opt.num_sacks;
4373         int this_sack;
4374 
4375         /* Empty ofo queue, hence, all the SACKs are eaten. Clear. */
4376         if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {
4377                 tp->rx_opt.num_sacks = 0;
4378                 return;
4379         }
4380 
4381         for (this_sack = 0; this_sack < num_sacks;) {
4382                 /* Check if the start of the sack is covered by RCV.NXT. */
4383                 if (!before(tp->rcv_nxt, sp->start_seq)) {
4384                         int i;
4385 
4386                         /* RCV.NXT must cover all the block! */
4387                         WARN_ON(before(tp->rcv_nxt, sp->end_seq));
4388 
4389                         /* Zap this SACK, by moving forward any other SACKS. */
4390                         for (i = this_sack+1; i < num_sacks; i++)
4391                                 tp->selective_acks[i-1] = tp->selective_acks[i];
4392                         num_sacks--;
4393                         continue;
4394                 }
4395                 this_sack++;
4396                 sp++;
4397         }
4398         tp->rx_opt.num_sacks = num_sacks;
4399 }
4400 
4401 /**
4402  * tcp_try_coalesce - try to merge skb to prior one
4403  * @sk: socket
4404  * @dest: destination queue
4405  * @to: prior buffer
4406  * @from: buffer to add in queue
4407  * @fragstolen: pointer to boolean
4408  *
4409  * Before queueing skb @from after @to, try to merge them
4410  * to reduce overall memory use and queue lengths, if cost is small.
4411  * Packets in ofo or receive queues can stay a long time.
4412  * Better try to coalesce them right now to avoid future collapses.
4413  * Returns true if caller should free @from instead of queueing it
4414  */
4415 static bool tcp_try_coalesce(struct sock *sk,
4416                              struct sk_buff *to,
4417                              struct sk_buff *from,
4418                              bool *fragstolen)
4419 {
4420         int delta;
4421 
4422         *fragstolen = false;
4423 
4424         /* Its possible this segment overlaps with prior segment in queue */
4425         if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq)
4426                 return false;
4427 
4428 #ifdef CONFIG_TLS_DEVICE
4429         if (from->decrypted != to->decrypted)
4430                 return false;
4431 #endif
4432 
4433         if (!skb_try_coalesce(to, from, fragstolen, &delta))
4434                 return false;
4435 
4436         atomic_add(delta, &sk->sk_rmem_alloc);
4437         sk_mem_charge(sk, delta);
4438         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVCOALESCE);
4439         TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq;
4440         TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq;
4441         TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags;
4442 
4443         if (TCP_SKB_CB(from)->has_rxtstamp) {
4444                 TCP_SKB_CB(to)->has_rxtstamp = true;
4445                 to->tstamp = from->tstamp;
4446                 skb_hwtstamps(to)->hwtstamp = skb_hwtstamps(from)->hwtstamp;
4447         }
4448 
4449         return true;
4450 }
4451 
4452 static bool tcp_ooo_try_coalesce(struct sock *sk,
4453                              struct sk_buff *to,
4454                              struct sk_buff *from,
4455                              bool *fragstolen)
4456 {
4457         bool res = tcp_try_coalesce(sk, to, from, fragstolen);
4458 
4459         /* In case tcp_drop() is called later, update to->gso_segs */
4460         if (res) {
4461                 u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) +
4462                                max_t(u16, 1, skb_shinfo(from)->gso_segs);
4463 
4464                 skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF);
4465         }
4466         return res;
4467 }
4468 
4469 static void tcp_drop(struct sock *sk, struct sk_buff *skb)
4470 {
4471         sk_drops_add(sk, skb);
4472         __kfree_skb(skb);
4473 }
4474 
4475 /* This one checks to see if we can put data from the
4476  * out_of_order queue into the receive_queue.
4477  */
4478 static void tcp_ofo_queue(struct sock *sk)
4479 {
4480         struct tcp_sock *tp = tcp_sk(sk);
4481         __u32 dsack_high = tp->rcv_nxt;
4482         bool fin, fragstolen, eaten;
4483         struct sk_buff *skb, *tail;
4484         struct rb_node *p;
4485 
4486         p = rb_first(&tp->out_of_order_queue);
4487         while (p) {
4488                 skb = rb_to_skb(p);
4489                 if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt))
4490                         break;
4491 
4492                 if (before(TCP_SKB_CB(skb)->seq, dsack_high)) {
4493                         __u32 dsack = dsack_high;
4494                         if (before(TCP_SKB_CB(skb)->end_seq, dsack_high))
4495                                 dsack_high = TCP_SKB_CB(skb)->end_seq;
4496                         tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack);
4497                 }
4498                 p = rb_next(p);
4499                 rb_erase(&skb->rbnode, &tp->out_of_order_queue);
4500 
4501                 if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) {
4502                         tcp_drop(sk, skb);
4503                         continue;
4504                 }
4505 
4506                 tail = skb_peek_tail(&sk->sk_receive_queue);
4507                 eaten = tail && tcp_try_coalesce(sk, tail, skb, &fragstolen);
4508                 tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq);
4509                 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
4510                 if (!eaten)
4511                         __skb_queue_tail(&sk->sk_receive_queue, skb);
4512                 else
4513                         kfree_skb_partial(skb, fragstolen);
4514 
4515                 if (unlikely(fin)) {
4516                         tcp_fin(sk);
4517                         /* tcp_fin() purges tp->out_of_order_queue,
4518                          * so we must end this loop right now.
4519                          */
4520                         break;
4521                 }
4522         }
4523 }
4524 
4525 static bool tcp_prune_ofo_queue(struct sock *sk);
4526 static int tcp_prune_queue(struct sock *sk);
4527 
4528 static int tcp_try_rmem_schedule(struct sock *sk, struct sk_buff *skb,
4529                                  unsigned int size)
4530 {
4531         if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
4532             !sk_rmem_schedule(sk, skb, size)) {
4533 
4534                 if (tcp_prune_queue(sk) < 0)
4535                         return -1;
4536 
4537                 while (!sk_rmem_schedule(sk, skb, size)) {
4538                         if (!tcp_prune_ofo_queue(sk))
4539                                 return -1;
4540                 }
4541         }
4542         return 0;
4543 }
4544 
4545 static void tcp_data_queue_ofo(struct sock *sk, struct sk_buff *skb)
4546 {
4547         struct tcp_sock *tp = tcp_sk(sk);
4548         struct rb_node **p, *parent;
4549         struct sk_buff *skb1;
4550         u32 seq, end_seq;
4551         bool fragstolen;
4552 
4553         tcp_ecn_check_ce(sk, skb);
4554 
4555         if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) {
4556                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPOFODROP);
4557                 tcp_drop(sk, skb);
4558                 return;
4559         }
4560 
4561         /* Disable header prediction. */
4562         tp->pred_flags = 0;
4563         inet_csk_schedule_ack(sk);
4564 
4565         tp->rcv_ooopack += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
4566         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPOFOQUEUE);
4567         seq = TCP_SKB_CB(skb)->seq;
4568         end_seq = TCP_SKB_CB(skb)->end_seq;
4569 
4570         p = &tp->out_of_order_queue.rb_node;
4571         if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {
4572                 /* Initial out of order segment, build 1 SACK. */
4573                 if (tcp_is_sack(tp)) {
4574                         tp->rx_opt.num_sacks = 1;
4575                         tp->selective_acks[0].start_seq = seq;
4576                         tp->selective_acks[0].end_seq = end_seq;
4577                 }
4578                 rb_link_node(&skb->rbnode, NULL, p);
4579                 rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);
4580                 tp->ooo_last_skb = skb;
4581                 goto end;
4582         }
4583 
4584         /* In the typical case, we are adding an skb to the end of the list.
4585          * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
4586          */
4587         if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb,
4588                                  skb, &fragstolen)) {
4589 coalesce_done:
4590                 tcp_grow_window(sk, skb);
4591                 kfree_skb_partial(skb, fragstolen);
4592                 skb = NULL;
4593                 goto add_sack;
4594         }
4595         /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
4596         if (!before(seq, TCP_SKB_CB(tp->ooo_last_skb)->end_seq)) {
4597                 parent = &tp->ooo_last_skb->rbnode;
4598                 p = &parent->rb_right;
4599                 goto insert;
4600         }
4601 
4602         /* Find place to insert this segment. Handle overlaps on the way. */
4603         parent = NULL;
4604         while (*p) {
4605                 parent = *p;
4606                 skb1 = rb_to_skb(parent);
4607                 if (before(seq, TCP_SKB_CB(skb1)->seq)) {
4608                         p = &parent->rb_left;
4609                         continue;
4610                 }
4611                 if (before(seq, TCP_SKB_CB(skb1)->end_seq)) {
4612                         if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
4613                                 /* All the bits are present. Drop. */
4614                                 NET_INC_STATS(sock_net(sk),
4615                                               LINUX_MIB_TCPOFOMERGE);
4616                                 tcp_drop(sk, skb);
4617                                 skb = NULL;
4618                                 tcp_dsack_set(sk, seq, end_seq);
4619                                 goto add_sack;
4620                         }
4621                         if (after(seq, TCP_SKB_CB(skb1)->seq)) {
4622                                 /* Partial overlap. */
4623                                 tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq);
4624                         } else {
4625                                 /* skb's seq == skb1's seq and skb covers skb1.
4626                                  * Replace skb1 with skb.
4627                                  */
4628                                 rb_replace_node(&skb1->rbnode, &skb->rbnode,
4629                                                 &tp->out_of_order_queue);
4630                                 tcp_dsack_extend(sk,
4631                                                  TCP_SKB_CB(skb1)->seq,
4632                                                  TCP_SKB_CB(skb1)->end_seq);
4633                                 NET_INC_STATS(sock_net(sk),
4634                                               LINUX_MIB_TCPOFOMERGE);
4635                                 tcp_drop(sk, skb1);
4636                                 goto merge_right;
4637                         }
4638                 } else if (tcp_ooo_try_coalesce(sk, skb1,
4639                                                 skb, &fragstolen)) {
4640                         goto coalesce_done;
4641                 }
4642                 p = &parent->rb_right;
4643         }
4644 insert:
4645         /* Insert segment into RB tree. */
4646         rb_link_node(&skb->rbnode, parent, p);
4647         rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);
4648 
4649 merge_right:
4650         /* Remove other segments covered by skb. */
4651         while ((skb1 = skb_rb_next(skb)) != NULL) {
4652                 if (!after(end_seq, TCP_SKB_CB(skb1)->seq))
4653                         break;
4654                 if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
4655                         tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,
4656                                          end_seq);
4657                         break;
4658                 }
4659                 rb_erase(&skb1->rbnode, &tp->out_of_order_queue);
4660                 tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,
4661                                  TCP_SKB_CB(skb1)->end_seq);
4662                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPOFOMERGE);
4663                 tcp_drop(sk, skb1);
4664         }
4665         /* If there is no skb after us, we are the last_skb ! */
4666         if (!skb1)
4667                 tp->ooo_last_skb = skb;
4668 
4669 add_sack:
4670         if (tcp_is_sack(tp))
4671                 tcp_sack_new_ofo_skb(sk, seq, end_seq);
4672 end:
4673         if (skb) {
4674                 tcp_grow_window(sk, skb);
4675                 skb_condense(skb);
4676                 skb_set_owner_r(skb, sk);
4677         }
4678 }
4679 
4680 static int __must_check tcp_queue_rcv(struct sock *sk, struct sk_buff *skb,
4681                                       bool *fragstolen)
4682 {
4683         int eaten;
4684         struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue);
4685 
4686         eaten = (tail &&
4687                  tcp_try_coalesce(sk, tail,
4688                                   skb, fragstolen)) ? 1 : 0;
4689         tcp_rcv_nxt_update(tcp_sk(sk), TCP_SKB_CB(skb)->end_seq);
4690         if (!eaten) {
4691                 __skb_queue_tail(&sk->sk_receive_queue, skb);
4692                 skb_set_owner_r(skb, sk);
4693         }
4694         return eaten;
4695 }
4696 
4697 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size)
4698 {
4699         struct sk_buff *skb;
4700         int err = -ENOMEM;
4701         int data_len = 0;
4702         bool fragstolen;
4703 
4704         if (size == 0)
4705                 return 0;
4706 
4707         if (size > PAGE_SIZE) {
4708                 int npages = min_t(size_t, size >> PAGE_SHIFT, MAX_SKB_FRAGS);
4709 
4710                 data_len = npages << PAGE_SHIFT;
4711                 size = data_len + (size & ~PAGE_MASK);
4712         }
4713         skb = alloc_skb_with_frags(size - data_len, data_len,
4714                                    PAGE_ALLOC_COSTLY_ORDER,
4715                                    &err, sk->sk_allocation);
4716         if (!skb)
4717                 goto err;
4718 
4719         skb_put(skb, size - data_len);
4720         skb->data_len = data_len;
4721         skb->len = size;
4722 
4723         if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {
4724                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVQDROP);
4725                 goto err_free;
4726         }
4727 
4728         err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);
4729         if (err)
4730                 goto err_free;
4731 
4732         TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt;
4733         TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size;
4734         TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1;
4735 
4736         if (tcp_queue_rcv(sk, skb, &fragstolen)) {
4737                 WARN_ON_ONCE(fragstolen); /* should not happen */
4738                 __kfree_skb(skb);
4739         }
4740         return size;
4741 
4742 err_free:
4743         kfree_skb(skb);
4744 err:
4745         return err;
4746 
4747 }
4748 
4749 void tcp_data_ready(struct sock *sk)
4750 {
4751         const struct tcp_sock *tp = tcp_sk(sk);
4752         int avail = tp->rcv_nxt - tp->copied_seq;
4753 
4754         if (avail < sk->sk_rcvlowat && !tcp_rmem_pressure(sk) &&
4755             !sock_flag(sk, SOCK_DONE))
4756                 return;
4757 
4758         sk->sk_data_ready(sk);
4759 }
4760 
4761 static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
4762 {
4763         struct tcp_sock *tp = tcp_sk(sk);
4764         bool fragstolen;
4765         int eaten;
4766 
4767         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
4768                 __kfree_skb(skb);
4769                 return;
4770         }
4771         skb_dst_drop(skb);
4772         __skb_pull(skb, tcp_hdr(skb)->doff * 4);
4773 
4774         tcp_ecn_accept_cwr(sk, skb);
4775 
4776         tp->rx_opt.dsack = 0;
4777 
4778         /*  Queue data for delivery to the user.
4779          *  Packets in sequence go to the receive queue.
4780          *  Out of sequence packets to the out_of_order_queue.
4781          */
4782         if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) {
4783                 if (tcp_receive_window(tp) == 0) {
4784                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPZEROWINDOWDROP);
4785                         goto out_of_window;
4786                 }
4787 
4788                 /* Ok. In sequence. In window. */
4789 queue_and_out:
4790                 if (skb_queue_len(&sk->sk_receive_queue) == 0)
4791                         sk_forced_mem_schedule(sk, skb->truesize);
4792                 else if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {
4793                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVQDROP);
4794                         goto drop;
4795                 }
4796 
4797                 eaten = tcp_queue_rcv(sk, skb, &fragstolen);
4798                 if (skb->len)
4799                         tcp_event_data_recv(sk, skb);
4800                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
4801                         tcp_fin(sk);
4802 
4803                 if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) {
4804                         tcp_ofo_queue(sk);
4805 
4806                         /* RFC5681. 4.2. SHOULD send immediate ACK, when
4807                          * gap in queue is filled.
4808                          */
4809                         if (RB_EMPTY_ROOT(&tp->out_of_order_queue))
4810                                 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
4811                 }
4812 
4813                 if (tp->rx_opt.num_sacks)
4814                         tcp_sack_remove(tp);
4815 
4816                 tcp_fast_path_check(sk);
4817 
4818                 if (eaten > 0)
4819                         kfree_skb_partial(skb, fragstolen);
4820                 if (!sock_flag(sk, SOCK_DEAD))
4821                         tcp_data_ready(sk);
4822                 return;
4823         }
4824 
4825         if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) {
4826                 tcp_rcv_spurious_retrans(sk, skb);
4827                 /* A retransmit, 2nd most common case.  Force an immediate ack. */
4828                 NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOST);
4829                 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);
4830 
4831 out_of_window:
4832                 tcp_enter_quickack_mode(sk, TCP_MAX_QUICKACKS);
4833                 inet_csk_schedule_ack(sk);
4834 drop:
4835                 tcp_drop(sk, skb);
4836                 return;
4837         }
4838 
4839         /* Out of window. F.e. zero window probe. */
4840         if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt + tcp_receive_window(tp)))
4841                 goto out_of_window;
4842 
4843         if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
4844                 /* Partial packet, seq < rcv_next < end_seq */
4845                 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt);
4846 
4847                 /* If window is closed, drop tail of packet. But after
4848                  * remembering D-SACK for its head made in previous line.
4849                  */
4850                 if (!tcp_receive_window(tp)) {
4851                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPZEROWINDOWDROP);
4852                         goto out_of_window;
4853                 }
4854                 goto queue_and_out;
4855         }
4856 
4857         tcp_data_queue_ofo(sk, skb);
4858 }
4859 
4860 static struct sk_buff *tcp_skb_next(struct sk_buff *skb, struct sk_buff_head *list)
4861 {
4862         if (list)
4863                 return !skb_queue_is_last(list, skb) ? skb->next : NULL;
4864 
4865         return skb_rb_next(skb);
4866 }
4867 
4868 static struct sk_buff *tcp_collapse_one(struct sock *sk, struct sk_buff *skb,
4869                                         struct sk_buff_head *list,
4870                                         struct rb_root *root)
4871 {
4872         struct sk_buff *next = tcp_skb_next(skb, list);
4873 
4874         if (list)
4875                 __skb_unlink(skb, list);
4876         else
4877                 rb_erase(&skb->rbnode, root);
4878 
4879         __kfree_skb(skb);
4880         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVCOLLAPSED);
4881 
4882         return next;
4883 }
4884 
4885 /* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */
4886 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb)
4887 {
4888         struct rb_node **p = &root->rb_node;
4889         struct rb_node *parent = NULL;
4890         struct sk_buff *skb1;
4891 
4892         while (*p) {
4893                 parent = *p;
4894                 skb1 = rb_to_skb(parent);
4895                 if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq))
4896                         p = &parent->rb_left;
4897                 else
4898                         p = &parent->rb_right;
4899         }
4900         rb_link_node(&skb->rbnode, parent, p);
4901         rb_insert_color(&skb->rbnode, root);
4902 }
4903 
4904 /* Collapse contiguous sequence of skbs head..tail with
4905  * sequence numbers start..end.
4906  *
4907  * If tail is NULL, this means until the end of the queue.
4908  *
4909  * Segments with FIN/SYN are not collapsed (only because this
4910  * simplifies code)
4911  */
4912 static void
4913 tcp_collapse(struct sock *sk, struct sk_buff_head *list, struct rb_root *root,
4914              struct sk_buff *head, struct sk_buff *tail, u32 start, u32 end)
4915 {
4916         struct sk_buff *skb = head, *n;
4917         struct sk_buff_head tmp;
4918         bool end_of_skbs;
4919 
4920         /* First, check that queue is collapsible and find
4921          * the point where collapsing can be useful.
4922          */
4923 restart:
4924         for (end_of_skbs = true; skb != NULL && skb != tail; skb = n) {
4925                 n = tcp_skb_next(skb, list);
4926 
4927                 /* No new bits? It is possible on ofo queue. */
4928                 if (!before(start, TCP_SKB_CB(skb)->end_seq)) {
4929                         skb = tcp_collapse_one(sk, skb, list, root);
4930                         if (!skb)
4931                                 break;
4932                         goto restart;
4933                 }
4934 
4935                 /* The first skb to collapse is:
4936                  * - not SYN/FIN and
4937                  * - bloated or contains data before "start" or
4938                  *   overlaps to the next one.
4939                  */
4940                 if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) &&
4941                     (tcp_win_from_space(sk, skb->truesize) > skb->len ||
4942                      before(TCP_SKB_CB(skb)->seq, start))) {
4943                         end_of_skbs = false;
4944                         break;
4945                 }
4946 
4947                 if (n && n != tail &&
4948                     TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) {
4949                         end_of_skbs = false;
4950                         break;
4951                 }
4952 
4953                 /* Decided to skip this, advance start seq. */
4954                 start = TCP_SKB_CB(skb)->end_seq;
4955         }
4956         if (end_of_skbs ||
4957             (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)))
4958                 return;
4959 
4960         __skb_queue_head_init(&tmp);
4961 
4962         while (before(start, end)) {
4963                 int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start);
4964                 struct sk_buff *nskb;
4965 
4966                 nskb = alloc_skb(copy, GFP_ATOMIC);
4967                 if (!nskb)
4968                         break;
4969 
4970                 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
4971 #ifdef CONFIG_TLS_DEVICE
4972                 nskb->decrypted = skb->decrypted;
4973 #endif
4974                 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start;
4975                 if (list)
4976                         __skb_queue_before(list, skb, nskb);
4977                 else
4978                         __skb_queue_tail(&tmp, nskb); /* defer rbtree insertion */
4979                 skb_set_owner_r(nskb, sk);
4980 
4981                 /* Copy data, releasing collapsed skbs. */
4982                 while (copy > 0) {
4983                         int offset = start - TCP_SKB_CB(skb)->seq;
4984                         int size = TCP_SKB_CB(skb)->end_seq - start;
4985 
4986                         BUG_ON(offset < 0);
4987                         if (size > 0) {
4988                                 size = min(copy, size);
4989                                 if (skb_copy_bits(skb, offset, skb_put(nskb, size), size))
4990                                         BUG();
4991                                 TCP_SKB_CB(nskb)->end_seq += size;
4992                                 copy -= size;
4993                                 start += size;
4994                         }
4995                         if (!before(start, TCP_SKB_CB(skb)->end_seq)) {
4996                                 skb = tcp_collapse_one(sk, skb, list, root);
4997                                 if (!skb ||
4998                                     skb == tail ||
4999                                     (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)))
5000                                         goto end;
5001 #ifdef CONFIG_TLS_DEVICE
5002                                 if (skb->decrypted != nskb->decrypted)
5003                                         goto end;
5004 #endif
5005                         }
5006                 }
5007         }
5008 end:
5009         skb_queue_walk_safe(&tmp, skb, n)
5010                 tcp_rbtree_insert(root, skb);
5011 }
5012 
5013 /* Collapse ofo queue. Algorithm: select contiguous sequence of skbs
5014  * and tcp_collapse() them until all the queue is collapsed.
5015  */
5016 static void tcp_collapse_ofo_queue(struct sock *sk)
5017 {
5018         struct tcp_sock *tp = tcp_sk(sk);
5019         u32 range_truesize, sum_tiny = 0;
5020         struct sk_buff *skb, *head;
5021         u32 start, end;
5022 
5023         skb = skb_rb_first(&tp->out_of_order_queue);
5024 new_range:
5025         if (!skb) {
5026                 tp->ooo_last_skb = skb_rb_last(&tp->out_of_order_queue);
5027                 return;
5028         }
5029         start = TCP_SKB_CB(skb)->seq;
5030         end = TCP_SKB_CB(skb)->end_seq;
5031         range_truesize = skb->truesize;
5032 
5033         for (head = skb;;) {
5034                 skb = skb_rb_next(skb);
5035 
5036                 /* Range is terminated when we see a gap or when
5037                  * we are at the queue end.
5038                  */
5039                 if (!skb ||
5040                     after(TCP_SKB_CB(skb)->seq, end) ||
5041                     before(TCP_SKB_CB(skb)->end_seq, start)) {
5042                         /* Do not attempt collapsing tiny skbs */
5043                         if (range_truesize != head->truesize ||
5044                             end - start >= SKB_WITH_OVERHEAD(SK_MEM_QUANTUM)) {
5045                                 tcp_collapse(sk, NULL, &tp->out_of_order_queue,
5046                                              head, skb, start, end);
5047                         } else {
5048                                 sum_tiny += range_truesize;
5049                                 if (sum_tiny > sk->sk_rcvbuf >> 3)
5050                                         return;
5051                         }
5052                         goto new_range;
5053                 }
5054 
5055                 range_truesize += skb->truesize;
5056                 if (unlikely(before(TCP_SKB_CB(skb)->seq, start)))
5057                         start = TCP_SKB_CB(skb)->seq;
5058                 if (after(TCP_SKB_CB(skb)->end_seq, end))
5059                         end = TCP_SKB_CB(skb)->end_seq;
5060         }
5061 }
5062 
5063 /*
5064  * Clean the out-of-order queue to make room.
5065  * We drop high sequences packets to :
5066  * 1) Let a chance for holes to be filled.
5067  * 2) not add too big latencies if thousands of packets sit there.
5068  *    (But if application shrinks SO_RCVBUF, we could still end up
5069  *     freeing whole queue here)
5070  * 3) Drop at least 12.5 % of sk_rcvbuf to avoid malicious attacks.
5071  *
5072  * Return true if queue has shrunk.
5073  */
5074 static bool tcp_prune_ofo_queue(struct sock *sk)
5075 {
5076         struct tcp_sock *tp = tcp_sk(sk);
5077         struct rb_node *node, *prev;
5078         int goal;
5079 
5080         if (RB_EMPTY_ROOT(&tp->out_of_order_queue))
5081                 return false;
5082 
5083         NET_INC_STATS(sock_net(sk), LINUX_MIB_OFOPRUNED);
5084         goal = sk->sk_rcvbuf >> 3;
5085         node = &tp->ooo_last_skb->rbnode;
5086         do {
5087                 prev = rb_prev(node);
5088                 rb_erase(node, &tp->out_of_order_queue);
5089                 goal -= rb_to_skb(node)->truesize;
5090                 tcp_drop(sk, rb_to_skb(node));
5091                 if (!prev || goal <= 0) {
5092                         sk_mem_reclaim(sk);
5093                         if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
5094                             !tcp_under_memory_pressure(sk))
5095                                 break;
5096                         goal = sk->sk_rcvbuf >> 3;
5097                 }
5098                 node = prev;
5099         } while (node);
5100         tp->ooo_last_skb = rb_to_skb(prev);
5101 
5102         /* Reset SACK state.  A conforming SACK implementation will
5103          * do the same at a timeout based retransmit.  When a connection
5104          * is in a sad state like this, we care only about integrity
5105          * of the connection not performance.
5106          */
5107         if (tp->rx_opt.sack_ok)
5108                 tcp_sack_reset(&tp->rx_opt);
5109         return true;
5110 }
5111 
5112 /* Reduce allocated memory if we can, trying to get
5113  * the socket within its memory limits again.
5114  *
5115  * Return less than zero if we should start dropping frames
5116  * until the socket owning process reads some of the data
5117  * to stabilize the situation.
5118  */
5119 static int tcp_prune_queue(struct sock *sk)
5120 {
5121         struct tcp_sock *tp = tcp_sk(sk);
5122 
5123         NET_INC_STATS(sock_net(sk), LINUX_MIB_PRUNECALLED);
5124 
5125         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
5126                 tcp_clamp_window(sk);
5127         else if (tcp_under_memory_pressure(sk))
5128                 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
5129 
5130         if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)
5131                 return 0;
5132 
5133         tcp_collapse_ofo_queue(sk);
5134         if (!skb_queue_empty(&sk->sk_receive_queue))
5135                 tcp_collapse(sk, &sk->sk_receive_queue, NULL,
5136                              skb_peek(&sk->sk_receive_queue),
5137                              NULL,
5138                              tp->copied_seq, tp->rcv_nxt);
5139         sk_mem_reclaim(sk);
5140 
5141         if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)
5142                 return 0;
5143 
5144         /* Collapsing did not help, destructive actions follow.
5145          * This must not ever occur. */
5146 
5147         tcp_prune_ofo_queue(sk);
5148 
5149         if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)
5150                 return 0;
5151 
5152         /* If we are really being abused, tell the caller to silently
5153          * drop receive data on the floor.  It will get retransmitted
5154          * and hopefully then we'll have sufficient space.
5155          */
5156         NET_INC_STATS(sock_net(sk), LINUX_MIB_RCVPRUNED);
5157 
5158         /* Massive buffer overcommit. */
5159         tp->pred_flags = 0;
5160         return -1;
5161 }
5162 
5163 static bool tcp_should_expand_sndbuf(const struct sock *sk)
5164 {
5165         const struct tcp_sock *tp = tcp_sk(sk);
5166 
5167         /* If the user specified a specific send buffer setting, do
5168          * not modify it.
5169          */
5170         if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
5171                 return false;
5172 
5173         /* If we are under global TCP memory pressure, do not expand.  */
5174         if (tcp_under_memory_pressure(sk))
5175                 return false;
5176 
5177         /* If we are under soft global TCP memory pressure, do not expand.  */
5178         if (sk_memory_allocated(sk) >= sk_prot_mem_limits(sk, 0))
5179                 return false;
5180 
5181         /* If we filled the congestion window, do not expand.  */
5182         if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
5183                 return false;
5184 
5185         return true;
5186 }
5187 
5188 /* When incoming ACK allowed to free some skb from write_queue,
5189  * we remember this event in flag SOCK_QUEUE_SHRUNK and wake up socket
5190  * on the exit from tcp input handler.
5191  *
5192  * PROBLEM: sndbuf expansion does not work well with largesend.
5193  */
5194 static void tcp_new_space(struct sock *sk)
5195 {
5196         struct tcp_sock *tp = tcp_sk(sk);
5197 
5198         if (tcp_should_expand_sndbuf(sk)) {
5199                 tcp_sndbuf_expand(sk);
5200                 tp->snd_cwnd_stamp = tcp_jiffies32;
5201         }
5202 
5203         sk->sk_write_space(sk);
5204 }
5205 
5206 static void tcp_check_space(struct sock *sk)
5207 {
5208         if (sock_flag(sk, SOCK_QUEUE_SHRUNK)) {
5209                 sock_reset_flag(sk, SOCK_QUEUE_SHRUNK);
5210                 /* pairs with tcp_poll() */
5211                 smp_mb();
5212                 if (sk->sk_socket &&
5213                     test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
5214                         tcp_new_space(sk);
5215                         if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))
5216                                 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
5217                 }
5218         }
5219 }
5220 
5221 static inline void tcp_data_snd_check(struct sock *sk)
5222 {
5223         tcp_push_pending_frames(sk);
5224         tcp_check_space(sk);
5225 }
5226 
5227 /*
5228  * Check if sending an ack is needed.
5229  */
5230 static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
5231 {
5232         struct tcp_sock *tp = tcp_sk(sk);
5233         unsigned long rtt, delay;
5234 
5235             /* More than one full frame received... */
5236         if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&
5237              /* ... and right edge of window advances far enough.
5238               * (tcp_recvmsg() will send ACK otherwise).
5239               * If application uses SO_RCVLOWAT, we want send ack now if
5240               * we have not received enough bytes to satisfy the condition.
5241               */
5242             (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||
5243              __tcp_select_window(sk) >= tp->rcv_wnd)) ||
5244             /* We ACK each frame or... */
5245             tcp_in_quickack_mode(sk) ||
5246             /* Protocol state mandates a one-time immediate ACK */
5247             inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOW) {
5248 send_now:
5249                 tcp_send_ack(sk);
5250                 return;
5251         }
5252 
5253         if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) {
5254                 tcp_send_delayed_ack(sk);
5255                 return;
5256         }
5257 
5258         if (!tcp_is_sack(tp) ||
5259             tp->compressed_ack >= sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr)
5260                 goto send_now;
5261 
5262         if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) {
5263                 tp->compressed_ack_rcv_nxt = tp->rcv_nxt;
5264                 if (tp->compressed_ack > TCP_FASTRETRANS_THRESH)
5265                         NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPACKCOMPRESSED,
5266                                       tp->compressed_ack - TCP_FASTRETRANS_THRESH);
5267                 tp->compressed_ack = 0;
5268         }
5269 
5270         if (++tp->compressed_ack <= TCP_FASTRETRANS_THRESH)
5271                 goto send_now;
5272 
5273         if (hrtimer_is_queued(&tp->compressed_ack_timer))
5274                 return;
5275 
5276         /* compress ack timer : 5 % of rtt, but no more than tcp_comp_sack_delay_ns */
5277 
5278         rtt = tp->rcv_rtt_est.rtt_us;
5279         if (tp->srtt_us && tp->srtt_us < rtt)
5280                 rtt = tp->srtt_us;
5281 
5282         delay = min_t(unsigned long, sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns,
5283                       rtt * (NSEC_PER_USEC >> 3)/20);
5284         sock_hold(sk);
5285         hrtimer_start(&tp->compressed_ack_timer, ns_to_ktime(delay),
5286                       HRTIMER_MODE_REL_PINNED_SOFT);
5287 }
5288 
5289 static inline void tcp_ack_snd_check(struct sock *sk)
5290 {
5291         if (!inet_csk_ack_scheduled(sk)) {
5292                 /* We sent a data segment already. */
5293                 return;
5294         }
5295         __tcp_ack_snd_check(sk, 1);
5296 }
5297 
5298 /*
5299  *      This routine is only called when we have urgent data
5300  *      signaled. Its the 'slow' part of tcp_urg. It could be
5301  *      moved inline now as tcp_urg is only called from one
5302  *      place. We handle URGent data wrong. We have to - as
5303  *      BSD still doesn't use the correction from RFC961.
5304  *      For 1003.1g we should support a new option TCP_STDURG to permit
5305  *      either form (or just set the sysctl tcp_stdurg).
5306  */
5307 
5308 static void tcp_check_urg(struct sock *sk, const struct tcphdr *th)
5309 {
5310         struct tcp_sock *tp = tcp_sk(sk);
5311         u32 ptr = ntohs(th->urg_ptr);
5312 
5313         if (ptr && !sock_net(sk)->ipv4.sysctl_tcp_stdurg)
5314                 ptr--;
5315         ptr += ntohl(th->seq);
5316 
5317         /* Ignore urgent data that we've already seen and read. */
5318         if (after(tp->copied_seq, ptr))
5319                 return;
5320 
5321         /* Do not replay urg ptr.
5322          *
5323          * NOTE: interesting situation not covered by specs.
5324          * Misbehaving sender may send urg ptr, pointing to segment,
5325          * which we already have in ofo queue. We are not able to fetch
5326          * such data and will stay in TCP_URG_NOTYET until will be eaten
5327          * by recvmsg(). Seems, we are not obliged to handle such wicked
5328          * situations. But it is worth to think about possibility of some
5329          * DoSes using some hypothetical application level deadlock.
5330          */
5331         if (before(ptr, tp->rcv_nxt))
5332                 return;
5333 
5334         /* Do we already have a newer (or duplicate) urgent pointer? */
5335         if (tp->urg_data && !after(ptr, tp->urg_seq))
5336                 return;
5337 
5338         /* Tell the world about our new urgent pointer. */
5339         sk_send_sigurg(sk);
5340 
5341         /* We may be adding urgent data when the last byte read was
5342          * urgent. To do this requires some care. We cannot just ignore
5343          * tp->copied_seq since we would read the last urgent byte again
5344          * as data, nor can we alter copied_seq until this data arrives
5345          * or we break the semantics of SIOCATMARK (and thus sockatmark())
5346          *
5347          * NOTE. Double Dutch. Rendering to plain English: author of comment
5348          * above did something sort of  send("A", MSG_OOB); send("B", MSG_OOB);
5349          * and expect that both A and B disappear from stream. This is _wrong_.
5350          * Though this happens in BSD with high probability, this is occasional.
5351          * Any application relying on this is buggy. Note also, that fix "works"
5352          * only in this artificial test. Insert some normal data between A and B and we will
5353          * decline of BSD again. Verdict: it is better to remove to trap
5354          * buggy users.
5355          */
5356         if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
5357             !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) {
5358                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
5359                 tp->copied_seq++;
5360                 if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) {
5361                         __skb_unlink(skb, &sk->sk_receive_queue);
5362                         __kfree_skb(skb);
5363                 }
5364         }
5365 
5366         tp->urg_data = TCP_URG_NOTYET;
5367         WRITE_ONCE(tp->urg_seq, ptr);
5368 
5369         /* Disable header prediction. */
5370         tp->pred_flags = 0;
5371 }
5372 
5373 /* This is the 'fast' part of urgent handling. */
5374 static void tcp_urg(struct sock *sk, struct sk_buff *skb, const struct tcphdr *th)
5375 {
5376         struct tcp_sock *tp = tcp_sk(sk);
5377 
5378         /* Check if we get a new urgent pointer - normally not. */
5379         if (th->urg)
5380                 tcp_check_urg(sk, th);
5381 
5382         /* Do we wait for any urgent data? - normally not... */
5383         if (tp->urg_data == TCP_URG_NOTYET) {
5384                 u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) -
5385                           th->syn;
5386 
5387                 /* Is the urgent pointer pointing into this packet? */
5388                 if (ptr < skb->len) {
5389                         u8 tmp;
5390                         if (skb_copy_bits(skb, ptr, &tmp, 1))
5391                                 BUG();
5392                         tp->urg_data = TCP_URG_VALID | tmp;
5393                         if (!sock_flag(sk, SOCK_DEAD))
5394                                 sk->sk_data_ready(sk);
5395                 }
5396         }
5397 }
5398 
5399 /* Accept RST for rcv_nxt - 1 after a FIN.
5400  * When tcp connections are abruptly terminated from Mac OSX (via ^C), a
5401  * FIN is sent followed by a RST packet. The RST is sent with the same
5402  * sequence number as the FIN, and thus according to RFC 5961 a challenge
5403  * ACK should be sent. However, Mac OSX rate limits replies to challenge
5404  * ACKs on the closed socket. In addition middleboxes can drop either the
5405  * challenge ACK or a subsequent RST.
5406  */
5407 static bool tcp_reset_check(const struct sock *sk, const struct sk_buff *skb)
5408 {
5409         struct tcp_sock *tp = tcp_sk(sk);
5410 
5411         return unlikely(TCP_SKB_CB(skb)->seq == (tp->rcv_nxt - 1) &&
5412                         (1 << sk->sk_state) & (TCPF_CLOSE_WAIT | TCPF_LAST_ACK |
5413                                                TCPF_CLOSING));
5414 }
5415 
5416 /* Does PAWS and seqno based validation of an incoming segment, flags will
5417  * play significant role here.
5418  */
5419 static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb,
5420                                   const struct tcphdr *th, int syn_inerr)
5421 {
5422         struct tcp_sock *tp = tcp_sk(sk);
5423         bool rst_seq_match = false;
5424 
5425         /* RFC1323: H1. Apply PAWS check first. */
5426         if (tcp_fast_parse_options(sock_net(sk), skb, th, tp) &&
5427             tp->rx_opt.saw_tstamp &&
5428             tcp_paws_discard(sk, skb)) {
5429                 if (!th->rst) {
5430                         NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
5431                         if (!tcp_oow_rate_limited(sock_net(sk), skb,
5432                                                   LINUX_MIB_TCPACKSKIPPEDPAWS,
5433                                                   &tp->last_oow_ack_time))
5434                                 tcp_send_dupack(sk, skb);
5435                         goto discard;
5436                 }
5437                 /* Reset is accepted even if it did not pass PAWS. */
5438         }
5439 
5440         /* Step 1: check sequence number */
5441         if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)) {
5442                 /* RFC793, page 37: "In all states except SYN-SENT, all reset
5443                  * (RST) segments are validated by checking their SEQ-fields."
5444                  * And page 69: "If an incoming segment is not acceptable,
5445                  * an acknowledgment should be sent in reply (unless the RST
5446                  * bit is set, if so drop the segment and return)".
5447                  */
5448                 if (!th->rst) {
5449                         if (th->syn)
5450                                 goto syn_challenge;
5451                         if (!tcp_oow_rate_limited(sock_net(sk), skb,
5452                                                   LINUX_MIB_TCPACKSKIPPEDSEQ,
5453                                                   &tp->last_oow_ack_time))
5454                                 tcp_send_dupack(sk, skb);
5455                 } else if (tcp_reset_check(sk, skb)) {
5456                         tcp_reset(sk);
5457                 }
5458                 goto discard;
5459         }
5460 
5461         /* Step 2: check RST bit */
5462         if (th->rst) {
5463                 /* RFC 5961 3.2 (extend to match against (RCV.NXT - 1) after a
5464                  * FIN and SACK too if available):
5465                  * If seq num matches RCV.NXT or (RCV.NXT - 1) after a FIN, or
5466                  * the right-most SACK block,
5467                  * then
5468                  *     RESET the connection
5469                  * else
5470                  *     Send a challenge ACK
5471                  */
5472                 if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt ||
5473                     tcp_reset_check(sk, skb)) {
5474                         rst_seq_match = true;
5475                 } else if (tcp_is_sack(tp) && tp->rx_opt.num_sacks > 0) {
5476                         struct tcp_sack_block *sp = &tp->selective_acks[0];
5477                         int max_sack = sp[0].end_seq;
5478                         int this_sack;
5479 
5480                         for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;
5481                              ++this_sack) {
5482                                 max_sack = after(sp[this_sack].end_seq,
5483                                                  max_sack) ?
5484                                         sp[this_sack].end_seq : max_sack;
5485                         }
5486 
5487                         if (TCP_SKB_CB(skb)->seq == max_sack)
5488                                 rst_seq_match = true;
5489                 }
5490 
5491                 if (rst_seq_match)
5492                         tcp_reset(sk);
5493                 else {
5494                         /* Disable TFO if RST is out-of-order
5495                          * and no data has been received
5496                          * for current active TFO socket
5497                          */
5498                         if (tp->syn_fastopen && !tp->data_segs_in &&
5499                             sk->sk_state == TCP_ESTABLISHED)
5500                                 tcp_fastopen_active_disable(sk);
5501                         tcp_send_challenge_ack(sk, skb);
5502                 }
5503                 goto discard;
5504         }
5505 
5506         /* step 3: check security and precedence [ignored] */
5507 
5508         /* step 4: Check for a SYN
5509          * RFC 5961 4.2 : Send a challenge ack
5510          */
5511         if (th->syn) {
5512 syn_challenge:
5513                 if (syn_inerr)
5514                         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
5515                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNCHALLENGE);
5516                 tcp_send_challenge_ack(sk, skb);
5517                 goto discard;
5518         }
5519 
5520         return true;
5521 
5522 discard:
5523         tcp_drop(sk, skb);
5524         return false;
5525 }
5526 
5527 /*
5528  *      TCP receive function for the ESTABLISHED state.
5529  *
5530  *      It is split into a fast path and a slow path. The fast path is
5531  *      disabled when:
5532  *      - A zero window was announced from us - zero window probing
5533  *        is only handled properly in the slow path.
5534  *      - Out of order segments arrived.
5535  *      - Urgent data is expected.
5536  *      - There is no buffer space left
5537  *      - Unexpected TCP flags/window values/header lengths are received
5538  *        (detected by checking the TCP header against pred_flags)
5539  *      - Data is sent in both directions. Fast path only supports pure senders
5540  *        or pure receivers (this means either the sequence number or the ack
5541  *        value must stay constant)
5542  *      - Unexpected TCP option.
5543  *
5544  *      When these conditions are not satisfied it drops into a standard
5545  *      receive procedure patterned after RFC793 to handle all cases.
5546  *      The first three cases are guaranteed by proper pred_flags setting,
5547  *      the rest is checked inline. Fast processing is turned on in
5548  *      tcp_data_queue when everything is OK.
5549  */
5550 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb)
5551 {
5552         const struct tcphdr *th = (const struct tcphdr *)skb->data;
5553         struct tcp_sock *tp = tcp_sk(sk);
5554         unsigned int len = skb->len;
5555 
5556         /* TCP congestion window tracking */
5557         trace_tcp_probe(sk, skb);
5558 
5559         tcp_mstamp_refresh(tp);
5560         if (unlikely(!sk->sk_rx_dst))
5561                 inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb);
5562         /*
5563          *      Header prediction.
5564          *      The code loosely follows the one in the famous
5565          *      "30 instruction TCP receive" Van Jacobson mail.
5566          *
5567          *      Van's trick is to deposit buffers into socket queue
5568          *      on a device interrupt, to call tcp_recv function
5569          *      on the receive process context and checksum and copy
5570          *      the buffer to user space. smart...
5571          *
5572          *      Our current scheme is not silly either but we take the
5573          *      extra cost of the net_bh soft interrupt processing...
5574          *      We do checksum and copy also but from device to kernel.
5575          */
5576 
5577         tp->rx_opt.saw_tstamp = 0;
5578 
5579         /*      pred_flags is 0xS?10 << 16 + snd_wnd
5580          *      if header_prediction is to be made
5581          *      'S' will always be tp->tcp_header_len >> 2
5582          *      '?' will be 0 for the fast path, otherwise pred_flags is 0 to
5583          *  turn it off (when there are holes in the receive
5584          *       space for instance)
5585          *      PSH flag is ignored.
5586          */
5587 
5588         if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags &&
5589             TCP_SKB_CB(skb)->seq == tp->rcv_nxt &&
5590             !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {
5591                 int tcp_header_len = tp->tcp_header_len;
5592 
5593                 /* Timestamp header prediction: tcp_header_len
5594                  * is automatically equal to th->doff*4 due to pred_flags
5595                  * match.
5596                  */
5597 
5598                 /* Check timestamp */
5599                 if (tcp_header_len == sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) {
5600                         /* No? Slow path! */
5601                         if (!tcp_parse_aligned_timestamp(tp, th))
5602                                 goto slow_path;
5603 
5604                         /* If PAWS failed, check it more carefully in slow path */
5605                         if ((s32)(tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent) < 0)
5606                                 goto slow_path;
5607 
5608                         /* DO NOT update ts_recent here, if checksum fails
5609                          * and timestamp was corrupted part, it will result
5610                          * in a hung connection since we will drop all
5611                          * future packets due to the PAWS test.
5612                          */
5613                 }
5614 
5615                 if (len <= tcp_header_len) {
5616                         /* Bulk data transfer: sender */
5617                         if (len == tcp_header_len) {
5618                                 /* Predicted packet is in window by definition.
5619                                  * seq == rcv_nxt and rcv_wup <= rcv_nxt.
5620                                  * Hence, check seq<=rcv_wup reduces to:
5621                                  */
5622                                 if (tcp_header_len ==
5623                                     (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) &&
5624                                     tp->rcv_nxt == tp->rcv_wup)
5625                                         tcp_store_ts_recent(tp);
5626 
5627                                 /* We know that such packets are checksummed
5628                                  * on entry.
5629                                  */
5630                                 tcp_ack(sk, skb, 0);
5631                                 __kfree_skb(skb);
5632                                 tcp_data_snd_check(sk);
5633                                 /* When receiving pure ack in fast path, update
5634                                  * last ts ecr directly instead of calling
5635                                  * tcp_rcv_rtt_measure_ts()
5636                                  */
5637                                 tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;
5638                                 return;
5639                         } else { /* Header too small */
5640                                 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
5641                                 goto discard;
5642                         }
5643                 } else {
5644                         int eaten = 0;
5645                         bool fragstolen = false;
5646 
5647                         if (tcp_checksum_complete(skb))
5648                                 goto csum_error;
5649 
5650                         if ((int)skb->truesize > sk->sk_forward_alloc)
5651                                 goto step5;
5652 
5653                         /* Predicted packet is in window by definition.
5654                          * seq == rcv_nxt and rcv_wup <= rcv_nxt.
5655                          * Hence, check seq<=rcv_wup reduces to:
5656                          */
5657                         if (tcp_header_len ==
5658                             (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED) &&
5659                             tp->rcv_nxt == tp->rcv_wup)
5660                                 tcp_store_ts_recent(tp);
5661 
5662                         tcp_rcv_rtt_measure_ts(sk, skb);
5663 
5664                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPHPHITS);
5665 
5666                         /* Bulk data transfer: receiver */
5667                         __skb_pull(skb, tcp_header_len);
5668                         eaten = tcp_queue_rcv(sk, skb, &fragstolen);
5669 
5670                         tcp_event_data_recv(sk, skb);
5671 
5672                         if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) {
5673                                 /* Well, only one small jumplet in fast path... */
5674                                 tcp_ack(sk, skb, FLAG_DATA);
5675                                 tcp_data_snd_check(sk);
5676                                 if (!inet_csk_ack_scheduled(sk))
5677                                         goto no_ack;
5678                         }
5679 
5680                         __tcp_ack_snd_check(sk, 0);
5681 no_ack:
5682                         if (eaten)
5683                                 kfree_skb_partial(skb, fragstolen);
5684                         tcp_data_ready(sk);
5685                         return;
5686                 }
5687         }
5688 
5689 slow_path:
5690         if (len < (th->doff << 2) || tcp_checksum_complete(skb))
5691                 goto csum_error;
5692 
5693         if (!th->ack && !th->rst && !th->syn)
5694                 goto discard;
5695 
5696         /*
5697          *      Standard slow path.
5698          */
5699 
5700         if (!tcp_validate_incoming(sk, skb, th, 1))
5701                 return;
5702 
5703 step5:
5704         if (tcp_ack(sk, skb, FLAG_SLOWPATH | FLAG_UPDATE_TS_RECENT) < 0)
5705                 goto discard;
5706 
5707         tcp_rcv_rtt_measure_ts(sk, skb);
5708 
5709         /* Process urgent data. */
5710         tcp_urg(sk, skb, th);
5711 
5712         /* step 7: process the segment text */
5713         tcp_data_queue(sk, skb);
5714 
5715         tcp_data_snd_check(sk);
5716         tcp_ack_snd_check(sk);
5717         return;
5718 
5719 csum_error:
5720         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
5721         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
5722 
5723 discard:
5724         tcp_drop(sk, skb);
5725 }
5726 EXPORT_SYMBOL(tcp_rcv_established);
5727 
5728 void tcp_init_transfer(struct sock *sk, int bpf_op)
5729 {
5730         struct inet_connection_sock *icsk = inet_csk(sk);
5731         struct tcp_sock *tp = tcp_sk(sk);
5732 
5733         tcp_mtup_init(sk);
5734         icsk->icsk_af_ops->rebuild_header(sk);
5735         tcp_init_metrics(sk);
5736 
5737         /* Initialize the congestion window to start the transfer.
5738          * Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
5739          * retransmitted. In light of RFC6298 more aggressive 1sec
5740          * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
5741          * retransmission has occurred.
5742          */
5743         if (tp->total_retrans > 1 && tp->undo_marker)
5744                 tp->snd_cwnd = 1;
5745         else
5746                 tp->snd_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
5747         tp->snd_cwnd_stamp = tcp_jiffies32;
5748 
5749         tcp_call_bpf(sk, bpf_op, 0, NULL);
5750         tcp_init_congestion_control(sk);
5751         tcp_init_buffer_space(sk);
5752 }
5753 
5754 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb)
5755 {
5756         struct tcp_sock *tp = tcp_sk(sk);
5757         struct inet_connection_sock *icsk = inet_csk(sk);
5758 
5759         tcp_set_state(sk, TCP_ESTABLISHED);
5760         icsk->icsk_ack.lrcvtime = tcp_jiffies32;
5761 
5762         if (skb) {
5763                 icsk->icsk_af_ops->sk_rx_dst_set(sk, skb);
5764                 security_inet_conn_established(sk, skb);
5765                 sk_mark_napi_id(sk, skb);
5766         }
5767 
5768         tcp_init_transfer(sk, BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB);
5769 
5770         /* Prevent spurious tcp_cwnd_restart() on first data
5771          * packet.
5772          */
5773         tp->lsndtime = tcp_jiffies32;
5774 
5775         if (sock_flag(sk, SOCK_KEEPOPEN))
5776                 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tp));
5777 
5778         if (!tp->rx_opt.snd_wscale)
5779                 __tcp_fast_path_on(tp, tp->snd_wnd);
5780         else
5781                 tp->pred_flags = 0;
5782 }
5783 
5784 static bool tcp_rcv_fastopen_synack(struct sock *sk, struct sk_buff *synack,
5785                                     struct tcp_fastopen_cookie *cookie)
5786 {
5787         struct tcp_sock *tp = tcp_sk(sk);
5788         struct sk_buff *data = tp->syn_data ? tcp_rtx_queue_head(sk) : NULL;
5789         u16 mss = tp->rx_opt.mss_clamp, try_exp = 0;
5790         bool syn_drop = false;
5791 
5792         if (mss == tp->rx_opt.user_mss) {
5793                 struct tcp_options_received opt;
5794 
5795                 /* Get original SYNACK MSS value if user MSS sets mss_clamp */
5796                 tcp_clear_options(&opt);
5797                 opt.user_mss = opt.mss_clamp = 0;
5798                 tcp_parse_options(sock_net(sk), synack, &opt, 0, NULL);
5799                 mss = opt.mss_clamp;
5800         }
5801 
5802         if (!tp->syn_fastopen) {
5803                 /* Ignore an unsolicited cookie */
5804                 cookie->len = -1;
5805         } else if (tp->total_retrans) {
5806                 /* SYN timed out and the SYN-ACK neither has a cookie nor
5807                  * acknowledges data. Presumably the remote received only
5808                  * the retransmitted (regular) SYNs: either the original
5809                  * SYN-data or the corresponding SYN-ACK was dropped.
5810                  */
5811                 syn_drop = (cookie->len < 0 && data);
5812         } else if (cookie->len < 0 && !tp->syn_data) {
5813                 /* We requested a cookie but didn't get it. If we did not use
5814                  * the (old) exp opt format then try so next time (try_exp=1).
5815                  * Otherwise we go back to use the RFC7413 opt (try_exp=2).
5816                  */
5817                 try_exp = tp->syn_fastopen_exp ? 2 : 1;
5818         }
5819 
5820         tcp_fastopen_cache_set(sk, mss, cookie, syn_drop, try_exp);
5821 
5822         if (data) { /* Retransmit unacked data in SYN */
5823                 skb_rbtree_walk_from(data) {
5824                         if (__tcp_retransmit_skb(sk, data, 1))
5825                                 break;
5826                 }
5827                 tcp_rearm_rto(sk);
5828                 NET_INC_STATS(sock_net(sk),
5829                                 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
5830                 return true;
5831         }
5832         tp->syn_data_acked = tp->syn_data;
5833         if (tp->syn_data_acked) {
5834                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFASTOPENACTIVE);
5835                 /* SYN-data is counted as two separate packets in tcp_ack() */
5836                 if (tp->delivered > 1)
5837                         --tp->delivered;
5838         }
5839 
5840         tcp_fastopen_add_skb(sk, synack);
5841 
5842         return false;
5843 }
5844 
5845 static void smc_check_reset_syn(struct tcp_sock *tp)
5846 {
5847 #if IS_ENABLED(CONFIG_SMC)
5848         if (static_branch_unlikely(&tcp_have_smc)) {
5849                 if (tp->syn_smc && !tp->rx_opt.smc_ok)
5850                         tp->syn_smc = 0;
5851         }
5852 #endif
5853 }
5854 
5855 static void tcp_try_undo_spurious_syn(struct sock *sk)
5856 {
5857         struct tcp_sock *tp = tcp_sk(sk);
5858         u32 syn_stamp;
5859 
5860         /* undo_marker is set when SYN or SYNACK times out. The timeout is
5861          * spurious if the ACK's timestamp option echo value matches the
5862          * original SYN timestamp.
5863          */
5864         syn_stamp = tp->retrans_stamp;
5865         if (tp->undo_marker && syn_stamp && tp->rx_opt.saw_tstamp &&
5866             syn_stamp == tp->rx_opt.rcv_tsecr)
5867                 tp->undo_marker = 0;
5868 }
5869 
5870 static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
5871                                          const struct tcphdr *th)
5872 {
5873         struct inet_connection_sock *icsk = inet_csk(sk);
5874         struct tcp_sock *tp = tcp_sk(sk);
5875         struct tcp_fastopen_cookie foc = { .len = -1 };
5876         int saved_clamp = tp->rx_opt.mss_clamp;
5877         bool fastopen_fail;
5878 
5879         tcp_parse_options(sock_net(sk), skb, &tp->rx_opt, 0, &foc);
5880         if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
5881                 tp->rx_opt.rcv_tsecr -= tp->tsoffset;
5882 
5883         if (th->ack) {
5884                 /* rfc793:
5885                  * "If the state is SYN-SENT then
5886                  *    first check the ACK bit
5887                  *      If the ACK bit is set
5888                  *        If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send
5889                  *        a reset (unless the RST bit is set, if so drop
5890                  *        the segment and return)"
5891                  */
5892                 if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) ||
5893                     after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt))
5894                         goto reset_and_undo;
5895 
5896                 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&
5897                     !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp,
5898                              tcp_time_stamp(tp))) {
5899                         NET_INC_STATS(sock_net(sk),
5900                                         LINUX_MIB_PAWSACTIVEREJECTED);
5901                         goto reset_and_undo;
5902                 }
5903 
5904                 /* Now ACK is acceptable.
5905                  *
5906                  * "If the RST bit is set
5907                  *    If the ACK was acceptable then signal the user "error:
5908                  *    connection reset", drop the segment, enter CLOSED state,
5909                  *    delete TCB, and return."
5910                  */
5911 
5912                 if (th->rst) {
5913                         tcp_reset(sk);
5914                         goto discard;
5915                 }
5916 
5917                 /* rfc793:
5918                  *   "fifth, if neither of the SYN or RST bits is set then
5919                  *    drop the segment and return."
5920                  *
5921                  *    See note below!
5922                  *                                        --ANK(990513)
5923                  */
5924                 if (!th->syn)
5925                         goto discard_and_undo;
5926 
5927                 /* rfc793:
5928                  *   "If the SYN bit is on ...
5929                  *    are acceptable then ...
5930                  *    (our SYN has been ACKed), change the connection
5931                  *    state to ESTABLISHED..."
5932                  */
5933 
5934                 tcp_ecn_rcv_synack(tp, th);
5935 
5936                 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);
5937                 tcp_try_undo_spurious_syn(sk);
5938                 tcp_ack(sk, skb, FLAG_SLOWPATH);
5939 
5940                 /* Ok.. it's good. Set up sequence numbers and
5941                  * move to established.
5942                  */
5943                 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);
5944                 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;
5945 
5946                 /* RFC1323: The window in SYN & SYN/ACK segments is
5947                  * never scaled.
5948                  */
5949                 tp->snd_wnd = ntohs(th->window);
5950 
5951                 if (!tp->rx_opt.wscale_ok) {
5952                         tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0;
5953                         tp->window_clamp = min(tp->window_clamp, 65535U);
5954                 }
5955 
5956                 if (tp->rx_opt.saw_tstamp) {
5957                         tp->rx_opt.tstamp_ok       = 1;
5958                         tp->tcp_header_len =
5959                                 sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
5960                         tp->advmss          -= TCPOLEN_TSTAMP_ALIGNED;
5961                         tcp_store_ts_recent(tp);
5962                 } else {
5963                         tp->tcp_header_len = sizeof(struct tcphdr);
5964                 }
5965 
5966                 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
5967                 tcp_initialize_rcv_mss(sk);
5968 
5969                 /* Remember, tcp_poll() does not lock socket!
5970                  * Change state from SYN-SENT only after copied_seq
5971                  * is initialized. */
5972                 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
5973 
5974                 smc_check_reset_syn(tp);
5975 
5976                 smp_mb();
5977 
5978                 tcp_finish_connect(sk, skb);
5979 
5980                 fastopen_fail = (tp->syn_fastopen || tp->syn_data) &&
5981                                 tcp_rcv_fastopen_synack(sk, skb, &foc);
5982 
5983                 if (!sock_flag(sk, SOCK_DEAD)) {
5984                         sk->sk_state_change(sk);
5985                         sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
5986                 }
5987                 if (fastopen_fail)
5988                         return -1;
5989                 if (sk->sk_write_pending ||
5990                     icsk->icsk_accept_queue.rskq_defer_accept ||
5991                     inet_csk_in_pingpong_mode(sk)) {
5992                         /* Save one ACK. Data will be ready after
5993                          * several ticks, if write_pending is set.
5994                          *
5995                          * It may be deleted, but with this feature tcpdumps
5996                          * look so _wonderfully_ clever, that I was not able
5997                          * to stand against the temptation 8)     --ANK
5998                          */
5999                         inet_csk_schedule_ack(sk);
6000                         tcp_enter_quickack_mode(sk, TCP_MAX_QUICKACKS);
6001                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
6002                                                   TCP_DELACK_MAX, TCP_RTO_MAX);
6003 
6004 discard:
6005                         tcp_drop(sk, skb);
6006                         return 0;
6007                 } else {
6008                         tcp_send_ack(sk);
6009                 }
6010                 return -1;
6011         }
6012 
6013         /* No ACK in the segment */
6014 
6015         if (th->rst) {
6016                 /* rfc793:
6017                  * "If the RST bit is set
6018                  *
6019                  *      Otherwise (no ACK) drop the segment and return."
6020                  */
6021 
6022                 goto discard_and_undo;
6023         }
6024 
6025         /* PAWS check. */
6026         if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp &&
6027             tcp_paws_reject(&tp->rx_opt, 0))
6028                 goto discard_and_undo;
6029 
6030         if (th->syn) {
6031                 /* We see SYN without ACK. It is attempt of
6032                  * simultaneous connect with crossed SYNs.
6033                  * Particularly, it can be connect to self.
6034                  */
6035                 tcp_set_state(sk, TCP_SYN_RECV);
6036 
6037                 if (tp->rx_opt.saw_tstamp) {
6038                         tp->rx_opt.tstamp_ok = 1;
6039                         tcp_store_ts_recent(tp);
6040                         tp->tcp_header_len =
6041                                 sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
6042                 } else {
6043                         tp->tcp_header_len = sizeof(struct tcphdr);
6044                 }
6045 
6046                 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);
6047                 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
6048                 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;
6049 
6050                 /* RFC1323: The window in SYN & SYN/ACK segments is
6051                  * never scaled.
6052                  */
6053                 tp->snd_wnd    = ntohs(th->window);
6054                 tp->snd_wl1    = TCP_SKB_CB(skb)->seq;
6055                 tp->max_window = tp->snd_wnd;
6056 
6057                 tcp_ecn_rcv_syn(tp, th);
6058 
6059                 tcp_mtup_init(sk);
6060                 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
6061                 tcp_initialize_rcv_mss(sk);
6062 
6063                 tcp_send_synack(sk);
6064 #if 0
6065                 /* Note, we could accept data and URG from this segment.
6066                  * There are no obstacles to make this (except that we must
6067                  * either change tcp_recvmsg() to prevent it from returning data
6068                  * before 3WHS completes per RFC793, or employ TCP Fast Open).
6069                  *
6070                  * However, if we ignore data in ACKless segments sometimes,
6071                  * we have no reasons to accept it sometimes.
6072                  * Also, seems the code doing it in step6 of tcp_rcv_state_process
6073                  * is not flawless. So, discard packet for sanity.
6074                  * Uncomment this return to process the data.
6075                  */
6076                 return -1;
6077 #else
6078                 goto discard;
6079 #endif
6080         }
6081         /* "fifth, if neither of the SYN or RST bits is set then
6082          * drop the segment and return."
6083          */
6084 
6085 discard_and_undo:
6086         tcp_clear_options(&tp->rx_opt);
6087         tp->rx_opt.mss_clamp = saved_clamp;
6088         goto discard;
6089 
6090 reset_and_undo:
6091         tcp_clear_options(&tp->rx_opt);
6092         tp->rx_opt.mss_clamp = saved_clamp;
6093         return 1;
6094 }
6095 
6096 static void tcp_rcv_synrecv_state_fastopen(struct sock *sk)
6097 {
6098         struct request_sock *req;
6099 
6100         /* If we are still handling the SYNACK RTO, see if timestamp ECR allows
6101          * undo. If peer SACKs triggered fast recovery, we can't undo here.
6102          */
6103         if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss)
6104                 tcp_try_undo_loss(sk, false);
6105 
6106         /* Reset rtx states to prevent spurious retransmits_timed_out() */
6107         tcp_sk(sk)->retrans_stamp = 0;
6108         inet_csk(sk)->icsk_retransmits = 0;
6109 
6110         /* Once we leave TCP_SYN_RECV or TCP_FIN_WAIT_1,
6111          * we no longer need req so release it.
6112          */
6113         req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk,
6114                                         lockdep_sock_is_held(sk));
6115         reqsk_fastopen_remove(sk, req, false);
6116 
6117         /* Re-arm the timer because data may have been sent out.
6118          * This is similar to the regular data transmission case
6119          * when new data has just been ack'ed.
6120          *
6121          * (TFO) - we could try to be more aggressive and
6122          * retransmitting any data sooner based on when they
6123          * are sent out.
6124          */
6125         tcp_rearm_rto(sk);
6126 }
6127 
6128 /*
6129  *      This function implements the receiving procedure of RFC 793 for
6130  *      all states except ESTABLISHED and TIME_WAIT.
6131  *      It's called from both tcp_v4_rcv and tcp_v6_rcv and should be
6132  *      address independent.
6133  */
6134 
6135 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb)
6136 {
6137         struct tcp_sock *tp = tcp_sk(sk);
6138         struct inet_connection_sock *icsk = inet_csk(sk);
6139         const struct tcphdr *th = tcp_hdr(skb);
6140         struct request_sock *req;
6141         int queued = 0;
6142         bool acceptable;
6143 
6144         switch (sk->sk_state) {
6145         case TCP_CLOSE:
6146                 goto discard;
6147 
6148         case TCP_LISTEN:
6149                 if (th->ack)
6150                         return 1;
6151 
6152                 if (th->rst)
6153                         goto discard;
6154 
6155                 if (th->syn) {
6156                         if (th->fin)
6157                                 goto discard;
6158                         /* It is possible that we process SYN packets from backlog,
6159                          * so we need to make sure to disable BH and RCU right there.
6160                          */
6161                         rcu_read_lock();
6162                         local_bh_disable();
6163                         acceptable = icsk->icsk_af_ops->conn_request(sk, skb) >= 0;
6164                         local_bh_enable();
6165                         rcu_read_unlock();
6166 
6167                         if (!acceptable)
6168                                 return 1;
6169                         consume_skb(skb);
6170                         return 0;
6171                 }
6172                 goto discard;
6173 
6174         case TCP_SYN_SENT:
6175                 tp->rx_opt.saw_tstamp = 0;
6176                 tcp_mstamp_refresh(tp);
6177                 queued = tcp_rcv_synsent_state_process(sk, skb, th);
6178                 if (queued >= 0)
6179                         return queued;
6180 
6181                 /* Do step6 onward by hand. */
6182                 tcp_urg(sk, skb, th);
6183                 __kfree_skb(skb);
6184                 tcp_data_snd_check(sk);
6185                 return 0;
6186         }
6187 
6188         tcp_mstamp_refresh(tp);
6189         tp->rx_opt.saw_tstamp = 0;
6190         req = rcu_dereference_protected(tp->fastopen_rsk,
6191                                         lockdep_sock_is_held(sk));
6192         if (req) {
6193                 bool req_stolen;
6194 
6195                 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
6196                     sk->sk_state != TCP_FIN_WAIT1);
6197 
6198                 if (!tcp_check_req(sk, skb, req, true, &req_stolen))
6199                         goto discard;
6200         }
6201 
6202         if (!th->ack && !th->rst && !th->syn)
6203                 goto discard;
6204 
6205         if (!tcp_validate_incoming(sk, skb, th, 0))
6206                 return 0;
6207 
6208         /* step 5: check the ACK field */
6209         acceptable = tcp_ack(sk, skb, FLAG_SLOWPATH |
6210                                       FLAG_UPDATE_TS_RECENT |
6211                                       FLAG_NO_CHALLENGE_ACK) > 0;
6212 
6213         if (!acceptable) {
6214                 if (sk->sk_state == TCP_SYN_RECV)
6215                         return 1;       /* send one RST */
6216                 tcp_send_challenge_ack(sk, skb);
6217                 goto discard;
6218         }
6219         switch (sk->sk_state) {
6220         case TCP_SYN_RECV:
6221                 tp->delivered++; /* SYN-ACK delivery isn't tracked in tcp_ack */
6222                 if (!tp->srtt_us)
6223                         tcp_synack_rtt_meas(sk, req);
6224 
6225                 if (req) {
6226                         tcp_rcv_synrecv_state_fastopen(sk);
6227                 } else {
6228                         tcp_try_undo_spurious_syn(sk);
6229                         tp->retrans_stamp = 0;
6230                         tcp_init_transfer(sk, BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB);
6231                         WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
6232                 }
6233                 smp_mb();
6234                 tcp_set_state(sk, TCP_ESTABLISHED);
6235                 sk->sk_state_change(sk);
6236 
6237                 /* Note, that this wakeup is only for marginal crossed SYN case.
6238                  * Passively open sockets are not waked up, because
6239                  * sk->sk_sleep == NULL and sk->sk_socket == NULL.
6240                  */
6241                 if (sk->sk_socket)
6242                         sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
6243 
6244                 tp->snd_una = TCP_SKB_CB(skb)->ack_seq;
6245                 tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale;
6246                 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);
6247 
6248                 if (tp->rx_opt.tstamp_ok)
6249                         tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
6250 
6251                 if (!inet_csk(sk)->icsk_ca_ops->cong_control)
6252                         tcp_update_pacing_rate(sk);
6253 
6254                 /* Prevent spurious tcp_cwnd_restart() on first data packet */
6255                 tp->lsndtime = tcp_jiffies32;
6256 
6257                 tcp_initialize_rcv_mss(sk);
6258                 tcp_fast_path_on(tp);
6259                 break;
6260 
6261         case TCP_FIN_WAIT1: {
6262                 int tmo;
6263 
6264                 if (req)
6265                         tcp_rcv_synrecv_state_fastopen(sk);
6266 
6267                 if (tp->snd_una != tp->write_seq)
6268                         break;
6269 
6270                 tcp_set_state(sk, TCP_FIN_WAIT2);
6271                 sk->sk_shutdown |= SEND_SHUTDOWN;
6272 
6273                 sk_dst_confirm(sk);
6274 
6275                 if (!sock_flag(sk, SOCK_DEAD)) {
6276                         /* Wake up lingering close() */
6277                         sk->sk_state_change(sk);
6278                         break;
6279                 }
6280 
6281                 if (tp->linger2 < 0) {
6282                         tcp_done(sk);
6283                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
6284                         return 1;
6285                 }
6286                 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
6287                     after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {
6288                         /* Receive out of order FIN after close() */
6289                         if (tp->syn_fastopen && th->fin)
6290                                 tcp_fastopen_active_disable(sk);
6291                         tcp_done(sk);
6292                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
6293                         return 1;
6294                 }
6295 
6296                 tmo = tcp_fin_time(sk);
6297                 if (tmo > TCP_TIMEWAIT_LEN) {
6298                         inet_csk_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN);
6299                 } else if (th->fin || sock_owned_by_user(sk)) {
6300                         /* Bad case. We could lose such FIN otherwise.
6301                          * It is not a big problem, but it looks confusing
6302                          * and not so rare event. We still can lose it now,
6303                          * if it spins in bh_lock_sock(), but it is really
6304                          * marginal case.
6305                          */
6306                         inet_csk_reset_keepalive_timer(sk, tmo);
6307                 } else {
6308                         tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
6309                         goto discard;
6310                 }
6311                 break;
6312         }
6313 
6314         case TCP_CLOSING:
6315                 if (tp->snd_una == tp->write_seq) {
6316                         tcp_time_wait(sk, TCP_TIME_WAIT, 0);
6317                         goto discard;
6318                 }
6319                 break;
6320 
6321         case TCP_LAST_ACK:
6322                 if (tp->snd_una == tp->write_seq) {
6323                         tcp_update_metrics(sk);
6324                         tcp_done(sk);
6325                         goto discard;
6326                 }
6327                 break;
6328         }
6329 
6330         /* step 6: check the URG bit */
6331         tcp_urg(sk, skb, th);
6332 
6333         /* step 7: process the segment text */
6334         switch (sk->sk_state) {
6335         case TCP_CLOSE_WAIT:
6336         case TCP_CLOSING:
6337         case TCP_LAST_ACK:
6338                 if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt))
6339                         break;
6340                 /* fall through */
6341         case TCP_FIN_WAIT1:
6342         case TCP_FIN_WAIT2:
6343                 /* RFC 793 says to queue data in these states,
6344                  * RFC 1122 says we MUST send a reset.
6345                  * BSD 4.4 also does reset.
6346                  */
6347                 if (sk->sk_shutdown & RCV_SHUTDOWN) {
6348                         if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
6349                             after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {
6350                                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
6351                                 tcp_reset(sk);
6352                                 return 1;
6353                         }
6354                 }
6355                 /* Fall through */
6356         case TCP_ESTABLISHED:
6357                 tcp_data_queue(sk, skb);
6358                 queued = 1;
6359                 break;
6360         }
6361 
6362         /* tcp_data could move socket to TIME-WAIT */
6363         if (sk->sk_state != TCP_CLOSE) {
6364                 tcp_data_snd_check(sk);
6365                 tcp_ack_snd_check(sk);
6366         }
6367 
6368         if (!queued) {
6369 discard:
6370                 tcp_drop(sk, skb);
6371         }
6372         return 0;
6373 }
6374 EXPORT_SYMBOL(tcp_rcv_state_process);
6375 
6376 static inline void pr_drop_req(struct request_sock *req, __u16 port, int family)
6377 {
6378         struct inet_request_sock *ireq = inet_rsk(req);
6379 
6380         if (family == AF_INET)
6381                 net_dbg_ratelimited("drop open request from %pI4/%u\n",
6382                                     &ireq->ir_rmt_addr, port);
6383 #if IS_ENABLED(CONFIG_IPV6)
6384         else if (family == AF_INET6)
6385                 net_dbg_ratelimited("drop open request from %pI6/%u\n",
6386                                     &ireq->ir_v6_rmt_addr, port);
6387 #endif
6388 }
6389 
6390 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
6391  *
6392  * If we receive a SYN packet with these bits set, it means a
6393  * network is playing bad games with TOS bits. In order to
6394  * avoid possible false congestion notifications, we disable
6395  * TCP ECN negotiation.
6396  *
6397  * Exception: tcp_ca wants ECN. This is required for DCTCP
6398  * congestion control: Linux DCTCP asserts ECT on all packets,
6399  * including SYN, which is most optimal solution; however,
6400  * others, such as FreeBSD do not.
6401  *
6402  * Exception: At least one of the reserved bits of the TCP header (th->res1) is
6403  * set, indicating the use of a future TCP extension (such as AccECN). See
6404  * RFC8311 §4.3 which updates RFC3168 to allow the development of such
6405  * extensions.
6406  */
6407 static void tcp_ecn_create_request(struct request_sock *req,
6408                                    const struct sk_buff *skb,
6409                                    const struct sock *listen_sk,
6410                                    const struct dst_entry *dst)
6411 {
6412         const struct tcphdr *th = tcp_hdr(skb);
6413         const struct net *net = sock_net(listen_sk);
6414         bool th_ecn = th->ece && th->cwr;
6415         bool ect, ecn_ok;
6416         u32 ecn_ok_dst;
6417 
6418         if (!th_ecn)
6419                 return;
6420 
6421         ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield);
6422         ecn_ok_dst = dst_feature(dst, DST_FEATURE_ECN_MASK);
6423         ecn_ok = net->ipv4.sysctl_tcp_ecn || ecn_ok_dst;
6424 
6425         if (((!ect || th->res1) && ecn_ok) || tcp_ca_needs_ecn(listen_sk) ||
6426             (ecn_ok_dst & DST_FEATURE_ECN_CA) ||
6427             tcp_bpf_ca_needs_ecn((struct sock *)req))
6428                 inet_rsk(req)->ecn_ok = 1;
6429 }
6430 
6431 static void tcp_openreq_init(struct request_sock *req,
6432                              const struct tcp_options_received *rx_opt,
6433                              struct sk_buff *skb, const struct sock *sk)
6434 {
6435         struct inet_request_sock *ireq = inet_rsk(req);
6436 
6437         req->rsk_rcv_wnd = 0;           /* So that tcp_send_synack() knows! */
6438         req->cookie_ts = 0;
6439         tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
6440         tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
6441         tcp_rsk(req)->snt_synack = 0;
6442         tcp_rsk(req)->last_oow_ack_time = 0;
6443         req->mss = rx_opt->mss_clamp;
6444         req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
6445         ireq->tstamp_ok = rx_opt->tstamp_ok;
6446         ireq->sack_ok = rx_opt->sack_ok;
6447         ireq->snd_wscale = rx_opt->snd_wscale;
6448         ireq->wscale_ok = rx_opt->wscale_ok;
6449         ireq->acked = 0;
6450         ireq->ecn_ok = 0;
6451         ireq->ir_rmt_port = tcp_hdr(skb)->source;
6452         ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
6453         ireq->ir_mark = inet_request_mark(sk, skb);
6454 #if IS_ENABLED(CONFIG_SMC)
6455         ireq->smc_ok = rx_opt->smc_ok;
6456 #endif
6457 }
6458 
6459 struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
6460                                       struct sock *sk_listener,
6461                                       bool attach_listener)
6462 {
6463         struct request_sock *req = reqsk_alloc(ops, sk_listener,
6464                                                attach_listener);
6465 
6466         if (req) {
6467                 struct inet_request_sock *ireq = inet_rsk(req);
6468 
6469                 ireq->ireq_opt = NULL;
6470 #if IS_ENABLED(CONFIG_IPV6)
6471                 ireq->pktopts = NULL;
6472 #endif
6473                 atomic64_set(&ireq->ir_cookie, 0);
6474                 ireq->ireq_state = TCP_NEW_SYN_RECV;
6475                 write_pnet(&ireq->ireq_net, sock_net(sk_listener));
6476                 ireq->ireq_family = sk_listener->sk_family;
6477         }
6478 
6479         return req;
6480 }
6481 EXPORT_SYMBOL(inet_reqsk_alloc);
6482 
6483 /*
6484  * Return true if a syncookie should be sent
6485  */
6486 static bool tcp_syn_flood_action(const struct sock *sk, const char *proto)
6487 {
6488         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
6489         const char *msg = "Dropping request";
6490         bool want_cookie = false;
6491         struct net *net = sock_net(sk);
6492 
6493 #ifdef CONFIG_SYN_COOKIES
6494         if (net->ipv4.sysctl_tcp_syncookies) {
6495                 msg = "Sending cookies";
6496                 want_cookie = true;
6497                 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
6498         } else
6499 #endif
6500                 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
6501 
6502         if (!queue->synflood_warned &&
6503             net->ipv4.sysctl_tcp_syncookies != 2 &&
6504             xchg(&queue->synflood_warned, 1) == 0)
6505                 net_info_ratelimited("%s: Possible SYN flooding on port %d. %s.  Check SNMP counters.\n",
6506                                      proto, sk->sk_num, msg);
6507 
6508         return want_cookie;
6509 }
6510 
6511 static void tcp_reqsk_record_syn(const struct sock *sk,
6512                                  struct request_sock *req,
6513                                  const struct sk_buff *skb)
6514 {
6515         if (tcp_sk(sk)->save_syn) {
6516                 u32 len = skb_network_header_len(skb) + tcp_hdrlen(skb);
6517                 u32 *copy;
6518 
6519                 copy = kmalloc(len + sizeof(u32), GFP_ATOMIC);
6520                 if (copy) {
6521                         copy[0] = len;
6522                         memcpy(&copy[1], skb_network_header(skb), len);
6523                         req->saved_syn = copy;
6524                 }
6525         }
6526 }
6527 
6528 /* If a SYN cookie is required and supported, returns a clamped MSS value to be
6529  * used for SYN cookie generation.
6530  */
6531 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
6532                           const struct tcp_request_sock_ops *af_ops,
6533                           struct sock *sk, struct tcphdr *th)
6534 {
6535         struct tcp_sock *tp = tcp_sk(sk);
6536         u16 mss;
6537 
6538         if (sock_net(sk)->ipv4.sysctl_tcp_syncookies != 2 &&
6539             !inet_csk_reqsk_queue_is_full(sk))
6540                 return 0;
6541 
6542         if (!tcp_syn_flood_action(sk, rsk_ops->slab_name))
6543                 return 0;
6544 
6545         if (sk_acceptq_is_full(sk)) {
6546                 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
6547                 return 0;
6548         }
6549 
6550         mss = tcp_parse_mss_option(th, tp->rx_opt.user_mss);
6551         if (!mss)
6552                 mss = af_ops->mss_clamp;
6553 
6554         return mss;
6555 }
6556 EXPORT_SYMBOL_GPL(tcp_get_syncookie_mss);
6557 
6558 int tcp_conn_request(struct request_sock_ops *rsk_ops,
6559                      const struct tcp_request_sock_ops *af_ops,
6560                      struct sock *sk, struct sk_buff *skb)
6561 {
6562         struct tcp_fastopen_cookie foc = { .len = -1 };
6563         __u32 isn = TCP_SKB_CB(skb)->tcp_tw_isn;
6564         struct tcp_options_received tmp_opt;
6565         struct tcp_sock *tp = tcp_sk(sk);
6566         struct net *net = sock_net(sk);
6567         struct sock *fastopen_sk = NULL;
6568         struct request_sock *req;
6569         bool want_cookie = false;
6570         struct dst_entry *dst;
6571         struct flowi fl;
6572 
6573         /* TW buckets are converted to open requests without
6574          * limitations, they conserve resources and peer is
6575          * evidently real one.
6576          */
6577         if ((net->ipv4.sysctl_tcp_syncookies == 2 ||
6578              inet_csk_reqsk_queue_is_full(sk)) && !isn) {
6579                 want_cookie = tcp_syn_flood_action(sk, rsk_ops->slab_name);
6580                 if (!want_cookie)
6581                         goto drop;
6582         }
6583 
6584         if (sk_acceptq_is_full(sk)) {
6585                 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
6586                 goto drop;
6587         }
6588 
6589         req = inet_reqsk_alloc(rsk_ops, sk, !want_cookie);
6590         if (!req)
6591                 goto drop;
6592 
6593         tcp_rsk(req)->af_specific = af_ops;
6594         tcp_rsk(req)->ts_off = 0;
6595 
6596         tcp_clear_options(&tmp_opt);
6597         tmp_opt.mss_clamp = af_ops->mss_clamp;
6598         tmp_opt.user_mss  = tp->rx_opt.user_mss;
6599         tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0,
6600                           want_cookie ? NULL : &foc);
6601 
6602         if (want_cookie && !tmp_opt.saw_tstamp)
6603                 tcp_clear_options(&tmp_opt);
6604 
6605         if (IS_ENABLED(CONFIG_SMC) && want_cookie)
6606                 tmp_opt.smc_ok = 0;
6607 
6608         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
6609         tcp_openreq_init(req, &tmp_opt, skb, sk);
6610         inet_rsk(req)->no_srccheck = inet_sk(sk)->transparent;
6611 
6612         /* Note: tcp_v6_init_req() might override ir_iif for link locals */
6613         inet_rsk(req)->ir_iif = inet_request_bound_dev_if(sk, skb);
6614 
6615         af_ops->init_req(req, sk, skb);
6616 
6617         if (security_inet_conn_request(sk, skb, req))
6618                 goto drop_and_free;
6619 
6620         if (tmp_opt.tstamp_ok)
6621                 tcp_rsk(req)->ts_off = af_ops->init_ts_off(net, skb);
6622 
6623         dst = af_ops->route_req(sk, &fl, req);
6624         if (!dst)
6625                 goto drop_and_free;
6626 
6627         if (!want_cookie && !isn) {
6628                 /* Kill the following clause, if you dislike this way. */
6629                 if (!net->ipv4.sysctl_tcp_syncookies &&
6630                     (net->ipv4.sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
6631                      (net->ipv4.sysctl_max_syn_backlog >> 2)) &&
6632                     !tcp_peer_is_proven(req, dst)) {
6633                         /* Without syncookies last quarter of
6634                          * backlog is filled with destinations,
6635                          * proven to be alive.
6636                          * It means that we continue to communicate
6637                          * to destinations, already remembered
6638                          * to the moment of synflood.
6639                          */
6640                         pr_drop_req(req, ntohs(tcp_hdr(skb)->source),
6641                                     rsk_ops->family);
6642                         goto drop_and_release;
6643                 }
6644 
6645                 isn = af_ops->init_seq(skb);
6646         }
6647 
6648         tcp_ecn_create_request(req, skb, sk, dst);
6649 
6650         if (want_cookie) {
6651                 isn = cookie_init_sequence(af_ops, sk, skb, &req->mss);
6652                 req->cookie_ts = tmp_opt.tstamp_ok;
6653                 if (!tmp_opt.tstamp_ok)
6654                         inet_rsk(req)->ecn_ok = 0;
6655         }
6656 
6657         tcp_rsk(req)->snt_isn = isn;
6658         tcp_rsk(req)->txhash = net_tx_rndhash();
6659         tcp_openreq_init_rwin(req, sk, dst);
6660         sk_rx_queue_set(req_to_sk(req), skb);
6661         if (!want_cookie) {
6662                 tcp_reqsk_record_syn(sk, req, skb);
6663                 fastopen_sk = tcp_try_fastopen(sk, skb, req, &foc, dst);
6664         }
6665         if (fastopen_sk) {
6666                 af_ops->send_synack(fastopen_sk, dst, &fl, req,
6667                                     &foc, TCP_SYNACK_FASTOPEN);
6668                 /* Add the child socket directly into the accept queue */
6669                 if (!inet_csk_reqsk_queue_add(sk, req, fastopen_sk)) {
6670                         reqsk_fastopen_remove(fastopen_sk, req, false);
6671                         bh_unlock_sock(fastopen_sk);
6672                         sock_put(fastopen_sk);
6673                         goto drop_and_free;
6674                 }
6675                 sk->sk_data_ready(sk);
6676                 bh_unlock_sock(fastopen_sk);
6677                 sock_put(fastopen_sk);
6678         } else {
6679                 tcp_rsk(req)->tfo_listener = false;
6680                 if (!want_cookie)
6681                         inet_csk_reqsk_queue_hash_add(sk, req,
6682                                 tcp_timeout_init((struct sock *)req));
6683                 af_ops->send_synack(sk, dst, &fl, req, &foc,
6684                                     !want_cookie ? TCP_SYNACK_NORMAL :
6685                                                    TCP_SYNACK_COOKIE);
6686                 if (want_cookie) {
6687                         reqsk_free(req);
6688                         return 0;
6689                 }
6690         }
6691         reqsk_put(req);
6692         return 0;
6693 
6694 drop_and_release:
6695         dst_release(dst);
6696 drop_and_free:
6697         __reqsk_free(req);
6698 drop:
6699         tcp_listendrop(sk);
6700         return 0;
6701 }
6702 EXPORT_SYMBOL(tcp_conn_request);