root/net/ipv4/tcp.c

DEFINITIONS

1. tcp_enter_memory_pressure
2. tcp_leave_memory_pressure
3. secs_to_retrans
4. retrans_to_secs
5. tcp_compute_delivery_rate
6. tcp_init_sock
7. tcp_tx_timestamp
8. tcp_stream_is_readable
9. tcp_poll
10. tcp_ioctl
11. tcp_mark_push
12. forced_push
13. skb_entail
14. tcp_mark_urg
15. tcp_should_autocork
16. tcp_push
17. tcp_splice_data_recv
18. __tcp_splice_read
19. tcp_splice_read
20. sk_stream_alloc_skb
21. tcp_xmit_size_goal
22. tcp_send_mss
23. tcp_remove_empty_skb
24. do_tcp_sendpages
25. tcp_sendpage_locked
26. tcp_sendpage
27. tcp_free_fastopen_req
28. tcp_sendmsg_fastopen
29. tcp_sendmsg_locked
30. tcp_sendmsg
31. tcp_recv_urg
32. tcp_peek_sndq
33. tcp_cleanup_rbuf
34. tcp_recv_skb
35. tcp_read_sock
36. tcp_peek_len
37. tcp_set_rcvlowat
38. tcp_mmap
39. tcp_zerocopy_receive
40. tcp_update_recv_tstamps
41. tcp_recv_timestamp
42. tcp_inq_hint
43. tcp_recvmsg
44. tcp_set_state
45. tcp_close_state
46. tcp_shutdown
47. tcp_check_oom
48. tcp_close
49. tcp_need_reset
50. tcp_rtx_queue_purge
51. tcp_write_queue_purge
52. tcp_disconnect
53. tcp_can_repair_sock
54. tcp_repair_set_window
55. tcp_repair_options_est
56. tcp_enable_tx_delay
57. do_tcp_setsockopt
58. tcp_setsockopt
59. compat_tcp_setsockopt
60. tcp_get_info_chrono_stats
61. tcp_get_info
62. tcp_opt_stats_get_size
63. tcp_get_timestamping_opt_stats
64. do_tcp_getsockopt
65. tcp_getsockopt
66. compat_tcp_getsockopt
67. __tcp_alloc_md5sig_pool
68. tcp_alloc_md5sig_pool
69. tcp_get_md5sig_pool
70. tcp_md5_hash_skb_data
71. tcp_md5_hash_key
72. tcp_done
73. tcp_abort
74. set_thash_entries
75. tcp_init_mem
76. tcp_init

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              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  * Fixes:
  22  *              Alan Cox        :       Numerous verify_area() calls
  23  *              Alan Cox        :       Set the ACK bit on a reset
  24  *              Alan Cox        :       Stopped it crashing if it closed while
  25  *                                      sk->inuse=1 and was trying to connect
  26  *                                      (tcp_err()).
  27  *              Alan Cox        :       All icmp error handling was broken
  28  *                                      pointers passed where wrong and the
  29  *                                      socket was looked up backwards. Nobody
  30  *                                      tested any icmp error code obviously.
  31  *              Alan Cox        :       tcp_err() now handled properly. It
  32  *                                      wakes people on errors. poll
  33  *                                      behaves and the icmp error race
  34  *                                      has gone by moving it into sock.c
  35  *              Alan Cox        :       tcp_send_reset() fixed to work for
  36  *                                      everything not just packets for
  37  *                                      unknown sockets.
  38  *              Alan Cox        :       tcp option processing.
  39  *              Alan Cox        :       Reset tweaked (still not 100%) [Had
  40  *                                      syn rule wrong]
  41  *              Herp Rosmanith  :       More reset fixes
  42  *              Alan Cox        :       No longer acks invalid rst frames.
  43  *                                      Acking any kind of RST is right out.
  44  *              Alan Cox        :       Sets an ignore me flag on an rst
  45  *                                      receive otherwise odd bits of prattle
  46  *                                      escape still
  47  *              Alan Cox        :       Fixed another acking RST frame bug.
  48  *                                      Should stop LAN workplace lockups.
  49  *              Alan Cox        :       Some tidyups using the new skb list
  50  *                                      facilities
  51  *              Alan Cox        :       sk->keepopen now seems to work
  52  *              Alan Cox        :       Pulls options out correctly on accepts
  53  *              Alan Cox        :       Fixed assorted sk->rqueue->next errors
  54  *              Alan Cox        :       PSH doesn't end a TCP read. Switched a
  55  *                                      bit to skb ops.
  56  *              Alan Cox        :       Tidied tcp_data to avoid a potential
  57  *                                      nasty.
  58  *              Alan Cox        :       Added some better commenting, as the
  59  *                                      tcp is hard to follow
  60  *              Alan Cox        :       Removed incorrect check for 20 * psh
  61  *      Michael O'Reilly        :       ack < copied bug fix.
  62  *      Johannes Stille         :       Misc tcp fixes (not all in yet).
  63  *              Alan Cox        :       FIN with no memory -> CRASH
  64  *              Alan Cox        :       Added socket option proto entries.
  65  *                                      Also added awareness of them to accept.
  66  *              Alan Cox        :       Added TCP options (SOL_TCP)
  67  *              Alan Cox        :       Switched wakeup calls to callbacks,
  68  *                                      so the kernel can layer network
  69  *                                      sockets.
  70  *              Alan Cox        :       Use ip_tos/ip_ttl settings.
  71  *              Alan Cox        :       Handle FIN (more) properly (we hope).
  72  *              Alan Cox        :       RST frames sent on unsynchronised
  73  *                                      state ack error.
  74  *              Alan Cox        :       Put in missing check for SYN bit.
  75  *              Alan Cox        :       Added tcp_select_window() aka NET2E
  76  *                                      window non shrink trick.
  77  *              Alan Cox        :       Added a couple of small NET2E timer
  78  *                                      fixes
  79  *              Charles Hedrick :       TCP fixes
  80  *              Toomas Tamm     :       TCP window fixes
  81  *              Alan Cox        :       Small URG fix to rlogin ^C ack fight
  82  *              Charles Hedrick :       Rewrote most of it to actually work
  83  *              Linus           :       Rewrote tcp_read() and URG handling
  84  *                                      completely
  85  *              Gerhard Koerting:       Fixed some missing timer handling
  86  *              Matthew Dillon  :       Reworked TCP machine states as per RFC
  87  *              Gerhard Koerting:       PC/TCP workarounds
  88  *              Adam Caldwell   :       Assorted timer/timing errors
  89  *              Matthew Dillon  :       Fixed another RST bug
  90  *              Alan Cox        :       Move to kernel side addressing changes.
  91  *              Alan Cox        :       Beginning work on TCP fastpathing
  92  *                                      (not yet usable)
  93  *              Arnt Gulbrandsen:       Turbocharged tcp_check() routine.
  94  *              Alan Cox        :       TCP fast path debugging
  95  *              Alan Cox        :       Window clamping
  96  *              Michael Riepe   :       Bug in tcp_check()
  97  *              Matt Dillon     :       More TCP improvements and RST bug fixes
  98  *              Matt Dillon     :       Yet more small nasties remove from the
  99  *                                      TCP code (Be very nice to this man if
 100  *                                      tcp finally works 100%) 8)
 101  *              Alan Cox        :       BSD accept semantics.
 102  *              Alan Cox        :       Reset on closedown bug.
 103  *      Peter De Schrijver      :       ENOTCONN check missing in tcp_sendto().
 104  *              Michael Pall    :       Handle poll() after URG properly in
 105  *                                      all cases.
 106  *              Michael Pall    :       Undo the last fix in tcp_read_urg()
 107  *                                      (multi URG PUSH broke rlogin).
 108  *              Michael Pall    :       Fix the multi URG PUSH problem in
 109  *                                      tcp_readable(), poll() after URG
 110  *                                      works now.
 111  *              Michael Pall    :       recv(...,MSG_OOB) never blocks in the
 112  *                                      BSD api.
 113  *              Alan Cox        :       Changed the semantics of sk->socket to
 114  *                                      fix a race and a signal problem with
 115  *                                      accept() and async I/O.
 116  *              Alan Cox        :       Relaxed the rules on tcp_sendto().
 117  *              Yury Shevchuk   :       Really fixed accept() blocking problem.
 118  *              Craig I. Hagan  :       Allow for BSD compatible TIME_WAIT for
 119  *                                      clients/servers which listen in on
 120  *                                      fixed ports.
 121  *              Alan Cox        :       Cleaned the above up and shrank it to
 122  *                                      a sensible code size.
 123  *              Alan Cox        :       Self connect lockup fix.
 124  *              Alan Cox        :       No connect to multicast.
 125  *              Ross Biro       :       Close unaccepted children on master
 126  *                                      socket close.
 127  *              Alan Cox        :       Reset tracing code.
 128  *              Alan Cox        :       Spurious resets on shutdown.
 129  *              Alan Cox        :       Giant 15 minute/60 second timer error
 130  *              Alan Cox        :       Small whoops in polling before an
 131  *                                      accept.
 132  *              Alan Cox        :       Kept the state trace facility since
 133  *                                      it's handy for debugging.
 134  *              Alan Cox        :       More reset handler fixes.
 135  *              Alan Cox        :       Started rewriting the code based on
 136  *                                      the RFC's for other useful protocol
 137  *                                      references see: Comer, KA9Q NOS, and
 138  *                                      for a reference on the difference
 139  *                                      between specifications and how BSD
 140  *                                      works see the 4.4lite source.
 141  *              A.N.Kuznetsov   :       Don't time wait on completion of tidy
 142  *                                      close.
 143  *              Linus Torvalds  :       Fin/Shutdown & copied_seq changes.
 144  *              Linus Torvalds  :       Fixed BSD port reuse to work first syn
 145  *              Alan Cox        :       Reimplemented timers as per the RFC
 146  *                                      and using multiple timers for sanity.
 147  *              Alan Cox        :       Small bug fixes, and a lot of new
 148  *                                      comments.
 149  *              Alan Cox        :       Fixed dual reader crash by locking
 150  *                                      the buffers (much like datagram.c)
 151  *              Alan Cox        :       Fixed stuck sockets in probe. A probe
 152  *                                      now gets fed up of retrying without
 153  *                                      (even a no space) answer.
 154  *              Alan Cox        :       Extracted closing code better
 155  *              Alan Cox        :       Fixed the closing state machine to
 156  *                                      resemble the RFC.
 157  *              Alan Cox        :       More 'per spec' fixes.
 158  *              Jorge Cwik      :       Even faster checksumming.
 159  *              Alan Cox        :       tcp_data() doesn't ack illegal PSH
 160  *                                      only frames. At least one pc tcp stack
 161  *                                      generates them.
 162  *              Alan Cox        :       Cache last socket.
 163  *              Alan Cox        :       Per route irtt.
 164  *              Matt Day        :       poll()->select() match BSD precisely on error
 165  *              Alan Cox        :       New buffers
 166  *              Marc Tamsky     :       Various sk->prot->retransmits and
 167  *                                      sk->retransmits misupdating fixed.
 168  *                                      Fixed tcp_write_timeout: stuck close,
 169  *                                      and TCP syn retries gets used now.
 170  *              Mark Yarvis     :       In tcp_read_wakeup(), don't send an
 171  *                                      ack if state is TCP_CLOSED.
 172  *              Alan Cox        :       Look up device on a retransmit - routes may
 173  *                                      change. Doesn't yet cope with MSS shrink right
 174  *                                      but it's a start!
 175  *              Marc Tamsky     :       Closing in closing fixes.
 176  *              Mike Shaver     :       RFC1122 verifications.
 177  *              Alan Cox        :       rcv_saddr errors.
 178  *              Alan Cox        :       Block double connect().
 179  *              Alan Cox        :       Small hooks for enSKIP.
 180  *              Alexey Kuznetsov:       Path MTU discovery.
 181  *              Alan Cox        :       Support soft errors.
 182  *              Alan Cox        :       Fix MTU discovery pathological case
 183  *                                      when the remote claims no mtu!
 184  *              Marc Tamsky     :       TCP_CLOSE fix.
 185  *              Colin (G3TNE)   :       Send a reset on syn ack replies in
 186  *                                      window but wrong (fixes NT lpd problems)
 187  *              Pedro Roque     :       Better TCP window handling, delayed ack.
 188  *              Joerg Reuter    :       No modification of locked buffers in
 189  *                                      tcp_do_retransmit()
 190  *              Eric Schenk     :       Changed receiver side silly window
 191  *                                      avoidance algorithm to BSD style
 192  *                                      algorithm. This doubles throughput
 193  *                                      against machines running Solaris,
 194  *                                      and seems to result in general
 195  *                                      improvement.
 196  *      Stefan Magdalinski      :       adjusted tcp_readable() to fix FIONREAD
 197  *      Willy Konynenberg       :       Transparent proxying support.
 198  *      Mike McLagan            :       Routing by source
 199  *              Keith Owens     :       Do proper merging with partial SKB's in
 200  *                                      tcp_do_sendmsg to avoid burstiness.
 201  *              Eric Schenk     :       Fix fast close down bug with
 202  *                                      shutdown() followed by close().
 203  *              Andi Kleen      :       Make poll agree with SIGIO
 204  *      Salvatore Sanfilippo    :       Support SO_LINGER with linger == 1 and
 205  *                                      lingertime == 0 (RFC 793 ABORT Call)
 206  *      Hirokazu Takahashi      :       Use copy_from_user() instead of
 207  *                                      csum_and_copy_from_user() if possible.
 208  *
 209  * Description of States:
 210  *
 211  *      TCP_SYN_SENT            sent a connection request, waiting for ack
 212  *
 213  *      TCP_SYN_RECV            received a connection request, sent ack,
 214  *                              waiting for final ack in three-way handshake.
 215  *
 216  *      TCP_ESTABLISHED         connection established
 217  *
 218  *      TCP_FIN_WAIT1           our side has shutdown, waiting to complete
 219  *                              transmission of remaining buffered data
 220  *
 221  *      TCP_FIN_WAIT2           all buffered data sent, waiting for remote
 222  *                              to shutdown
 223  *
 224  *      TCP_CLOSING             both sides have shutdown but we still have
 225  *                              data we have to finish sending
 226  *
 227  *      TCP_TIME_WAIT           timeout to catch resent junk before entering
 228  *                              closed, can only be entered from FIN_WAIT2
 229  *                              or CLOSING.  Required because the other end
 230  *                              may not have gotten our last ACK causing it
 231  *                              to retransmit the data packet (which we ignore)
 232  *
 233  *      TCP_CLOSE_WAIT          remote side has shutdown and is waiting for
 234  *                              us to finish writing our data and to shutdown
 235  *                              (we have to close() to move on to LAST_ACK)
 236  *
 237  *      TCP_LAST_ACK            out side has shutdown after remote has
 238  *                              shutdown.  There may still be data in our
 239  *                              buffer that we have to finish sending
 240  *
 241  *      TCP_CLOSE               socket is finished
 242  */
 243 
 244 #define pr_fmt(fmt) "TCP: " fmt
 245 
 246 #include <crypto/hash.h>
 247 #include <linux/kernel.h>
 248 #include <linux/module.h>
 249 #include <linux/types.h>
 250 #include <linux/fcntl.h>
 251 #include <linux/poll.h>
 252 #include <linux/inet_diag.h>
 253 #include <linux/init.h>
 254 #include <linux/fs.h>
 255 #include <linux/skbuff.h>
 256 #include <linux/scatterlist.h>
 257 #include <linux/splice.h>
 258 #include <linux/net.h>
 259 #include <linux/socket.h>
 260 #include <linux/random.h>
 261 #include <linux/memblock.h>
 262 #include <linux/highmem.h>
 263 #include <linux/swap.h>
 264 #include <linux/cache.h>
 265 #include <linux/err.h>
 266 #include <linux/time.h>
 267 #include <linux/slab.h>
 268 #include <linux/errqueue.h>
 269 #include <linux/static_key.h>
 270 
 271 #include <net/icmp.h>
 272 #include <net/inet_common.h>
 273 #include <net/tcp.h>
 274 #include <net/xfrm.h>
 275 #include <net/ip.h>
 276 #include <net/sock.h>
 277 
 278 #include <linux/uaccess.h>
 279 #include <asm/ioctls.h>
 280 #include <net/busy_poll.h>
 281 
 282 struct percpu_counter tcp_orphan_count;
 283 EXPORT_SYMBOL_GPL(tcp_orphan_count);
 284 
 285 long sysctl_tcp_mem[3] __read_mostly;
 286 EXPORT_SYMBOL(sysctl_tcp_mem);
 287 
 288 atomic_long_t tcp_memory_allocated;     /* Current allocated memory. */
 289 EXPORT_SYMBOL(tcp_memory_allocated);
 290 
 291 #if IS_ENABLED(CONFIG_SMC)
 292 DEFINE_STATIC_KEY_FALSE(tcp_have_smc);
 293 EXPORT_SYMBOL(tcp_have_smc);
 294 #endif
 295 
 296 /*
 297  * Current number of TCP sockets.
 298  */
 299 struct percpu_counter tcp_sockets_allocated;
 300 EXPORT_SYMBOL(tcp_sockets_allocated);
 301 
 302 /*
 303  * TCP splice context
 304  */
 305 struct tcp_splice_state {
 306         struct pipe_inode_info *pipe;
 307         size_t len;
 308         unsigned int flags;
 309 };
 310 
 311 /*
 312  * Pressure flag: try to collapse.
 313  * Technical note: it is used by multiple contexts non atomically.
 314  * All the __sk_mem_schedule() is of this nature: accounting
 315  * is strict, actions are advisory and have some latency.
 316  */
 317 unsigned long tcp_memory_pressure __read_mostly;
 318 EXPORT_SYMBOL_GPL(tcp_memory_pressure);
 319 
 320 DEFINE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
 321 EXPORT_SYMBOL(tcp_rx_skb_cache_key);
 322 
 323 DEFINE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
 324 
 325 void tcp_enter_memory_pressure(struct sock *sk)
 326 {
 327         unsigned long val;
 328 
 329         if (READ_ONCE(tcp_memory_pressure))
 330                 return;
 331         val = jiffies;
 332 
 333         if (!val)
 334                 val--;
 335         if (!cmpxchg(&tcp_memory_pressure, 0, val))
 336                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
 337 }
 338 EXPORT_SYMBOL_GPL(tcp_enter_memory_pressure);
 339 
 340 void tcp_leave_memory_pressure(struct sock *sk)
 341 {
 342         unsigned long val;
 343 
 344         if (!READ_ONCE(tcp_memory_pressure))
 345                 return;
 346         val = xchg(&tcp_memory_pressure, 0);
 347         if (val)
 348                 NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURESCHRONO,
 349                               jiffies_to_msecs(jiffies - val));
 350 }
 351 EXPORT_SYMBOL_GPL(tcp_leave_memory_pressure);
 352 
 353 /* Convert seconds to retransmits based on initial and max timeout */
 354 static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
 355 {
 356         u8 res = 0;
 357 
 358         if (seconds > 0) {
 359                 int period = timeout;
 360 
 361                 res = 1;
 362                 while (seconds > period && res < 255) {
 363                         res++;
 364                         timeout <<= 1;
 365                         if (timeout > rto_max)
 366                                 timeout = rto_max;
 367                         period += timeout;
 368                 }
 369         }
 370         return res;
 371 }
 372 
 373 /* Convert retransmits to seconds based on initial and max timeout */
 374 static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
 375 {
 376         int period = 0;
 377 
 378         if (retrans > 0) {
 379                 period = timeout;
 380                 while (--retrans) {
 381                         timeout <<= 1;
 382                         if (timeout > rto_max)
 383                                 timeout = rto_max;
 384                         period += timeout;
 385                 }
 386         }
 387         return period;
 388 }
 389 
 390 static u64 tcp_compute_delivery_rate(const struct tcp_sock *tp)
 391 {
 392         u32 rate = READ_ONCE(tp->rate_delivered);
 393         u32 intv = READ_ONCE(tp->rate_interval_us);
 394         u64 rate64 = 0;
 395 
 396         if (rate && intv) {
 397                 rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
 398                 do_div(rate64, intv);
 399         }
 400         return rate64;
 401 }
 402 
 403 /* Address-family independent initialization for a tcp_sock.
 404  *
 405  * NOTE: A lot of things set to zero explicitly by call to
 406  *       sk_alloc() so need not be done here.
 407  */
 408 void tcp_init_sock(struct sock *sk)
 409 {
 410         struct inet_connection_sock *icsk = inet_csk(sk);
 411         struct tcp_sock *tp = tcp_sk(sk);
 412 
 413         tp->out_of_order_queue = RB_ROOT;
 414         sk->tcp_rtx_queue = RB_ROOT;
 415         tcp_init_xmit_timers(sk);
 416         INIT_LIST_HEAD(&tp->tsq_node);
 417         INIT_LIST_HEAD(&tp->tsorted_sent_queue);
 418 
 419         icsk->icsk_rto = TCP_TIMEOUT_INIT;
 420         tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
 421         minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U);
 422 
 423         /* So many TCP implementations out there (incorrectly) count the
 424          * initial SYN frame in their delayed-ACK and congestion control
 425          * algorithms that we must have the following bandaid to talk
 426          * efficiently to them.  -DaveM
 427          */
 428         tp->snd_cwnd = TCP_INIT_CWND;
 429 
 430         /* There's a bubble in the pipe until at least the first ACK. */
 431         tp->app_limited = ~0U;
 432 
 433         /* See draft-stevens-tcpca-spec-01 for discussion of the
 434          * initialization of these values.
 435          */
 436         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
 437         tp->snd_cwnd_clamp = ~0;
 438         tp->mss_cache = TCP_MSS_DEFAULT;
 439 
 440         tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
 441         tcp_assign_congestion_control(sk);
 442 
 443         tp->tsoffset = 0;
 444         tp->rack.reo_wnd_steps = 1;
 445 
 446         sk->sk_state = TCP_CLOSE;
 447 
 448         sk->sk_write_space = sk_stream_write_space;
 449         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 450 
 451         icsk->icsk_sync_mss = tcp_sync_mss;
 452 
 453         WRITE_ONCE(sk->sk_sndbuf, sock_net(sk)->ipv4.sysctl_tcp_wmem[1]);
 454         WRITE_ONCE(sk->sk_rcvbuf, sock_net(sk)->ipv4.sysctl_tcp_rmem[1]);
 455 
 456         sk_sockets_allocated_inc(sk);
 457         sk->sk_route_forced_caps = NETIF_F_GSO;
 458 }
 459 EXPORT_SYMBOL(tcp_init_sock);
 460 
 461 static void tcp_tx_timestamp(struct sock *sk, u16 tsflags)
 462 {
 463         struct sk_buff *skb = tcp_write_queue_tail(sk);
 464 
 465         if (tsflags && skb) {
 466                 struct skb_shared_info *shinfo = skb_shinfo(skb);
 467                 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 468 
 469                 sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags);
 470                 if (tsflags & SOF_TIMESTAMPING_TX_ACK)
 471                         tcb->txstamp_ack = 1;
 472                 if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
 473                         shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
 474         }
 475 }
 476 
 477 static inline bool tcp_stream_is_readable(const struct tcp_sock *tp,
 478                                           int target, struct sock *sk)
 479 {
 480         int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
 481 
 482         if (avail > 0) {
 483                 if (avail >= target)
 484                         return true;
 485                 if (tcp_rmem_pressure(sk))
 486                         return true;
 487         }
 488         if (sk->sk_prot->stream_memory_read)
 489                 return sk->sk_prot->stream_memory_read(sk);
 490         return false;
 491 }
 492 
 493 /*
 494  *      Wait for a TCP event.
 495  *
 496  *      Note that we don't need to lock the socket, as the upper poll layers
 497  *      take care of normal races (between the test and the event) and we don't
 498  *      go look at any of the socket buffers directly.
 499  */
 500 __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
 501 {
 502         __poll_t mask;
 503         struct sock *sk = sock->sk;
 504         const struct tcp_sock *tp = tcp_sk(sk);
 505         int state;
 506 
 507         sock_poll_wait(file, sock, wait);
 508 
 509         state = inet_sk_state_load(sk);
 510         if (state == TCP_LISTEN)
 511                 return inet_csk_listen_poll(sk);
 512 
 513         /* Socket is not locked. We are protected from async events
 514          * by poll logic and correct handling of state changes
 515          * made by other threads is impossible in any case.
 516          */
 517 
 518         mask = 0;
 519 
 520         /*
 521          * EPOLLHUP is certainly not done right. But poll() doesn't
 522          * have a notion of HUP in just one direction, and for a
 523          * socket the read side is more interesting.
 524          *
 525          * Some poll() documentation says that EPOLLHUP is incompatible
 526          * with the EPOLLOUT/POLLWR flags, so somebody should check this
 527          * all. But careful, it tends to be safer to return too many
 528          * bits than too few, and you can easily break real applications
 529          * if you don't tell them that something has hung up!
 530          *
 531          * Check-me.
 532          *
 533          * Check number 1. EPOLLHUP is _UNMASKABLE_ event (see UNIX98 and
 534          * our fs/select.c). It means that after we received EOF,
 535          * poll always returns immediately, making impossible poll() on write()
 536          * in state CLOSE_WAIT. One solution is evident --- to set EPOLLHUP
 537          * if and only if shutdown has been made in both directions.
 538          * Actually, it is interesting to look how Solaris and DUX
 539          * solve this dilemma. I would prefer, if EPOLLHUP were maskable,
 540          * then we could set it on SND_SHUTDOWN. BTW examples given
 541          * in Stevens' books assume exactly this behaviour, it explains
 542          * why EPOLLHUP is incompatible with EPOLLOUT.  --ANK
 543          *
 544          * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
 545          * blocking on fresh not-connected or disconnected socket. --ANK
 546          */
 547         if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
 548                 mask |= EPOLLHUP;
 549         if (sk->sk_shutdown & RCV_SHUTDOWN)
 550                 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
 551 
 552         /* Connected or passive Fast Open socket? */
 553         if (state != TCP_SYN_SENT &&
 554             (state != TCP_SYN_RECV || rcu_access_pointer(tp->fastopen_rsk))) {
 555                 int target = sock_rcvlowat(sk, 0, INT_MAX);
 556 
 557                 if (READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq) &&
 558                     !sock_flag(sk, SOCK_URGINLINE) &&
 559                     tp->urg_data)
 560                         target++;
 561 
 562                 if (tcp_stream_is_readable(tp, target, sk))
 563                         mask |= EPOLLIN | EPOLLRDNORM;
 564 
 565                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
 566                         if (sk_stream_is_writeable(sk)) {
 567                                 mask |= EPOLLOUT | EPOLLWRNORM;
 568                         } else {  /* send SIGIO later */
 569                                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 570                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 571 
 572                                 /* Race breaker. If space is freed after
 573                                  * wspace test but before the flags are set,
 574                                  * IO signal will be lost. Memory barrier
 575                                  * pairs with the input side.
 576                                  */
 577                                 smp_mb__after_atomic();
 578                                 if (sk_stream_is_writeable(sk))
 579                                         mask |= EPOLLOUT | EPOLLWRNORM;
 580                         }
 581                 } else
 582                         mask |= EPOLLOUT | EPOLLWRNORM;
 583 
 584                 if (tp->urg_data & TCP_URG_VALID)
 585                         mask |= EPOLLPRI;
 586         } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
 587                 /* Active TCP fastopen socket with defer_connect
 588                  * Return EPOLLOUT so application can call write()
 589                  * in order for kernel to generate SYN+data
 590                  */
 591                 mask |= EPOLLOUT | EPOLLWRNORM;
 592         }
 593         /* This barrier is coupled with smp_wmb() in tcp_reset() */
 594         smp_rmb();
 595         if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
 596                 mask |= EPOLLERR;
 597 
 598         return mask;
 599 }
 600 EXPORT_SYMBOL(tcp_poll);
 601 
 602 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
 603 {
 604         struct tcp_sock *tp = tcp_sk(sk);
 605         int answ;
 606         bool slow;
 607 
 608         switch (cmd) {
 609         case SIOCINQ:
 610                 if (sk->sk_state == TCP_LISTEN)
 611                         return -EINVAL;
 612 
 613                 slow = lock_sock_fast(sk);
 614                 answ = tcp_inq(sk);
 615                 unlock_sock_fast(sk, slow);
 616                 break;
 617         case SIOCATMARK:
 618                 answ = tp->urg_data &&
 619                        READ_ONCE(tp->urg_seq) == READ_ONCE(tp->copied_seq);
 620                 break;
 621         case SIOCOUTQ:
 622                 if (sk->sk_state == TCP_LISTEN)
 623                         return -EINVAL;
 624 
 625                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
 626                         answ = 0;
 627                 else
 628                         answ = READ_ONCE(tp->write_seq) - tp->snd_una;
 629                 break;
 630         case SIOCOUTQNSD:
 631                 if (sk->sk_state == TCP_LISTEN)
 632                         return -EINVAL;
 633 
 634                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
 635                         answ = 0;
 636                 else
 637                         answ = READ_ONCE(tp->write_seq) -
 638                                READ_ONCE(tp->snd_nxt);
 639                 break;
 640         default:
 641                 return -ENOIOCTLCMD;
 642         }
 643 
 644         return put_user(answ, (int __user *)arg);
 645 }
 646 EXPORT_SYMBOL(tcp_ioctl);
 647 
 648 static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
 649 {
 650         TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
 651         tp->pushed_seq = tp->write_seq;
 652 }
 653 
 654 static inline bool forced_push(const struct tcp_sock *tp)
 655 {
 656         return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
 657 }
 658 
 659 static void skb_entail(struct sock *sk, struct sk_buff *skb)
 660 {
 661         struct tcp_sock *tp = tcp_sk(sk);
 662         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 663 
 664         skb->csum    = 0;
 665         tcb->seq     = tcb->end_seq = tp->write_seq;
 666         tcb->tcp_flags = TCPHDR_ACK;
 667         tcb->sacked  = 0;
 668         __skb_header_release(skb);
 669         tcp_add_write_queue_tail(sk, skb);
 670         sk_wmem_queued_add(sk, skb->truesize);
 671         sk_mem_charge(sk, skb->truesize);
 672         if (tp->nonagle & TCP_NAGLE_PUSH)
 673                 tp->nonagle &= ~TCP_NAGLE_PUSH;
 674 
 675         tcp_slow_start_after_idle_check(sk);
 676 }
 677 
 678 static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
 679 {
 680         if (flags & MSG_OOB)
 681                 tp->snd_up = tp->write_seq;
 682 }
 683 
 684 /* If a not yet filled skb is pushed, do not send it if
 685  * we have data packets in Qdisc or NIC queues :
 686  * Because TX completion will happen shortly, it gives a chance
 687  * to coalesce future sendmsg() payload into this skb, without
 688  * need for a timer, and with no latency trade off.
 689  * As packets containing data payload have a bigger truesize
 690  * than pure acks (dataless) packets, the last checks prevent
 691  * autocorking if we only have an ACK in Qdisc/NIC queues,
 692  * or if TX completion was delayed after we processed ACK packet.
 693  */
 694 static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
 695                                 int size_goal)
 696 {
 697         return skb->len < size_goal &&
 698                sock_net(sk)->ipv4.sysctl_tcp_autocorking &&
 699                !tcp_rtx_queue_empty(sk) &&
 700                refcount_read(&sk->sk_wmem_alloc) > skb->truesize;
 701 }
 702 
 703 static void tcp_push(struct sock *sk, int flags, int mss_now,
 704                      int nonagle, int size_goal)
 705 {
 706         struct tcp_sock *tp = tcp_sk(sk);
 707         struct sk_buff *skb;
 708 
 709         skb = tcp_write_queue_tail(sk);
 710         if (!skb)
 711                 return;
 712         if (!(flags & MSG_MORE) || forced_push(tp))
 713                 tcp_mark_push(tp, skb);
 714 
 715         tcp_mark_urg(tp, flags);
 716 
 717         if (tcp_should_autocork(sk, skb, size_goal)) {
 718 
 719                 /* avoid atomic op if TSQ_THROTTLED bit is already set */
 720                 if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) {
 721                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
 722                         set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
 723                 }
 724                 /* It is possible TX completion already happened
 725                  * before we set TSQ_THROTTLED.
 726                  */
 727                 if (refcount_read(&sk->sk_wmem_alloc) > skb->truesize)
 728                         return;
 729         }
 730 
 731         if (flags & MSG_MORE)
 732                 nonagle = TCP_NAGLE_CORK;
 733 
 734         __tcp_push_pending_frames(sk, mss_now, nonagle);
 735 }
 736 
 737 static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
 738                                 unsigned int offset, size_t len)
 739 {
 740         struct tcp_splice_state *tss = rd_desc->arg.data;
 741         int ret;
 742 
 743         ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
 744                               min(rd_desc->count, len), tss->flags);
 745         if (ret > 0)
 746                 rd_desc->count -= ret;
 747         return ret;
 748 }
 749 
 750 static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
 751 {
 752         /* Store TCP splice context information in read_descriptor_t. */
 753         read_descriptor_t rd_desc = {
 754                 .arg.data = tss,
 755                 .count    = tss->len,
 756         };
 757 
 758         return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
 759 }
 760 
 761 /**
 762  *  tcp_splice_read - splice data from TCP socket to a pipe
 763  * @sock:       socket to splice from
 764  * @ppos:       position (not valid)
 765  * @pipe:       pipe to splice to
 766  * @len:        number of bytes to splice
 767  * @flags:      splice modifier flags
 768  *
 769  * Description:
 770  *    Will read pages from given socket and fill them into a pipe.
 771  *
 772  **/
 773 ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
 774                         struct pipe_inode_info *pipe, size_t len,
 775                         unsigned int flags)
 776 {
 777         struct sock *sk = sock->sk;
 778         struct tcp_splice_state tss = {
 779                 .pipe = pipe,
 780                 .len = len,
 781                 .flags = flags,
 782         };
 783         long timeo;
 784         ssize_t spliced;
 785         int ret;
 786 
 787         sock_rps_record_flow(sk);
 788         /*
 789          * We can't seek on a socket input
 790          */
 791         if (unlikely(*ppos))
 792                 return -ESPIPE;
 793 
 794         ret = spliced = 0;
 795 
 796         lock_sock(sk);
 797 
 798         timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
 799         while (tss.len) {
 800                 ret = __tcp_splice_read(sk, &tss);
 801                 if (ret < 0)
 802                         break;
 803                 else if (!ret) {
 804                         if (spliced)
 805                                 break;
 806                         if (sock_flag(sk, SOCK_DONE))
 807                                 break;
 808                         if (sk->sk_err) {
 809                                 ret = sock_error(sk);
 810                                 break;
 811                         }
 812                         if (sk->sk_shutdown & RCV_SHUTDOWN)
 813                                 break;
 814                         if (sk->sk_state == TCP_CLOSE) {
 815                                 /*
 816                                  * This occurs when user tries to read
 817                                  * from never connected socket.
 818                                  */
 819                                 ret = -ENOTCONN;
 820                                 break;
 821                         }
 822                         if (!timeo) {
 823                                 ret = -EAGAIN;
 824                                 break;
 825                         }
 826                         /* if __tcp_splice_read() got nothing while we have
 827                          * an skb in receive queue, we do not want to loop.
 828                          * This might happen with URG data.
 829                          */
 830                         if (!skb_queue_empty(&sk->sk_receive_queue))
 831                                 break;
 832                         sk_wait_data(sk, &timeo, NULL);
 833                         if (signal_pending(current)) {
 834                                 ret = sock_intr_errno(timeo);
 835                                 break;
 836                         }
 837                         continue;
 838                 }
 839                 tss.len -= ret;
 840                 spliced += ret;
 841 
 842                 if (!timeo)
 843                         break;
 844                 release_sock(sk);
 845                 lock_sock(sk);
 846 
 847                 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
 848                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
 849                     signal_pending(current))
 850                         break;
 851         }
 852 
 853         release_sock(sk);
 854 
 855         if (spliced)
 856                 return spliced;
 857 
 858         return ret;
 859 }
 860 EXPORT_SYMBOL(tcp_splice_read);
 861 
 862 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
 863                                     bool force_schedule)
 864 {
 865         struct sk_buff *skb;
 866 
 867         if (likely(!size)) {
 868                 skb = sk->sk_tx_skb_cache;
 869                 if (skb) {
 870                         skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
 871                         sk->sk_tx_skb_cache = NULL;
 872                         pskb_trim(skb, 0);
 873                         INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
 874                         skb_shinfo(skb)->tx_flags = 0;
 875                         memset(TCP_SKB_CB(skb), 0, sizeof(struct tcp_skb_cb));
 876                         return skb;
 877                 }
 878         }
 879         /* The TCP header must be at least 32-bit aligned.  */
 880         size = ALIGN(size, 4);
 881 
 882         if (unlikely(tcp_under_memory_pressure(sk)))
 883                 sk_mem_reclaim_partial(sk);
 884 
 885         skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
 886         if (likely(skb)) {
 887                 bool mem_scheduled;
 888 
 889                 if (force_schedule) {
 890                         mem_scheduled = true;
 891                         sk_forced_mem_schedule(sk, skb->truesize);
 892                 } else {
 893                         mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
 894                 }
 895                 if (likely(mem_scheduled)) {
 896                         skb_reserve(skb, sk->sk_prot->max_header);
 897                         /*
 898                          * Make sure that we have exactly size bytes
 899                          * available to the caller, no more, no less.
 900                          */
 901                         skb->reserved_tailroom = skb->end - skb->tail - size;
 902                         INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
 903                         return skb;
 904                 }
 905                 __kfree_skb(skb);
 906         } else {
 907                 sk->sk_prot->enter_memory_pressure(sk);
 908                 sk_stream_moderate_sndbuf(sk);
 909         }
 910         return NULL;
 911 }
 912 
 913 static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
 914                                        int large_allowed)
 915 {
 916         struct tcp_sock *tp = tcp_sk(sk);
 917         u32 new_size_goal, size_goal;
 918 
 919         if (!large_allowed)
 920                 return mss_now;
 921 
 922         /* Note : tcp_tso_autosize() will eventually split this later */
 923         new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
 924         new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
 925 
 926         /* We try hard to avoid divides here */
 927         size_goal = tp->gso_segs * mss_now;
 928         if (unlikely(new_size_goal < size_goal ||
 929                      new_size_goal >= size_goal + mss_now)) {
 930                 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
 931                                      sk->sk_gso_max_segs);
 932                 size_goal = tp->gso_segs * mss_now;
 933         }
 934 
 935         return max(size_goal, mss_now);
 936 }
 937 
 938 static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
 939 {
 940         int mss_now;
 941 
 942         mss_now = tcp_current_mss(sk);
 943         *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
 944 
 945         return mss_now;
 946 }
 947 
 948 /* In some cases, both sendpage() and sendmsg() could have added
 949  * an skb to the write queue, but failed adding payload on it.
 950  * We need to remove it to consume less memory, but more
 951  * importantly be able to generate EPOLLOUT for Edge Trigger epoll()
 952  * users.
 953  */
 954 static void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb)
 955 {
 956         if (skb && !skb->len) {
 957                 tcp_unlink_write_queue(skb, sk);
 958                 if (tcp_write_queue_empty(sk))
 959                         tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
 960                 sk_wmem_free_skb(sk, skb);
 961         }
 962 }
 963 
 964 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
 965                          size_t size, int flags)
 966 {
 967         struct tcp_sock *tp = tcp_sk(sk);
 968         int mss_now, size_goal;
 969         int err;
 970         ssize_t copied;
 971         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 972 
 973         if (IS_ENABLED(CONFIG_DEBUG_VM) &&
 974             WARN_ONCE(PageSlab(page), "page must not be a Slab one"))
 975                 return -EINVAL;
 976 
 977         /* Wait for a connection to finish. One exception is TCP Fast Open
 978          * (passive side) where data is allowed to be sent before a connection
 979          * is fully established.
 980          */
 981         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
 982             !tcp_passive_fastopen(sk)) {
 983                 err = sk_stream_wait_connect(sk, &timeo);
 984                 if (err != 0)
 985                         goto out_err;
 986         }
 987 
 988         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 989 
 990         mss_now = tcp_send_mss(sk, &size_goal, flags);
 991         copied = 0;
 992 
 993         err = -EPIPE;
 994         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 995                 goto out_err;
 996 
 997         while (size > 0) {
 998                 struct sk_buff *skb = tcp_write_queue_tail(sk);
 999                 int copy, i;
1000                 bool can_coalesce;
1001 
1002                 if (!skb || (copy = size_goal - skb->len) <= 0 ||
1003                     !tcp_skb_can_collapse_to(skb)) {
1004 new_segment:
1005                         if (!sk_stream_memory_free(sk))
1006                                 goto wait_for_sndbuf;
1007 
1008                         skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
1009                                         tcp_rtx_and_write_queues_empty(sk));
1010                         if (!skb)
1011                                 goto wait_for_memory;
1012 
1013 #ifdef CONFIG_TLS_DEVICE
1014                         skb->decrypted = !!(flags & MSG_SENDPAGE_DECRYPTED);
1015 #endif
1016                         skb_entail(sk, skb);
1017                         copy = size_goal;
1018                 }
1019 
1020                 if (copy > size)
1021                         copy = size;
1022 
1023                 i = skb_shinfo(skb)->nr_frags;
1024                 can_coalesce = skb_can_coalesce(skb, i, page, offset);
1025                 if (!can_coalesce && i >= sysctl_max_skb_frags) {
1026                         tcp_mark_push(tp, skb);
1027                         goto new_segment;
1028                 }
1029                 if (!sk_wmem_schedule(sk, copy))
1030                         goto wait_for_memory;
1031 
1032                 if (can_coalesce) {
1033                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1034                 } else {
1035                         get_page(page);
1036                         skb_fill_page_desc(skb, i, page, offset, copy);
1037                 }
1038 
1039                 if (!(flags & MSG_NO_SHARED_FRAGS))
1040                         skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1041 
1042                 skb->len += copy;
1043                 skb->data_len += copy;
1044                 skb->truesize += copy;
1045                 sk_wmem_queued_add(sk, copy);
1046                 sk_mem_charge(sk, copy);
1047                 skb->ip_summed = CHECKSUM_PARTIAL;
1048                 WRITE_ONCE(tp->write_seq, tp->write_seq + copy);
1049                 TCP_SKB_CB(skb)->end_seq += copy;
1050                 tcp_skb_pcount_set(skb, 0);
1051 
1052                 if (!copied)
1053                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1054 
1055                 copied += copy;
1056                 offset += copy;
1057                 size -= copy;
1058                 if (!size)
1059                         goto out;
1060 
1061                 if (skb->len < size_goal || (flags & MSG_OOB))
1062                         continue;
1063 
1064                 if (forced_push(tp)) {
1065                         tcp_mark_push(tp, skb);
1066                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1067                 } else if (skb == tcp_send_head(sk))
1068                         tcp_push_one(sk, mss_now);
1069                 continue;
1070 
1071 wait_for_sndbuf:
1072                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1073 wait_for_memory:
1074                 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1075                          TCP_NAGLE_PUSH, size_goal);
1076 
1077                 err = sk_stream_wait_memory(sk, &timeo);
1078                 if (err != 0)
1079                         goto do_error;
1080 
1081                 mss_now = tcp_send_mss(sk, &size_goal, flags);
1082         }
1083 
1084 out:
1085         if (copied) {
1086                 tcp_tx_timestamp(sk, sk->sk_tsflags);
1087                 if (!(flags & MSG_SENDPAGE_NOTLAST))
1088                         tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1089         }
1090         return copied;
1091 
1092 do_error:
1093         tcp_remove_empty_skb(sk, tcp_write_queue_tail(sk));
1094         if (copied)
1095                 goto out;
1096 out_err:
1097         /* make sure we wake any epoll edge trigger waiter */
1098         if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) {
1099                 sk->sk_write_space(sk);
1100                 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1101         }
1102         return sk_stream_error(sk, flags, err);
1103 }
1104 EXPORT_SYMBOL_GPL(do_tcp_sendpages);
1105 
1106 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
1107                         size_t size, int flags)
1108 {
1109         if (!(sk->sk_route_caps & NETIF_F_SG))
1110                 return sock_no_sendpage_locked(sk, page, offset, size, flags);
1111 
1112         tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1113 
1114         return do_tcp_sendpages(sk, page, offset, size, flags);
1115 }
1116 EXPORT_SYMBOL_GPL(tcp_sendpage_locked);
1117 
1118 int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1119                  size_t size, int flags)
1120 {
1121         int ret;
1122 
1123         lock_sock(sk);
1124         ret = tcp_sendpage_locked(sk, page, offset, size, flags);
1125         release_sock(sk);
1126 
1127         return ret;
1128 }
1129 EXPORT_SYMBOL(tcp_sendpage);
1130 
1131 void tcp_free_fastopen_req(struct tcp_sock *tp)
1132 {
1133         if (tp->fastopen_req) {
1134                 kfree(tp->fastopen_req);
1135                 tp->fastopen_req = NULL;
1136         }
1137 }
1138 
1139 static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1140                                 int *copied, size_t size,
1141                                 struct ubuf_info *uarg)
1142 {
1143         struct tcp_sock *tp = tcp_sk(sk);
1144         struct inet_sock *inet = inet_sk(sk);
1145         struct sockaddr *uaddr = msg->msg_name;
1146         int err, flags;
1147 
1148         if (!(sock_net(sk)->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
1149             (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1150              uaddr->sa_family == AF_UNSPEC))
1151                 return -EOPNOTSUPP;
1152         if (tp->fastopen_req)
1153                 return -EALREADY; /* Another Fast Open is in progress */
1154 
1155         tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1156                                    sk->sk_allocation);
1157         if (unlikely(!tp->fastopen_req))
1158                 return -ENOBUFS;
1159         tp->fastopen_req->data = msg;
1160         tp->fastopen_req->size = size;
1161         tp->fastopen_req->uarg = uarg;
1162 
1163         if (inet->defer_connect) {
1164                 err = tcp_connect(sk);
1165                 /* Same failure procedure as in tcp_v4/6_connect */
1166                 if (err) {
1167                         tcp_set_state(sk, TCP_CLOSE);
1168                         inet->inet_dport = 0;
1169                         sk->sk_route_caps = 0;
1170                 }
1171         }
1172         flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1173         err = __inet_stream_connect(sk->sk_socket, uaddr,
1174                                     msg->msg_namelen, flags, 1);
1175         /* fastopen_req could already be freed in __inet_stream_connect
1176          * if the connection times out or gets rst
1177          */
1178         if (tp->fastopen_req) {
1179                 *copied = tp->fastopen_req->copied;
1180                 tcp_free_fastopen_req(tp);
1181                 inet->defer_connect = 0;
1182         }
1183         return err;
1184 }
1185 
1186 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
1187 {
1188         struct tcp_sock *tp = tcp_sk(sk);
1189         struct ubuf_info *uarg = NULL;
1190         struct sk_buff *skb;
1191         struct sockcm_cookie sockc;
1192         int flags, err, copied = 0;
1193         int mss_now = 0, size_goal, copied_syn = 0;
1194         int process_backlog = 0;
1195         bool zc = false;
1196         long timeo;
1197 
1198         flags = msg->msg_flags;
1199 
1200         if (flags & MSG_ZEROCOPY && size && sock_flag(sk, SOCK_ZEROCOPY)) {
1201                 skb = tcp_write_queue_tail(sk);
1202                 uarg = sock_zerocopy_realloc(sk, size, skb_zcopy(skb));
1203                 if (!uarg) {
1204                         err = -ENOBUFS;
1205                         goto out_err;
1206                 }
1207 
1208                 zc = sk->sk_route_caps & NETIF_F_SG;
1209                 if (!zc)
1210                         uarg->zerocopy = 0;
1211         }
1212 
1213         if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect) &&
1214             !tp->repair) {
1215                 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size, uarg);
1216                 if (err == -EINPROGRESS && copied_syn > 0)
1217                         goto out;
1218                 else if (err)
1219                         goto out_err;
1220         }
1221 
1222         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1223 
1224         tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1225 
1226         /* Wait for a connection to finish. One exception is TCP Fast Open
1227          * (passive side) where data is allowed to be sent before a connection
1228          * is fully established.
1229          */
1230         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1231             !tcp_passive_fastopen(sk)) {
1232                 err = sk_stream_wait_connect(sk, &timeo);
1233                 if (err != 0)
1234                         goto do_error;
1235         }
1236 
1237         if (unlikely(tp->repair)) {
1238                 if (tp->repair_queue == TCP_RECV_QUEUE) {
1239                         copied = tcp_send_rcvq(sk, msg, size);
1240                         goto out_nopush;
1241                 }
1242 
1243                 err = -EINVAL;
1244                 if (tp->repair_queue == TCP_NO_QUEUE)
1245                         goto out_err;
1246 
1247                 /* 'common' sending to sendq */
1248         }
1249 
1250         sockcm_init(&sockc, sk);
1251         if (msg->msg_controllen) {
1252                 err = sock_cmsg_send(sk, msg, &sockc);
1253                 if (unlikely(err)) {
1254                         err = -EINVAL;
1255                         goto out_err;
1256                 }
1257         }
1258 
1259         /* This should be in poll */
1260         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1261 
1262         /* Ok commence sending. */
1263         copied = 0;
1264 
1265 restart:
1266         mss_now = tcp_send_mss(sk, &size_goal, flags);
1267 
1268         err = -EPIPE;
1269         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1270                 goto do_error;
1271 
1272         while (msg_data_left(msg)) {
1273                 int copy = 0;
1274 
1275                 skb = tcp_write_queue_tail(sk);
1276                 if (skb)
1277                         copy = size_goal - skb->len;
1278 
1279                 if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
1280                         bool first_skb;
1281 
1282 new_segment:
1283                         if (!sk_stream_memory_free(sk))
1284                                 goto wait_for_sndbuf;
1285 
1286                         if (unlikely(process_backlog >= 16)) {
1287                                 process_backlog = 0;
1288                                 if (sk_flush_backlog(sk))
1289                                         goto restart;
1290                         }
1291                         first_skb = tcp_rtx_and_write_queues_empty(sk);
1292                         skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
1293                                                   first_skb);
1294                         if (!skb)
1295                                 goto wait_for_memory;
1296 
1297                         process_backlog++;
1298                         skb->ip_summed = CHECKSUM_PARTIAL;
1299 
1300                         skb_entail(sk, skb);
1301                         copy = size_goal;
1302 
1303                         /* All packets are restored as if they have
1304                          * already been sent. skb_mstamp_ns isn't set to
1305                          * avoid wrong rtt estimation.
1306                          */
1307                         if (tp->repair)
1308                                 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1309                 }
1310 
1311                 /* Try to append data to the end of skb. */
1312                 if (copy > msg_data_left(msg))
1313                         copy = msg_data_left(msg);
1314 
1315                 /* Where to copy to? */
1316                 if (skb_availroom(skb) > 0 && !zc) {
1317                         /* We have some space in skb head. Superb! */
1318                         copy = min_t(int, copy, skb_availroom(skb));
1319                         err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1320                         if (err)
1321                                 goto do_fault;
1322                 } else if (!zc) {
1323                         bool merge = true;
1324                         int i = skb_shinfo(skb)->nr_frags;
1325                         struct page_frag *pfrag = sk_page_frag(sk);
1326 
1327                         if (!sk_page_frag_refill(sk, pfrag))
1328                                 goto wait_for_memory;
1329 
1330                         if (!skb_can_coalesce(skb, i, pfrag->page,
1331                                               pfrag->offset)) {
1332                                 if (i >= sysctl_max_skb_frags) {
1333                                         tcp_mark_push(tp, skb);
1334                                         goto new_segment;
1335                                 }
1336                                 merge = false;
1337                         }
1338 
1339                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1340 
1341                         if (!sk_wmem_schedule(sk, copy))
1342                                 goto wait_for_memory;
1343 
1344                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1345                                                        pfrag->page,
1346                                                        pfrag->offset,
1347                                                        copy);
1348                         if (err)
1349                                 goto do_error;
1350 
1351                         /* Update the skb. */
1352                         if (merge) {
1353                                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1354                         } else {
1355                                 skb_fill_page_desc(skb, i, pfrag->page,
1356                                                    pfrag->offset, copy);
1357                                 page_ref_inc(pfrag->page);
1358                         }
1359                         pfrag->offset += copy;
1360                 } else {
1361                         err = skb_zerocopy_iter_stream(sk, skb, msg, copy, uarg);
1362                         if (err == -EMSGSIZE || err == -EEXIST) {
1363                                 tcp_mark_push(tp, skb);
1364                                 goto new_segment;
1365                         }
1366                         if (err < 0)
1367                                 goto do_error;
1368                         copy = err;
1369                 }
1370 
1371                 if (!copied)
1372                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1373 
1374                 WRITE_ONCE(tp->write_seq, tp->write_seq + copy);
1375                 TCP_SKB_CB(skb)->end_seq += copy;
1376                 tcp_skb_pcount_set(skb, 0);
1377 
1378                 copied += copy;
1379                 if (!msg_data_left(msg)) {
1380                         if (unlikely(flags & MSG_EOR))
1381                                 TCP_SKB_CB(skb)->eor = 1;
1382                         goto out;
1383                 }
1384 
1385                 if (skb->len < size_goal || (flags & MSG_OOB) || unlikely(tp->repair))
1386                         continue;
1387 
1388                 if (forced_push(tp)) {
1389                         tcp_mark_push(tp, skb);
1390                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1391                 } else if (skb == tcp_send_head(sk))
1392                         tcp_push_one(sk, mss_now);
1393                 continue;
1394 
1395 wait_for_sndbuf:
1396                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1397 wait_for_memory:
1398                 if (copied)
1399                         tcp_push(sk, flags & ~MSG_MORE, mss_now,
1400                                  TCP_NAGLE_PUSH, size_goal);
1401 
1402                 err = sk_stream_wait_memory(sk, &timeo);
1403                 if (err != 0)
1404                         goto do_error;
1405 
1406                 mss_now = tcp_send_mss(sk, &size_goal, flags);
1407         }
1408 
1409 out:
1410         if (copied) {
1411                 tcp_tx_timestamp(sk, sockc.tsflags);
1412                 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1413         }
1414 out_nopush:
1415         sock_zerocopy_put(uarg);
1416         return copied + copied_syn;
1417 
1418 do_error:
1419         skb = tcp_write_queue_tail(sk);
1420 do_fault:
1421         tcp_remove_empty_skb(sk, skb);
1422 
1423         if (copied + copied_syn)
1424                 goto out;
1425 out_err:
1426         sock_zerocopy_put_abort(uarg, true);
1427         err = sk_stream_error(sk, flags, err);
1428         /* make sure we wake any epoll edge trigger waiter */
1429         if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) {
1430                 sk->sk_write_space(sk);
1431                 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1432         }
1433         return err;
1434 }
1435 EXPORT_SYMBOL_GPL(tcp_sendmsg_locked);
1436 
1437 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1438 {
1439         int ret;
1440 
1441         lock_sock(sk);
1442         ret = tcp_sendmsg_locked(sk, msg, size);
1443         release_sock(sk);
1444 
1445         return ret;
1446 }
1447 EXPORT_SYMBOL(tcp_sendmsg);
1448 
1449 /*
1450  *      Handle reading urgent data. BSD has very simple semantics for
1451  *      this, no blocking and very strange errors 8)
1452  */
1453 
1454 static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1455 {
1456         struct tcp_sock *tp = tcp_sk(sk);
1457 
1458         /* No URG data to read. */
1459         if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1460             tp->urg_data == TCP_URG_READ)
1461                 return -EINVAL; /* Yes this is right ! */
1462 
1463         if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1464                 return -ENOTCONN;
1465 
1466         if (tp->urg_data & TCP_URG_VALID) {
1467                 int err = 0;
1468                 char c = tp->urg_data;
1469 
1470                 if (!(flags & MSG_PEEK))
1471                         tp->urg_data = TCP_URG_READ;
1472 
1473                 /* Read urgent data. */
1474                 msg->msg_flags |= MSG_OOB;
1475 
1476                 if (len > 0) {
1477                         if (!(flags & MSG_TRUNC))
1478                                 err = memcpy_to_msg(msg, &c, 1);
1479                         len = 1;
1480                 } else
1481                         msg->msg_flags |= MSG_TRUNC;
1482 
1483                 return err ? -EFAULT : len;
1484         }
1485 
1486         if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1487                 return 0;
1488 
1489         /* Fixed the recv(..., MSG_OOB) behaviour.  BSD docs and
1490          * the available implementations agree in this case:
1491          * this call should never block, independent of the
1492          * blocking state of the socket.
1493          * Mike <pall@rz.uni-karlsruhe.de>
1494          */
1495         return -EAGAIN;
1496 }
1497 
1498 static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1499 {
1500         struct sk_buff *skb;
1501         int copied = 0, err = 0;
1502 
1503         /* XXX -- need to support SO_PEEK_OFF */
1504 
1505         skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {
1506                 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1507                 if (err)
1508                         return err;
1509                 copied += skb->len;
1510         }
1511 
1512         skb_queue_walk(&sk->sk_write_queue, skb) {
1513                 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1514                 if (err)
1515                         break;
1516 
1517                 copied += skb->len;
1518         }
1519 
1520         return err ?: copied;
1521 }
1522 
1523 /* Clean up the receive buffer for full frames taken by the user,
1524  * then send an ACK if necessary.  COPIED is the number of bytes
1525  * tcp_recvmsg has given to the user so far, it speeds up the
1526  * calculation of whether or not we must ACK for the sake of
1527  * a window update.
1528  */
1529 static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1530 {
1531         struct tcp_sock *tp = tcp_sk(sk);
1532         bool time_to_ack = false;
1533 
1534         struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1535 
1536         WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1537              "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1538              tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1539 
1540         if (inet_csk_ack_scheduled(sk)) {
1541                 const struct inet_connection_sock *icsk = inet_csk(sk);
1542                    /* Delayed ACKs frequently hit locked sockets during bulk
1543                     * receive. */
1544                 if (icsk->icsk_ack.blocked ||
1545                     /* Once-per-two-segments ACK was not sent by tcp_input.c */
1546                     tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1547                     /*
1548                      * If this read emptied read buffer, we send ACK, if
1549                      * connection is not bidirectional, user drained
1550                      * receive buffer and there was a small segment
1551                      * in queue.
1552                      */
1553                     (copied > 0 &&
1554                      ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1555                       ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1556                        !inet_csk_in_pingpong_mode(sk))) &&
1557                       !atomic_read(&sk->sk_rmem_alloc)))
1558                         time_to_ack = true;
1559         }
1560 
1561         /* We send an ACK if we can now advertise a non-zero window
1562          * which has been raised "significantly".
1563          *
1564          * Even if window raised up to infinity, do not send window open ACK
1565          * in states, where we will not receive more. It is useless.
1566          */
1567         if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1568                 __u32 rcv_window_now = tcp_receive_window(tp);
1569 
1570                 /* Optimize, __tcp_select_window() is not cheap. */
1571                 if (2*rcv_window_now <= tp->window_clamp) {
1572                         __u32 new_window = __tcp_select_window(sk);
1573 
1574                         /* Send ACK now, if this read freed lots of space
1575                          * in our buffer. Certainly, new_window is new window.
1576                          * We can advertise it now, if it is not less than current one.
1577                          * "Lots" means "at least twice" here.
1578                          */
1579                         if (new_window && new_window >= 2 * rcv_window_now)
1580                                 time_to_ack = true;
1581                 }
1582         }
1583         if (time_to_ack)
1584                 tcp_send_ack(sk);
1585 }
1586 
1587 static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1588 {
1589         struct sk_buff *skb;
1590         u32 offset;
1591 
1592         while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1593                 offset = seq - TCP_SKB_CB(skb)->seq;
1594                 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1595                         pr_err_once("%s: found a SYN, please report !\n", __func__);
1596                         offset--;
1597                 }
1598                 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1599                         *off = offset;
1600                         return skb;
1601                 }
1602                 /* This looks weird, but this can happen if TCP collapsing
1603                  * splitted a fat GRO packet, while we released socket lock
1604                  * in skb_splice_bits()
1605                  */
1606                 sk_eat_skb(sk, skb);
1607         }
1608         return NULL;
1609 }
1610 
1611 /*
1612  * This routine provides an alternative to tcp_recvmsg() for routines
1613  * that would like to handle copying from skbuffs directly in 'sendfile'
1614  * fashion.
1615  * Note:
1616  *      - It is assumed that the socket was locked by the caller.
1617  *      - The routine does not block.
1618  *      - At present, there is no support for reading OOB data
1619  *        or for 'peeking' the socket using this routine
1620  *        (although both would be easy to implement).
1621  */
1622 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1623                   sk_read_actor_t recv_actor)
1624 {
1625         struct sk_buff *skb;
1626         struct tcp_sock *tp = tcp_sk(sk);
1627         u32 seq = tp->copied_seq;
1628         u32 offset;
1629         int copied = 0;
1630 
1631         if (sk->sk_state == TCP_LISTEN)
1632                 return -ENOTCONN;
1633         while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1634                 if (offset < skb->len) {
1635                         int used;
1636                         size_t len;
1637 
1638                         len = skb->len - offset;
1639                         /* Stop reading if we hit a patch of urgent data */
1640                         if (tp->urg_data) {
1641                                 u32 urg_offset = tp->urg_seq - seq;
1642                                 if (urg_offset < len)
1643                                         len = urg_offset;
1644                                 if (!len)
1645                                         break;
1646                         }
1647                         used = recv_actor(desc, skb, offset, len);
1648                         if (used <= 0) {
1649                                 if (!copied)
1650                                         copied = used;
1651                                 break;
1652                         } else if (used <= len) {
1653                                 seq += used;
1654                                 copied += used;
1655                                 offset += used;
1656                         }
1657                         /* If recv_actor drops the lock (e.g. TCP splice
1658                          * receive) the skb pointer might be invalid when
1659                          * getting here: tcp_collapse might have deleted it
1660                          * while aggregating skbs from the socket queue.
1661                          */
1662                         skb = tcp_recv_skb(sk, seq - 1, &offset);
1663                         if (!skb)
1664                                 break;
1665                         /* TCP coalescing might have appended data to the skb.
1666                          * Try to splice more frags
1667                          */
1668                         if (offset + 1 != skb->len)
1669                                 continue;
1670                 }
1671                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
1672                         sk_eat_skb(sk, skb);
1673                         ++seq;
1674                         break;
1675                 }
1676                 sk_eat_skb(sk, skb);
1677                 if (!desc->count)
1678                         break;
1679                 WRITE_ONCE(tp->copied_seq, seq);
1680         }
1681         WRITE_ONCE(tp->copied_seq, seq);
1682 
1683         tcp_rcv_space_adjust(sk);
1684 
1685         /* Clean up data we have read: This will do ACK frames. */
1686         if (copied > 0) {
1687                 tcp_recv_skb(sk, seq, &offset);
1688                 tcp_cleanup_rbuf(sk, copied);
1689         }
1690         return copied;
1691 }
1692 EXPORT_SYMBOL(tcp_read_sock);
1693 
1694 int tcp_peek_len(struct socket *sock)
1695 {
1696         return tcp_inq(sock->sk);
1697 }
1698 EXPORT_SYMBOL(tcp_peek_len);
1699 
1700 /* Make sure sk_rcvbuf is big enough to satisfy SO_RCVLOWAT hint */
1701 int tcp_set_rcvlowat(struct sock *sk, int val)
1702 {
1703         int cap;
1704 
1705         if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1706                 cap = sk->sk_rcvbuf >> 1;
1707         else
1708                 cap = sock_net(sk)->ipv4.sysctl_tcp_rmem[2] >> 1;
1709         val = min(val, cap);
1710         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1711 
1712         /* Check if we need to signal EPOLLIN right now */
1713         tcp_data_ready(sk);
1714 
1715         if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1716                 return 0;
1717 
1718         val <<= 1;
1719         if (val > sk->sk_rcvbuf) {
1720                 WRITE_ONCE(sk->sk_rcvbuf, val);
1721                 tcp_sk(sk)->window_clamp = tcp_win_from_space(sk, val);
1722         }
1723         return 0;
1724 }
1725 EXPORT_SYMBOL(tcp_set_rcvlowat);
1726 
1727 #ifdef CONFIG_MMU
1728 static const struct vm_operations_struct tcp_vm_ops = {
1729 };
1730 
1731 int tcp_mmap(struct file *file, struct socket *sock,
1732              struct vm_area_struct *vma)
1733 {
1734         if (vma->vm_flags & (VM_WRITE | VM_EXEC))
1735                 return -EPERM;
1736         vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
1737 
1738         /* Instruct vm_insert_page() to not down_read(mmap_sem) */
1739         vma->vm_flags |= VM_MIXEDMAP;
1740 
1741         vma->vm_ops = &tcp_vm_ops;
1742         return 0;
1743 }
1744 EXPORT_SYMBOL(tcp_mmap);
1745 
1746 static int tcp_zerocopy_receive(struct sock *sk,
1747                                 struct tcp_zerocopy_receive *zc)
1748 {
1749         unsigned long address = (unsigned long)zc->address;
1750         const skb_frag_t *frags = NULL;
1751         u32 length = 0, seq, offset;
1752         struct vm_area_struct *vma;
1753         struct sk_buff *skb = NULL;
1754         struct tcp_sock *tp;
1755         int inq;
1756         int ret;
1757 
1758         if (address & (PAGE_SIZE - 1) || address != zc->address)
1759                 return -EINVAL;
1760 
1761         if (sk->sk_state == TCP_LISTEN)
1762                 return -ENOTCONN;
1763 
1764         sock_rps_record_flow(sk);
1765 
1766         down_read(&current->mm->mmap_sem);
1767 
1768         vma = find_vma(current->mm, address);
1769         if (!vma || vma->vm_start > address || vma->vm_ops != &tcp_vm_ops) {
1770                 up_read(&current->mm->mmap_sem);
1771                 return -EINVAL;
1772         }
1773         zc->length = min_t(unsigned long, zc->length, vma->vm_end - address);
1774 
1775         tp = tcp_sk(sk);
1776         seq = tp->copied_seq;
1777         inq = tcp_inq(sk);
1778         zc->length = min_t(u32, zc->length, inq);
1779         zc->length &= ~(PAGE_SIZE - 1);
1780         if (zc->length) {
1781                 zap_page_range(vma, address, zc->length);
1782                 zc->recv_skip_hint = 0;
1783         } else {
1784                 zc->recv_skip_hint = inq;
1785         }
1786         ret = 0;
1787         while (length + PAGE_SIZE <= zc->length) {
1788                 if (zc->recv_skip_hint < PAGE_SIZE) {
1789                         if (skb) {
1790                                 skb = skb->next;
1791                                 offset = seq - TCP_SKB_CB(skb)->seq;
1792                         } else {
1793                                 skb = tcp_recv_skb(sk, seq, &offset);
1794                         }
1795 
1796                         zc->recv_skip_hint = skb->len - offset;
1797                         offset -= skb_headlen(skb);
1798                         if ((int)offset < 0 || skb_has_frag_list(skb))
1799                                 break;
1800                         frags = skb_shinfo(skb)->frags;
1801                         while (offset) {
1802                                 if (skb_frag_size(frags) > offset)
1803                                         goto out;
1804                                 offset -= skb_frag_size(frags);
1805                                 frags++;
1806                         }
1807                 }
1808                 if (skb_frag_size(frags) != PAGE_SIZE || skb_frag_off(frags)) {
1809                         int remaining = zc->recv_skip_hint;
1810 
1811                         while (remaining && (skb_frag_size(frags) != PAGE_SIZE ||
1812                                              skb_frag_off(frags))) {
1813                                 remaining -= skb_frag_size(frags);
1814                                 frags++;
1815                         }
1816                         zc->recv_skip_hint -= remaining;
1817                         break;
1818                 }
1819                 ret = vm_insert_page(vma, address + length,
1820                                      skb_frag_page(frags));
1821                 if (ret)
1822                         break;
1823                 length += PAGE_SIZE;
1824                 seq += PAGE_SIZE;
1825                 zc->recv_skip_hint -= PAGE_SIZE;
1826                 frags++;
1827         }
1828 out:
1829         up_read(&current->mm->mmap_sem);
1830         if (length) {
1831                 WRITE_ONCE(tp->copied_seq, seq);
1832                 tcp_rcv_space_adjust(sk);
1833 
1834                 /* Clean up data we have read: This will do ACK frames. */
1835                 tcp_recv_skb(sk, seq, &offset);
1836                 tcp_cleanup_rbuf(sk, length);
1837                 ret = 0;
1838                 if (length == zc->length)
1839                         zc->recv_skip_hint = 0;
1840         } else {
1841                 if (!zc->recv_skip_hint && sock_flag(sk, SOCK_DONE))
1842                         ret = -EIO;
1843         }
1844         zc->length = length;
1845         return ret;
1846 }
1847 #endif
1848 
1849 static void tcp_update_recv_tstamps(struct sk_buff *skb,
1850                                     struct scm_timestamping_internal *tss)
1851 {
1852         if (skb->tstamp)
1853                 tss->ts[0] = ktime_to_timespec64(skb->tstamp);
1854         else
1855                 tss->ts[0] = (struct timespec64) {0};
1856 
1857         if (skb_hwtstamps(skb)->hwtstamp)
1858                 tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp);
1859         else
1860                 tss->ts[2] = (struct timespec64) {0};
1861 }
1862 
1863 /* Similar to __sock_recv_timestamp, but does not require an skb */
1864 static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
1865                                struct scm_timestamping_internal *tss)
1866 {
1867         int new_tstamp = sock_flag(sk, SOCK_TSTAMP_NEW);
1868         bool has_timestamping = false;
1869 
1870         if (tss->ts[0].tv_sec || tss->ts[0].tv_nsec) {
1871                 if (sock_flag(sk, SOCK_RCVTSTAMP)) {
1872                         if (sock_flag(sk, SOCK_RCVTSTAMPNS)) {
1873                                 if (new_tstamp) {
1874                                         struct __kernel_timespec kts = {tss->ts[0].tv_sec, tss->ts[0].tv_nsec};
1875 
1876                                         put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_NEW,
1877                                                  sizeof(kts), &kts);
1878                                 } else {
1879                                         struct timespec ts_old = timespec64_to_timespec(tss->ts[0]);
1880 
1881                                         put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
1882                                                  sizeof(ts_old), &ts_old);
1883                                 }
1884                         } else {
1885                                 if (new_tstamp) {
1886                                         struct __kernel_sock_timeval stv;
1887 
1888                                         stv.tv_sec = tss->ts[0].tv_sec;
1889                                         stv.tv_usec = tss->ts[0].tv_nsec / 1000;
1890                                         put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_NEW,
1891                                                  sizeof(stv), &stv);
1892                                 } else {
1893                                         struct __kernel_old_timeval tv;
1894 
1895                                         tv.tv_sec = tss->ts[0].tv_sec;
1896                                         tv.tv_usec = tss->ts[0].tv_nsec / 1000;
1897                                         put_cmsg(msg, SOL_SOCKET, SO_TIMESTAMP_OLD,
1898                                                  sizeof(tv), &tv);
1899                                 }
1900                         }
1901                 }
1902 
1903                 if (sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE)
1904                         has_timestamping = true;
1905                 else
1906                         tss->ts[0] = (struct timespec64) {0};
1907         }
1908 
1909         if (tss->ts[2].tv_sec || tss->ts[2].tv_nsec) {
1910                 if (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)
1911                         has_timestamping = true;
1912                 else
1913                         tss->ts[2] = (struct timespec64) {0};
1914         }
1915 
1916         if (has_timestamping) {
1917                 tss->ts[1] = (struct timespec64) {0};
1918                 if (sock_flag(sk, SOCK_TSTAMP_NEW))
1919                         put_cmsg_scm_timestamping64(msg, tss);
1920                 else
1921                         put_cmsg_scm_timestamping(msg, tss);
1922         }
1923 }
1924 
1925 static int tcp_inq_hint(struct sock *sk)
1926 {
1927         const struct tcp_sock *tp = tcp_sk(sk);
1928         u32 copied_seq = READ_ONCE(tp->copied_seq);
1929         u32 rcv_nxt = READ_ONCE(tp->rcv_nxt);
1930         int inq;
1931 
1932         inq = rcv_nxt - copied_seq;
1933         if (unlikely(inq < 0 || copied_seq != READ_ONCE(tp->copied_seq))) {
1934                 lock_sock(sk);
1935                 inq = tp->rcv_nxt - tp->copied_seq;
1936                 release_sock(sk);
1937         }
1938         /* After receiving a FIN, tell the user-space to continue reading
1939          * by returning a non-zero inq.
1940          */
1941         if (inq == 0 && sock_flag(sk, SOCK_DONE))
1942                 inq = 1;
1943         return inq;
1944 }
1945 
1946 /*
1947  *      This routine copies from a sock struct into the user buffer.
1948  *
1949  *      Technical note: in 2.3 we work on _locked_ socket, so that
1950  *      tricks with *seq access order and skb->users are not required.
1951  *      Probably, code can be easily improved even more.
1952  */
1953 
1954 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1955                 int flags, int *addr_len)
1956 {
1957         struct tcp_sock *tp = tcp_sk(sk);
1958         int copied = 0;
1959         u32 peek_seq;
1960         u32 *seq;
1961         unsigned long used;
1962         int err, inq;
1963         int target;             /* Read at least this many bytes */
1964         long timeo;
1965         struct sk_buff *skb, *last;
1966         u32 urg_hole = 0;
1967         struct scm_timestamping_internal tss;
1968         int cmsg_flags;
1969 
1970         if (unlikely(flags & MSG_ERRQUEUE))
1971                 return inet_recv_error(sk, msg, len, addr_len);
1972 
1973         if (sk_can_busy_loop(sk) && skb_queue_empty_lockless(&sk->sk_receive_queue) &&
1974             (sk->sk_state == TCP_ESTABLISHED))
1975                 sk_busy_loop(sk, nonblock);
1976 
1977         lock_sock(sk);
1978 
1979         err = -ENOTCONN;
1980         if (sk->sk_state == TCP_LISTEN)
1981                 goto out;
1982 
1983         cmsg_flags = tp->recvmsg_inq ? 1 : 0;
1984         timeo = sock_rcvtimeo(sk, nonblock);
1985 
1986         /* Urgent data needs to be handled specially. */
1987         if (flags & MSG_OOB)
1988                 goto recv_urg;
1989 
1990         if (unlikely(tp->repair)) {
1991                 err = -EPERM;
1992                 if (!(flags & MSG_PEEK))
1993                         goto out;
1994 
1995                 if (tp->repair_queue == TCP_SEND_QUEUE)
1996                         goto recv_sndq;
1997 
1998                 err = -EINVAL;
1999                 if (tp->repair_queue == TCP_NO_QUEUE)
2000                         goto out;
2001 
2002                 /* 'common' recv queue MSG_PEEK-ing */
2003         }
2004 
2005         seq = &tp->copied_seq;
2006         if (flags & MSG_PEEK) {
2007                 peek_seq = tp->copied_seq;
2008                 seq = &peek_seq;
2009         }
2010 
2011         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2012 
2013         do {
2014                 u32 offset;
2015 
2016                 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
2017                 if (tp->urg_data && tp->urg_seq == *seq) {
2018                         if (copied)
2019                                 break;
2020                         if (signal_pending(current)) {
2021                                 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
2022                                 break;
2023                         }
2024                 }
2025 
2026                 /* Next get a buffer. */
2027 
2028                 last = skb_peek_tail(&sk->sk_receive_queue);
2029                 skb_queue_walk(&sk->sk_receive_queue, skb) {
2030                         last = skb;
2031                         /* Now that we have two receive queues this
2032                          * shouldn't happen.
2033                          */
2034                         if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
2035                                  "TCP recvmsg seq # bug: copied %X, seq %X, rcvnxt %X, fl %X\n",
2036                                  *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
2037                                  flags))
2038                                 break;
2039 
2040                         offset = *seq - TCP_SKB_CB(skb)->seq;
2041                         if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
2042                                 pr_err_once("%s: found a SYN, please report !\n", __func__);
2043                                 offset--;
2044                         }
2045                         if (offset < skb->len)
2046                                 goto found_ok_skb;
2047                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2048                                 goto found_fin_ok;
2049                         WARN(!(flags & MSG_PEEK),
2050                              "TCP recvmsg seq # bug 2: copied %X, seq %X, rcvnxt %X, fl %X\n",
2051                              *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
2052                 }
2053 
2054                 /* Well, if we have backlog, try to process it now yet. */
2055 
2056                 if (copied >= target && !sk->sk_backlog.tail)
2057                         break;
2058 
2059                 if (copied) {
2060                         if (sk->sk_err ||
2061                             sk->sk_state == TCP_CLOSE ||
2062                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
2063                             !timeo ||
2064                             signal_pending(current))
2065                                 break;
2066                 } else {
2067                         if (sock_flag(sk, SOCK_DONE))
2068                                 break;
2069 
2070                         if (sk->sk_err) {
2071                                 copied = sock_error(sk);
2072                                 break;
2073                         }
2074 
2075                         if (sk->sk_shutdown & RCV_SHUTDOWN)
2076                                 break;
2077 
2078                         if (sk->sk_state == TCP_CLOSE) {
2079                                 /* This occurs when user tries to read
2080                                  * from never connected socket.
2081                                  */
2082                                 copied = -ENOTCONN;
2083                                 break;
2084                         }
2085 
2086                         if (!timeo) {
2087                                 copied = -EAGAIN;
2088                                 break;
2089                         }
2090 
2091                         if (signal_pending(current)) {
2092                                 copied = sock_intr_errno(timeo);
2093                                 break;
2094                         }
2095                 }
2096 
2097                 tcp_cleanup_rbuf(sk, copied);
2098 
2099                 if (copied >= target) {
2100                         /* Do not sleep, just process backlog. */
2101                         release_sock(sk);
2102                         lock_sock(sk);
2103                 } else {
2104                         sk_wait_data(sk, &timeo, last);
2105                 }
2106 
2107                 if ((flags & MSG_PEEK) &&
2108                     (peek_seq - copied - urg_hole != tp->copied_seq)) {
2109                         net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
2110                                             current->comm,
2111                                             task_pid_nr(current));
2112                         peek_seq = tp->copied_seq;
2113                 }
2114                 continue;
2115 
2116 found_ok_skb:
2117                 /* Ok so how much can we use? */
2118                 used = skb->len - offset;
2119                 if (len < used)
2120                         used = len;
2121 
2122                 /* Do we have urgent data here? */
2123                 if (tp->urg_data) {
2124                         u32 urg_offset = tp->urg_seq - *seq;
2125                         if (urg_offset < used) {
2126                                 if (!urg_offset) {
2127                                         if (!sock_flag(sk, SOCK_URGINLINE)) {
2128                                                 WRITE_ONCE(*seq, *seq + 1);
2129                                                 urg_hole++;
2130                                                 offset++;
2131                                                 used--;
2132                                                 if (!used)
2133                                                         goto skip_copy;
2134                                         }
2135                                 } else
2136                                         used = urg_offset;
2137                         }
2138                 }
2139 
2140                 if (!(flags & MSG_TRUNC)) {
2141                         err = skb_copy_datagram_msg(skb, offset, msg, used);
2142                         if (err) {
2143                                 /* Exception. Bailout! */
2144                                 if (!copied)
2145                                         copied = -EFAULT;
2146                                 break;
2147                         }
2148                 }
2149 
2150                 WRITE_ONCE(*seq, *seq + used);
2151                 copied += used;
2152                 len -= used;
2153 
2154                 tcp_rcv_space_adjust(sk);
2155 
2156 skip_copy:
2157                 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
2158                         tp->urg_data = 0;
2159                         tcp_fast_path_check(sk);
2160                 }
2161 
2162                 if (TCP_SKB_CB(skb)->has_rxtstamp) {
2163                         tcp_update_recv_tstamps(skb, &tss);
2164                         cmsg_flags |= 2;
2165                 }
2166 
2167                 if (used + offset < skb->len)
2168                         continue;
2169 
2170                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2171                         goto found_fin_ok;
2172                 if (!(flags & MSG_PEEK))
2173                         sk_eat_skb(sk, skb);
2174                 continue;
2175 
2176 found_fin_ok:
2177                 /* Process the FIN. */
2178                 WRITE_ONCE(*seq, *seq + 1);
2179                 if (!(flags & MSG_PEEK))
2180                         sk_eat_skb(sk, skb);
2181                 break;
2182         } while (len > 0);
2183 
2184         /* According to UNIX98, msg_name/msg_namelen are ignored
2185          * on connected socket. I was just happy when found this 8) --ANK
2186          */
2187 
2188         /* Clean up data we have read: This will do ACK frames. */
2189         tcp_cleanup_rbuf(sk, copied);
2190 
2191         release_sock(sk);
2192 
2193         if (cmsg_flags) {
2194                 if (cmsg_flags & 2)
2195                         tcp_recv_timestamp(msg, sk, &tss);
2196                 if (cmsg_flags & 1) {
2197                         inq = tcp_inq_hint(sk);
2198                         put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2199                 }
2200         }
2201 
2202         return copied;
2203 
2204 out:
2205         release_sock(sk);
2206         return err;
2207 
2208 recv_urg:
2209         err = tcp_recv_urg(sk, msg, len, flags);
2210         goto out;
2211 
2212 recv_sndq:
2213         err = tcp_peek_sndq(sk, msg, len);
2214         goto out;
2215 }
2216 EXPORT_SYMBOL(tcp_recvmsg);
2217 
2218 void tcp_set_state(struct sock *sk, int state)
2219 {
2220         int oldstate = sk->sk_state;
2221 
2222         /* We defined a new enum for TCP states that are exported in BPF
2223          * so as not force the internal TCP states to be frozen. The
2224          * following checks will detect if an internal state value ever
2225          * differs from the BPF value. If this ever happens, then we will
2226          * need to remap the internal value to the BPF value before calling
2227          * tcp_call_bpf_2arg.
2228          */
2229         BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED);
2230         BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT);
2231         BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV);
2232         BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1);
2233         BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2);
2234         BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT);
2235         BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
2236         BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT);
2237         BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK);
2238         BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN);
2239         BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING);
2240         BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV);
2241         BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES);
2242 
2243         if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG))
2244                 tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state);
2245 
2246         switch (state) {
2247         case TCP_ESTABLISHED:
2248                 if (oldstate != TCP_ESTABLISHED)
2249                         TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2250                 break;
2251 
2252         case TCP_CLOSE:
2253                 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
2254                         TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
2255 
2256                 sk->sk_prot->unhash(sk);
2257                 if (inet_csk(sk)->icsk_bind_hash &&
2258                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2259                         inet_put_port(sk);
2260                 /* fall through */
2261         default:
2262                 if (oldstate == TCP_ESTABLISHED)
2263                         TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2264         }
2265 
2266         /* Change state AFTER socket is unhashed to avoid closed
2267          * socket sitting in hash tables.
2268          */
2269         inet_sk_state_store(sk, state);
2270 }
2271 EXPORT_SYMBOL_GPL(tcp_set_state);
2272 
2273 /*
2274  *      State processing on a close. This implements the state shift for
2275  *      sending our FIN frame. Note that we only send a FIN for some
2276  *      states. A shutdown() may have already sent the FIN, or we may be
2277  *      closed.
2278  */
2279 
2280 static const unsigned char new_state[16] = {
2281   /* current state:        new state:      action:      */
2282   [0 /* (Invalid) */]   = TCP_CLOSE,
2283   [TCP_ESTABLISHED]     = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2284   [TCP_SYN_SENT]        = TCP_CLOSE,
2285   [TCP_SYN_RECV]        = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2286   [TCP_FIN_WAIT1]       = TCP_FIN_WAIT1,
2287   [TCP_FIN_WAIT2]       = TCP_FIN_WAIT2,
2288   [TCP_TIME_WAIT]       = TCP_CLOSE,
2289   [TCP_CLOSE]           = TCP_CLOSE,
2290   [TCP_CLOSE_WAIT]      = TCP_LAST_ACK  | TCP_ACTION_FIN,
2291   [TCP_LAST_ACK]        = TCP_LAST_ACK,
2292   [TCP_LISTEN]          = TCP_CLOSE,
2293   [TCP_CLOSING]         = TCP_CLOSING,
2294   [TCP_NEW_SYN_RECV]    = TCP_CLOSE,    /* should not happen ! */
2295 };
2296 
2297 static int tcp_close_state(struct sock *sk)
2298 {
2299         int next = (int)new_state[sk->sk_state];
2300         int ns = next & TCP_STATE_MASK;
2301 
2302         tcp_set_state(sk, ns);
2303 
2304         return next & TCP_ACTION_FIN;
2305 }
2306 
2307 /*
2308  *      Shutdown the sending side of a connection. Much like close except
2309  *      that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
2310  */
2311 
2312 void tcp_shutdown(struct sock *sk, int how)
2313 {
2314         /*      We need to grab some memory, and put together a FIN,
2315          *      and then put it into the queue to be sent.
2316          *              Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2317          */
2318         if (!(how & SEND_SHUTDOWN))
2319                 return;
2320 
2321         /* If we've already sent a FIN, or it's a closed state, skip this. */
2322         if ((1 << sk->sk_state) &
2323             (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2324              TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2325                 /* Clear out any half completed packets.  FIN if needed. */
2326                 if (tcp_close_state(sk))
2327                         tcp_send_fin(sk);
2328         }
2329 }
2330 EXPORT_SYMBOL(tcp_shutdown);
2331 
2332 bool tcp_check_oom(struct sock *sk, int shift)
2333 {
2334         bool too_many_orphans, out_of_socket_memory;
2335 
2336         too_many_orphans = tcp_too_many_orphans(sk, shift);
2337         out_of_socket_memory = tcp_out_of_memory(sk);
2338 
2339         if (too_many_orphans)
2340                 net_info_ratelimited("too many orphaned sockets\n");
2341         if (out_of_socket_memory)
2342                 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
2343         return too_many_orphans || out_of_socket_memory;
2344 }
2345 
2346 void tcp_close(struct sock *sk, long timeout)
2347 {
2348         struct sk_buff *skb;
2349         int data_was_unread = 0;
2350         int state;
2351 
2352         lock_sock(sk);
2353         sk->sk_shutdown = SHUTDOWN_MASK;
2354 
2355         if (sk->sk_state == TCP_LISTEN) {
2356                 tcp_set_state(sk, TCP_CLOSE);
2357 
2358                 /* Special case. */
2359                 inet_csk_listen_stop(sk);
2360 
2361                 goto adjudge_to_death;
2362         }
2363 
2364         /*  We need to flush the recv. buffs.  We do this only on the
2365          *  descriptor close, not protocol-sourced closes, because the
2366          *  reader process may not have drained the data yet!
2367          */
2368         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2369                 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2370 
2371                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2372                         len--;
2373                 data_was_unread += len;
2374                 __kfree_skb(skb);
2375         }
2376 
2377         sk_mem_reclaim(sk);
2378 
2379         /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2380         if (sk->sk_state == TCP_CLOSE)
2381                 goto adjudge_to_death;
2382 
2383         /* As outlined in RFC 2525, section 2.17, we send a RST here because
2384          * data was lost. To witness the awful effects of the old behavior of
2385          * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2386          * GET in an FTP client, suspend the process, wait for the client to
2387          * advertise a zero window, then kill -9 the FTP client, wheee...
2388          * Note: timeout is always zero in such a case.
2389          */
2390         if (unlikely(tcp_sk(sk)->repair)) {
2391                 sk->sk_prot->disconnect(sk, 0);
2392         } else if (data_was_unread) {
2393                 /* Unread data was tossed, zap the connection. */
2394                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
2395                 tcp_set_state(sk, TCP_CLOSE);
2396                 tcp_send_active_reset(sk, sk->sk_allocation);
2397         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2398                 /* Check zero linger _after_ checking for unread data. */
2399                 sk->sk_prot->disconnect(sk, 0);
2400                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
2401         } else if (tcp_close_state(sk)) {
2402                 /* We FIN if the application ate all the data before
2403                  * zapping the connection.
2404                  */
2405 
2406                 /* RED-PEN. Formally speaking, we have broken TCP state
2407                  * machine. State transitions:
2408                  *
2409                  * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2410                  * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2411                  * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2412                  *
2413                  * are legal only when FIN has been sent (i.e. in window),
2414                  * rather than queued out of window. Purists blame.
2415                  *
2416                  * F.e. "RFC state" is ESTABLISHED,
2417                  * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2418                  *
2419                  * The visible declinations are that sometimes
2420                  * we enter time-wait state, when it is not required really
2421                  * (harmless), do not send active resets, when they are
2422                  * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2423                  * they look as CLOSING or LAST_ACK for Linux)
2424                  * Probably, I missed some more holelets.
2425                  *                                              --ANK
2426                  * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2427                  * in a single packet! (May consider it later but will
2428                  * probably need API support or TCP_CORK SYN-ACK until
2429                  * data is written and socket is closed.)
2430                  */
2431                 tcp_send_fin(sk);
2432         }
2433 
2434         sk_stream_wait_close(sk, timeout);
2435 
2436 adjudge_to_death:
2437         state = sk->sk_state;
2438         sock_hold(sk);
2439         sock_orphan(sk);
2440 
2441         local_bh_disable();
2442         bh_lock_sock(sk);
2443         /* remove backlog if any, without releasing ownership. */
2444         __release_sock(sk);
2445 
2446         percpu_counter_inc(sk->sk_prot->orphan_count);
2447 
2448         /* Have we already been destroyed by a softirq or backlog? */
2449         if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2450                 goto out;
2451 
2452         /*      This is a (useful) BSD violating of the RFC. There is a
2453          *      problem with TCP as specified in that the other end could
2454          *      keep a socket open forever with no application left this end.
2455          *      We use a 1 minute timeout (about the same as BSD) then kill
2456          *      our end. If they send after that then tough - BUT: long enough
2457          *      that we won't make the old 4*rto = almost no time - whoops
2458          *      reset mistake.
2459          *
2460          *      Nope, it was not mistake. It is really desired behaviour
2461          *      f.e. on http servers, when such sockets are useless, but
2462          *      consume significant resources. Let's do it with special
2463          *      linger2 option.                                 --ANK
2464          */
2465 
2466         if (sk->sk_state == TCP_FIN_WAIT2) {
2467                 struct tcp_sock *tp = tcp_sk(sk);
2468                 if (tp->linger2 < 0) {
2469                         tcp_set_state(sk, TCP_CLOSE);
2470                         tcp_send_active_reset(sk, GFP_ATOMIC);
2471                         __NET_INC_STATS(sock_net(sk),
2472                                         LINUX_MIB_TCPABORTONLINGER);
2473                 } else {
2474                         const int tmo = tcp_fin_time(sk);
2475 
2476                         if (tmo > TCP_TIMEWAIT_LEN) {
2477                                 inet_csk_reset_keepalive_timer(sk,
2478                                                 tmo - TCP_TIMEWAIT_LEN);
2479                         } else {
2480                                 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2481                                 goto out;
2482                         }
2483                 }
2484         }
2485         if (sk->sk_state != TCP_CLOSE) {
2486                 sk_mem_reclaim(sk);
2487                 if (tcp_check_oom(sk, 0)) {
2488                         tcp_set_state(sk, TCP_CLOSE);
2489                         tcp_send_active_reset(sk, GFP_ATOMIC);
2490                         __NET_INC_STATS(sock_net(sk),
2491                                         LINUX_MIB_TCPABORTONMEMORY);
2492                 } else if (!check_net(sock_net(sk))) {
2493                         /* Not possible to send reset; just close */
2494                         tcp_set_state(sk, TCP_CLOSE);
2495                 }
2496         }
2497 
2498         if (sk->sk_state == TCP_CLOSE) {
2499                 struct request_sock *req;
2500 
2501                 req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk,
2502                                                 lockdep_sock_is_held(sk));
2503                 /* We could get here with a non-NULL req if the socket is
2504                  * aborted (e.g., closed with unread data) before 3WHS
2505                  * finishes.
2506                  */
2507                 if (req)
2508                         reqsk_fastopen_remove(sk, req, false);
2509                 inet_csk_destroy_sock(sk);
2510         }
2511         /* Otherwise, socket is reprieved until protocol close. */
2512 
2513 out:
2514         bh_unlock_sock(sk);
2515         local_bh_enable();
2516         release_sock(sk);
2517         sock_put(sk);
2518 }
2519 EXPORT_SYMBOL(tcp_close);
2520 
2521 /* These states need RST on ABORT according to RFC793 */
2522 
2523 static inline bool tcp_need_reset(int state)
2524 {
2525         return (1 << state) &
2526                (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2527                 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2528 }
2529 
2530 static void tcp_rtx_queue_purge(struct sock *sk)
2531 {
2532         struct rb_node *p = rb_first(&sk->tcp_rtx_queue);
2533 
2534         tcp_sk(sk)->highest_sack = NULL;
2535         while (p) {
2536                 struct sk_buff *skb = rb_to_skb(p);
2537 
2538                 p = rb_next(p);
2539                 /* Since we are deleting whole queue, no need to
2540                  * list_del(&skb->tcp_tsorted_anchor)
2541                  */
2542                 tcp_rtx_queue_unlink(skb, sk);
2543                 sk_wmem_free_skb(sk, skb);
2544         }
2545 }
2546 
2547 void tcp_write_queue_purge(struct sock *sk)
2548 {
2549         struct sk_buff *skb;
2550 
2551         tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
2552         while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
2553                 tcp_skb_tsorted_anchor_cleanup(skb);
2554                 sk_wmem_free_skb(sk, skb);
2555         }
2556         tcp_rtx_queue_purge(sk);
2557         skb = sk->sk_tx_skb_cache;
2558         if (skb) {
2559                 __kfree_skb(skb);
2560                 sk->sk_tx_skb_cache = NULL;
2561         }
2562         INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue);
2563         sk_mem_reclaim(sk);
2564         tcp_clear_all_retrans_hints(tcp_sk(sk));
2565         tcp_sk(sk)->packets_out = 0;
2566         inet_csk(sk)->icsk_backoff = 0;
2567 }
2568 
2569 int tcp_disconnect(struct sock *sk, int flags)
2570 {
2571         struct inet_sock *inet = inet_sk(sk);
2572         struct inet_connection_sock *icsk = inet_csk(sk);
2573         struct tcp_sock *tp = tcp_sk(sk);
2574         int old_state = sk->sk_state;
2575         u32 seq;
2576 
2577         if (old_state != TCP_CLOSE)
2578                 tcp_set_state(sk, TCP_CLOSE);
2579 
2580         /* ABORT function of RFC793 */
2581         if (old_state == TCP_LISTEN) {
2582                 inet_csk_listen_stop(sk);
2583         } else if (unlikely(tp->repair)) {
2584                 sk->sk_err = ECONNABORTED;
2585         } else if (tcp_need_reset(old_state) ||
2586                    (tp->snd_nxt != tp->write_seq &&
2587                     (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2588                 /* The last check adjusts for discrepancy of Linux wrt. RFC
2589                  * states
2590                  */
2591                 tcp_send_active_reset(sk, gfp_any());
2592                 sk->sk_err = ECONNRESET;
2593         } else if (old_state == TCP_SYN_SENT)
2594                 sk->sk_err = ECONNRESET;
2595 
2596         tcp_clear_xmit_timers(sk);
2597         __skb_queue_purge(&sk->sk_receive_queue);
2598         if (sk->sk_rx_skb_cache) {
2599                 __kfree_skb(sk->sk_rx_skb_cache);
2600                 sk->sk_rx_skb_cache = NULL;
2601         }
2602         WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);
2603         tp->urg_data = 0;
2604         tcp_write_queue_purge(sk);
2605         tcp_fastopen_active_disable_ofo_check(sk);
2606         skb_rbtree_purge(&tp->out_of_order_queue);
2607 
2608         inet->inet_dport = 0;
2609 
2610         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2611                 inet_reset_saddr(sk);
2612 
2613         sk->sk_shutdown = 0;
2614         sock_reset_flag(sk, SOCK_DONE);
2615         tp->srtt_us = 0;
2616         tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
2617         tp->rcv_rtt_last_tsecr = 0;
2618 
2619         seq = tp->write_seq + tp->max_window + 2;
2620         if (!seq)
2621                 seq = 1;
2622         WRITE_ONCE(tp->write_seq, seq);
2623 
2624         icsk->icsk_backoff = 0;
2625         tp->snd_cwnd = 2;
2626         icsk->icsk_probes_out = 0;
2627         icsk->icsk_rto = TCP_TIMEOUT_INIT;
2628         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2629         tp->snd_cwnd = TCP_INIT_CWND;
2630         tp->snd_cwnd_cnt = 0;
2631         tp->window_clamp = 0;
2632         tp->delivered = 0;
2633         tp->delivered_ce = 0;
2634         tcp_set_ca_state(sk, TCP_CA_Open);
2635         tp->is_sack_reneg = 0;
2636         tcp_clear_retrans(tp);
2637         tp->total_retrans = 0;
2638         inet_csk_delack_init(sk);
2639         /* Initialize rcv_mss to TCP_MIN_MSS to avoid division by 0
2640          * issue in __tcp_select_window()
2641          */
2642         icsk->icsk_ack.rcv_mss = TCP_MIN_MSS;
2643         memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2644         __sk_dst_reset(sk);
2645         dst_release(sk->sk_rx_dst);
2646         sk->sk_rx_dst = NULL;
2647         tcp_saved_syn_free(tp);
2648         tp->compressed_ack = 0;
2649         tp->segs_in = 0;
2650         tp->segs_out = 0;
2651         tp->bytes_sent = 0;
2652         tp->bytes_acked = 0;
2653         tp->bytes_received = 0;
2654         tp->bytes_retrans = 0;
2655         tp->data_segs_in = 0;
2656         tp->data_segs_out = 0;
2657         tp->duplicate_sack[0].start_seq = 0;
2658         tp->duplicate_sack[0].end_seq = 0;
2659         tp->dsack_dups = 0;
2660         tp->reord_seen = 0;
2661         tp->retrans_out = 0;
2662         tp->sacked_out = 0;
2663         tp->tlp_high_seq = 0;
2664         tp->last_oow_ack_time = 0;
2665         /* There's a bubble in the pipe until at least the first ACK. */
2666         tp->app_limited = ~0U;
2667         tp->rack.mstamp = 0;
2668         tp->rack.advanced = 0;
2669         tp->rack.reo_wnd_steps = 1;
2670         tp->rack.last_delivered = 0;
2671         tp->rack.reo_wnd_persist = 0;
2672         tp->rack.dsack_seen = 0;
2673         tp->syn_data_acked = 0;
2674         tp->rx_opt.saw_tstamp = 0;
2675         tp->rx_opt.dsack = 0;
2676         tp->rx_opt.num_sacks = 0;
2677         tp->rcv_ooopack = 0;
2678 
2679 
2680         /* Clean up fastopen related fields */
2681         tcp_free_fastopen_req(tp);
2682         inet->defer_connect = 0;
2683 
2684         WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
2685 
2686         if (sk->sk_frag.page) {
2687                 put_page(sk->sk_frag.page);
2688                 sk->sk_frag.page = NULL;
2689                 sk->sk_frag.offset = 0;
2690         }
2691 
2692         sk->sk_error_report(sk);
2693         return 0;
2694 }
2695 EXPORT_SYMBOL(tcp_disconnect);
2696 
2697 static inline bool tcp_can_repair_sock(const struct sock *sk)
2698 {
2699         return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
2700                 (sk->sk_state != TCP_LISTEN);
2701 }
2702 
2703 static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2704 {
2705         struct tcp_repair_window opt;
2706 
2707         if (!tp->repair)
2708                 return -EPERM;
2709 
2710         if (len != sizeof(opt))
2711                 return -EINVAL;
2712 
2713         if (copy_from_user(&opt, optbuf, sizeof(opt)))
2714                 return -EFAULT;
2715 
2716         if (opt.max_window < opt.snd_wnd)
2717                 return -EINVAL;
2718 
2719         if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2720                 return -EINVAL;
2721 
2722         if (after(opt.rcv_wup, tp->rcv_nxt))
2723                 return -EINVAL;
2724 
2725         tp->snd_wl1     = opt.snd_wl1;
2726         tp->snd_wnd     = opt.snd_wnd;
2727         tp->max_window  = opt.max_window;
2728 
2729         tp->rcv_wnd     = opt.rcv_wnd;
2730         tp->rcv_wup     = opt.rcv_wup;
2731 
2732         return 0;
2733 }
2734 
2735 static int tcp_repair_options_est(struct sock *sk,
2736                 struct tcp_repair_opt __user *optbuf, unsigned int len)
2737 {
2738         struct tcp_sock *tp = tcp_sk(sk);
2739         struct tcp_repair_opt opt;
2740 
2741         while (len >= sizeof(opt)) {
2742                 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2743                         return -EFAULT;
2744 
2745                 optbuf++;
2746                 len -= sizeof(opt);
2747 
2748                 switch (opt.opt_code) {
2749                 case TCPOPT_MSS:
2750                         tp->rx_opt.mss_clamp = opt.opt_val;
2751                         tcp_mtup_init(sk);
2752                         break;
2753                 case TCPOPT_WINDOW:
2754                         {
2755                                 u16 snd_wscale = opt.opt_val & 0xFFFF;
2756                                 u16 rcv_wscale = opt.opt_val >> 16;
2757 
2758                                 if (snd_wscale > TCP_MAX_WSCALE || rcv_wscale > TCP_MAX_WSCALE)
2759                                         return -EFBIG;
2760 
2761                                 tp->rx_opt.snd_wscale = snd_wscale;
2762                                 tp->rx_opt.rcv_wscale = rcv_wscale;
2763                                 tp->rx_opt.wscale_ok = 1;
2764                         }
2765                         break;
2766                 case TCPOPT_SACK_PERM:
2767                         if (opt.opt_val != 0)
2768                                 return -EINVAL;
2769 
2770                         tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2771                         break;
2772                 case TCPOPT_TIMESTAMP:
2773                         if (opt.opt_val != 0)
2774                                 return -EINVAL;
2775 
2776                         tp->rx_opt.tstamp_ok = 1;
2777                         break;
2778                 }
2779         }
2780 
2781         return 0;
2782 }
2783 
2784 DEFINE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2785 EXPORT_SYMBOL(tcp_tx_delay_enabled);
2786 
2787 static void tcp_enable_tx_delay(void)
2788 {
2789         if (!static_branch_unlikely(&tcp_tx_delay_enabled)) {
2790                 static int __tcp_tx_delay_enabled = 0;
2791 
2792                 if (cmpxchg(&__tcp_tx_delay_enabled, 0, 1) == 0) {
2793                         static_branch_enable(&tcp_tx_delay_enabled);
2794                         pr_info("TCP_TX_DELAY enabled\n");
2795                 }
2796         }
2797 }
2798 
2799 /*
2800  *      Socket option code for TCP.
2801  */
2802 static int do_tcp_setsockopt(struct sock *sk, int level,
2803                 int optname, char __user *optval, unsigned int optlen)
2804 {
2805         struct tcp_sock *tp = tcp_sk(sk);
2806         struct inet_connection_sock *icsk = inet_csk(sk);
2807         struct net *net = sock_net(sk);
2808         int val;
2809         int err = 0;
2810 
2811         /* These are data/string values, all the others are ints */
2812         switch (optname) {
2813         case TCP_CONGESTION: {
2814                 char name[TCP_CA_NAME_MAX];
2815 
2816                 if (optlen < 1)
2817                         return -EINVAL;
2818 
2819                 val = strncpy_from_user(name, optval,
2820                                         min_t(long, TCP_CA_NAME_MAX-1, optlen));
2821                 if (val < 0)
2822                         return -EFAULT;
2823                 name[val] = 0;
2824 
2825                 lock_sock(sk);
2826                 err = tcp_set_congestion_control(sk, name, true, true,
2827                                                  ns_capable(sock_net(sk)->user_ns,
2828                                                             CAP_NET_ADMIN));
2829                 release_sock(sk);
2830                 return err;
2831         }
2832         case TCP_ULP: {
2833                 char name[TCP_ULP_NAME_MAX];
2834 
2835                 if (optlen < 1)
2836                         return -EINVAL;
2837 
2838                 val = strncpy_from_user(name, optval,
2839                                         min_t(long, TCP_ULP_NAME_MAX - 1,
2840                                               optlen));
2841                 if (val < 0)
2842                         return -EFAULT;
2843                 name[val] = 0;
2844 
2845                 lock_sock(sk);
2846                 err = tcp_set_ulp(sk, name);
2847                 release_sock(sk);
2848                 return err;
2849         }
2850         case TCP_FASTOPEN_KEY: {
2851                 __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH];
2852                 __u8 *backup_key = NULL;
2853 
2854                 /* Allow a backup key as well to facilitate key rotation
2855                  * First key is the active one.
2856                  */
2857                 if (optlen != TCP_FASTOPEN_KEY_LENGTH &&
2858                     optlen != TCP_FASTOPEN_KEY_BUF_LENGTH)
2859                         return -EINVAL;
2860 
2861                 if (copy_from_user(key, optval, optlen))
2862                         return -EFAULT;
2863 
2864                 if (optlen == TCP_FASTOPEN_KEY_BUF_LENGTH)
2865                         backup_key = key + TCP_FASTOPEN_KEY_LENGTH;
2866 
2867                 return tcp_fastopen_reset_cipher(net, sk, key, backup_key);
2868         }
2869         default:
2870                 /* fallthru */
2871                 break;
2872         }
2873 
2874         if (optlen < sizeof(int))
2875                 return -EINVAL;
2876 
2877         if (get_user(val, (int __user *)optval))
2878                 return -EFAULT;
2879 
2880         lock_sock(sk);
2881 
2882         switch (optname) {
2883         case TCP_MAXSEG:
2884                 /* Values greater than interface MTU won't take effect. However
2885                  * at the point when this call is done we typically don't yet
2886                  * know which interface is going to be used
2887                  */
2888                 if (val && (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW)) {
2889                         err = -EINVAL;
2890                         break;
2891                 }
2892                 tp->rx_opt.user_mss = val;
2893                 break;
2894 
2895         case TCP_NODELAY:
2896                 if (val) {
2897                         /* TCP_NODELAY is weaker than TCP_CORK, so that
2898                          * this option on corked socket is remembered, but
2899                          * it is not activated until cork is cleared.
2900                          *
2901                          * However, when TCP_NODELAY is set we make
2902                          * an explicit push, which overrides even TCP_CORK
2903                          * for currently queued segments.
2904                          */
2905                         tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2906                         tcp_push_pending_frames(sk);
2907                 } else {
2908                         tp->nonagle &= ~TCP_NAGLE_OFF;
2909                 }
2910                 break;
2911 
2912         case TCP_THIN_LINEAR_TIMEOUTS:
2913                 if (val < 0 || val > 1)
2914                         err = -EINVAL;
2915                 else
2916                         tp->thin_lto = val;
2917                 break;
2918 
2919         case TCP_THIN_DUPACK:
2920                 if (val < 0 || val > 1)
2921                         err = -EINVAL;
2922                 break;
2923 
2924         case TCP_REPAIR:
2925                 if (!tcp_can_repair_sock(sk))
2926                         err = -EPERM;
2927                 else if (val == TCP_REPAIR_ON) {
2928                         tp->repair = 1;
2929                         sk->sk_reuse = SK_FORCE_REUSE;
2930                         tp->repair_queue = TCP_NO_QUEUE;
2931                 } else if (val == TCP_REPAIR_OFF) {
2932                         tp->repair = 0;
2933                         sk->sk_reuse = SK_NO_REUSE;
2934                         tcp_send_window_probe(sk);
2935                 } else if (val == TCP_REPAIR_OFF_NO_WP) {
2936                         tp->repair = 0;
2937                         sk->sk_reuse = SK_NO_REUSE;
2938                 } else
2939                         err = -EINVAL;
2940 
2941                 break;
2942 
2943         case TCP_REPAIR_QUEUE:
2944                 if (!tp->repair)
2945                         err = -EPERM;
2946                 else if ((unsigned int)val < TCP_QUEUES_NR)
2947                         tp->repair_queue = val;
2948                 else
2949                         err = -EINVAL;
2950                 break;
2951 
2952         case TCP_QUEUE_SEQ:
2953                 if (sk->sk_state != TCP_CLOSE)
2954                         err = -EPERM;
2955                 else if (tp->repair_queue == TCP_SEND_QUEUE)
2956                         WRITE_ONCE(tp->write_seq, val);
2957                 else if (tp->repair_queue == TCP_RECV_QUEUE) {
2958                         WRITE_ONCE(tp->rcv_nxt, val);
2959                         WRITE_ONCE(tp->copied_seq, val);
2960                 }
2961                 else
2962                         err = -EINVAL;
2963                 break;
2964 
2965         case TCP_REPAIR_OPTIONS:
2966                 if (!tp->repair)
2967                         err = -EINVAL;
2968                 else if (sk->sk_state == TCP_ESTABLISHED)
2969                         err = tcp_repair_options_est(sk,
2970                                         (struct tcp_repair_opt __user *)optval,
2971                                         optlen);
2972                 else
2973                         err = -EPERM;
2974                 break;
2975 
2976         case TCP_CORK:
2977                 /* When set indicates to always queue non-full frames.
2978                  * Later the user clears this option and we transmit
2979                  * any pending partial frames in the queue.  This is
2980                  * meant to be used alongside sendfile() to get properly
2981                  * filled frames when the user (for example) must write
2982                  * out headers with a write() call first and then use
2983                  * sendfile to send out the data parts.
2984                  *
2985                  * TCP_CORK can be set together with TCP_NODELAY and it is
2986                  * stronger than TCP_NODELAY.
2987                  */
2988                 if (val) {
2989                         tp->nonagle |= TCP_NAGLE_CORK;
2990                 } else {
2991                         tp->nonagle &= ~TCP_NAGLE_CORK;
2992                         if (tp->nonagle&TCP_NAGLE_OFF)
2993                                 tp->nonagle |= TCP_NAGLE_PUSH;
2994                         tcp_push_pending_frames(sk);
2995                 }
2996                 break;
2997 
2998         case TCP_KEEPIDLE:
2999                 if (val < 1 || val > MAX_TCP_KEEPIDLE)
3000                         err = -EINVAL;
3001                 else {
3002                         tp->keepalive_time = val * HZ;
3003                         if (sock_flag(sk, SOCK_KEEPOPEN) &&
3004                             !((1 << sk->sk_state) &
3005                               (TCPF_CLOSE | TCPF_LISTEN))) {
3006                                 u32 elapsed = keepalive_time_elapsed(tp);
3007                                 if (tp->keepalive_time > elapsed)
3008                                         elapsed = tp->keepalive_time - elapsed;
3009                                 else
3010                                         elapsed = 0;
3011                                 inet_csk_reset_keepalive_timer(sk, elapsed);
3012                         }
3013                 }
3014                 break;
3015         case TCP_KEEPINTVL:
3016                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
3017                         err = -EINVAL;
3018                 else
3019                         tp->keepalive_intvl = val * HZ;
3020                 break;
3021         case TCP_KEEPCNT:
3022                 if (val < 1 || val > MAX_TCP_KEEPCNT)
3023                         err = -EINVAL;
3024                 else
3025                         tp->keepalive_probes = val;
3026                 break;
3027         case TCP_SYNCNT:
3028                 if (val < 1 || val > MAX_TCP_SYNCNT)
3029                         err = -EINVAL;
3030                 else
3031                         icsk->icsk_syn_retries = val;
3032                 break;
3033 
3034         case TCP_SAVE_SYN:
3035                 if (val < 0 || val > 1)
3036                         err = -EINVAL;
3037                 else
3038                         tp->save_syn = val;
3039                 break;
3040 
3041         case TCP_LINGER2:
3042                 if (val < 0)
3043                         tp->linger2 = -1;
3044                 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
3045                         tp->linger2 = 0;
3046                 else
3047                         tp->linger2 = val * HZ;
3048                 break;
3049 
3050         case TCP_DEFER_ACCEPT:
3051                 /* Translate value in seconds to number of retransmits */
3052                 icsk->icsk_accept_queue.rskq_defer_accept =
3053                         secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
3054                                         TCP_RTO_MAX / HZ);
3055                 break;
3056 
3057         case TCP_WINDOW_CLAMP:
3058                 if (!val) {
3059                         if (sk->sk_state != TCP_CLOSE) {
3060                                 err = -EINVAL;
3061                                 break;
3062                         }
3063                         tp->window_clamp = 0;
3064                 } else
3065                         tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
3066                                                 SOCK_MIN_RCVBUF / 2 : val;
3067                 break;
3068 
3069         case TCP_QUICKACK:
3070                 if (!val) {
3071                         inet_csk_enter_pingpong_mode(sk);
3072                 } else {
3073                         inet_csk_exit_pingpong_mode(sk);
3074                         if ((1 << sk->sk_state) &
3075                             (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
3076                             inet_csk_ack_scheduled(sk)) {
3077                                 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
3078                                 tcp_cleanup_rbuf(sk, 1);
3079                                 if (!(val & 1))
3080                                         inet_csk_enter_pingpong_mode(sk);
3081                         }
3082                 }
3083                 break;
3084 
3085 #ifdef CONFIG_TCP_MD5SIG
3086         case TCP_MD5SIG:
3087         case TCP_MD5SIG_EXT:
3088                 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
3089                         err = tp->af_specific->md5_parse(sk, optname, optval, optlen);
3090                 else
3091                         err = -EINVAL;
3092                 break;
3093 #endif
3094         case TCP_USER_TIMEOUT:
3095                 /* Cap the max time in ms TCP will retry or probe the window
3096                  * before giving up and aborting (ETIMEDOUT) a connection.
3097                  */
3098                 if (val < 0)
3099                         err = -EINVAL;
3100                 else
3101                         icsk->icsk_user_timeout = val;
3102                 break;
3103 
3104         case TCP_FASTOPEN:
3105                 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
3106                     TCPF_LISTEN))) {
3107                         tcp_fastopen_init_key_once(net);
3108 
3109                         fastopen_queue_tune(sk, val);
3110                 } else {
3111                         err = -EINVAL;
3112                 }
3113                 break;
3114         case TCP_FASTOPEN_CONNECT:
3115                 if (val > 1 || val < 0) {
3116                         err = -EINVAL;
3117                 } else if (net->ipv4.sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
3118                         if (sk->sk_state == TCP_CLOSE)
3119                                 tp->fastopen_connect = val;
3120                         else
3121                                 err = -EINVAL;
3122                 } else {
3123                         err = -EOPNOTSUPP;
3124                 }
3125                 break;
3126         case TCP_FASTOPEN_NO_COOKIE:
3127                 if (val > 1 || val < 0)
3128                         err = -EINVAL;
3129                 else if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3130                         err = -EINVAL;
3131                 else
3132                         tp->fastopen_no_cookie = val;
3133                 break;
3134         case TCP_TIMESTAMP:
3135                 if (!tp->repair)
3136                         err = -EPERM;
3137                 else
3138                         tp->tsoffset = val - tcp_time_stamp_raw();
3139                 break;
3140         case TCP_REPAIR_WINDOW:
3141                 err = tcp_repair_set_window(tp, optval, optlen);
3142                 break;
3143         case TCP_NOTSENT_LOWAT:
3144                 tp->notsent_lowat = val;
3145                 sk->sk_write_space(sk);
3146                 break;
3147         case TCP_INQ:
3148                 if (val > 1 || val < 0)
3149                         err = -EINVAL;
3150                 else
3151                         tp->recvmsg_inq = val;
3152                 break;
3153         case TCP_TX_DELAY:
3154                 if (val)
3155                         tcp_enable_tx_delay();
3156                 tp->tcp_tx_delay = val;
3157                 break;
3158         default:
3159                 err = -ENOPROTOOPT;
3160                 break;
3161         }
3162 
3163         release_sock(sk);
3164         return err;
3165 }
3166 
3167 int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
3168                    unsigned int optlen)
3169 {
3170         const struct inet_connection_sock *icsk = inet_csk(sk);
3171 
3172         if (level != SOL_TCP)
3173                 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
3174                                                      optval, optlen);
3175         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3176 }
3177 EXPORT_SYMBOL(tcp_setsockopt);
3178 
3179 #ifdef CONFIG_COMPAT
3180 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
3181                           char __user *optval, unsigned int optlen)
3182 {
3183         if (level != SOL_TCP)
3184                 return inet_csk_compat_setsockopt(sk, level, optname,
3185                                                   optval, optlen);
3186         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
3187 }
3188 EXPORT_SYMBOL(compat_tcp_setsockopt);
3189 #endif
3190 
3191 static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
3192                                       struct tcp_info *info)
3193 {
3194         u64 stats[__TCP_CHRONO_MAX], total = 0;
3195         enum tcp_chrono i;
3196 
3197         for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
3198                 stats[i] = tp->chrono_stat[i - 1];
3199                 if (i == tp->chrono_type)
3200                         stats[i] += tcp_jiffies32 - tp->chrono_start;
3201                 stats[i] *= USEC_PER_SEC / HZ;
3202                 total += stats[i];
3203         }
3204 
3205         info->tcpi_busy_time = total;
3206         info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
3207         info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
3208 }
3209 
3210 /* Return information about state of tcp endpoint in API format. */
3211 void tcp_get_info(struct sock *sk, struct tcp_info *info)
3212 {
3213         const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
3214         const struct inet_connection_sock *icsk = inet_csk(sk);
3215         unsigned long rate;
3216         u32 now;
3217         u64 rate64;
3218         bool slow;
3219 
3220         memset(info, 0, sizeof(*info));
3221         if (sk->sk_type != SOCK_STREAM)
3222                 return;
3223 
3224         info->tcpi_state = inet_sk_state_load(sk);
3225 
3226         /* Report meaningful fields for all TCP states, including listeners */
3227         rate = READ_ONCE(sk->sk_pacing_rate);
3228         rate64 = (rate != ~0UL) ? rate : ~0ULL;
3229         info->tcpi_pacing_rate = rate64;
3230 
3231         rate = READ_ONCE(sk->sk_max_pacing_rate);
3232         rate64 = (rate != ~0UL) ? rate : ~0ULL;
3233         info->tcpi_max_pacing_rate = rate64;
3234 
3235         info->tcpi_reordering = tp->reordering;
3236         info->tcpi_snd_cwnd = tp->snd_cwnd;
3237 
3238         if (info->tcpi_state == TCP_LISTEN) {
3239                 /* listeners aliased fields :
3240                  * tcpi_unacked -> Number of children ready for accept()
3241                  * tcpi_sacked  -> max backlog
3242                  */
3243                 info->tcpi_unacked = sk->sk_ack_backlog;
3244                 info->tcpi_sacked = sk->sk_max_ack_backlog;
3245                 return;
3246         }
3247 
3248         slow = lock_sock_fast(sk);
3249 
3250         info->tcpi_ca_state = icsk->icsk_ca_state;
3251         info->tcpi_retransmits = icsk->icsk_retransmits;
3252         info->tcpi_probes = icsk->icsk_probes_out;
3253         info->tcpi_backoff = icsk->icsk_backoff;
3254 
3255         if (tp->rx_opt.tstamp_ok)
3256                 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
3257         if (tcp_is_sack(tp))
3258                 info->tcpi_options |= TCPI_OPT_SACK;
3259         if (tp->rx_opt.wscale_ok) {
3260                 info->tcpi_options |= TCPI_OPT_WSCALE;
3261                 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
3262                 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
3263         }
3264 
3265         if (tp->ecn_flags & TCP_ECN_OK)
3266                 info->tcpi_options |= TCPI_OPT_ECN;
3267         if (tp->ecn_flags & TCP_ECN_SEEN)
3268                 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
3269         if (tp->syn_data_acked)
3270                 info->tcpi_options |= TCPI_OPT_SYN_DATA;
3271 
3272         info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
3273         info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
3274         info->tcpi_snd_mss = tp->mss_cache;
3275         info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
3276 
3277         info->tcpi_unacked = tp->packets_out;
3278         info->tcpi_sacked = tp->sacked_out;
3279 
3280         info->tcpi_lost = tp->lost_out;
3281         info->tcpi_retrans = tp->retrans_out;
3282 
3283         now = tcp_jiffies32;
3284         info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
3285         info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
3286         info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
3287 
3288         info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
3289         info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
3290         info->tcpi_rtt = tp->srtt_us >> 3;
3291         info->tcpi_rttvar = tp->mdev_us >> 2;
3292         info->tcpi_snd_ssthresh = tp->snd_ssthresh;
3293         info->tcpi_advmss = tp->advmss;
3294 
3295         info->tcpi_rcv_rtt = tp->rcv_rtt_est.rtt_us >> 3;
3296         info->tcpi_rcv_space = tp->rcvq_space.space;
3297 
3298         info->tcpi_total_retrans = tp->total_retrans;
3299 
3300         info->tcpi_bytes_acked = tp->bytes_acked;
3301         info->tcpi_bytes_received = tp->bytes_received;
3302         info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt);
3303         tcp_get_info_chrono_stats(tp, info);
3304 
3305         info->tcpi_segs_out = tp->segs_out;
3306         info->tcpi_segs_in = tp->segs_in;
3307 
3308         info->tcpi_min_rtt = tcp_min_rtt(tp);
3309         info->tcpi_data_segs_in = tp->data_segs_in;
3310         info->tcpi_data_segs_out = tp->data_segs_out;
3311 
3312         info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
3313         rate64 = tcp_compute_delivery_rate(tp);
3314         if (rate64)
3315                 info->tcpi_delivery_rate = rate64;
3316         info->tcpi_delivered = tp->delivered;
3317         info->tcpi_delivered_ce = tp->delivered_ce;
3318         info->tcpi_bytes_sent = tp->bytes_sent;
3319         info->tcpi_bytes_retrans = tp->bytes_retrans;
3320         info->tcpi_dsack_dups = tp->dsack_dups;
3321         info->tcpi_reord_seen = tp->reord_seen;
3322         info->tcpi_rcv_ooopack = tp->rcv_ooopack;
3323         info->tcpi_snd_wnd = tp->snd_wnd;
3324         unlock_sock_fast(sk, slow);
3325 }
3326 EXPORT_SYMBOL_GPL(tcp_get_info);
3327 
3328 static size_t tcp_opt_stats_get_size(void)
3329 {
3330         return
3331                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BUSY */
3332                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_RWND_LIMITED */
3333                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_SNDBUF_LIMITED */
3334                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DATA_SEGS_OUT */
3335                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_TOTAL_RETRANS */
3336                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_PACING_RATE */
3337                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_DELIVERY_RATE */
3338                 nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_CWND */
3339                 nla_total_size(sizeof(u32)) + /* TCP_NLA_REORDERING */
3340                 nla_total_size(sizeof(u32)) + /* TCP_NLA_MIN_RTT */
3341                 nla_total_size(sizeof(u8)) + /* TCP_NLA_RECUR_RETRANS */
3342                 nla_total_size(sizeof(u8)) + /* TCP_NLA_DELIVERY_RATE_APP_LMT */
3343                 nla_total_size(sizeof(u32)) + /* TCP_NLA_SNDQ_SIZE */
3344                 nla_total_size(sizeof(u8)) + /* TCP_NLA_CA_STATE */
3345                 nla_total_size(sizeof(u32)) + /* TCP_NLA_SND_SSTHRESH */
3346                 nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED */
3347                 nla_total_size(sizeof(u32)) + /* TCP_NLA_DELIVERED_CE */
3348                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_SENT */
3349                 nla_total_size_64bit(sizeof(u64)) + /* TCP_NLA_BYTES_RETRANS */
3350                 nla_total_size(sizeof(u32)) + /* TCP_NLA_DSACK_DUPS */
3351                 nla_total_size(sizeof(u32)) + /* TCP_NLA_REORD_SEEN */
3352                 nla_total_size(sizeof(u32)) + /* TCP_NLA_SRTT */
3353                 0;
3354 }
3355 
3356 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
3357 {
3358         const struct tcp_sock *tp = tcp_sk(sk);
3359         struct sk_buff *stats;
3360         struct tcp_info info;
3361         unsigned long rate;
3362         u64 rate64;
3363 
3364         stats = alloc_skb(tcp_opt_stats_get_size(), GFP_ATOMIC);
3365         if (!stats)
3366                 return NULL;
3367 
3368         tcp_get_info_chrono_stats(tp, &info);
3369         nla_put_u64_64bit(stats, TCP_NLA_BUSY,
3370                           info.tcpi_busy_time, TCP_NLA_PAD);
3371         nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED,
3372                           info.tcpi_rwnd_limited, TCP_NLA_PAD);
3373         nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED,
3374                           info.tcpi_sndbuf_limited, TCP_NLA_PAD);
3375         nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT,
3376                           tp->data_segs_out, TCP_NLA_PAD);
3377         nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS,
3378                           tp->total_retrans, TCP_NLA_PAD);
3379 
3380         rate = READ_ONCE(sk->sk_pacing_rate);
3381         rate64 = (rate != ~0UL) ? rate : ~0ULL;
3382         nla_put_u64_64bit(stats, TCP_NLA_PACING_RATE, rate64, TCP_NLA_PAD);
3383 
3384         rate64 = tcp_compute_delivery_rate(tp);
3385         nla_put_u64_64bit(stats, TCP_NLA_DELIVERY_RATE, rate64, TCP_NLA_PAD);
3386 
3387         nla_put_u32(stats, TCP_NLA_SND_CWND, tp->snd_cwnd);
3388         nla_put_u32(stats, TCP_NLA_REORDERING, tp->reordering);
3389         nla_put_u32(stats, TCP_NLA_MIN_RTT, tcp_min_rtt(tp));
3390 
3391         nla_put_u8(stats, TCP_NLA_RECUR_RETRANS, inet_csk(sk)->icsk_retransmits);
3392         nla_put_u8(stats, TCP_NLA_DELIVERY_RATE_APP_LMT, !!tp->rate_app_limited);
3393         nla_put_u32(stats, TCP_NLA_SND_SSTHRESH, tp->snd_ssthresh);
3394         nla_put_u32(stats, TCP_NLA_DELIVERED, tp->delivered);
3395         nla_put_u32(stats, TCP_NLA_DELIVERED_CE, tp->delivered_ce);
3396 
3397         nla_put_u32(stats, TCP_NLA_SNDQ_SIZE, tp->write_seq - tp->snd_una);
3398         nla_put_u8(stats, TCP_NLA_CA_STATE, inet_csk(sk)->icsk_ca_state);
3399 
3400         nla_put_u64_64bit(stats, TCP_NLA_BYTES_SENT, tp->bytes_sent,
3401                           TCP_NLA_PAD);
3402         nla_put_u64_64bit(stats, TCP_NLA_BYTES_RETRANS, tp->bytes_retrans,
3403                           TCP_NLA_PAD);
3404         nla_put_u32(stats, TCP_NLA_DSACK_DUPS, tp->dsack_dups);
3405         nla_put_u32(stats, TCP_NLA_REORD_SEEN, tp->reord_seen);
3406         nla_put_u32(stats, TCP_NLA_SRTT, tp->srtt_us >> 3);
3407 
3408         return stats;
3409 }
3410 
3411 static int do_tcp_getsockopt(struct sock *sk, int level,
3412                 int optname, char __user *optval, int __user *optlen)
3413 {
3414         struct inet_connection_sock *icsk = inet_csk(sk);
3415         struct tcp_sock *tp = tcp_sk(sk);
3416         struct net *net = sock_net(sk);
3417         int val, len;
3418 
3419         if (get_user(len, optlen))
3420                 return -EFAULT;
3421 
3422         len = min_t(unsigned int, len, sizeof(int));
3423 
3424         if (len < 0)
3425                 return -EINVAL;
3426 
3427         switch (optname) {
3428         case TCP_MAXSEG:
3429                 val = tp->mss_cache;
3430                 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
3431                         val = tp->rx_opt.user_mss;
3432                 if (tp->repair)
3433                         val = tp->rx_opt.mss_clamp;
3434                 break;
3435         case TCP_NODELAY:
3436                 val = !!(tp->nonagle&TCP_NAGLE_OFF);
3437                 break;
3438         case TCP_CORK:
3439                 val = !!(tp->nonagle&TCP_NAGLE_CORK);
3440                 break;
3441         case TCP_KEEPIDLE:
3442                 val = keepalive_time_when(tp) / HZ;
3443                 break;
3444         case TCP_KEEPINTVL:
3445                 val = keepalive_intvl_when(tp) / HZ;
3446                 break;
3447         case TCP_KEEPCNT:
3448                 val = keepalive_probes(tp);
3449                 break;
3450         case TCP_SYNCNT:
3451                 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
3452                 break;
3453         case TCP_LINGER2:
3454                 val = tp->linger2;
3455                 if (val >= 0)
3456                         val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
3457                 break;
3458         case TCP_DEFER_ACCEPT:
3459                 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
3460                                       TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
3461                 break;
3462         case TCP_WINDOW_CLAMP:
3463                 val = tp->window_clamp;
3464                 break;
3465         case TCP_INFO: {
3466                 struct tcp_info info;
3467 
3468                 if (get_user(len, optlen))
3469                         return -EFAULT;
3470 
3471                 tcp_get_info(sk, &info);
3472 
3473                 len = min_t(unsigned int, len, sizeof(info));
3474                 if (put_user(len, optlen))
3475                         return -EFAULT;
3476                 if (copy_to_user(optval, &info, len))
3477                         return -EFAULT;
3478                 return 0;
3479         }
3480         case TCP_CC_INFO: {
3481                 const struct tcp_congestion_ops *ca_ops;
3482                 union tcp_cc_info info;
3483                 size_t sz = 0;
3484                 int attr;
3485 
3486                 if (get_user(len, optlen))
3487                         return -EFAULT;
3488 
3489                 ca_ops = icsk->icsk_ca_ops;
3490                 if (ca_ops && ca_ops->get_info)
3491                         sz = ca_ops->get_info(sk, ~0U, &attr, &info);
3492 
3493                 len = min_t(unsigned int, len, sz);
3494                 if (put_user(len, optlen))
3495                         return -EFAULT;
3496                 if (copy_to_user(optval, &info, len))
3497                         return -EFAULT;
3498                 return 0;
3499         }
3500         case TCP_QUICKACK:
3501                 val = !inet_csk_in_pingpong_mode(sk);
3502                 break;
3503 
3504         case TCP_CONGESTION:
3505                 if (get_user(len, optlen))
3506                         return -EFAULT;
3507                 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
3508                 if (put_user(len, optlen))
3509                         return -EFAULT;
3510                 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
3511                         return -EFAULT;
3512                 return 0;
3513 
3514         case TCP_ULP:
3515                 if (get_user(len, optlen))
3516                         return -EFAULT;
3517                 len = min_t(unsigned int, len, TCP_ULP_NAME_MAX);
3518                 if (!icsk->icsk_ulp_ops) {
3519                         if (put_user(0, optlen))
3520                                 return -EFAULT;
3521                         return 0;
3522                 }
3523                 if (put_user(len, optlen))
3524                         return -EFAULT;
3525                 if (copy_to_user(optval, icsk->icsk_ulp_ops->name, len))
3526                         return -EFAULT;
3527                 return 0;
3528 
3529         case TCP_FASTOPEN_KEY: {
3530                 __u8 key[TCP_FASTOPEN_KEY_BUF_LENGTH];
3531                 struct tcp_fastopen_context *ctx;
3532                 unsigned int key_len = 0;
3533 
3534                 if (get_user(len, optlen))
3535                         return -EFAULT;
3536 
3537                 rcu_read_lock();
3538                 ctx = rcu_dereference(icsk->icsk_accept_queue.fastopenq.ctx);
3539                 if (ctx) {
3540                         key_len = tcp_fastopen_context_len(ctx) *
3541                                         TCP_FASTOPEN_KEY_LENGTH;
3542                         memcpy(&key[0], &ctx->key[0], key_len);
3543                 }
3544                 rcu_read_unlock();
3545 
3546                 len = min_t(unsigned int, len, key_len);
3547                 if (put_user(len, optlen))
3548                         return -EFAULT;
3549                 if (copy_to_user(optval, key, len))
3550                         return -EFAULT;
3551                 return 0;
3552         }
3553         case TCP_THIN_LINEAR_TIMEOUTS:
3554                 val = tp->thin_lto;
3555                 break;
3556 
3557         case TCP_THIN_DUPACK:
3558                 val = 0;
3559                 break;
3560 
3561         case TCP_REPAIR:
3562                 val = tp->repair;
3563                 break;
3564 
3565         case TCP_REPAIR_QUEUE:
3566                 if (tp->repair)
3567                         val = tp->repair_queue;
3568                 else
3569                         return -EINVAL;
3570                 break;
3571 
3572         case TCP_REPAIR_WINDOW: {
3573                 struct tcp_repair_window opt;
3574 
3575                 if (get_user(len, optlen))
3576                         return -EFAULT;
3577 
3578                 if (len != sizeof(opt))
3579                         return -EINVAL;
3580 
3581                 if (!tp->repair)
3582                         return -EPERM;
3583 
3584                 opt.snd_wl1     = tp->snd_wl1;
3585                 opt.snd_wnd     = tp->snd_wnd;
3586                 opt.max_window  = tp->max_window;
3587                 opt.rcv_wnd     = tp->rcv_wnd;
3588                 opt.rcv_wup     = tp->rcv_wup;
3589 
3590                 if (copy_to_user(optval, &opt, len))
3591                         return -EFAULT;
3592                 return 0;
3593         }
3594         case TCP_QUEUE_SEQ:
3595                 if (tp->repair_queue == TCP_SEND_QUEUE)
3596                         val = tp->write_seq;
3597                 else if (tp->repair_queue == TCP_RECV_QUEUE)
3598                         val = tp->rcv_nxt;
3599                 else
3600                         return -EINVAL;
3601                 break;
3602 
3603         case TCP_USER_TIMEOUT:
3604                 val = icsk->icsk_user_timeout;
3605                 break;
3606 
3607         case TCP_FASTOPEN:
3608                 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
3609                 break;
3610 
3611         case TCP_FASTOPEN_CONNECT:
3612                 val = tp->fastopen_connect;
3613                 break;
3614 
3615         case TCP_FASTOPEN_NO_COOKIE:
3616                 val = tp->fastopen_no_cookie;
3617                 break;
3618 
3619         case TCP_TX_DELAY:
3620                 val = tp->tcp_tx_delay;
3621                 break;
3622 
3623         case TCP_TIMESTAMP:
3624                 val = tcp_time_stamp_raw() + tp->tsoffset;
3625                 break;
3626         case TCP_NOTSENT_LOWAT:
3627                 val = tp->notsent_lowat;
3628                 break;
3629         case TCP_INQ:
3630                 val = tp->recvmsg_inq;
3631                 break;
3632         case TCP_SAVE_SYN:
3633                 val = tp->save_syn;
3634                 break;
3635         case TCP_SAVED_SYN: {
3636                 if (get_user(len, optlen))
3637                         return -EFAULT;
3638 
3639                 lock_sock(sk);
3640                 if (tp->saved_syn) {
3641                         if (len < tp->saved_syn[0]) {
3642                                 if (put_user(tp->saved_syn[0], optlen)) {
3643                                         release_sock(sk);
3644                                         return -EFAULT;
3645                                 }
3646                                 release_sock(sk);
3647                                 return -EINVAL;
3648                         }
3649                         len = tp->saved_syn[0];
3650                         if (put_user(len, optlen)) {
3651                                 release_sock(sk);
3652                                 return -EFAULT;
3653                         }
3654                         if (copy_to_user(optval, tp->saved_syn + 1, len)) {
3655                                 release_sock(sk);
3656                                 return -EFAULT;
3657                         }
3658                         tcp_saved_syn_free(tp);
3659                         release_sock(sk);
3660                 } else {
3661                         release_sock(sk);
3662                         len = 0;
3663                         if (put_user(len, optlen))
3664                                 return -EFAULT;
3665                 }
3666                 return 0;
3667         }
3668 #ifdef CONFIG_MMU
3669         case TCP_ZEROCOPY_RECEIVE: {
3670                 struct tcp_zerocopy_receive zc;
3671                 int err;
3672 
3673                 if (get_user(len, optlen))
3674                         return -EFAULT;
3675                 if (len != sizeof(zc))
3676                         return -EINVAL;
3677                 if (copy_from_user(&zc, optval, len))
3678                         return -EFAULT;
3679                 lock_sock(sk);
3680                 err = tcp_zerocopy_receive(sk, &zc);
3681                 release_sock(sk);
3682                 if (!err && copy_to_user(optval, &zc, len))
3683                         err = -EFAULT;
3684                 return err;
3685         }
3686 #endif
3687         default:
3688                 return -ENOPROTOOPT;
3689         }
3690 
3691         if (put_user(len, optlen))
3692                 return -EFAULT;
3693         if (copy_to_user(optval, &val, len))
3694                 return -EFAULT;
3695         return 0;
3696 }
3697 
3698 int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
3699                    int __user *optlen)
3700 {
3701         struct inet_connection_sock *icsk = inet_csk(sk);
3702 
3703         if (level != SOL_TCP)
3704                 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
3705                                                      optval, optlen);
3706         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3707 }
3708 EXPORT_SYMBOL(tcp_getsockopt);
3709 
3710 #ifdef CONFIG_COMPAT
3711 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
3712                           char __user *optval, int __user *optlen)
3713 {
3714         if (level != SOL_TCP)
3715                 return inet_csk_compat_getsockopt(sk, level, optname,
3716                                                   optval, optlen);
3717         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3718 }
3719 EXPORT_SYMBOL(compat_tcp_getsockopt);
3720 #endif
3721 
3722 #ifdef CONFIG_TCP_MD5SIG
3723 static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
3724 static DEFINE_MUTEX(tcp_md5sig_mutex);
3725 static bool tcp_md5sig_pool_populated = false;
3726 
3727 static void __tcp_alloc_md5sig_pool(void)
3728 {
3729         struct crypto_ahash *hash;
3730         int cpu;
3731 
3732         hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
3733         if (IS_ERR(hash))
3734                 return;
3735 
3736         for_each_possible_cpu(cpu) {
3737                 void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch;
3738                 struct ahash_request *req;
3739 
3740                 if (!scratch) {
3741                         scratch = kmalloc_node(sizeof(union tcp_md5sum_block) +
3742                                                sizeof(struct tcphdr),
3743                                                GFP_KERNEL,
3744                                                cpu_to_node(cpu));
3745                         if (!scratch)
3746                                 return;
3747                         per_cpu(tcp_md5sig_pool, cpu).scratch = scratch;
3748                 }
3749                 if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
3750                         continue;
3751 
3752                 req = ahash_request_alloc(hash, GFP_KERNEL);
3753                 if (!req)
3754                         return;
3755 
3756                 ahash_request_set_callback(req, 0, NULL, NULL);
3757 
3758                 per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
3759         }
3760         /* before setting tcp_md5sig_pool_populated, we must commit all writes
3761          * to memory. See smp_rmb() in tcp_get_md5sig_pool()
3762          */
3763         smp_wmb();
3764         tcp_md5sig_pool_populated = true;
3765 }
3766 
3767 bool tcp_alloc_md5sig_pool(void)
3768 {
3769         if (unlikely(!tcp_md5sig_pool_populated)) {
3770                 mutex_lock(&tcp_md5sig_mutex);
3771 
3772                 if (!tcp_md5sig_pool_populated) {
3773                         __tcp_alloc_md5sig_pool();
3774                         if (tcp_md5sig_pool_populated)
3775                                 static_branch_inc(&tcp_md5_needed);
3776                 }
3777 
3778                 mutex_unlock(&tcp_md5sig_mutex);
3779         }
3780         return tcp_md5sig_pool_populated;
3781 }
3782 EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3783 
3784 
3785 /**
3786  *      tcp_get_md5sig_pool - get md5sig_pool for this user
3787  *
3788  *      We use percpu structure, so if we succeed, we exit with preemption
3789  *      and BH disabled, to make sure another thread or softirq handling
3790  *      wont try to get same context.
3791  */
3792 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
3793 {
3794         local_bh_disable();
3795 
3796         if (tcp_md5sig_pool_populated) {
3797                 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3798                 smp_rmb();
3799                 return this_cpu_ptr(&tcp_md5sig_pool);
3800         }
3801         local_bh_enable();
3802         return NULL;
3803 }
3804 EXPORT_SYMBOL(tcp_get_md5sig_pool);
3805 
3806 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
3807                           const struct sk_buff *skb, unsigned int header_len)
3808 {
3809         struct scatterlist sg;
3810         const struct tcphdr *tp = tcp_hdr(skb);
3811         struct ahash_request *req = hp->md5_req;
3812         unsigned int i;
3813         const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3814                                            skb_headlen(skb) - header_len : 0;
3815         const struct skb_shared_info *shi = skb_shinfo(skb);
3816         struct sk_buff *frag_iter;
3817 
3818         sg_init_table(&sg, 1);
3819 
3820         sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3821         ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3822         if (crypto_ahash_update(req))
3823                 return 1;
3824 
3825         for (i = 0; i < shi->nr_frags; ++i) {
3826                 const skb_frag_t *f = &shi->frags[i];
3827                 unsigned int offset = skb_frag_off(f);
3828                 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3829 
3830                 sg_set_page(&sg, page, skb_frag_size(f),
3831                             offset_in_page(offset));
3832                 ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3833                 if (crypto_ahash_update(req))
3834                         return 1;
3835         }
3836 
3837         skb_walk_frags(skb, frag_iter)
3838                 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3839                         return 1;
3840 
3841         return 0;
3842 }
3843 EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3844 
3845 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
3846 {
3847         struct scatterlist sg;
3848 
3849         sg_init_one(&sg, key->key, key->keylen);
3850         ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3851         return crypto_ahash_update(hp->md5_req);
3852 }
3853 EXPORT_SYMBOL(tcp_md5_hash_key);
3854 
3855 #endif
3856 
3857 void tcp_done(struct sock *sk)
3858 {
3859         struct request_sock *req;
3860 
3861         /* We might be called with a new socket, after
3862          * inet_csk_prepare_forced_close() has been called
3863          * so we can not use lockdep_sock_is_held(sk)
3864          */
3865         req = rcu_dereference_protected(tcp_sk(sk)->fastopen_rsk, 1);
3866 
3867         if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
3868                 TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
3869 
3870         tcp_set_state(sk, TCP_CLOSE);
3871         tcp_clear_xmit_timers(sk);
3872         if (req)
3873                 reqsk_fastopen_remove(sk, req, false);
3874 
3875         sk->sk_shutdown = SHUTDOWN_MASK;
3876 
3877         if (!sock_flag(sk, SOCK_DEAD))
3878                 sk->sk_state_change(sk);
3879         else
3880                 inet_csk_destroy_sock(sk);
3881 }
3882 EXPORT_SYMBOL_GPL(tcp_done);
3883 
3884 int tcp_abort(struct sock *sk, int err)
3885 {
3886         if (!sk_fullsock(sk)) {
3887                 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3888                         struct request_sock *req = inet_reqsk(sk);
3889 
3890                         local_bh_disable();
3891                         inet_csk_reqsk_queue_drop(req->rsk_listener, req);
3892                         local_bh_enable();
3893                         return 0;
3894                 }
3895                 return -EOPNOTSUPP;
3896         }
3897 
3898         /* Don't race with userspace socket closes such as tcp_close. */
3899         lock_sock(sk);
3900 
3901         if (sk->sk_state == TCP_LISTEN) {
3902                 tcp_set_state(sk, TCP_CLOSE);
3903                 inet_csk_listen_stop(sk);
3904         }
3905 
3906         /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3907         local_bh_disable();
3908         bh_lock_sock(sk);
3909 
3910         if (!sock_flag(sk, SOCK_DEAD)) {
3911                 sk->sk_err = err;
3912                 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3913                 smp_wmb();
3914                 sk->sk_error_report(sk);
3915                 if (tcp_need_reset(sk->sk_state))
3916                         tcp_send_active_reset(sk, GFP_ATOMIC);
3917                 tcp_done(sk);
3918         }
3919 
3920         bh_unlock_sock(sk);
3921         local_bh_enable();
3922         tcp_write_queue_purge(sk);
3923         release_sock(sk);
3924         return 0;
3925 }
3926 EXPORT_SYMBOL_GPL(tcp_abort);
3927 
3928 extern struct tcp_congestion_ops tcp_reno;
3929 
3930 static __initdata unsigned long thash_entries;
3931 static int __init set_thash_entries(char *str)
3932 {
3933         ssize_t ret;
3934 
3935         if (!str)
3936                 return 0;
3937 
3938         ret = kstrtoul(str, 0, &thash_entries);
3939         if (ret)
3940                 return 0;
3941 
3942         return 1;
3943 }
3944 __setup("thash_entries=", set_thash_entries);
3945 
3946 static void __init tcp_init_mem(void)
3947 {
3948         unsigned long limit = nr_free_buffer_pages() / 16;
3949 
3950         limit = max(limit, 128UL);
3951         sysctl_tcp_mem[0] = limit / 4 * 3;              /* 4.68 % */
3952         sysctl_tcp_mem[1] = limit;                      /* 6.25 % */
3953         sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;      /* 9.37 % */
3954 }
3955 
3956 void __init tcp_init(void)
3957 {
3958         int max_rshare, max_wshare, cnt;
3959         unsigned long limit;
3960         unsigned int i;
3961 
3962         BUILD_BUG_ON(TCP_MIN_SND_MSS <= MAX_TCP_OPTION_SPACE);
3963         BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
3964                      FIELD_SIZEOF(struct sk_buff, cb));
3965 
3966         percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3967         percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
3968         inet_hashinfo_init(&tcp_hashinfo);
3969         inet_hashinfo2_init(&tcp_hashinfo, "tcp_listen_portaddr_hash",
3970                             thash_entries, 21,  /* one slot per 2 MB*/
3971                             0, 64 * 1024);
3972         tcp_hashinfo.bind_bucket_cachep =
3973                 kmem_cache_create("tcp_bind_bucket",
3974                                   sizeof(struct inet_bind_bucket), 0,
3975                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3976 
3977         /* Size and allocate the main established and bind bucket
3978          * hash tables.
3979          *
3980          * The methodology is similar to that of the buffer cache.
3981          */
3982         tcp_hashinfo.ehash =
3983                 alloc_large_system_hash("TCP established",
3984                                         sizeof(struct inet_ehash_bucket),
3985                                         thash_entries,
3986                                         17, /* one slot per 128 KB of memory */
3987                                         0,
3988                                         NULL,
3989                                         &tcp_hashinfo.ehash_mask,
3990                                         0,
3991                                         thash_entries ? 0 : 512 * 1024);
3992         for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3993                 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3994 
3995         if (inet_ehash_locks_alloc(&tcp_hashinfo))
3996                 panic("TCP: failed to alloc ehash_locks");
3997         tcp_hashinfo.bhash =
3998                 alloc_large_system_hash("TCP bind",
3999                                         sizeof(struct inet_bind_hashbucket),
4000                                         tcp_hashinfo.ehash_mask + 1,
4001                                         17, /* one slot per 128 KB of memory */
4002                                         0,
4003                                         &tcp_hashinfo.bhash_size,
4004                                         NULL,
4005                                         0,
4006                                         64 * 1024);
4007         tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
4008         for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
4009                 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
4010                 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
4011         }
4012 
4013 
4014         cnt = tcp_hashinfo.ehash_mask + 1;
4015         sysctl_tcp_max_orphans = cnt / 2;
4016 
4017         tcp_init_mem();
4018         /* Set per-socket limits to no more than 1/128 the pressure threshold */
4019         limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
4020         max_wshare = min(4UL*1024*1024, limit);
4021         max_rshare = min(6UL*1024*1024, limit);
4022 
4023         init_net.ipv4.sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
4024         init_net.ipv4.sysctl_tcp_wmem[1] = 16*1024;
4025         init_net.ipv4.sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
4026 
4027         init_net.ipv4.sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
4028         init_net.ipv4.sysctl_tcp_rmem[1] = 131072;
4029         init_net.ipv4.sysctl_tcp_rmem[2] = max(131072, max_rshare);
4030 
4031         pr_info("Hash tables configured (established %u bind %u)\n",
4032                 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
4033 
4034         tcp_v4_init();
4035         tcp_metrics_init();
4036         BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
4037         tcp_tasklet_init();
4038 }