Version:  2.0.40 2.2.26 2.4.37 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6

Linux/net/ipv4/tcp.c

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              Implementation of the Transmission Control Protocol(TCP).
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Mark Evans, <evansmp@uhura.aston.ac.uk>
 11  *              Corey Minyard <wf-rch!minyard@relay.EU.net>
 12  *              Florian La Roche, <flla@stud.uni-sb.de>
 13  *              Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
 14  *              Linus Torvalds, <torvalds@cs.helsinki.fi>
 15  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
 16  *              Matthew Dillon, <dillon@apollo.west.oic.com>
 17  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 18  *              Jorge Cwik, <jorge@laser.satlink.net>
 19  *
 20  * Fixes:
 21  *              Alan Cox        :       Numerous verify_area() calls
 22  *              Alan Cox        :       Set the ACK bit on a reset
 23  *              Alan Cox        :       Stopped it crashing if it closed while
 24  *                                      sk->inuse=1 and was trying to connect
 25  *                                      (tcp_err()).
 26  *              Alan Cox        :       All icmp error handling was broken
 27  *                                      pointers passed where wrong and the
 28  *                                      socket was looked up backwards. Nobody
 29  *                                      tested any icmp error code obviously.
 30  *              Alan Cox        :       tcp_err() now handled properly. It
 31  *                                      wakes people on errors. poll
 32  *                                      behaves and the icmp error race
 33  *                                      has gone by moving it into sock.c
 34  *              Alan Cox        :       tcp_send_reset() fixed to work for
 35  *                                      everything not just packets for
 36  *                                      unknown sockets.
 37  *              Alan Cox        :       tcp option processing.
 38  *              Alan Cox        :       Reset tweaked (still not 100%) [Had
 39  *                                      syn rule wrong]
 40  *              Herp Rosmanith  :       More reset fixes
 41  *              Alan Cox        :       No longer acks invalid rst frames.
 42  *                                      Acking any kind of RST is right out.
 43  *              Alan Cox        :       Sets an ignore me flag on an rst
 44  *                                      receive otherwise odd bits of prattle
 45  *                                      escape still
 46  *              Alan Cox        :       Fixed another acking RST frame bug.
 47  *                                      Should stop LAN workplace lockups.
 48  *              Alan Cox        :       Some tidyups using the new skb list
 49  *                                      facilities
 50  *              Alan Cox        :       sk->keepopen now seems to work
 51  *              Alan Cox        :       Pulls options out correctly on accepts
 52  *              Alan Cox        :       Fixed assorted sk->rqueue->next errors
 53  *              Alan Cox        :       PSH doesn't end a TCP read. Switched a
 54  *                                      bit to skb ops.
 55  *              Alan Cox        :       Tidied tcp_data to avoid a potential
 56  *                                      nasty.
 57  *              Alan Cox        :       Added some better commenting, as the
 58  *                                      tcp is hard to follow
 59  *              Alan Cox        :       Removed incorrect check for 20 * psh
 60  *      Michael O'Reilly        :       ack < copied bug fix.
 61  *      Johannes Stille         :       Misc tcp fixes (not all in yet).
 62  *              Alan Cox        :       FIN with no memory -> CRASH
 63  *              Alan Cox        :       Added socket option proto entries.
 64  *                                      Also added awareness of them to accept.
 65  *              Alan Cox        :       Added TCP options (SOL_TCP)
 66  *              Alan Cox        :       Switched wakeup calls to callbacks,
 67  *                                      so the kernel can layer network
 68  *                                      sockets.
 69  *              Alan Cox        :       Use ip_tos/ip_ttl settings.
 70  *              Alan Cox        :       Handle FIN (more) properly (we hope).
 71  *              Alan Cox        :       RST frames sent on unsynchronised
 72  *                                      state ack error.
 73  *              Alan Cox        :       Put in missing check for SYN bit.
 74  *              Alan Cox        :       Added tcp_select_window() aka NET2E
 75  *                                      window non shrink trick.
 76  *              Alan Cox        :       Added a couple of small NET2E timer
 77  *                                      fixes
 78  *              Charles Hedrick :       TCP fixes
 79  *              Toomas Tamm     :       TCP window fixes
 80  *              Alan Cox        :       Small URG fix to rlogin ^C ack fight
 81  *              Charles Hedrick :       Rewrote most of it to actually work
 82  *              Linus           :       Rewrote tcp_read() and URG handling
 83  *                                      completely
 84  *              Gerhard Koerting:       Fixed some missing timer handling
 85  *              Matthew Dillon  :       Reworked TCP machine states as per RFC
 86  *              Gerhard Koerting:       PC/TCP workarounds
 87  *              Adam Caldwell   :       Assorted timer/timing errors
 88  *              Matthew Dillon  :       Fixed another RST bug
 89  *              Alan Cox        :       Move to kernel side addressing changes.
 90  *              Alan Cox        :       Beginning work on TCP fastpathing
 91  *                                      (not yet usable)
 92  *              Arnt Gulbrandsen:       Turbocharged tcp_check() routine.
 93  *              Alan Cox        :       TCP fast path debugging
 94  *              Alan Cox        :       Window clamping
 95  *              Michael Riepe   :       Bug in tcp_check()
 96  *              Matt Dillon     :       More TCP improvements and RST bug fixes
 97  *              Matt Dillon     :       Yet more small nasties remove from the
 98  *                                      TCP code (Be very nice to this man if
 99  *                                      tcp finally works 100%) 8)
100  *              Alan Cox        :       BSD accept semantics.
101  *              Alan Cox        :       Reset on closedown bug.
102  *      Peter De Schrijver      :       ENOTCONN check missing in tcp_sendto().
103  *              Michael Pall    :       Handle poll() after URG properly in
104  *                                      all cases.
105  *              Michael Pall    :       Undo the last fix in tcp_read_urg()
106  *                                      (multi URG PUSH broke rlogin).
107  *              Michael Pall    :       Fix the multi URG PUSH problem in
108  *                                      tcp_readable(), poll() after URG
109  *                                      works now.
110  *              Michael Pall    :       recv(...,MSG_OOB) never blocks in the
111  *                                      BSD api.
112  *              Alan Cox        :       Changed the semantics of sk->socket to
113  *                                      fix a race and a signal problem with
114  *                                      accept() and async I/O.
115  *              Alan Cox        :       Relaxed the rules on tcp_sendto().
116  *              Yury Shevchuk   :       Really fixed accept() blocking problem.
117  *              Craig I. Hagan  :       Allow for BSD compatible TIME_WAIT for
118  *                                      clients/servers which listen in on
119  *                                      fixed ports.
120  *              Alan Cox        :       Cleaned the above up and shrank it to
121  *                                      a sensible code size.
122  *              Alan Cox        :       Self connect lockup fix.
123  *              Alan Cox        :       No connect to multicast.
124  *              Ross Biro       :       Close unaccepted children on master
125  *                                      socket close.
126  *              Alan Cox        :       Reset tracing code.
127  *              Alan Cox        :       Spurious resets on shutdown.
128  *              Alan Cox        :       Giant 15 minute/60 second timer error
129  *              Alan Cox        :       Small whoops in polling before an
130  *                                      accept.
131  *              Alan Cox        :       Kept the state trace facility since
132  *                                      it's handy for debugging.
133  *              Alan Cox        :       More reset handler fixes.
134  *              Alan Cox        :       Started rewriting the code based on
135  *                                      the RFC's for other useful protocol
136  *                                      references see: Comer, KA9Q NOS, and
137  *                                      for a reference on the difference
138  *                                      between specifications and how BSD
139  *                                      works see the 4.4lite source.
140  *              A.N.Kuznetsov   :       Don't time wait on completion of tidy
141  *                                      close.
142  *              Linus Torvalds  :       Fin/Shutdown & copied_seq changes.
143  *              Linus Torvalds  :       Fixed BSD port reuse to work first syn
144  *              Alan Cox        :       Reimplemented timers as per the RFC
145  *                                      and using multiple timers for sanity.
146  *              Alan Cox        :       Small bug fixes, and a lot of new
147  *                                      comments.
148  *              Alan Cox        :       Fixed dual reader crash by locking
149  *                                      the buffers (much like datagram.c)
150  *              Alan Cox        :       Fixed stuck sockets in probe. A probe
151  *                                      now gets fed up of retrying without
152  *                                      (even a no space) answer.
153  *              Alan Cox        :       Extracted closing code better
154  *              Alan Cox        :       Fixed the closing state machine to
155  *                                      resemble the RFC.
156  *              Alan Cox        :       More 'per spec' fixes.
157  *              Jorge Cwik      :       Even faster checksumming.
158  *              Alan Cox        :       tcp_data() doesn't ack illegal PSH
159  *                                      only frames. At least one pc tcp stack
160  *                                      generates them.
161  *              Alan Cox        :       Cache last socket.
162  *              Alan Cox        :       Per route irtt.
163  *              Matt Day        :       poll()->select() match BSD precisely on error
164  *              Alan Cox        :       New buffers
165  *              Marc Tamsky     :       Various sk->prot->retransmits and
166  *                                      sk->retransmits misupdating fixed.
167  *                                      Fixed tcp_write_timeout: stuck close,
168  *                                      and TCP syn retries gets used now.
169  *              Mark Yarvis     :       In tcp_read_wakeup(), don't send an
170  *                                      ack if state is TCP_CLOSED.
171  *              Alan Cox        :       Look up device on a retransmit - routes may
172  *                                      change. Doesn't yet cope with MSS shrink right
173  *                                      but it's a start!
174  *              Marc Tamsky     :       Closing in closing fixes.
175  *              Mike Shaver     :       RFC1122 verifications.
176  *              Alan Cox        :       rcv_saddr errors.
177  *              Alan Cox        :       Block double connect().
178  *              Alan Cox        :       Small hooks for enSKIP.
179  *              Alexey Kuznetsov:       Path MTU discovery.
180  *              Alan Cox        :       Support soft errors.
181  *              Alan Cox        :       Fix MTU discovery pathological case
182  *                                      when the remote claims no mtu!
183  *              Marc Tamsky     :       TCP_CLOSE fix.
184  *              Colin (G3TNE)   :       Send a reset on syn ack replies in
185  *                                      window but wrong (fixes NT lpd problems)
186  *              Pedro Roque     :       Better TCP window handling, delayed ack.
187  *              Joerg Reuter    :       No modification of locked buffers in
188  *                                      tcp_do_retransmit()
189  *              Eric Schenk     :       Changed receiver side silly window
190  *                                      avoidance algorithm to BSD style
191  *                                      algorithm. This doubles throughput
192  *                                      against machines running Solaris,
193  *                                      and seems to result in general
194  *                                      improvement.
195  *      Stefan Magdalinski      :       adjusted tcp_readable() to fix FIONREAD
196  *      Willy Konynenberg       :       Transparent proxying support.
197  *      Mike McLagan            :       Routing by source
198  *              Keith Owens     :       Do proper merging with partial SKB's in
199  *                                      tcp_do_sendmsg to avoid burstiness.
200  *              Eric Schenk     :       Fix fast close down bug with
201  *                                      shutdown() followed by close().
202  *              Andi Kleen      :       Make poll agree with SIGIO
203  *      Salvatore Sanfilippo    :       Support SO_LINGER with linger == 1 and
204  *                                      lingertime == 0 (RFC 793 ABORT Call)
205  *      Hirokazu Takahashi      :       Use copy_from_user() instead of
206  *                                      csum_and_copy_from_user() if possible.
207  *
208  *              This program is free software; you can redistribute it and/or
209  *              modify it under the terms of the GNU General Public License
210  *              as published by the Free Software Foundation; either version
211  *              2 of the License, or(at your option) any later version.
212  *
213  * Description of States:
214  *
215  *      TCP_SYN_SENT            sent a connection request, waiting for ack
216  *
217  *      TCP_SYN_RECV            received a connection request, sent ack,
218  *                              waiting for final ack in three-way handshake.
219  *
220  *      TCP_ESTABLISHED         connection established
221  *
222  *      TCP_FIN_WAIT1           our side has shutdown, waiting to complete
223  *                              transmission of remaining buffered data
224  *
225  *      TCP_FIN_WAIT2           all buffered data sent, waiting for remote
226  *                              to shutdown
227  *
228  *      TCP_CLOSING             both sides have shutdown but we still have
229  *                              data we have to finish sending
230  *
231  *      TCP_TIME_WAIT           timeout to catch resent junk before entering
232  *                              closed, can only be entered from FIN_WAIT2
233  *                              or CLOSING.  Required because the other end
234  *                              may not have gotten our last ACK causing it
235  *                              to retransmit the data packet (which we ignore)
236  *
237  *      TCP_CLOSE_WAIT          remote side has shutdown and is waiting for
238  *                              us to finish writing our data and to shutdown
239  *                              (we have to close() to move on to LAST_ACK)
240  *
241  *      TCP_LAST_ACK            out side has shutdown after remote has
242  *                              shutdown.  There may still be data in our
243  *                              buffer that we have to finish sending
244  *
245  *      TCP_CLOSE               socket is finished
246  */
247 
248 #define pr_fmt(fmt) "TCP: " fmt
249 
250 #include <crypto/hash.h>
251 #include <linux/kernel.h>
252 #include <linux/module.h>
253 #include <linux/types.h>
254 #include <linux/fcntl.h>
255 #include <linux/poll.h>
256 #include <linux/inet_diag.h>
257 #include <linux/init.h>
258 #include <linux/fs.h>
259 #include <linux/skbuff.h>
260 #include <linux/scatterlist.h>
261 #include <linux/splice.h>
262 #include <linux/net.h>
263 #include <linux/socket.h>
264 #include <linux/random.h>
265 #include <linux/bootmem.h>
266 #include <linux/highmem.h>
267 #include <linux/swap.h>
268 #include <linux/cache.h>
269 #include <linux/err.h>
270 #include <linux/time.h>
271 #include <linux/slab.h>
272 
273 #include <net/icmp.h>
274 #include <net/inet_common.h>
275 #include <net/tcp.h>
276 #include <net/xfrm.h>
277 #include <net/ip.h>
278 #include <net/sock.h>
279 
280 #include <asm/uaccess.h>
281 #include <asm/ioctls.h>
282 #include <asm/unaligned.h>
283 #include <net/busy_poll.h>
284 
285 int sysctl_tcp_min_tso_segs __read_mostly = 2;
286 
287 int sysctl_tcp_autocorking __read_mostly = 1;
288 
289 struct percpu_counter tcp_orphan_count;
290 EXPORT_SYMBOL_GPL(tcp_orphan_count);
291 
292 long sysctl_tcp_mem[3] __read_mostly;
293 int sysctl_tcp_wmem[3] __read_mostly;
294 int sysctl_tcp_rmem[3] __read_mostly;
295 
296 EXPORT_SYMBOL(sysctl_tcp_mem);
297 EXPORT_SYMBOL(sysctl_tcp_rmem);
298 EXPORT_SYMBOL(sysctl_tcp_wmem);
299 
300 atomic_long_t tcp_memory_allocated;     /* Current allocated memory. */
301 EXPORT_SYMBOL(tcp_memory_allocated);
302 
303 /*
304  * Current number of TCP sockets.
305  */
306 struct percpu_counter tcp_sockets_allocated;
307 EXPORT_SYMBOL(tcp_sockets_allocated);
308 
309 /*
310  * TCP splice context
311  */
312 struct tcp_splice_state {
313         struct pipe_inode_info *pipe;
314         size_t len;
315         unsigned int flags;
316 };
317 
318 /*
319  * Pressure flag: try to collapse.
320  * Technical note: it is used by multiple contexts non atomically.
321  * All the __sk_mem_schedule() is of this nature: accounting
322  * is strict, actions are advisory and have some latency.
323  */
324 int tcp_memory_pressure __read_mostly;
325 EXPORT_SYMBOL(tcp_memory_pressure);
326 
327 void tcp_enter_memory_pressure(struct sock *sk)
328 {
329         if (!tcp_memory_pressure) {
330                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
331                 tcp_memory_pressure = 1;
332         }
333 }
334 EXPORT_SYMBOL(tcp_enter_memory_pressure);
335 
336 /* Convert seconds to retransmits based on initial and max timeout */
337 static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
338 {
339         u8 res = 0;
340 
341         if (seconds > 0) {
342                 int period = timeout;
343 
344                 res = 1;
345                 while (seconds > period && res < 255) {
346                         res++;
347                         timeout <<= 1;
348                         if (timeout > rto_max)
349                                 timeout = rto_max;
350                         period += timeout;
351                 }
352         }
353         return res;
354 }
355 
356 /* Convert retransmits to seconds based on initial and max timeout */
357 static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
358 {
359         int period = 0;
360 
361         if (retrans > 0) {
362                 period = timeout;
363                 while (--retrans) {
364                         timeout <<= 1;
365                         if (timeout > rto_max)
366                                 timeout = rto_max;
367                         period += timeout;
368                 }
369         }
370         return period;
371 }
372 
373 /* Address-family independent initialization for a tcp_sock.
374  *
375  * NOTE: A lot of things set to zero explicitly by call to
376  *       sk_alloc() so need not be done here.
377  */
378 void tcp_init_sock(struct sock *sk)
379 {
380         struct inet_connection_sock *icsk = inet_csk(sk);
381         struct tcp_sock *tp = tcp_sk(sk);
382 
383         __skb_queue_head_init(&tp->out_of_order_queue);
384         tcp_init_xmit_timers(sk);
385         tcp_prequeue_init(tp);
386         INIT_LIST_HEAD(&tp->tsq_node);
387 
388         icsk->icsk_rto = TCP_TIMEOUT_INIT;
389         tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
390         tp->rtt_min[0].rtt = ~0U;
391 
392         /* So many TCP implementations out there (incorrectly) count the
393          * initial SYN frame in their delayed-ACK and congestion control
394          * algorithms that we must have the following bandaid to talk
395          * efficiently to them.  -DaveM
396          */
397         tp->snd_cwnd = TCP_INIT_CWND;
398 
399         /* See draft-stevens-tcpca-spec-01 for discussion of the
400          * initialization of these values.
401          */
402         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
403         tp->snd_cwnd_clamp = ~0;
404         tp->mss_cache = TCP_MSS_DEFAULT;
405         u64_stats_init(&tp->syncp);
406 
407         tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
408         tcp_enable_early_retrans(tp);
409         tcp_assign_congestion_control(sk);
410 
411         tp->tsoffset = 0;
412 
413         sk->sk_state = TCP_CLOSE;
414 
415         sk->sk_write_space = sk_stream_write_space;
416         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
417 
418         icsk->icsk_sync_mss = tcp_sync_mss;
419 
420         sk->sk_sndbuf = sysctl_tcp_wmem[1];
421         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
422 
423         local_bh_disable();
424         if (mem_cgroup_sockets_enabled)
425                 sock_update_memcg(sk);
426         sk_sockets_allocated_inc(sk);
427         local_bh_enable();
428 }
429 EXPORT_SYMBOL(tcp_init_sock);
430 
431 static void tcp_tx_timestamp(struct sock *sk, struct sk_buff *skb)
432 {
433         if (sk->sk_tsflags) {
434                 struct skb_shared_info *shinfo = skb_shinfo(skb);
435 
436                 sock_tx_timestamp(sk, &shinfo->tx_flags);
437                 if (shinfo->tx_flags & SKBTX_ANY_TSTAMP)
438                         shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
439         }
440 }
441 
442 /*
443  *      Wait for a TCP event.
444  *
445  *      Note that we don't need to lock the socket, as the upper poll layers
446  *      take care of normal races (between the test and the event) and we don't
447  *      go look at any of the socket buffers directly.
448  */
449 unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
450 {
451         unsigned int mask;
452         struct sock *sk = sock->sk;
453         const struct tcp_sock *tp = tcp_sk(sk);
454         int state;
455 
456         sock_rps_record_flow(sk);
457 
458         sock_poll_wait(file, sk_sleep(sk), wait);
459 
460         state = sk_state_load(sk);
461         if (state == TCP_LISTEN)
462                 return inet_csk_listen_poll(sk);
463 
464         /* Socket is not locked. We are protected from async events
465          * by poll logic and correct handling of state changes
466          * made by other threads is impossible in any case.
467          */
468 
469         mask = 0;
470 
471         /*
472          * POLLHUP is certainly not done right. But poll() doesn't
473          * have a notion of HUP in just one direction, and for a
474          * socket the read side is more interesting.
475          *
476          * Some poll() documentation says that POLLHUP is incompatible
477          * with the POLLOUT/POLLWR flags, so somebody should check this
478          * all. But careful, it tends to be safer to return too many
479          * bits than too few, and you can easily break real applications
480          * if you don't tell them that something has hung up!
481          *
482          * Check-me.
483          *
484          * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
485          * our fs/select.c). It means that after we received EOF,
486          * poll always returns immediately, making impossible poll() on write()
487          * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
488          * if and only if shutdown has been made in both directions.
489          * Actually, it is interesting to look how Solaris and DUX
490          * solve this dilemma. I would prefer, if POLLHUP were maskable,
491          * then we could set it on SND_SHUTDOWN. BTW examples given
492          * in Stevens' books assume exactly this behaviour, it explains
493          * why POLLHUP is incompatible with POLLOUT.    --ANK
494          *
495          * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
496          * blocking on fresh not-connected or disconnected socket. --ANK
497          */
498         if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
499                 mask |= POLLHUP;
500         if (sk->sk_shutdown & RCV_SHUTDOWN)
501                 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
502 
503         /* Connected or passive Fast Open socket? */
504         if (state != TCP_SYN_SENT &&
505             (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
506                 int target = sock_rcvlowat(sk, 0, INT_MAX);
507 
508                 if (tp->urg_seq == tp->copied_seq &&
509                     !sock_flag(sk, SOCK_URGINLINE) &&
510                     tp->urg_data)
511                         target++;
512 
513                 if (tp->rcv_nxt - tp->copied_seq >= target)
514                         mask |= POLLIN | POLLRDNORM;
515 
516                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
517                         if (sk_stream_is_writeable(sk)) {
518                                 mask |= POLLOUT | POLLWRNORM;
519                         } else {  /* send SIGIO later */
520                                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
521                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
522 
523                                 /* Race breaker. If space is freed after
524                                  * wspace test but before the flags are set,
525                                  * IO signal will be lost. Memory barrier
526                                  * pairs with the input side.
527                                  */
528                                 smp_mb__after_atomic();
529                                 if (sk_stream_is_writeable(sk))
530                                         mask |= POLLOUT | POLLWRNORM;
531                         }
532                 } else
533                         mask |= POLLOUT | POLLWRNORM;
534 
535                 if (tp->urg_data & TCP_URG_VALID)
536                         mask |= POLLPRI;
537         }
538         /* This barrier is coupled with smp_wmb() in tcp_reset() */
539         smp_rmb();
540         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
541                 mask |= POLLERR;
542 
543         return mask;
544 }
545 EXPORT_SYMBOL(tcp_poll);
546 
547 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
548 {
549         struct tcp_sock *tp = tcp_sk(sk);
550         int answ;
551         bool slow;
552 
553         switch (cmd) {
554         case SIOCINQ:
555                 if (sk->sk_state == TCP_LISTEN)
556                         return -EINVAL;
557 
558                 slow = lock_sock_fast(sk);
559                 answ = tcp_inq(sk);
560                 unlock_sock_fast(sk, slow);
561                 break;
562         case SIOCATMARK:
563                 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
564                 break;
565         case SIOCOUTQ:
566                 if (sk->sk_state == TCP_LISTEN)
567                         return -EINVAL;
568 
569                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
570                         answ = 0;
571                 else
572                         answ = tp->write_seq - tp->snd_una;
573                 break;
574         case SIOCOUTQNSD:
575                 if (sk->sk_state == TCP_LISTEN)
576                         return -EINVAL;
577 
578                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
579                         answ = 0;
580                 else
581                         answ = tp->write_seq - tp->snd_nxt;
582                 break;
583         default:
584                 return -ENOIOCTLCMD;
585         }
586 
587         return put_user(answ, (int __user *)arg);
588 }
589 EXPORT_SYMBOL(tcp_ioctl);
590 
591 static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
592 {
593         TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
594         tp->pushed_seq = tp->write_seq;
595 }
596 
597 static inline bool forced_push(const struct tcp_sock *tp)
598 {
599         return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
600 }
601 
602 static void skb_entail(struct sock *sk, struct sk_buff *skb)
603 {
604         struct tcp_sock *tp = tcp_sk(sk);
605         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
606 
607         skb->csum    = 0;
608         tcb->seq     = tcb->end_seq = tp->write_seq;
609         tcb->tcp_flags = TCPHDR_ACK;
610         tcb->sacked  = 0;
611         __skb_header_release(skb);
612         tcp_add_write_queue_tail(sk, skb);
613         sk->sk_wmem_queued += skb->truesize;
614         sk_mem_charge(sk, skb->truesize);
615         if (tp->nonagle & TCP_NAGLE_PUSH)
616                 tp->nonagle &= ~TCP_NAGLE_PUSH;
617 
618         tcp_slow_start_after_idle_check(sk);
619 }
620 
621 static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
622 {
623         if (flags & MSG_OOB)
624                 tp->snd_up = tp->write_seq;
625 }
626 
627 /* If a not yet filled skb is pushed, do not send it if
628  * we have data packets in Qdisc or NIC queues :
629  * Because TX completion will happen shortly, it gives a chance
630  * to coalesce future sendmsg() payload into this skb, without
631  * need for a timer, and with no latency trade off.
632  * As packets containing data payload have a bigger truesize
633  * than pure acks (dataless) packets, the last checks prevent
634  * autocorking if we only have an ACK in Qdisc/NIC queues,
635  * or if TX completion was delayed after we processed ACK packet.
636  */
637 static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
638                                 int size_goal)
639 {
640         return skb->len < size_goal &&
641                sysctl_tcp_autocorking &&
642                skb != tcp_write_queue_head(sk) &&
643                atomic_read(&sk->sk_wmem_alloc) > skb->truesize;
644 }
645 
646 static void tcp_push(struct sock *sk, int flags, int mss_now,
647                      int nonagle, int size_goal)
648 {
649         struct tcp_sock *tp = tcp_sk(sk);
650         struct sk_buff *skb;
651 
652         if (!tcp_send_head(sk))
653                 return;
654 
655         skb = tcp_write_queue_tail(sk);
656         if (!(flags & MSG_MORE) || forced_push(tp))
657                 tcp_mark_push(tp, skb);
658 
659         tcp_mark_urg(tp, flags);
660 
661         if (tcp_should_autocork(sk, skb, size_goal)) {
662 
663                 /* avoid atomic op if TSQ_THROTTLED bit is already set */
664                 if (!test_bit(TSQ_THROTTLED, &tp->tsq_flags)) {
665                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
666                         set_bit(TSQ_THROTTLED, &tp->tsq_flags);
667                 }
668                 /* It is possible TX completion already happened
669                  * before we set TSQ_THROTTLED.
670                  */
671                 if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize)
672                         return;
673         }
674 
675         if (flags & MSG_MORE)
676                 nonagle = TCP_NAGLE_CORK;
677 
678         __tcp_push_pending_frames(sk, mss_now, nonagle);
679 }
680 
681 static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
682                                 unsigned int offset, size_t len)
683 {
684         struct tcp_splice_state *tss = rd_desc->arg.data;
685         int ret;
686 
687         ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
688                               min(rd_desc->count, len), tss->flags,
689                               skb_socket_splice);
690         if (ret > 0)
691                 rd_desc->count -= ret;
692         return ret;
693 }
694 
695 static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
696 {
697         /* Store TCP splice context information in read_descriptor_t. */
698         read_descriptor_t rd_desc = {
699                 .arg.data = tss,
700                 .count    = tss->len,
701         };
702 
703         return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
704 }
705 
706 /**
707  *  tcp_splice_read - splice data from TCP socket to a pipe
708  * @sock:       socket to splice from
709  * @ppos:       position (not valid)
710  * @pipe:       pipe to splice to
711  * @len:        number of bytes to splice
712  * @flags:      splice modifier flags
713  *
714  * Description:
715  *    Will read pages from given socket and fill them into a pipe.
716  *
717  **/
718 ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
719                         struct pipe_inode_info *pipe, size_t len,
720                         unsigned int flags)
721 {
722         struct sock *sk = sock->sk;
723         struct tcp_splice_state tss = {
724                 .pipe = pipe,
725                 .len = len,
726                 .flags = flags,
727         };
728         long timeo;
729         ssize_t spliced;
730         int ret;
731 
732         sock_rps_record_flow(sk);
733         /*
734          * We can't seek on a socket input
735          */
736         if (unlikely(*ppos))
737                 return -ESPIPE;
738 
739         ret = spliced = 0;
740 
741         lock_sock(sk);
742 
743         timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
744         while (tss.len) {
745                 ret = __tcp_splice_read(sk, &tss);
746                 if (ret < 0)
747                         break;
748                 else if (!ret) {
749                         if (spliced)
750                                 break;
751                         if (sock_flag(sk, SOCK_DONE))
752                                 break;
753                         if (sk->sk_err) {
754                                 ret = sock_error(sk);
755                                 break;
756                         }
757                         if (sk->sk_shutdown & RCV_SHUTDOWN)
758                                 break;
759                         if (sk->sk_state == TCP_CLOSE) {
760                                 /*
761                                  * This occurs when user tries to read
762                                  * from never connected socket.
763                                  */
764                                 if (!sock_flag(sk, SOCK_DONE))
765                                         ret = -ENOTCONN;
766                                 break;
767                         }
768                         if (!timeo) {
769                                 ret = -EAGAIN;
770                                 break;
771                         }
772                         sk_wait_data(sk, &timeo, NULL);
773                         if (signal_pending(current)) {
774                                 ret = sock_intr_errno(timeo);
775                                 break;
776                         }
777                         continue;
778                 }
779                 tss.len -= ret;
780                 spliced += ret;
781 
782                 if (!timeo)
783                         break;
784                 release_sock(sk);
785                 lock_sock(sk);
786 
787                 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
788                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
789                     signal_pending(current))
790                         break;
791         }
792 
793         release_sock(sk);
794 
795         if (spliced)
796                 return spliced;
797 
798         return ret;
799 }
800 EXPORT_SYMBOL(tcp_splice_read);
801 
802 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
803                                     bool force_schedule)
804 {
805         struct sk_buff *skb;
806 
807         /* The TCP header must be at least 32-bit aligned.  */
808         size = ALIGN(size, 4);
809 
810         if (unlikely(tcp_under_memory_pressure(sk)))
811                 sk_mem_reclaim_partial(sk);
812 
813         skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
814         if (likely(skb)) {
815                 bool mem_scheduled;
816 
817                 if (force_schedule) {
818                         mem_scheduled = true;
819                         sk_forced_mem_schedule(sk, skb->truesize);
820                 } else {
821                         mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
822                 }
823                 if (likely(mem_scheduled)) {
824                         skb_reserve(skb, sk->sk_prot->max_header);
825                         /*
826                          * Make sure that we have exactly size bytes
827                          * available to the caller, no more, no less.
828                          */
829                         skb->reserved_tailroom = skb->end - skb->tail - size;
830                         return skb;
831                 }
832                 __kfree_skb(skb);
833         } else {
834                 sk->sk_prot->enter_memory_pressure(sk);
835                 sk_stream_moderate_sndbuf(sk);
836         }
837         return NULL;
838 }
839 
840 static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
841                                        int large_allowed)
842 {
843         struct tcp_sock *tp = tcp_sk(sk);
844         u32 new_size_goal, size_goal;
845 
846         if (!large_allowed || !sk_can_gso(sk))
847                 return mss_now;
848 
849         /* Note : tcp_tso_autosize() will eventually split this later */
850         new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
851         new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
852 
853         /* We try hard to avoid divides here */
854         size_goal = tp->gso_segs * mss_now;
855         if (unlikely(new_size_goal < size_goal ||
856                      new_size_goal >= size_goal + mss_now)) {
857                 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
858                                      sk->sk_gso_max_segs);
859                 size_goal = tp->gso_segs * mss_now;
860         }
861 
862         return max(size_goal, mss_now);
863 }
864 
865 static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
866 {
867         int mss_now;
868 
869         mss_now = tcp_current_mss(sk);
870         *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
871 
872         return mss_now;
873 }
874 
875 static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
876                                 size_t size, int flags)
877 {
878         struct tcp_sock *tp = tcp_sk(sk);
879         int mss_now, size_goal;
880         int err;
881         ssize_t copied;
882         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
883 
884         /* Wait for a connection to finish. One exception is TCP Fast Open
885          * (passive side) where data is allowed to be sent before a connection
886          * is fully established.
887          */
888         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
889             !tcp_passive_fastopen(sk)) {
890                 err = sk_stream_wait_connect(sk, &timeo);
891                 if (err != 0)
892                         goto out_err;
893         }
894 
895         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
896 
897         mss_now = tcp_send_mss(sk, &size_goal, flags);
898         copied = 0;
899 
900         err = -EPIPE;
901         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
902                 goto out_err;
903 
904         while (size > 0) {
905                 struct sk_buff *skb = tcp_write_queue_tail(sk);
906                 int copy, i;
907                 bool can_coalesce;
908 
909                 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
910 new_segment:
911                         if (!sk_stream_memory_free(sk))
912                                 goto wait_for_sndbuf;
913 
914                         skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
915                                                   skb_queue_empty(&sk->sk_write_queue));
916                         if (!skb)
917                                 goto wait_for_memory;
918 
919                         skb_entail(sk, skb);
920                         copy = size_goal;
921                 }
922 
923                 if (copy > size)
924                         copy = size;
925 
926                 i = skb_shinfo(skb)->nr_frags;
927                 can_coalesce = skb_can_coalesce(skb, i, page, offset);
928                 if (!can_coalesce && i >= sysctl_max_skb_frags) {
929                         tcp_mark_push(tp, skb);
930                         goto new_segment;
931                 }
932                 if (!sk_wmem_schedule(sk, copy))
933                         goto wait_for_memory;
934 
935                 if (can_coalesce) {
936                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
937                 } else {
938                         get_page(page);
939                         skb_fill_page_desc(skb, i, page, offset, copy);
940                 }
941                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
942 
943                 skb->len += copy;
944                 skb->data_len += copy;
945                 skb->truesize += copy;
946                 sk->sk_wmem_queued += copy;
947                 sk_mem_charge(sk, copy);
948                 skb->ip_summed = CHECKSUM_PARTIAL;
949                 tp->write_seq += copy;
950                 TCP_SKB_CB(skb)->end_seq += copy;
951                 tcp_skb_pcount_set(skb, 0);
952 
953                 if (!copied)
954                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
955 
956                 copied += copy;
957                 offset += copy;
958                 size -= copy;
959                 if (!size) {
960                         tcp_tx_timestamp(sk, skb);
961                         goto out;
962                 }
963 
964                 if (skb->len < size_goal || (flags & MSG_OOB))
965                         continue;
966 
967                 if (forced_push(tp)) {
968                         tcp_mark_push(tp, skb);
969                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
970                 } else if (skb == tcp_send_head(sk))
971                         tcp_push_one(sk, mss_now);
972                 continue;
973 
974 wait_for_sndbuf:
975                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
976 wait_for_memory:
977                 tcp_push(sk, flags & ~MSG_MORE, mss_now,
978                          TCP_NAGLE_PUSH, size_goal);
979 
980                 err = sk_stream_wait_memory(sk, &timeo);
981                 if (err != 0)
982                         goto do_error;
983 
984                 mss_now = tcp_send_mss(sk, &size_goal, flags);
985         }
986 
987 out:
988         if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
989                 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
990         return copied;
991 
992 do_error:
993         if (copied)
994                 goto out;
995 out_err:
996         /* make sure we wake any epoll edge trigger waiter */
997         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
998                 sk->sk_write_space(sk);
999         return sk_stream_error(sk, flags, err);
1000 }
1001 
1002 int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1003                  size_t size, int flags)
1004 {
1005         ssize_t res;
1006 
1007         if (!(sk->sk_route_caps & NETIF_F_SG) ||
1008             !sk_check_csum_caps(sk))
1009                 return sock_no_sendpage(sk->sk_socket, page, offset, size,
1010                                         flags);
1011 
1012         lock_sock(sk);
1013         res = do_tcp_sendpages(sk, page, offset, size, flags);
1014         release_sock(sk);
1015         return res;
1016 }
1017 EXPORT_SYMBOL(tcp_sendpage);
1018 
1019 static inline int select_size(const struct sock *sk, bool sg)
1020 {
1021         const struct tcp_sock *tp = tcp_sk(sk);
1022         int tmp = tp->mss_cache;
1023 
1024         if (sg) {
1025                 if (sk_can_gso(sk)) {
1026                         /* Small frames wont use a full page:
1027                          * Payload will immediately follow tcp header.
1028                          */
1029                         tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1030                 } else {
1031                         int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
1032 
1033                         if (tmp >= pgbreak &&
1034                             tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
1035                                 tmp = pgbreak;
1036                 }
1037         }
1038 
1039         return tmp;
1040 }
1041 
1042 void tcp_free_fastopen_req(struct tcp_sock *tp)
1043 {
1044         if (tp->fastopen_req) {
1045                 kfree(tp->fastopen_req);
1046                 tp->fastopen_req = NULL;
1047         }
1048 }
1049 
1050 static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1051                                 int *copied, size_t size)
1052 {
1053         struct tcp_sock *tp = tcp_sk(sk);
1054         int err, flags;
1055 
1056         if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
1057                 return -EOPNOTSUPP;
1058         if (tp->fastopen_req)
1059                 return -EALREADY; /* Another Fast Open is in progress */
1060 
1061         tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1062                                    sk->sk_allocation);
1063         if (unlikely(!tp->fastopen_req))
1064                 return -ENOBUFS;
1065         tp->fastopen_req->data = msg;
1066         tp->fastopen_req->size = size;
1067 
1068         flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1069         err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1070                                     msg->msg_namelen, flags);
1071         *copied = tp->fastopen_req->copied;
1072         tcp_free_fastopen_req(tp);
1073         return err;
1074 }
1075 
1076 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1077 {
1078         struct tcp_sock *tp = tcp_sk(sk);
1079         struct sk_buff *skb;
1080         int flags, err, copied = 0;
1081         int mss_now = 0, size_goal, copied_syn = 0;
1082         bool sg;
1083         long timeo;
1084 
1085         lock_sock(sk);
1086 
1087         flags = msg->msg_flags;
1088         if (flags & MSG_FASTOPEN) {
1089                 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size);
1090                 if (err == -EINPROGRESS && copied_syn > 0)
1091                         goto out;
1092                 else if (err)
1093                         goto out_err;
1094         }
1095 
1096         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1097 
1098         /* Wait for a connection to finish. One exception is TCP Fast Open
1099          * (passive side) where data is allowed to be sent before a connection
1100          * is fully established.
1101          */
1102         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1103             !tcp_passive_fastopen(sk)) {
1104                 err = sk_stream_wait_connect(sk, &timeo);
1105                 if (err != 0)
1106                         goto do_error;
1107         }
1108 
1109         if (unlikely(tp->repair)) {
1110                 if (tp->repair_queue == TCP_RECV_QUEUE) {
1111                         copied = tcp_send_rcvq(sk, msg, size);
1112                         goto out_nopush;
1113                 }
1114 
1115                 err = -EINVAL;
1116                 if (tp->repair_queue == TCP_NO_QUEUE)
1117                         goto out_err;
1118 
1119                 /* 'common' sending to sendq */
1120         }
1121 
1122         /* This should be in poll */
1123         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1124 
1125         mss_now = tcp_send_mss(sk, &size_goal, flags);
1126 
1127         /* Ok commence sending. */
1128         copied = 0;
1129 
1130         err = -EPIPE;
1131         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1132                 goto out_err;
1133 
1134         sg = !!(sk->sk_route_caps & NETIF_F_SG);
1135 
1136         while (msg_data_left(msg)) {
1137                 int copy = 0;
1138                 int max = size_goal;
1139 
1140                 skb = tcp_write_queue_tail(sk);
1141                 if (tcp_send_head(sk)) {
1142                         if (skb->ip_summed == CHECKSUM_NONE)
1143                                 max = mss_now;
1144                         copy = max - skb->len;
1145                 }
1146 
1147                 if (copy <= 0) {
1148 new_segment:
1149                         /* Allocate new segment. If the interface is SG,
1150                          * allocate skb fitting to single page.
1151                          */
1152                         if (!sk_stream_memory_free(sk))
1153                                 goto wait_for_sndbuf;
1154 
1155                         skb = sk_stream_alloc_skb(sk,
1156                                                   select_size(sk, sg),
1157                                                   sk->sk_allocation,
1158                                                   skb_queue_empty(&sk->sk_write_queue));
1159                         if (!skb)
1160                                 goto wait_for_memory;
1161 
1162                         /*
1163                          * Check whether we can use HW checksum.
1164                          */
1165                         if (sk_check_csum_caps(sk))
1166                                 skb->ip_summed = CHECKSUM_PARTIAL;
1167 
1168                         skb_entail(sk, skb);
1169                         copy = size_goal;
1170                         max = size_goal;
1171 
1172                         /* All packets are restored as if they have
1173                          * already been sent. skb_mstamp isn't set to
1174                          * avoid wrong rtt estimation.
1175                          */
1176                         if (tp->repair)
1177                                 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1178                 }
1179 
1180                 /* Try to append data to the end of skb. */
1181                 if (copy > msg_data_left(msg))
1182                         copy = msg_data_left(msg);
1183 
1184                 /* Where to copy to? */
1185                 if (skb_availroom(skb) > 0) {
1186                         /* We have some space in skb head. Superb! */
1187                         copy = min_t(int, copy, skb_availroom(skb));
1188                         err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1189                         if (err)
1190                                 goto do_fault;
1191                 } else {
1192                         bool merge = true;
1193                         int i = skb_shinfo(skb)->nr_frags;
1194                         struct page_frag *pfrag = sk_page_frag(sk);
1195 
1196                         if (!sk_page_frag_refill(sk, pfrag))
1197                                 goto wait_for_memory;
1198 
1199                         if (!skb_can_coalesce(skb, i, pfrag->page,
1200                                               pfrag->offset)) {
1201                                 if (i == sysctl_max_skb_frags || !sg) {
1202                                         tcp_mark_push(tp, skb);
1203                                         goto new_segment;
1204                                 }
1205                                 merge = false;
1206                         }
1207 
1208                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1209 
1210                         if (!sk_wmem_schedule(sk, copy))
1211                                 goto wait_for_memory;
1212 
1213                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1214                                                        pfrag->page,
1215                                                        pfrag->offset,
1216                                                        copy);
1217                         if (err)
1218                                 goto do_error;
1219 
1220                         /* Update the skb. */
1221                         if (merge) {
1222                                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1223                         } else {
1224                                 skb_fill_page_desc(skb, i, pfrag->page,
1225                                                    pfrag->offset, copy);
1226                                 get_page(pfrag->page);
1227                         }
1228                         pfrag->offset += copy;
1229                 }
1230 
1231                 if (!copied)
1232                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1233 
1234                 tp->write_seq += copy;
1235                 TCP_SKB_CB(skb)->end_seq += copy;
1236                 tcp_skb_pcount_set(skb, 0);
1237 
1238                 copied += copy;
1239                 if (!msg_data_left(msg)) {
1240                         tcp_tx_timestamp(sk, skb);
1241                         goto out;
1242                 }
1243 
1244                 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
1245                         continue;
1246 
1247                 if (forced_push(tp)) {
1248                         tcp_mark_push(tp, skb);
1249                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1250                 } else if (skb == tcp_send_head(sk))
1251                         tcp_push_one(sk, mss_now);
1252                 continue;
1253 
1254 wait_for_sndbuf:
1255                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1256 wait_for_memory:
1257                 if (copied)
1258                         tcp_push(sk, flags & ~MSG_MORE, mss_now,
1259                                  TCP_NAGLE_PUSH, size_goal);
1260 
1261                 err = sk_stream_wait_memory(sk, &timeo);
1262                 if (err != 0)
1263                         goto do_error;
1264 
1265                 mss_now = tcp_send_mss(sk, &size_goal, flags);
1266         }
1267 
1268 out:
1269         if (copied)
1270                 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1271 out_nopush:
1272         release_sock(sk);
1273         return copied + copied_syn;
1274 
1275 do_fault:
1276         if (!skb->len) {
1277                 tcp_unlink_write_queue(skb, sk);
1278                 /* It is the one place in all of TCP, except connection
1279                  * reset, where we can be unlinking the send_head.
1280                  */
1281                 tcp_check_send_head(sk, skb);
1282                 sk_wmem_free_skb(sk, skb);
1283         }
1284 
1285 do_error:
1286         if (copied + copied_syn)
1287                 goto out;
1288 out_err:
1289         err = sk_stream_error(sk, flags, err);
1290         /* make sure we wake any epoll edge trigger waiter */
1291         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1292                 sk->sk_write_space(sk);
1293         release_sock(sk);
1294         return err;
1295 }
1296 EXPORT_SYMBOL(tcp_sendmsg);
1297 
1298 /*
1299  *      Handle reading urgent data. BSD has very simple semantics for
1300  *      this, no blocking and very strange errors 8)
1301  */
1302 
1303 static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1304 {
1305         struct tcp_sock *tp = tcp_sk(sk);
1306 
1307         /* No URG data to read. */
1308         if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1309             tp->urg_data == TCP_URG_READ)
1310                 return -EINVAL; /* Yes this is right ! */
1311 
1312         if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1313                 return -ENOTCONN;
1314 
1315         if (tp->urg_data & TCP_URG_VALID) {
1316                 int err = 0;
1317                 char c = tp->urg_data;
1318 
1319                 if (!(flags & MSG_PEEK))
1320                         tp->urg_data = TCP_URG_READ;
1321 
1322                 /* Read urgent data. */
1323                 msg->msg_flags |= MSG_OOB;
1324 
1325                 if (len > 0) {
1326                         if (!(flags & MSG_TRUNC))
1327                                 err = memcpy_to_msg(msg, &c, 1);
1328                         len = 1;
1329                 } else
1330                         msg->msg_flags |= MSG_TRUNC;
1331 
1332                 return err ? -EFAULT : len;
1333         }
1334 
1335         if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1336                 return 0;
1337 
1338         /* Fixed the recv(..., MSG_OOB) behaviour.  BSD docs and
1339          * the available implementations agree in this case:
1340          * this call should never block, independent of the
1341          * blocking state of the socket.
1342          * Mike <pall@rz.uni-karlsruhe.de>
1343          */
1344         return -EAGAIN;
1345 }
1346 
1347 static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1348 {
1349         struct sk_buff *skb;
1350         int copied = 0, err = 0;
1351 
1352         /* XXX -- need to support SO_PEEK_OFF */
1353 
1354         skb_queue_walk(&sk->sk_write_queue, skb) {
1355                 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1356                 if (err)
1357                         break;
1358 
1359                 copied += skb->len;
1360         }
1361 
1362         return err ?: copied;
1363 }
1364 
1365 /* Clean up the receive buffer for full frames taken by the user,
1366  * then send an ACK if necessary.  COPIED is the number of bytes
1367  * tcp_recvmsg has given to the user so far, it speeds up the
1368  * calculation of whether or not we must ACK for the sake of
1369  * a window update.
1370  */
1371 static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1372 {
1373         struct tcp_sock *tp = tcp_sk(sk);
1374         bool time_to_ack = false;
1375 
1376         struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1377 
1378         WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1379              "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1380              tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1381 
1382         if (inet_csk_ack_scheduled(sk)) {
1383                 const struct inet_connection_sock *icsk = inet_csk(sk);
1384                    /* Delayed ACKs frequently hit locked sockets during bulk
1385                     * receive. */
1386                 if (icsk->icsk_ack.blocked ||
1387                     /* Once-per-two-segments ACK was not sent by tcp_input.c */
1388                     tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1389                     /*
1390                      * If this read emptied read buffer, we send ACK, if
1391                      * connection is not bidirectional, user drained
1392                      * receive buffer and there was a small segment
1393                      * in queue.
1394                      */
1395                     (copied > 0 &&
1396                      ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1397                       ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1398                        !icsk->icsk_ack.pingpong)) &&
1399                       !atomic_read(&sk->sk_rmem_alloc)))
1400                         time_to_ack = true;
1401         }
1402 
1403         /* We send an ACK if we can now advertise a non-zero window
1404          * which has been raised "significantly".
1405          *
1406          * Even if window raised up to infinity, do not send window open ACK
1407          * in states, where we will not receive more. It is useless.
1408          */
1409         if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1410                 __u32 rcv_window_now = tcp_receive_window(tp);
1411 
1412                 /* Optimize, __tcp_select_window() is not cheap. */
1413                 if (2*rcv_window_now <= tp->window_clamp) {
1414                         __u32 new_window = __tcp_select_window(sk);
1415 
1416                         /* Send ACK now, if this read freed lots of space
1417                          * in our buffer. Certainly, new_window is new window.
1418                          * We can advertise it now, if it is not less than current one.
1419                          * "Lots" means "at least twice" here.
1420                          */
1421                         if (new_window && new_window >= 2 * rcv_window_now)
1422                                 time_to_ack = true;
1423                 }
1424         }
1425         if (time_to_ack)
1426                 tcp_send_ack(sk);
1427 }
1428 
1429 static void tcp_prequeue_process(struct sock *sk)
1430 {
1431         struct sk_buff *skb;
1432         struct tcp_sock *tp = tcp_sk(sk);
1433 
1434         NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
1435 
1436         /* RX process wants to run with disabled BHs, though it is not
1437          * necessary */
1438         local_bh_disable();
1439         while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1440                 sk_backlog_rcv(sk, skb);
1441         local_bh_enable();
1442 
1443         /* Clear memory counter. */
1444         tp->ucopy.memory = 0;
1445 }
1446 
1447 static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1448 {
1449         struct sk_buff *skb;
1450         u32 offset;
1451 
1452         while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1453                 offset = seq - TCP_SKB_CB(skb)->seq;
1454                 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1455                         pr_err_once("%s: found a SYN, please report !\n", __func__);
1456                         offset--;
1457                 }
1458                 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1459                         *off = offset;
1460                         return skb;
1461                 }
1462                 /* This looks weird, but this can happen if TCP collapsing
1463                  * splitted a fat GRO packet, while we released socket lock
1464                  * in skb_splice_bits()
1465                  */
1466                 sk_eat_skb(sk, skb);
1467         }
1468         return NULL;
1469 }
1470 
1471 /*
1472  * This routine provides an alternative to tcp_recvmsg() for routines
1473  * that would like to handle copying from skbuffs directly in 'sendfile'
1474  * fashion.
1475  * Note:
1476  *      - It is assumed that the socket was locked by the caller.
1477  *      - The routine does not block.
1478  *      - At present, there is no support for reading OOB data
1479  *        or for 'peeking' the socket using this routine
1480  *        (although both would be easy to implement).
1481  */
1482 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1483                   sk_read_actor_t recv_actor)
1484 {
1485         struct sk_buff *skb;
1486         struct tcp_sock *tp = tcp_sk(sk);
1487         u32 seq = tp->copied_seq;
1488         u32 offset;
1489         int copied = 0;
1490 
1491         if (sk->sk_state == TCP_LISTEN)
1492                 return -ENOTCONN;
1493         while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1494                 if (offset < skb->len) {
1495                         int used;
1496                         size_t len;
1497 
1498                         len = skb->len - offset;
1499                         /* Stop reading if we hit a patch of urgent data */
1500                         if (tp->urg_data) {
1501                                 u32 urg_offset = tp->urg_seq - seq;
1502                                 if (urg_offset < len)
1503                                         len = urg_offset;
1504                                 if (!len)
1505                                         break;
1506                         }
1507                         used = recv_actor(desc, skb, offset, len);
1508                         if (used <= 0) {
1509                                 if (!copied)
1510                                         copied = used;
1511                                 break;
1512                         } else if (used <= len) {
1513                                 seq += used;
1514                                 copied += used;
1515                                 offset += used;
1516                         }
1517                         /* If recv_actor drops the lock (e.g. TCP splice
1518                          * receive) the skb pointer might be invalid when
1519                          * getting here: tcp_collapse might have deleted it
1520                          * while aggregating skbs from the socket queue.
1521                          */
1522                         skb = tcp_recv_skb(sk, seq - 1, &offset);
1523                         if (!skb)
1524                                 break;
1525                         /* TCP coalescing might have appended data to the skb.
1526                          * Try to splice more frags
1527                          */
1528                         if (offset + 1 != skb->len)
1529                                 continue;
1530                 }
1531                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
1532                         sk_eat_skb(sk, skb);
1533                         ++seq;
1534                         break;
1535                 }
1536                 sk_eat_skb(sk, skb);
1537                 if (!desc->count)
1538                         break;
1539                 tp->copied_seq = seq;
1540         }
1541         tp->copied_seq = seq;
1542 
1543         tcp_rcv_space_adjust(sk);
1544 
1545         /* Clean up data we have read: This will do ACK frames. */
1546         if (copied > 0) {
1547                 tcp_recv_skb(sk, seq, &offset);
1548                 tcp_cleanup_rbuf(sk, copied);
1549         }
1550         return copied;
1551 }
1552 EXPORT_SYMBOL(tcp_read_sock);
1553 
1554 /*
1555  *      This routine copies from a sock struct into the user buffer.
1556  *
1557  *      Technical note: in 2.3 we work on _locked_ socket, so that
1558  *      tricks with *seq access order and skb->users are not required.
1559  *      Probably, code can be easily improved even more.
1560  */
1561 
1562 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1563                 int flags, int *addr_len)
1564 {
1565         struct tcp_sock *tp = tcp_sk(sk);
1566         int copied = 0;
1567         u32 peek_seq;
1568         u32 *seq;
1569         unsigned long used;
1570         int err;
1571         int target;             /* Read at least this many bytes */
1572         long timeo;
1573         struct task_struct *user_recv = NULL;
1574         struct sk_buff *skb, *last;
1575         u32 urg_hole = 0;
1576 
1577         if (unlikely(flags & MSG_ERRQUEUE))
1578                 return inet_recv_error(sk, msg, len, addr_len);
1579 
1580         if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1581             (sk->sk_state == TCP_ESTABLISHED))
1582                 sk_busy_loop(sk, nonblock);
1583 
1584         lock_sock(sk);
1585 
1586         err = -ENOTCONN;
1587         if (sk->sk_state == TCP_LISTEN)
1588                 goto out;
1589 
1590         timeo = sock_rcvtimeo(sk, nonblock);
1591 
1592         /* Urgent data needs to be handled specially. */
1593         if (flags & MSG_OOB)
1594                 goto recv_urg;
1595 
1596         if (unlikely(tp->repair)) {
1597                 err = -EPERM;
1598                 if (!(flags & MSG_PEEK))
1599                         goto out;
1600 
1601                 if (tp->repair_queue == TCP_SEND_QUEUE)
1602                         goto recv_sndq;
1603 
1604                 err = -EINVAL;
1605                 if (tp->repair_queue == TCP_NO_QUEUE)
1606                         goto out;
1607 
1608                 /* 'common' recv queue MSG_PEEK-ing */
1609         }
1610 
1611         seq = &tp->copied_seq;
1612         if (flags & MSG_PEEK) {
1613                 peek_seq = tp->copied_seq;
1614                 seq = &peek_seq;
1615         }
1616 
1617         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1618 
1619         do {
1620                 u32 offset;
1621 
1622                 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1623                 if (tp->urg_data && tp->urg_seq == *seq) {
1624                         if (copied)
1625                                 break;
1626                         if (signal_pending(current)) {
1627                                 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1628                                 break;
1629                         }
1630                 }
1631 
1632                 /* Next get a buffer. */
1633 
1634                 last = skb_peek_tail(&sk->sk_receive_queue);
1635                 skb_queue_walk(&sk->sk_receive_queue, skb) {
1636                         last = skb;
1637                         /* Now that we have two receive queues this
1638                          * shouldn't happen.
1639                          */
1640                         if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
1641                                  "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1642                                  *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1643                                  flags))
1644                                 break;
1645 
1646                         offset = *seq - TCP_SKB_CB(skb)->seq;
1647                         if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1648                                 pr_err_once("%s: found a SYN, please report !\n", __func__);
1649                                 offset--;
1650                         }
1651                         if (offset < skb->len)
1652                                 goto found_ok_skb;
1653                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1654                                 goto found_fin_ok;
1655                         WARN(!(flags & MSG_PEEK),
1656                              "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1657                              *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
1658                 }
1659 
1660                 /* Well, if we have backlog, try to process it now yet. */
1661 
1662                 if (copied >= target && !sk->sk_backlog.tail)
1663                         break;
1664 
1665                 if (copied) {
1666                         if (sk->sk_err ||
1667                             sk->sk_state == TCP_CLOSE ||
1668                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1669                             !timeo ||
1670                             signal_pending(current))
1671                                 break;
1672                 } else {
1673                         if (sock_flag(sk, SOCK_DONE))
1674                                 break;
1675 
1676                         if (sk->sk_err) {
1677                                 copied = sock_error(sk);
1678                                 break;
1679                         }
1680 
1681                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1682                                 break;
1683 
1684                         if (sk->sk_state == TCP_CLOSE) {
1685                                 if (!sock_flag(sk, SOCK_DONE)) {
1686                                         /* This occurs when user tries to read
1687                                          * from never connected socket.
1688                                          */
1689                                         copied = -ENOTCONN;
1690                                         break;
1691                                 }
1692                                 break;
1693                         }
1694 
1695                         if (!timeo) {
1696                                 copied = -EAGAIN;
1697                                 break;
1698                         }
1699 
1700                         if (signal_pending(current)) {
1701                                 copied = sock_intr_errno(timeo);
1702                                 break;
1703                         }
1704                 }
1705 
1706                 tcp_cleanup_rbuf(sk, copied);
1707 
1708                 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
1709                         /* Install new reader */
1710                         if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1711                                 user_recv = current;
1712                                 tp->ucopy.task = user_recv;
1713                                 tp->ucopy.msg = msg;
1714                         }
1715 
1716                         tp->ucopy.len = len;
1717 
1718                         WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1719                                 !(flags & (MSG_PEEK | MSG_TRUNC)));
1720 
1721                         /* Ugly... If prequeue is not empty, we have to
1722                          * process it before releasing socket, otherwise
1723                          * order will be broken at second iteration.
1724                          * More elegant solution is required!!!
1725                          *
1726                          * Look: we have the following (pseudo)queues:
1727                          *
1728                          * 1. packets in flight
1729                          * 2. backlog
1730                          * 3. prequeue
1731                          * 4. receive_queue
1732                          *
1733                          * Each queue can be processed only if the next ones
1734                          * are empty. At this point we have empty receive_queue.
1735                          * But prequeue _can_ be not empty after 2nd iteration,
1736                          * when we jumped to start of loop because backlog
1737                          * processing added something to receive_queue.
1738                          * We cannot release_sock(), because backlog contains
1739                          * packets arrived _after_ prequeued ones.
1740                          *
1741                          * Shortly, algorithm is clear --- to process all
1742                          * the queues in order. We could make it more directly,
1743                          * requeueing packets from backlog to prequeue, if
1744                          * is not empty. It is more elegant, but eats cycles,
1745                          * unfortunately.
1746                          */
1747                         if (!skb_queue_empty(&tp->ucopy.prequeue))
1748                                 goto do_prequeue;
1749 
1750                         /* __ Set realtime policy in scheduler __ */
1751                 }
1752 
1753                 if (copied >= target) {
1754                         /* Do not sleep, just process backlog. */
1755                         release_sock(sk);
1756                         lock_sock(sk);
1757                 } else {
1758                         sk_wait_data(sk, &timeo, last);
1759                 }
1760 
1761                 if (user_recv) {
1762                         int chunk;
1763 
1764                         /* __ Restore normal policy in scheduler __ */
1765 
1766                         chunk = len - tp->ucopy.len;
1767                         if (chunk != 0) {
1768                                 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
1769                                 len -= chunk;
1770                                 copied += chunk;
1771                         }
1772 
1773                         if (tp->rcv_nxt == tp->copied_seq &&
1774                             !skb_queue_empty(&tp->ucopy.prequeue)) {
1775 do_prequeue:
1776                                 tcp_prequeue_process(sk);
1777 
1778                                 chunk = len - tp->ucopy.len;
1779                                 if (chunk != 0) {
1780                                         NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1781                                         len -= chunk;
1782                                         copied += chunk;
1783                                 }
1784                         }
1785                 }
1786                 if ((flags & MSG_PEEK) &&
1787                     (peek_seq - copied - urg_hole != tp->copied_seq)) {
1788                         net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1789                                             current->comm,
1790                                             task_pid_nr(current));
1791                         peek_seq = tp->copied_seq;
1792                 }
1793                 continue;
1794 
1795         found_ok_skb:
1796                 /* Ok so how much can we use? */
1797                 used = skb->len - offset;
1798                 if (len < used)
1799                         used = len;
1800 
1801                 /* Do we have urgent data here? */
1802                 if (tp->urg_data) {
1803                         u32 urg_offset = tp->urg_seq - *seq;
1804                         if (urg_offset < used) {
1805                                 if (!urg_offset) {
1806                                         if (!sock_flag(sk, SOCK_URGINLINE)) {
1807                                                 ++*seq;
1808                                                 urg_hole++;
1809                                                 offset++;
1810                                                 used--;
1811                                                 if (!used)
1812                                                         goto skip_copy;
1813                                         }
1814                                 } else
1815                                         used = urg_offset;
1816                         }
1817                 }
1818 
1819                 if (!(flags & MSG_TRUNC)) {
1820                         err = skb_copy_datagram_msg(skb, offset, msg, used);
1821                         if (err) {
1822                                 /* Exception. Bailout! */
1823                                 if (!copied)
1824                                         copied = -EFAULT;
1825                                 break;
1826                         }
1827                 }
1828 
1829                 *seq += used;
1830                 copied += used;
1831                 len -= used;
1832 
1833                 tcp_rcv_space_adjust(sk);
1834 
1835 skip_copy:
1836                 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1837                         tp->urg_data = 0;
1838                         tcp_fast_path_check(sk);
1839                 }
1840                 if (used + offset < skb->len)
1841                         continue;
1842 
1843                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1844                         goto found_fin_ok;
1845                 if (!(flags & MSG_PEEK))
1846                         sk_eat_skb(sk, skb);
1847                 continue;
1848 
1849         found_fin_ok:
1850                 /* Process the FIN. */
1851                 ++*seq;
1852                 if (!(flags & MSG_PEEK))
1853                         sk_eat_skb(sk, skb);
1854                 break;
1855         } while (len > 0);
1856 
1857         if (user_recv) {
1858                 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
1859                         int chunk;
1860 
1861                         tp->ucopy.len = copied > 0 ? len : 0;
1862 
1863                         tcp_prequeue_process(sk);
1864 
1865                         if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
1866                                 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1867                                 len -= chunk;
1868                                 copied += chunk;
1869                         }
1870                 }
1871 
1872                 tp->ucopy.task = NULL;
1873                 tp->ucopy.len = 0;
1874         }
1875 
1876         /* According to UNIX98, msg_name/msg_namelen are ignored
1877          * on connected socket. I was just happy when found this 8) --ANK
1878          */
1879 
1880         /* Clean up data we have read: This will do ACK frames. */
1881         tcp_cleanup_rbuf(sk, copied);
1882 
1883         release_sock(sk);
1884         return copied;
1885 
1886 out:
1887         release_sock(sk);
1888         return err;
1889 
1890 recv_urg:
1891         err = tcp_recv_urg(sk, msg, len, flags);
1892         goto out;
1893 
1894 recv_sndq:
1895         err = tcp_peek_sndq(sk, msg, len);
1896         goto out;
1897 }
1898 EXPORT_SYMBOL(tcp_recvmsg);
1899 
1900 void tcp_set_state(struct sock *sk, int state)
1901 {
1902         int oldstate = sk->sk_state;
1903 
1904         switch (state) {
1905         case TCP_ESTABLISHED:
1906                 if (oldstate != TCP_ESTABLISHED)
1907                         TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
1908                 break;
1909 
1910         case TCP_CLOSE:
1911                 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1912                         TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
1913 
1914                 sk->sk_prot->unhash(sk);
1915                 if (inet_csk(sk)->icsk_bind_hash &&
1916                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1917                         inet_put_port(sk);
1918                 /* fall through */
1919         default:
1920                 if (oldstate == TCP_ESTABLISHED)
1921                         TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
1922         }
1923 
1924         /* Change state AFTER socket is unhashed to avoid closed
1925          * socket sitting in hash tables.
1926          */
1927         sk_state_store(sk, state);
1928 
1929 #ifdef STATE_TRACE
1930         SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
1931 #endif
1932 }
1933 EXPORT_SYMBOL_GPL(tcp_set_state);
1934 
1935 /*
1936  *      State processing on a close. This implements the state shift for
1937  *      sending our FIN frame. Note that we only send a FIN for some
1938  *      states. A shutdown() may have already sent the FIN, or we may be
1939  *      closed.
1940  */
1941 
1942 static const unsigned char new_state[16] = {
1943   /* current state:        new state:      action:      */
1944   [0 /* (Invalid) */]   = TCP_CLOSE,
1945   [TCP_ESTABLISHED]     = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1946   [TCP_SYN_SENT]        = TCP_CLOSE,
1947   [TCP_SYN_RECV]        = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1948   [TCP_FIN_WAIT1]       = TCP_FIN_WAIT1,
1949   [TCP_FIN_WAIT2]       = TCP_FIN_WAIT2,
1950   [TCP_TIME_WAIT]       = TCP_CLOSE,
1951   [TCP_CLOSE]           = TCP_CLOSE,
1952   [TCP_CLOSE_WAIT]      = TCP_LAST_ACK  | TCP_ACTION_FIN,
1953   [TCP_LAST_ACK]        = TCP_LAST_ACK,
1954   [TCP_LISTEN]          = TCP_CLOSE,
1955   [TCP_CLOSING]         = TCP_CLOSING,
1956   [TCP_NEW_SYN_RECV]    = TCP_CLOSE,    /* should not happen ! */
1957 };
1958 
1959 static int tcp_close_state(struct sock *sk)
1960 {
1961         int next = (int)new_state[sk->sk_state];
1962         int ns = next & TCP_STATE_MASK;
1963 
1964         tcp_set_state(sk, ns);
1965 
1966         return next & TCP_ACTION_FIN;
1967 }
1968 
1969 /*
1970  *      Shutdown the sending side of a connection. Much like close except
1971  *      that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
1972  */
1973 
1974 void tcp_shutdown(struct sock *sk, int how)
1975 {
1976         /*      We need to grab some memory, and put together a FIN,
1977          *      and then put it into the queue to be sent.
1978          *              Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1979          */
1980         if (!(how & SEND_SHUTDOWN))
1981                 return;
1982 
1983         /* If we've already sent a FIN, or it's a closed state, skip this. */
1984         if ((1 << sk->sk_state) &
1985             (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1986              TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
1987                 /* Clear out any half completed packets.  FIN if needed. */
1988                 if (tcp_close_state(sk))
1989                         tcp_send_fin(sk);
1990         }
1991 }
1992 EXPORT_SYMBOL(tcp_shutdown);
1993 
1994 bool tcp_check_oom(struct sock *sk, int shift)
1995 {
1996         bool too_many_orphans, out_of_socket_memory;
1997 
1998         too_many_orphans = tcp_too_many_orphans(sk, shift);
1999         out_of_socket_memory = tcp_out_of_memory(sk);
2000 
2001         if (too_many_orphans)
2002                 net_info_ratelimited("too many orphaned sockets\n");
2003         if (out_of_socket_memory)
2004                 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
2005         return too_many_orphans || out_of_socket_memory;
2006 }
2007 
2008 void tcp_close(struct sock *sk, long timeout)
2009 {
2010         struct sk_buff *skb;
2011         int data_was_unread = 0;
2012         int state;
2013 
2014         lock_sock(sk);
2015         sk->sk_shutdown = SHUTDOWN_MASK;
2016 
2017         if (sk->sk_state == TCP_LISTEN) {
2018                 tcp_set_state(sk, TCP_CLOSE);
2019 
2020                 /* Special case. */
2021                 inet_csk_listen_stop(sk);
2022 
2023                 goto adjudge_to_death;
2024         }
2025 
2026         /*  We need to flush the recv. buffs.  We do this only on the
2027          *  descriptor close, not protocol-sourced closes, because the
2028          *  reader process may not have drained the data yet!
2029          */
2030         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2031                 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2032 
2033                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2034                         len--;
2035                 data_was_unread += len;
2036                 __kfree_skb(skb);
2037         }
2038 
2039         sk_mem_reclaim(sk);
2040 
2041         /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2042         if (sk->sk_state == TCP_CLOSE)
2043                 goto adjudge_to_death;
2044 
2045         /* As outlined in RFC 2525, section 2.17, we send a RST here because
2046          * data was lost. To witness the awful effects of the old behavior of
2047          * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2048          * GET in an FTP client, suspend the process, wait for the client to
2049          * advertise a zero window, then kill -9 the FTP client, wheee...
2050          * Note: timeout is always zero in such a case.
2051          */
2052         if (unlikely(tcp_sk(sk)->repair)) {
2053                 sk->sk_prot->disconnect(sk, 0);
2054         } else if (data_was_unread) {
2055                 /* Unread data was tossed, zap the connection. */
2056                 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
2057                 tcp_set_state(sk, TCP_CLOSE);
2058                 tcp_send_active_reset(sk, sk->sk_allocation);
2059         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2060                 /* Check zero linger _after_ checking for unread data. */
2061                 sk->sk_prot->disconnect(sk, 0);
2062                 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
2063         } else if (tcp_close_state(sk)) {
2064                 /* We FIN if the application ate all the data before
2065                  * zapping the connection.
2066                  */
2067 
2068                 /* RED-PEN. Formally speaking, we have broken TCP state
2069                  * machine. State transitions:
2070                  *
2071                  * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2072                  * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2073                  * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2074                  *
2075                  * are legal only when FIN has been sent (i.e. in window),
2076                  * rather than queued out of window. Purists blame.
2077                  *
2078                  * F.e. "RFC state" is ESTABLISHED,
2079                  * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2080                  *
2081                  * The visible declinations are that sometimes
2082                  * we enter time-wait state, when it is not required really
2083                  * (harmless), do not send active resets, when they are
2084                  * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2085                  * they look as CLOSING or LAST_ACK for Linux)
2086                  * Probably, I missed some more holelets.
2087                  *                                              --ANK
2088                  * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2089                  * in a single packet! (May consider it later but will
2090                  * probably need API support or TCP_CORK SYN-ACK until
2091                  * data is written and socket is closed.)
2092                  */
2093                 tcp_send_fin(sk);
2094         }
2095 
2096         sk_stream_wait_close(sk, timeout);
2097 
2098 adjudge_to_death:
2099         state = sk->sk_state;
2100         sock_hold(sk);
2101         sock_orphan(sk);
2102 
2103         /* It is the last release_sock in its life. It will remove backlog. */
2104         release_sock(sk);
2105 
2106 
2107         /* Now socket is owned by kernel and we acquire BH lock
2108            to finish close. No need to check for user refs.
2109          */
2110         local_bh_disable();
2111         bh_lock_sock(sk);
2112         WARN_ON(sock_owned_by_user(sk));
2113 
2114         percpu_counter_inc(sk->sk_prot->orphan_count);
2115 
2116         /* Have we already been destroyed by a softirq or backlog? */
2117         if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2118                 goto out;
2119 
2120         /*      This is a (useful) BSD violating of the RFC. There is a
2121          *      problem with TCP as specified in that the other end could
2122          *      keep a socket open forever with no application left this end.
2123          *      We use a 1 minute timeout (about the same as BSD) then kill
2124          *      our end. If they send after that then tough - BUT: long enough
2125          *      that we won't make the old 4*rto = almost no time - whoops
2126          *      reset mistake.
2127          *
2128          *      Nope, it was not mistake. It is really desired behaviour
2129          *      f.e. on http servers, when such sockets are useless, but
2130          *      consume significant resources. Let's do it with special
2131          *      linger2 option.                                 --ANK
2132          */
2133 
2134         if (sk->sk_state == TCP_FIN_WAIT2) {
2135                 struct tcp_sock *tp = tcp_sk(sk);
2136                 if (tp->linger2 < 0) {
2137                         tcp_set_state(sk, TCP_CLOSE);
2138                         tcp_send_active_reset(sk, GFP_ATOMIC);
2139                         NET_INC_STATS_BH(sock_net(sk),
2140                                         LINUX_MIB_TCPABORTONLINGER);
2141                 } else {
2142                         const int tmo = tcp_fin_time(sk);
2143 
2144                         if (tmo > TCP_TIMEWAIT_LEN) {
2145                                 inet_csk_reset_keepalive_timer(sk,
2146                                                 tmo - TCP_TIMEWAIT_LEN);
2147                         } else {
2148                                 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2149                                 goto out;
2150                         }
2151                 }
2152         }
2153         if (sk->sk_state != TCP_CLOSE) {
2154                 sk_mem_reclaim(sk);
2155                 if (tcp_check_oom(sk, 0)) {
2156                         tcp_set_state(sk, TCP_CLOSE);
2157                         tcp_send_active_reset(sk, GFP_ATOMIC);
2158                         NET_INC_STATS_BH(sock_net(sk),
2159                                         LINUX_MIB_TCPABORTONMEMORY);
2160                 }
2161         }
2162 
2163         if (sk->sk_state == TCP_CLOSE) {
2164                 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2165                 /* We could get here with a non-NULL req if the socket is
2166                  * aborted (e.g., closed with unread data) before 3WHS
2167                  * finishes.
2168                  */
2169                 if (req)
2170                         reqsk_fastopen_remove(sk, req, false);
2171                 inet_csk_destroy_sock(sk);
2172         }
2173         /* Otherwise, socket is reprieved until protocol close. */
2174 
2175 out:
2176         bh_unlock_sock(sk);
2177         local_bh_enable();
2178         sock_put(sk);
2179 }
2180 EXPORT_SYMBOL(tcp_close);
2181 
2182 /* These states need RST on ABORT according to RFC793 */
2183 
2184 static inline bool tcp_need_reset(int state)
2185 {
2186         return (1 << state) &
2187                (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2188                 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2189 }
2190 
2191 int tcp_disconnect(struct sock *sk, int flags)
2192 {
2193         struct inet_sock *inet = inet_sk(sk);
2194         struct inet_connection_sock *icsk = inet_csk(sk);
2195         struct tcp_sock *tp = tcp_sk(sk);
2196         int err = 0;
2197         int old_state = sk->sk_state;
2198 
2199         if (old_state != TCP_CLOSE)
2200                 tcp_set_state(sk, TCP_CLOSE);
2201 
2202         /* ABORT function of RFC793 */
2203         if (old_state == TCP_LISTEN) {
2204                 inet_csk_listen_stop(sk);
2205         } else if (unlikely(tp->repair)) {
2206                 sk->sk_err = ECONNABORTED;
2207         } else if (tcp_need_reset(old_state) ||
2208                    (tp->snd_nxt != tp->write_seq &&
2209                     (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2210                 /* The last check adjusts for discrepancy of Linux wrt. RFC
2211                  * states
2212                  */
2213                 tcp_send_active_reset(sk, gfp_any());
2214                 sk->sk_err = ECONNRESET;
2215         } else if (old_state == TCP_SYN_SENT)
2216                 sk->sk_err = ECONNRESET;
2217 
2218         tcp_clear_xmit_timers(sk);
2219         __skb_queue_purge(&sk->sk_receive_queue);
2220         tcp_write_queue_purge(sk);
2221         __skb_queue_purge(&tp->out_of_order_queue);
2222 
2223         inet->inet_dport = 0;
2224 
2225         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2226                 inet_reset_saddr(sk);
2227 
2228         sk->sk_shutdown = 0;
2229         sock_reset_flag(sk, SOCK_DONE);
2230         tp->srtt_us = 0;
2231         tp->write_seq += tp->max_window + 2;
2232         if (tp->write_seq == 0)
2233                 tp->write_seq = 1;
2234         icsk->icsk_backoff = 0;
2235         tp->snd_cwnd = 2;
2236         icsk->icsk_probes_out = 0;
2237         tp->packets_out = 0;
2238         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2239         tp->snd_cwnd_cnt = 0;
2240         tp->window_clamp = 0;
2241         tcp_set_ca_state(sk, TCP_CA_Open);
2242         tcp_clear_retrans(tp);
2243         inet_csk_delack_init(sk);
2244         tcp_init_send_head(sk);
2245         memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2246         __sk_dst_reset(sk);
2247 
2248         WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
2249 
2250         sk->sk_error_report(sk);
2251         return err;
2252 }
2253 EXPORT_SYMBOL(tcp_disconnect);
2254 
2255 static inline bool tcp_can_repair_sock(const struct sock *sk)
2256 {
2257         return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
2258                 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
2259 }
2260 
2261 static int tcp_repair_options_est(struct tcp_sock *tp,
2262                 struct tcp_repair_opt __user *optbuf, unsigned int len)
2263 {
2264         struct tcp_repair_opt opt;
2265 
2266         while (len >= sizeof(opt)) {
2267                 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2268                         return -EFAULT;
2269 
2270                 optbuf++;
2271                 len -= sizeof(opt);
2272 
2273                 switch (opt.opt_code) {
2274                 case TCPOPT_MSS:
2275                         tp->rx_opt.mss_clamp = opt.opt_val;
2276                         break;
2277                 case TCPOPT_WINDOW:
2278                         {
2279                                 u16 snd_wscale = opt.opt_val & 0xFFFF;
2280                                 u16 rcv_wscale = opt.opt_val >> 16;
2281 
2282                                 if (snd_wscale > 14 || rcv_wscale > 14)
2283                                         return -EFBIG;
2284 
2285                                 tp->rx_opt.snd_wscale = snd_wscale;
2286                                 tp->rx_opt.rcv_wscale = rcv_wscale;
2287                                 tp->rx_opt.wscale_ok = 1;
2288                         }
2289                         break;
2290                 case TCPOPT_SACK_PERM:
2291                         if (opt.opt_val != 0)
2292                                 return -EINVAL;
2293 
2294                         tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2295                         if (sysctl_tcp_fack)
2296                                 tcp_enable_fack(tp);
2297                         break;
2298                 case TCPOPT_TIMESTAMP:
2299                         if (opt.opt_val != 0)
2300                                 return -EINVAL;
2301 
2302                         tp->rx_opt.tstamp_ok = 1;
2303                         break;
2304                 }
2305         }
2306 
2307         return 0;
2308 }
2309 
2310 /*
2311  *      Socket option code for TCP.
2312  */
2313 static int do_tcp_setsockopt(struct sock *sk, int level,
2314                 int optname, char __user *optval, unsigned int optlen)
2315 {
2316         struct tcp_sock *tp = tcp_sk(sk);
2317         struct inet_connection_sock *icsk = inet_csk(sk);
2318         struct net *net = sock_net(sk);
2319         int val;
2320         int err = 0;
2321 
2322         /* These are data/string values, all the others are ints */
2323         switch (optname) {
2324         case TCP_CONGESTION: {
2325                 char name[TCP_CA_NAME_MAX];
2326 
2327                 if (optlen < 1)
2328                         return -EINVAL;
2329 
2330                 val = strncpy_from_user(name, optval,
2331                                         min_t(long, TCP_CA_NAME_MAX-1, optlen));
2332                 if (val < 0)
2333                         return -EFAULT;
2334                 name[val] = 0;
2335 
2336                 lock_sock(sk);
2337                 err = tcp_set_congestion_control(sk, name);
2338                 release_sock(sk);
2339                 return err;
2340         }
2341         default:
2342                 /* fallthru */
2343                 break;
2344         }
2345 
2346         if (optlen < sizeof(int))
2347                 return -EINVAL;
2348 
2349         if (get_user(val, (int __user *)optval))
2350                 return -EFAULT;
2351 
2352         lock_sock(sk);
2353 
2354         switch (optname) {
2355         case TCP_MAXSEG:
2356                 /* Values greater than interface MTU won't take effect. However
2357                  * at the point when this call is done we typically don't yet
2358                  * know which interface is going to be used */
2359                 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
2360                         err = -EINVAL;
2361                         break;
2362                 }
2363                 tp->rx_opt.user_mss = val;
2364                 break;
2365 
2366         case TCP_NODELAY:
2367                 if (val) {
2368                         /* TCP_NODELAY is weaker than TCP_CORK, so that
2369                          * this option on corked socket is remembered, but
2370                          * it is not activated until cork is cleared.
2371                          *
2372                          * However, when TCP_NODELAY is set we make
2373                          * an explicit push, which overrides even TCP_CORK
2374                          * for currently queued segments.
2375                          */
2376                         tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2377                         tcp_push_pending_frames(sk);
2378                 } else {
2379                         tp->nonagle &= ~TCP_NAGLE_OFF;
2380                 }
2381                 break;
2382 
2383         case TCP_THIN_LINEAR_TIMEOUTS:
2384                 if (val < 0 || val > 1)
2385                         err = -EINVAL;
2386                 else
2387                         tp->thin_lto = val;
2388                 break;
2389 
2390         case TCP_THIN_DUPACK:
2391                 if (val < 0 || val > 1)
2392                         err = -EINVAL;
2393                 else {
2394                         tp->thin_dupack = val;
2395                         if (tp->thin_dupack)
2396                                 tcp_disable_early_retrans(tp);
2397                 }
2398                 break;
2399 
2400         case TCP_REPAIR:
2401                 if (!tcp_can_repair_sock(sk))
2402                         err = -EPERM;
2403                 else if (val == 1) {
2404                         tp->repair = 1;
2405                         sk->sk_reuse = SK_FORCE_REUSE;
2406                         tp->repair_queue = TCP_NO_QUEUE;
2407                 } else if (val == 0) {
2408                         tp->repair = 0;
2409                         sk->sk_reuse = SK_NO_REUSE;
2410                         tcp_send_window_probe(sk);
2411                 } else
2412                         err = -EINVAL;
2413 
2414                 break;
2415 
2416         case TCP_REPAIR_QUEUE:
2417                 if (!tp->repair)
2418                         err = -EPERM;
2419                 else if (val < TCP_QUEUES_NR)
2420                         tp->repair_queue = val;
2421                 else
2422                         err = -EINVAL;
2423                 break;
2424 
2425         case TCP_QUEUE_SEQ:
2426                 if (sk->sk_state != TCP_CLOSE)
2427                         err = -EPERM;
2428                 else if (tp->repair_queue == TCP_SEND_QUEUE)
2429                         tp->write_seq = val;
2430                 else if (tp->repair_queue == TCP_RECV_QUEUE)
2431                         tp->rcv_nxt = val;
2432                 else
2433                         err = -EINVAL;
2434                 break;
2435 
2436         case TCP_REPAIR_OPTIONS:
2437                 if (!tp->repair)
2438                         err = -EINVAL;
2439                 else if (sk->sk_state == TCP_ESTABLISHED)
2440                         err = tcp_repair_options_est(tp,
2441                                         (struct tcp_repair_opt __user *)optval,
2442                                         optlen);
2443                 else
2444                         err = -EPERM;
2445                 break;
2446 
2447         case TCP_CORK:
2448                 /* When set indicates to always queue non-full frames.
2449                  * Later the user clears this option and we transmit
2450                  * any pending partial frames in the queue.  This is
2451                  * meant to be used alongside sendfile() to get properly
2452                  * filled frames when the user (for example) must write
2453                  * out headers with a write() call first and then use
2454                  * sendfile to send out the data parts.
2455                  *
2456                  * TCP_CORK can be set together with TCP_NODELAY and it is
2457                  * stronger than TCP_NODELAY.
2458                  */
2459                 if (val) {
2460                         tp->nonagle |= TCP_NAGLE_CORK;
2461                 } else {
2462                         tp->nonagle &= ~TCP_NAGLE_CORK;
2463                         if (tp->nonagle&TCP_NAGLE_OFF)
2464                                 tp->nonagle |= TCP_NAGLE_PUSH;
2465                         tcp_push_pending_frames(sk);
2466                 }
2467                 break;
2468 
2469         case TCP_KEEPIDLE:
2470                 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2471                         err = -EINVAL;
2472                 else {
2473                         tp->keepalive_time = val * HZ;
2474                         if (sock_flag(sk, SOCK_KEEPOPEN) &&
2475                             !((1 << sk->sk_state) &
2476                               (TCPF_CLOSE | TCPF_LISTEN))) {
2477                                 u32 elapsed = keepalive_time_elapsed(tp);
2478                                 if (tp->keepalive_time > elapsed)
2479                                         elapsed = tp->keepalive_time - elapsed;
2480                                 else
2481                                         elapsed = 0;
2482                                 inet_csk_reset_keepalive_timer(sk, elapsed);
2483                         }
2484                 }
2485                 break;
2486         case TCP_KEEPINTVL:
2487                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2488                         err = -EINVAL;
2489                 else
2490                         tp->keepalive_intvl = val * HZ;
2491                 break;
2492         case TCP_KEEPCNT:
2493                 if (val < 1 || val > MAX_TCP_KEEPCNT)
2494                         err = -EINVAL;
2495                 else
2496                         tp->keepalive_probes = val;
2497                 break;
2498         case TCP_SYNCNT:
2499                 if (val < 1 || val > MAX_TCP_SYNCNT)
2500                         err = -EINVAL;
2501                 else
2502                         icsk->icsk_syn_retries = val;
2503                 break;
2504 
2505         case TCP_SAVE_SYN:
2506                 if (val < 0 || val > 1)
2507                         err = -EINVAL;
2508                 else
2509                         tp->save_syn = val;
2510                 break;
2511 
2512         case TCP_LINGER2:
2513                 if (val < 0)
2514                         tp->linger2 = -1;
2515                 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
2516                         tp->linger2 = 0;
2517                 else
2518                         tp->linger2 = val * HZ;
2519                 break;
2520 
2521         case TCP_DEFER_ACCEPT:
2522                 /* Translate value in seconds to number of retransmits */
2523                 icsk->icsk_accept_queue.rskq_defer_accept =
2524                         secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2525                                         TCP_RTO_MAX / HZ);
2526                 break;
2527 
2528         case TCP_WINDOW_CLAMP:
2529                 if (!val) {
2530                         if (sk->sk_state != TCP_CLOSE) {
2531                                 err = -EINVAL;
2532                                 break;
2533                         }
2534                         tp->window_clamp = 0;
2535                 } else
2536                         tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2537                                                 SOCK_MIN_RCVBUF / 2 : val;
2538                 break;
2539 
2540         case TCP_QUICKACK:
2541                 if (!val) {
2542                         icsk->icsk_ack.pingpong = 1;
2543                 } else {
2544                         icsk->icsk_ack.pingpong = 0;
2545                         if ((1 << sk->sk_state) &
2546                             (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
2547                             inet_csk_ack_scheduled(sk)) {
2548                                 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
2549                                 tcp_cleanup_rbuf(sk, 1);
2550                                 if (!(val & 1))
2551                                         icsk->icsk_ack.pingpong = 1;
2552                         }
2553                 }
2554                 break;
2555 
2556 #ifdef CONFIG_TCP_MD5SIG
2557         case TCP_MD5SIG:
2558                 /* Read the IP->Key mappings from userspace */
2559                 err = tp->af_specific->md5_parse(sk, optval, optlen);
2560                 break;
2561 #endif
2562         case TCP_USER_TIMEOUT:
2563                 /* Cap the max time in ms TCP will retry or probe the window
2564                  * before giving up and aborting (ETIMEDOUT) a connection.
2565                  */
2566                 if (val < 0)
2567                         err = -EINVAL;
2568                 else
2569                         icsk->icsk_user_timeout = msecs_to_jiffies(val);
2570                 break;
2571 
2572         case TCP_FASTOPEN:
2573                 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
2574                     TCPF_LISTEN))) {
2575                         tcp_fastopen_init_key_once(true);
2576 
2577                         fastopen_queue_tune(sk, val);
2578                 } else {
2579                         err = -EINVAL;
2580                 }
2581                 break;
2582         case TCP_TIMESTAMP:
2583                 if (!tp->repair)
2584                         err = -EPERM;
2585                 else
2586                         tp->tsoffset = val - tcp_time_stamp;
2587                 break;
2588         case TCP_NOTSENT_LOWAT:
2589                 tp->notsent_lowat = val;
2590                 sk->sk_write_space(sk);
2591                 break;
2592         default:
2593                 err = -ENOPROTOOPT;
2594                 break;
2595         }
2596 
2597         release_sock(sk);
2598         return err;
2599 }
2600 
2601 int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
2602                    unsigned int optlen)
2603 {
2604         const struct inet_connection_sock *icsk = inet_csk(sk);
2605 
2606         if (level != SOL_TCP)
2607                 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2608                                                      optval, optlen);
2609         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2610 }
2611 EXPORT_SYMBOL(tcp_setsockopt);
2612 
2613 #ifdef CONFIG_COMPAT
2614 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
2615                           char __user *optval, unsigned int optlen)
2616 {
2617         if (level != SOL_TCP)
2618                 return inet_csk_compat_setsockopt(sk, level, optname,
2619                                                   optval, optlen);
2620         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2621 }
2622 EXPORT_SYMBOL(compat_tcp_setsockopt);
2623 #endif
2624 
2625 /* Return information about state of tcp endpoint in API format. */
2626 void tcp_get_info(struct sock *sk, struct tcp_info *info)
2627 {
2628         const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
2629         const struct inet_connection_sock *icsk = inet_csk(sk);
2630         u32 now = tcp_time_stamp;
2631         unsigned int start;
2632         int notsent_bytes;
2633         u64 rate64;
2634         u32 rate;
2635 
2636         memset(info, 0, sizeof(*info));
2637         if (sk->sk_type != SOCK_STREAM)
2638                 return;
2639 
2640         info->tcpi_state = sk_state_load(sk);
2641 
2642         info->tcpi_ca_state = icsk->icsk_ca_state;
2643         info->tcpi_retransmits = icsk->icsk_retransmits;
2644         info->tcpi_probes = icsk->icsk_probes_out;
2645         info->tcpi_backoff = icsk->icsk_backoff;
2646 
2647         if (tp->rx_opt.tstamp_ok)
2648                 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
2649         if (tcp_is_sack(tp))
2650                 info->tcpi_options |= TCPI_OPT_SACK;
2651         if (tp->rx_opt.wscale_ok) {
2652                 info->tcpi_options |= TCPI_OPT_WSCALE;
2653                 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2654                 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
2655         }
2656 
2657         if (tp->ecn_flags & TCP_ECN_OK)
2658                 info->tcpi_options |= TCPI_OPT_ECN;
2659         if (tp->ecn_flags & TCP_ECN_SEEN)
2660                 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
2661         if (tp->syn_data_acked)
2662                 info->tcpi_options |= TCPI_OPT_SYN_DATA;
2663 
2664         info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2665         info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
2666         info->tcpi_snd_mss = tp->mss_cache;
2667         info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
2668 
2669         if (info->tcpi_state == TCP_LISTEN) {
2670                 info->tcpi_unacked = sk->sk_ack_backlog;
2671                 info->tcpi_sacked = sk->sk_max_ack_backlog;
2672         } else {
2673                 info->tcpi_unacked = tp->packets_out;
2674                 info->tcpi_sacked = tp->sacked_out;
2675         }
2676         info->tcpi_lost = tp->lost_out;
2677         info->tcpi_retrans = tp->retrans_out;
2678         info->tcpi_fackets = tp->fackets_out;
2679 
2680         info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
2681         info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
2682         info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2683 
2684         info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
2685         info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2686         info->tcpi_rtt = tp->srtt_us >> 3;
2687         info->tcpi_rttvar = tp->mdev_us >> 2;
2688         info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2689         info->tcpi_snd_cwnd = tp->snd_cwnd;
2690         info->tcpi_advmss = tp->advmss;
2691         info->tcpi_reordering = tp->reordering;
2692 
2693         info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2694         info->tcpi_rcv_space = tp->rcvq_space.space;
2695 
2696         info->tcpi_total_retrans = tp->total_retrans;
2697 
2698         rate = READ_ONCE(sk->sk_pacing_rate);
2699         rate64 = rate != ~0U ? rate : ~0ULL;
2700         put_unaligned(rate64, &info->tcpi_pacing_rate);
2701 
2702         rate = READ_ONCE(sk->sk_max_pacing_rate);
2703         rate64 = rate != ~0U ? rate : ~0ULL;
2704         put_unaligned(rate64, &info->tcpi_max_pacing_rate);
2705 
2706         do {
2707                 start = u64_stats_fetch_begin_irq(&tp->syncp);
2708                 put_unaligned(tp->bytes_acked, &info->tcpi_bytes_acked);
2709                 put_unaligned(tp->bytes_received, &info->tcpi_bytes_received);
2710         } while (u64_stats_fetch_retry_irq(&tp->syncp, start));
2711         info->tcpi_segs_out = tp->segs_out;
2712         info->tcpi_segs_in = tp->segs_in;
2713 
2714         notsent_bytes = READ_ONCE(tp->write_seq) - READ_ONCE(tp->snd_nxt);
2715         info->tcpi_notsent_bytes = max(0, notsent_bytes);
2716 
2717         info->tcpi_min_rtt = tcp_min_rtt(tp);
2718         info->tcpi_data_segs_in = tp->data_segs_in;
2719         info->tcpi_data_segs_out = tp->data_segs_out;
2720 }
2721 EXPORT_SYMBOL_GPL(tcp_get_info);
2722 
2723 static int do_tcp_getsockopt(struct sock *sk, int level,
2724                 int optname, char __user *optval, int __user *optlen)
2725 {
2726         struct inet_connection_sock *icsk = inet_csk(sk);
2727         struct tcp_sock *tp = tcp_sk(sk);
2728         struct net *net = sock_net(sk);
2729         int val, len;
2730 
2731         if (get_user(len, optlen))
2732                 return -EFAULT;
2733 
2734         len = min_t(unsigned int, len, sizeof(int));
2735 
2736         if (len < 0)
2737                 return -EINVAL;
2738 
2739         switch (optname) {
2740         case TCP_MAXSEG:
2741                 val = tp->mss_cache;
2742                 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2743                         val = tp->rx_opt.user_mss;
2744                 if (tp->repair)
2745                         val = tp->rx_opt.mss_clamp;
2746                 break;
2747         case TCP_NODELAY:
2748                 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2749                 break;
2750         case TCP_CORK:
2751                 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2752                 break;
2753         case TCP_KEEPIDLE:
2754                 val = keepalive_time_when(tp) / HZ;
2755                 break;
2756         case TCP_KEEPINTVL:
2757                 val = keepalive_intvl_when(tp) / HZ;
2758                 break;
2759         case TCP_KEEPCNT:
2760                 val = keepalive_probes(tp);
2761                 break;
2762         case TCP_SYNCNT:
2763                 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
2764                 break;
2765         case TCP_LINGER2:
2766                 val = tp->linger2;
2767                 if (val >= 0)
2768                         val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
2769                 break;
2770         case TCP_DEFER_ACCEPT:
2771                 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2772                                       TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
2773                 break;
2774         case TCP_WINDOW_CLAMP:
2775                 val = tp->window_clamp;
2776                 break;
2777         case TCP_INFO: {
2778                 struct tcp_info info;
2779 
2780                 if (get_user(len, optlen))
2781                         return -EFAULT;
2782 
2783                 tcp_get_info(sk, &info);
2784 
2785                 len = min_t(unsigned int, len, sizeof(info));
2786                 if (put_user(len, optlen))
2787                         return -EFAULT;
2788                 if (copy_to_user(optval, &info, len))
2789                         return -EFAULT;
2790                 return 0;
2791         }
2792         case TCP_CC_INFO: {
2793                 const struct tcp_congestion_ops *ca_ops;
2794                 union tcp_cc_info info;
2795                 size_t sz = 0;
2796                 int attr;
2797 
2798                 if (get_user(len, optlen))
2799                         return -EFAULT;
2800 
2801                 ca_ops = icsk->icsk_ca_ops;
2802                 if (ca_ops && ca_ops->get_info)
2803                         sz = ca_ops->get_info(sk, ~0U, &attr, &info);
2804 
2805                 len = min_t(unsigned int, len, sz);
2806                 if (put_user(len, optlen))
2807                         return -EFAULT;
2808                 if (copy_to_user(optval, &info, len))
2809                         return -EFAULT;
2810                 return 0;
2811         }
2812         case TCP_QUICKACK:
2813                 val = !icsk->icsk_ack.pingpong;
2814                 break;
2815 
2816         case TCP_CONGESTION:
2817                 if (get_user(len, optlen))
2818                         return -EFAULT;
2819                 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2820                 if (put_user(len, optlen))
2821                         return -EFAULT;
2822                 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
2823                         return -EFAULT;
2824                 return 0;
2825 
2826         case TCP_THIN_LINEAR_TIMEOUTS:
2827                 val = tp->thin_lto;
2828                 break;
2829         case TCP_THIN_DUPACK:
2830                 val = tp->thin_dupack;
2831                 break;
2832 
2833         case TCP_REPAIR:
2834                 val = tp->repair;
2835                 break;
2836 
2837         case TCP_REPAIR_QUEUE:
2838                 if (tp->repair)
2839                         val = tp->repair_queue;
2840                 else
2841                         return -EINVAL;
2842                 break;
2843 
2844         case TCP_QUEUE_SEQ:
2845                 if (tp->repair_queue == TCP_SEND_QUEUE)
2846                         val = tp->write_seq;
2847                 else if (tp->repair_queue == TCP_RECV_QUEUE)
2848                         val = tp->rcv_nxt;
2849                 else
2850                         return -EINVAL;
2851                 break;
2852 
2853         case TCP_USER_TIMEOUT:
2854                 val = jiffies_to_msecs(icsk->icsk_user_timeout);
2855                 break;
2856 
2857         case TCP_FASTOPEN:
2858                 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
2859                 break;
2860 
2861         case TCP_TIMESTAMP:
2862                 val = tcp_time_stamp + tp->tsoffset;
2863                 break;
2864         case TCP_NOTSENT_LOWAT:
2865                 val = tp->notsent_lowat;
2866                 break;
2867         case TCP_SAVE_SYN:
2868                 val = tp->save_syn;
2869                 break;
2870         case TCP_SAVED_SYN: {
2871                 if (get_user(len, optlen))
2872                         return -EFAULT;
2873 
2874                 lock_sock(sk);
2875                 if (tp->saved_syn) {
2876                         if (len < tp->saved_syn[0]) {
2877                                 if (put_user(tp->saved_syn[0], optlen)) {
2878                                         release_sock(sk);
2879                                         return -EFAULT;
2880                                 }
2881                                 release_sock(sk);
2882                                 return -EINVAL;
2883                         }
2884                         len = tp->saved_syn[0];
2885                         if (put_user(len, optlen)) {
2886                                 release_sock(sk);
2887                                 return -EFAULT;
2888                         }
2889                         if (copy_to_user(optval, tp->saved_syn + 1, len)) {
2890                                 release_sock(sk);
2891                                 return -EFAULT;
2892                         }
2893                         tcp_saved_syn_free(tp);
2894                         release_sock(sk);
2895                 } else {
2896                         release_sock(sk);
2897                         len = 0;
2898                         if (put_user(len, optlen))
2899                                 return -EFAULT;
2900                 }
2901                 return 0;
2902         }
2903         default:
2904                 return -ENOPROTOOPT;
2905         }
2906 
2907         if (put_user(len, optlen))
2908                 return -EFAULT;
2909         if (copy_to_user(optval, &val, len))
2910                 return -EFAULT;
2911         return 0;
2912 }
2913 
2914 int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2915                    int __user *optlen)
2916 {
2917         struct inet_connection_sock *icsk = inet_csk(sk);
2918 
2919         if (level != SOL_TCP)
2920                 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2921                                                      optval, optlen);
2922         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2923 }
2924 EXPORT_SYMBOL(tcp_getsockopt);
2925 
2926 #ifdef CONFIG_COMPAT
2927 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2928                           char __user *optval, int __user *optlen)
2929 {
2930         if (level != SOL_TCP)
2931                 return inet_csk_compat_getsockopt(sk, level, optname,
2932                                                   optval, optlen);
2933         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2934 }
2935 EXPORT_SYMBOL(compat_tcp_getsockopt);
2936 #endif
2937 
2938 #ifdef CONFIG_TCP_MD5SIG
2939 static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
2940 static DEFINE_MUTEX(tcp_md5sig_mutex);
2941 static bool tcp_md5sig_pool_populated = false;
2942 
2943 static void __tcp_alloc_md5sig_pool(void)
2944 {
2945         struct crypto_ahash *hash;
2946         int cpu;
2947 
2948         hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
2949         if (IS_ERR(hash))
2950                 return;
2951 
2952         for_each_possible_cpu(cpu) {
2953                 struct ahash_request *req;
2954 
2955                 if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
2956                         continue;
2957 
2958                 req = ahash_request_alloc(hash, GFP_KERNEL);
2959                 if (!req)
2960                         return;
2961 
2962                 ahash_request_set_callback(req, 0, NULL, NULL);
2963 
2964                 per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
2965         }
2966         /* before setting tcp_md5sig_pool_populated, we must commit all writes
2967          * to memory. See smp_rmb() in tcp_get_md5sig_pool()
2968          */
2969         smp_wmb();
2970         tcp_md5sig_pool_populated = true;
2971 }
2972 
2973 bool tcp_alloc_md5sig_pool(void)
2974 {
2975         if (unlikely(!tcp_md5sig_pool_populated)) {
2976                 mutex_lock(&tcp_md5sig_mutex);
2977 
2978                 if (!tcp_md5sig_pool_populated)
2979                         __tcp_alloc_md5sig_pool();
2980 
2981                 mutex_unlock(&tcp_md5sig_mutex);
2982         }
2983         return tcp_md5sig_pool_populated;
2984 }
2985 EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2986 
2987 
2988 /**
2989  *      tcp_get_md5sig_pool - get md5sig_pool for this user
2990  *
2991  *      We use percpu structure, so if we succeed, we exit with preemption
2992  *      and BH disabled, to make sure another thread or softirq handling
2993  *      wont try to get same context.
2994  */
2995 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
2996 {
2997         local_bh_disable();
2998 
2999         if (tcp_md5sig_pool_populated) {
3000                 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3001                 smp_rmb();
3002                 return this_cpu_ptr(&tcp_md5sig_pool);
3003         }
3004         local_bh_enable();
3005         return NULL;
3006 }
3007 EXPORT_SYMBOL(tcp_get_md5sig_pool);
3008 
3009 int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
3010                         const struct tcphdr *th)
3011 {
3012         struct scatterlist sg;
3013         struct tcphdr hdr;
3014 
3015         /* We are not allowed to change tcphdr, make a local copy */
3016         memcpy(&hdr, th, sizeof(hdr));
3017         hdr.check = 0;
3018 
3019         /* options aren't included in the hash */
3020         sg_init_one(&sg, &hdr, sizeof(hdr));
3021         ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(hdr));
3022         return crypto_ahash_update(hp->md5_req);
3023 }
3024 EXPORT_SYMBOL(tcp_md5_hash_header);
3025 
3026 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
3027                           const struct sk_buff *skb, unsigned int header_len)
3028 {
3029         struct scatterlist sg;
3030         const struct tcphdr *tp = tcp_hdr(skb);
3031         struct ahash_request *req = hp->md5_req;
3032         unsigned int i;
3033         const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3034                                            skb_headlen(skb) - header_len : 0;
3035         const struct skb_shared_info *shi = skb_shinfo(skb);
3036         struct sk_buff *frag_iter;
3037 
3038         sg_init_table(&sg, 1);
3039 
3040         sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3041         ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3042         if (crypto_ahash_update(req))
3043                 return 1;
3044 
3045         for (i = 0; i < shi->nr_frags; ++i) {
3046                 const struct skb_frag_struct *f = &shi->frags[i];
3047                 unsigned int offset = f->page_offset;
3048                 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3049 
3050                 sg_set_page(&sg, page, skb_frag_size(f),
3051                             offset_in_page(offset));
3052                 ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3053                 if (crypto_ahash_update(req))
3054                         return 1;
3055         }
3056 
3057         skb_walk_frags(skb, frag_iter)
3058                 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3059                         return 1;
3060 
3061         return 0;
3062 }
3063 EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3064 
3065 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
3066 {
3067         struct scatterlist sg;
3068 
3069         sg_init_one(&sg, key->key, key->keylen);
3070         ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3071         return crypto_ahash_update(hp->md5_req);
3072 }
3073 EXPORT_SYMBOL(tcp_md5_hash_key);
3074 
3075 #endif
3076 
3077 void tcp_done(struct sock *sk)
3078 {
3079         struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3080 
3081         if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
3082                 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
3083 
3084         tcp_set_state(sk, TCP_CLOSE);
3085         tcp_clear_xmit_timers(sk);
3086         if (req)
3087                 reqsk_fastopen_remove(sk, req, false);
3088 
3089         sk->sk_shutdown = SHUTDOWN_MASK;
3090 
3091         if (!sock_flag(sk, SOCK_DEAD))
3092                 sk->sk_state_change(sk);
3093         else
3094                 inet_csk_destroy_sock(sk);
3095 }
3096 EXPORT_SYMBOL_GPL(tcp_done);
3097 
3098 int tcp_abort(struct sock *sk, int err)
3099 {
3100         if (!sk_fullsock(sk)) {
3101                 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3102                         struct request_sock *req = inet_reqsk(sk);
3103 
3104                         local_bh_disable();
3105                         inet_csk_reqsk_queue_drop_and_put(req->rsk_listener,
3106                                                           req);
3107                         local_bh_enable();
3108                         return 0;
3109                 }
3110                 sock_gen_put(sk);
3111                 return -EOPNOTSUPP;
3112         }
3113 
3114         /* Don't race with userspace socket closes such as tcp_close. */
3115         lock_sock(sk);
3116 
3117         if (sk->sk_state == TCP_LISTEN) {
3118                 tcp_set_state(sk, TCP_CLOSE);
3119                 inet_csk_listen_stop(sk);
3120         }
3121 
3122         /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3123         local_bh_disable();
3124         bh_lock_sock(sk);
3125 
3126         if (!sock_flag(sk, SOCK_DEAD)) {
3127                 sk->sk_err = err;
3128                 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3129                 smp_wmb();
3130                 sk->sk_error_report(sk);
3131                 if (tcp_need_reset(sk->sk_state))
3132                         tcp_send_active_reset(sk, GFP_ATOMIC);
3133                 tcp_done(sk);
3134         }
3135 
3136         bh_unlock_sock(sk);
3137         local_bh_enable();
3138         release_sock(sk);
3139         sock_put(sk);
3140         return 0;
3141 }
3142 EXPORT_SYMBOL_GPL(tcp_abort);
3143 
3144 extern struct tcp_congestion_ops tcp_reno;
3145 
3146 static __initdata unsigned long thash_entries;
3147 static int __init set_thash_entries(char *str)
3148 {
3149         ssize_t ret;
3150 
3151         if (!str)
3152                 return 0;
3153 
3154         ret = kstrtoul(str, 0, &thash_entries);
3155         if (ret)
3156                 return 0;
3157 
3158         return 1;
3159 }
3160 __setup("thash_entries=", set_thash_entries);
3161 
3162 static void __init tcp_init_mem(void)
3163 {
3164         unsigned long limit = nr_free_buffer_pages() / 16;
3165 
3166         limit = max(limit, 128UL);
3167         sysctl_tcp_mem[0] = limit / 4 * 3;              /* 4.68 % */
3168         sysctl_tcp_mem[1] = limit;                      /* 6.25 % */
3169         sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;      /* 9.37 % */
3170 }
3171 
3172 void __init tcp_init(void)
3173 {
3174         unsigned long limit;
3175         int max_rshare, max_wshare, cnt;
3176         unsigned int i;
3177 
3178         sock_skb_cb_check_size(sizeof(struct tcp_skb_cb));
3179 
3180         percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3181         percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
3182         tcp_hashinfo.bind_bucket_cachep =
3183                 kmem_cache_create("tcp_bind_bucket",
3184                                   sizeof(struct inet_bind_bucket), 0,
3185                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3186 
3187         /* Size and allocate the main established and bind bucket
3188          * hash tables.
3189          *
3190          * The methodology is similar to that of the buffer cache.
3191          */
3192         tcp_hashinfo.ehash =
3193                 alloc_large_system_hash("TCP established",
3194                                         sizeof(struct inet_ehash_bucket),
3195                                         thash_entries,
3196                                         17, /* one slot per 128 KB of memory */
3197                                         0,
3198                                         NULL,
3199                                         &tcp_hashinfo.ehash_mask,
3200                                         0,
3201                                         thash_entries ? 0 : 512 * 1024);
3202         for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3203                 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3204 
3205         if (inet_ehash_locks_alloc(&tcp_hashinfo))
3206                 panic("TCP: failed to alloc ehash_locks");
3207         tcp_hashinfo.bhash =
3208                 alloc_large_system_hash("TCP bind",
3209                                         sizeof(struct inet_bind_hashbucket),
3210                                         tcp_hashinfo.ehash_mask + 1,
3211                                         17, /* one slot per 128 KB of memory */
3212                                         0,
3213                                         &tcp_hashinfo.bhash_size,
3214                                         NULL,
3215                                         0,
3216                                         64 * 1024);
3217         tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
3218         for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3219                 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3220                 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
3221         }
3222 
3223 
3224         cnt = tcp_hashinfo.ehash_mask + 1;
3225 
3226         tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
3227         sysctl_tcp_max_orphans = cnt / 2;
3228         sysctl_max_syn_backlog = max(128, cnt / 256);
3229 
3230         tcp_init_mem();
3231         /* Set per-socket limits to no more than 1/128 the pressure threshold */
3232         limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
3233         max_wshare = min(4UL*1024*1024, limit);
3234         max_rshare = min(6UL*1024*1024, limit);
3235 
3236         sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
3237         sysctl_tcp_wmem[1] = 16*1024;
3238         sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
3239 
3240         sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
3241         sysctl_tcp_rmem[1] = 87380;
3242         sysctl_tcp_rmem[2] = max(87380, max_rshare);
3243 
3244         pr_info("Hash tables configured (established %u bind %u)\n",
3245                 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
3246 
3247         tcp_metrics_init();
3248         BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
3249         tcp_tasklet_init();
3250 }
3251 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us