~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/tcp.c

Version: ~ [ linux-5.8-rc5 ] ~ [ linux-5.7.8 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.51 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.132 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.188 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.230 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.230 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 <linux/uaccess.h>
281 #include <asm/ioctls.h>
282 #include <net/busy_poll.h>
283 
284 int sysctl_tcp_min_tso_segs __read_mostly = 2;
285 
286 int sysctl_tcp_autocorking __read_mostly = 1;
287 
288 struct percpu_counter tcp_orphan_count;
289 EXPORT_SYMBOL_GPL(tcp_orphan_count);
290 
291 long sysctl_tcp_mem[3] __read_mostly;
292 int sysctl_tcp_wmem[3] __read_mostly;
293 int sysctl_tcp_rmem[3] __read_mostly;
294 
295 EXPORT_SYMBOL(sysctl_tcp_mem);
296 EXPORT_SYMBOL(sysctl_tcp_rmem);
297 EXPORT_SYMBOL(sysctl_tcp_wmem);
298 
299 atomic_long_t tcp_memory_allocated;     /* Current allocated memory. */
300 EXPORT_SYMBOL(tcp_memory_allocated);
301 
302 /*
303  * Current number of TCP sockets.
304  */
305 struct percpu_counter tcp_sockets_allocated;
306 EXPORT_SYMBOL(tcp_sockets_allocated);
307 
308 /*
309  * TCP splice context
310  */
311 struct tcp_splice_state {
312         struct pipe_inode_info *pipe;
313         size_t len;
314         unsigned int flags;
315 };
316 
317 /*
318  * Pressure flag: try to collapse.
319  * Technical note: it is used by multiple contexts non atomically.
320  * All the __sk_mem_schedule() is of this nature: accounting
321  * is strict, actions are advisory and have some latency.
322  */
323 int tcp_memory_pressure __read_mostly;
324 EXPORT_SYMBOL(tcp_memory_pressure);
325 
326 void tcp_enter_memory_pressure(struct sock *sk)
327 {
328         if (!tcp_memory_pressure) {
329                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
330                 tcp_memory_pressure = 1;
331         }
332 }
333 EXPORT_SYMBOL(tcp_enter_memory_pressure);
334 
335 /* Convert seconds to retransmits based on initial and max timeout */
336 static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
337 {
338         u8 res = 0;
339 
340         if (seconds > 0) {
341                 int period = timeout;
342 
343                 res = 1;
344                 while (seconds > period && res < 255) {
345                         res++;
346                         timeout <<= 1;
347                         if (timeout > rto_max)
348                                 timeout = rto_max;
349                         period += timeout;
350                 }
351         }
352         return res;
353 }
354 
355 /* Convert retransmits to seconds based on initial and max timeout */
356 static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
357 {
358         int period = 0;
359 
360         if (retrans > 0) {
361                 period = timeout;
362                 while (--retrans) {
363                         timeout <<= 1;
364                         if (timeout > rto_max)
365                                 timeout = rto_max;
366                         period += timeout;
367                 }
368         }
369         return period;
370 }
371 
372 /* Address-family independent initialization for a tcp_sock.
373  *
374  * NOTE: A lot of things set to zero explicitly by call to
375  *       sk_alloc() so need not be done here.
376  */
377 void tcp_init_sock(struct sock *sk)
378 {
379         struct inet_connection_sock *icsk = inet_csk(sk);
380         struct tcp_sock *tp = tcp_sk(sk);
381 
382         tp->out_of_order_queue = RB_ROOT;
383         tcp_init_xmit_timers(sk);
384         tcp_prequeue_init(tp);
385         INIT_LIST_HEAD(&tp->tsq_node);
386 
387         icsk->icsk_rto = TCP_TIMEOUT_INIT;
388         tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
389         minmax_reset(&tp->rtt_min, tcp_time_stamp, ~0U);
390 
391         /* So many TCP implementations out there (incorrectly) count the
392          * initial SYN frame in their delayed-ACK and congestion control
393          * algorithms that we must have the following bandaid to talk
394          * efficiently to them.  -DaveM
395          */
396         tp->snd_cwnd = TCP_INIT_CWND;
397 
398         /* There's a bubble in the pipe until at least the first ACK. */
399         tp->app_limited = ~0U;
400 
401         /* See draft-stevens-tcpca-spec-01 for discussion of the
402          * initialization of these values.
403          */
404         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
405         tp->snd_cwnd_clamp = ~0;
406         tp->mss_cache = TCP_MSS_DEFAULT;
407 
408         tp->reordering = sock_net(sk)->ipv4.sysctl_tcp_reordering;
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         sk_sockets_allocated_inc(sk);
424 }
425 EXPORT_SYMBOL(tcp_init_sock);
426 
427 static void tcp_tx_timestamp(struct sock *sk, u16 tsflags, struct sk_buff *skb)
428 {
429         if (tsflags && skb) {
430                 struct skb_shared_info *shinfo = skb_shinfo(skb);
431                 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
432 
433                 sock_tx_timestamp(sk, tsflags, &shinfo->tx_flags);
434                 if (tsflags & SOF_TIMESTAMPING_TX_ACK)
435                         tcb->txstamp_ack = 1;
436                 if (tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
437                         shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
438         }
439 }
440 
441 /*
442  *      Wait for a TCP event.
443  *
444  *      Note that we don't need to lock the socket, as the upper poll layers
445  *      take care of normal races (between the test and the event) and we don't
446  *      go look at any of the socket buffers directly.
447  */
448 unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
449 {
450         unsigned int mask;
451         struct sock *sk = sock->sk;
452         const struct tcp_sock *tp = tcp_sk(sk);
453         int state;
454 
455         sock_rps_record_flow(sk);
456 
457         sock_poll_wait(file, sk_sleep(sk), wait);
458 
459         state = sk_state_load(sk);
460         if (state == TCP_LISTEN)
461                 return inet_csk_listen_poll(sk);
462 
463         /* Socket is not locked. We are protected from async events
464          * by poll logic and correct handling of state changes
465          * made by other threads is impossible in any case.
466          */
467 
468         mask = 0;
469 
470         /*
471          * POLLHUP is certainly not done right. But poll() doesn't
472          * have a notion of HUP in just one direction, and for a
473          * socket the read side is more interesting.
474          *
475          * Some poll() documentation says that POLLHUP is incompatible
476          * with the POLLOUT/POLLWR flags, so somebody should check this
477          * all. But careful, it tends to be safer to return too many
478          * bits than too few, and you can easily break real applications
479          * if you don't tell them that something has hung up!
480          *
481          * Check-me.
482          *
483          * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
484          * our fs/select.c). It means that after we received EOF,
485          * poll always returns immediately, making impossible poll() on write()
486          * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
487          * if and only if shutdown has been made in both directions.
488          * Actually, it is interesting to look how Solaris and DUX
489          * solve this dilemma. I would prefer, if POLLHUP were maskable,
490          * then we could set it on SND_SHUTDOWN. BTW examples given
491          * in Stevens' books assume exactly this behaviour, it explains
492          * why POLLHUP is incompatible with POLLOUT.    --ANK
493          *
494          * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
495          * blocking on fresh not-connected or disconnected socket. --ANK
496          */
497         if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
498                 mask |= POLLHUP;
499         if (sk->sk_shutdown & RCV_SHUTDOWN)
500                 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
501 
502         /* Connected or passive Fast Open socket? */
503         if (state != TCP_SYN_SENT &&
504             (state != TCP_SYN_RECV || tp->fastopen_rsk)) {
505                 int target = sock_rcvlowat(sk, 0, INT_MAX);
506 
507                 if (tp->urg_seq == tp->copied_seq &&
508                     !sock_flag(sk, SOCK_URGINLINE) &&
509                     tp->urg_data)
510                         target++;
511 
512                 if (tp->rcv_nxt - tp->copied_seq >= target)
513                         mask |= POLLIN | POLLRDNORM;
514 
515                 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
516                         if (sk_stream_is_writeable(sk)) {
517                                 mask |= POLLOUT | POLLWRNORM;
518                         } else {  /* send SIGIO later */
519                                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
520                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
521 
522                                 /* Race breaker. If space is freed after
523                                  * wspace test but before the flags are set,
524                                  * IO signal will be lost. Memory barrier
525                                  * pairs with the input side.
526                                  */
527                                 smp_mb__after_atomic();
528                                 if (sk_stream_is_writeable(sk))
529                                         mask |= POLLOUT | POLLWRNORM;
530                         }
531                 } else
532                         mask |= POLLOUT | POLLWRNORM;
533 
534                 if (tp->urg_data & TCP_URG_VALID)
535                         mask |= POLLPRI;
536         } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
537                 /* Active TCP fastopen socket with defer_connect
538                  * Return POLLOUT so application can call write()
539                  * in order for kernel to generate SYN+data
540                  */
541                 mask |= POLLOUT | POLLWRNORM;
542         }
543         /* This barrier is coupled with smp_wmb() in tcp_reset() */
544         smp_rmb();
545         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
546                 mask |= POLLERR;
547 
548         return mask;
549 }
550 EXPORT_SYMBOL(tcp_poll);
551 
552 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
553 {
554         struct tcp_sock *tp = tcp_sk(sk);
555         int answ;
556         bool slow;
557 
558         switch (cmd) {
559         case SIOCINQ:
560                 if (sk->sk_state == TCP_LISTEN)
561                         return -EINVAL;
562 
563                 slow = lock_sock_fast(sk);
564                 answ = tcp_inq(sk);
565                 unlock_sock_fast(sk, slow);
566                 break;
567         case SIOCATMARK:
568                 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
569                 break;
570         case SIOCOUTQ:
571                 if (sk->sk_state == TCP_LISTEN)
572                         return -EINVAL;
573 
574                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
575                         answ = 0;
576                 else
577                         answ = tp->write_seq - tp->snd_una;
578                 break;
579         case SIOCOUTQNSD:
580                 if (sk->sk_state == TCP_LISTEN)
581                         return -EINVAL;
582 
583                 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
584                         answ = 0;
585                 else
586                         answ = tp->write_seq - tp->snd_nxt;
587                 break;
588         default:
589                 return -ENOIOCTLCMD;
590         }
591 
592         return put_user(answ, (int __user *)arg);
593 }
594 EXPORT_SYMBOL(tcp_ioctl);
595 
596 static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
597 {
598         TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
599         tp->pushed_seq = tp->write_seq;
600 }
601 
602 static inline bool forced_push(const struct tcp_sock *tp)
603 {
604         return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
605 }
606 
607 static void skb_entail(struct sock *sk, struct sk_buff *skb)
608 {
609         struct tcp_sock *tp = tcp_sk(sk);
610         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
611 
612         skb->csum    = 0;
613         tcb->seq     = tcb->end_seq = tp->write_seq;
614         tcb->tcp_flags = TCPHDR_ACK;
615         tcb->sacked  = 0;
616         __skb_header_release(skb);
617         tcp_add_write_queue_tail(sk, skb);
618         sk->sk_wmem_queued += skb->truesize;
619         sk_mem_charge(sk, skb->truesize);
620         if (tp->nonagle & TCP_NAGLE_PUSH)
621                 tp->nonagle &= ~TCP_NAGLE_PUSH;
622 
623         tcp_slow_start_after_idle_check(sk);
624 }
625 
626 static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
627 {
628         if (flags & MSG_OOB)
629                 tp->snd_up = tp->write_seq;
630 }
631 
632 /* If a not yet filled skb is pushed, do not send it if
633  * we have data packets in Qdisc or NIC queues :
634  * Because TX completion will happen shortly, it gives a chance
635  * to coalesce future sendmsg() payload into this skb, without
636  * need for a timer, and with no latency trade off.
637  * As packets containing data payload have a bigger truesize
638  * than pure acks (dataless) packets, the last checks prevent
639  * autocorking if we only have an ACK in Qdisc/NIC queues,
640  * or if TX completion was delayed after we processed ACK packet.
641  */
642 static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
643                                 int size_goal)
644 {
645         return skb->len < size_goal &&
646                sysctl_tcp_autocorking &&
647                skb != tcp_write_queue_head(sk) &&
648                atomic_read(&sk->sk_wmem_alloc) > skb->truesize;
649 }
650 
651 static void tcp_push(struct sock *sk, int flags, int mss_now,
652                      int nonagle, int size_goal)
653 {
654         struct tcp_sock *tp = tcp_sk(sk);
655         struct sk_buff *skb;
656 
657         if (!tcp_send_head(sk))
658                 return;
659 
660         skb = tcp_write_queue_tail(sk);
661         if (!(flags & MSG_MORE) || forced_push(tp))
662                 tcp_mark_push(tp, skb);
663 
664         tcp_mark_urg(tp, flags);
665 
666         if (tcp_should_autocork(sk, skb, size_goal)) {
667 
668                 /* avoid atomic op if TSQ_THROTTLED bit is already set */
669                 if (!test_bit(TSQ_THROTTLED, &sk->sk_tsq_flags)) {
670                         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
671                         set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags);
672                 }
673                 /* It is possible TX completion already happened
674                  * before we set TSQ_THROTTLED.
675                  */
676                 if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize)
677                         return;
678         }
679 
680         if (flags & MSG_MORE)
681                 nonagle = TCP_NAGLE_CORK;
682 
683         __tcp_push_pending_frames(sk, mss_now, nonagle);
684 }
685 
686 static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
687                                 unsigned int offset, size_t len)
688 {
689         struct tcp_splice_state *tss = rd_desc->arg.data;
690         int ret;
691 
692         ret = skb_splice_bits(skb, skb->sk, offset, tss->pipe,
693                               min(rd_desc->count, len), tss->flags);
694         if (ret > 0)
695                 rd_desc->count -= ret;
696         return ret;
697 }
698 
699 static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
700 {
701         /* Store TCP splice context information in read_descriptor_t. */
702         read_descriptor_t rd_desc = {
703                 .arg.data = tss,
704                 .count    = tss->len,
705         };
706 
707         return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
708 }
709 
710 /**
711  *  tcp_splice_read - splice data from TCP socket to a pipe
712  * @sock:       socket to splice from
713  * @ppos:       position (not valid)
714  * @pipe:       pipe to splice to
715  * @len:        number of bytes to splice
716  * @flags:      splice modifier flags
717  *
718  * Description:
719  *    Will read pages from given socket and fill them into a pipe.
720  *
721  **/
722 ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
723                         struct pipe_inode_info *pipe, size_t len,
724                         unsigned int flags)
725 {
726         struct sock *sk = sock->sk;
727         struct tcp_splice_state tss = {
728                 .pipe = pipe,
729                 .len = len,
730                 .flags = flags,
731         };
732         long timeo;
733         ssize_t spliced;
734         int ret;
735 
736         sock_rps_record_flow(sk);
737         /*
738          * We can't seek on a socket input
739          */
740         if (unlikely(*ppos))
741                 return -ESPIPE;
742 
743         ret = spliced = 0;
744 
745         lock_sock(sk);
746 
747         timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
748         while (tss.len) {
749                 ret = __tcp_splice_read(sk, &tss);
750                 if (ret < 0)
751                         break;
752                 else if (!ret) {
753                         if (spliced)
754                                 break;
755                         if (sock_flag(sk, SOCK_DONE))
756                                 break;
757                         if (sk->sk_err) {
758                                 ret = sock_error(sk);
759                                 break;
760                         }
761                         if (sk->sk_shutdown & RCV_SHUTDOWN)
762                                 break;
763                         if (sk->sk_state == TCP_CLOSE) {
764                                 /*
765                                  * This occurs when user tries to read
766                                  * from never connected socket.
767                                  */
768                                 if (!sock_flag(sk, SOCK_DONE))
769                                         ret = -ENOTCONN;
770                                 break;
771                         }
772                         if (!timeo) {
773                                 ret = -EAGAIN;
774                                 break;
775                         }
776                         /* if __tcp_splice_read() got nothing while we have
777                          * an skb in receive queue, we do not want to loop.
778                          * This might happen with URG data.
779                          */
780                         if (!skb_queue_empty(&sk->sk_receive_queue))
781                                 break;
782                         sk_wait_data(sk, &timeo, NULL);
783                         if (signal_pending(current)) {
784                                 ret = sock_intr_errno(timeo);
785                                 break;
786                         }
787                         continue;
788                 }
789                 tss.len -= ret;
790                 spliced += ret;
791 
792                 if (!timeo)
793                         break;
794                 release_sock(sk);
795                 lock_sock(sk);
796 
797                 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
798                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
799                     signal_pending(current))
800                         break;
801         }
802 
803         release_sock(sk);
804 
805         if (spliced)
806                 return spliced;
807 
808         return ret;
809 }
810 EXPORT_SYMBOL(tcp_splice_read);
811 
812 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
813                                     bool force_schedule)
814 {
815         struct sk_buff *skb;
816 
817         /* The TCP header must be at least 32-bit aligned.  */
818         size = ALIGN(size, 4);
819 
820         if (unlikely(tcp_under_memory_pressure(sk)))
821                 sk_mem_reclaim_partial(sk);
822 
823         skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
824         if (likely(skb)) {
825                 bool mem_scheduled;
826 
827                 if (force_schedule) {
828                         mem_scheduled = true;
829                         sk_forced_mem_schedule(sk, skb->truesize);
830                 } else {
831                         mem_scheduled = sk_wmem_schedule(sk, skb->truesize);
832                 }
833                 if (likely(mem_scheduled)) {
834                         skb_reserve(skb, sk->sk_prot->max_header);
835                         /*
836                          * Make sure that we have exactly size bytes
837                          * available to the caller, no more, no less.
838                          */
839                         skb->reserved_tailroom = skb->end - skb->tail - size;
840                         return skb;
841                 }
842                 __kfree_skb(skb);
843         } else {
844                 sk->sk_prot->enter_memory_pressure(sk);
845                 sk_stream_moderate_sndbuf(sk);
846         }
847         return NULL;
848 }
849 
850 static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
851                                        int large_allowed)
852 {
853         struct tcp_sock *tp = tcp_sk(sk);
854         u32 new_size_goal, size_goal;
855 
856         if (!large_allowed || !sk_can_gso(sk))
857                 return mss_now;
858 
859         /* Note : tcp_tso_autosize() will eventually split this later */
860         new_size_goal = sk->sk_gso_max_size - 1 - MAX_TCP_HEADER;
861         new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
862 
863         /* We try hard to avoid divides here */
864         size_goal = tp->gso_segs * mss_now;
865         if (unlikely(new_size_goal < size_goal ||
866                      new_size_goal >= size_goal + mss_now)) {
867                 tp->gso_segs = min_t(u16, new_size_goal / mss_now,
868                                      sk->sk_gso_max_segs);
869                 size_goal = tp->gso_segs * mss_now;
870         }
871 
872         return max(size_goal, mss_now);
873 }
874 
875 static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
876 {
877         int mss_now;
878 
879         mss_now = tcp_current_mss(sk);
880         *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
881 
882         return mss_now;
883 }
884 
885 static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
886                                 size_t size, int flags)
887 {
888         struct tcp_sock *tp = tcp_sk(sk);
889         int mss_now, size_goal;
890         int err;
891         ssize_t copied;
892         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
893 
894         /* Wait for a connection to finish. One exception is TCP Fast Open
895          * (passive side) where data is allowed to be sent before a connection
896          * is fully established.
897          */
898         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
899             !tcp_passive_fastopen(sk)) {
900                 err = sk_stream_wait_connect(sk, &timeo);
901                 if (err != 0)
902                         goto out_err;
903         }
904 
905         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
906 
907         mss_now = tcp_send_mss(sk, &size_goal, flags);
908         copied = 0;
909 
910         err = -EPIPE;
911         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
912                 goto out_err;
913 
914         while (size > 0) {
915                 struct sk_buff *skb = tcp_write_queue_tail(sk);
916                 int copy, i;
917                 bool can_coalesce;
918 
919                 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0 ||
920                     !tcp_skb_can_collapse_to(skb)) {
921 new_segment:
922                         if (!sk_stream_memory_free(sk))
923                                 goto wait_for_sndbuf;
924 
925                         skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
926                                                   skb_queue_empty(&sk->sk_write_queue));
927                         if (!skb)
928                                 goto wait_for_memory;
929 
930                         skb_entail(sk, skb);
931                         copy = size_goal;
932                 }
933 
934                 if (copy > size)
935                         copy = size;
936 
937                 i = skb_shinfo(skb)->nr_frags;
938                 can_coalesce = skb_can_coalesce(skb, i, page, offset);
939                 if (!can_coalesce && i >= sysctl_max_skb_frags) {
940                         tcp_mark_push(tp, skb);
941                         goto new_segment;
942                 }
943                 if (!sk_wmem_schedule(sk, copy))
944                         goto wait_for_memory;
945 
946                 if (can_coalesce) {
947                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
948                 } else {
949                         get_page(page);
950                         skb_fill_page_desc(skb, i, page, offset, copy);
951                 }
952                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
953 
954                 skb->len += copy;
955                 skb->data_len += copy;
956                 skb->truesize += copy;
957                 sk->sk_wmem_queued += copy;
958                 sk_mem_charge(sk, copy);
959                 skb->ip_summed = CHECKSUM_PARTIAL;
960                 tp->write_seq += copy;
961                 TCP_SKB_CB(skb)->end_seq += copy;
962                 tcp_skb_pcount_set(skb, 0);
963 
964                 if (!copied)
965                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
966 
967                 copied += copy;
968                 offset += copy;
969                 size -= copy;
970                 if (!size)
971                         goto out;
972 
973                 if (skb->len < size_goal || (flags & MSG_OOB))
974                         continue;
975 
976                 if (forced_push(tp)) {
977                         tcp_mark_push(tp, skb);
978                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
979                 } else if (skb == tcp_send_head(sk))
980                         tcp_push_one(sk, mss_now);
981                 continue;
982 
983 wait_for_sndbuf:
984                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
985 wait_for_memory:
986                 tcp_push(sk, flags & ~MSG_MORE, mss_now,
987                          TCP_NAGLE_PUSH, size_goal);
988 
989                 err = sk_stream_wait_memory(sk, &timeo);
990                 if (err != 0)
991                         goto do_error;
992 
993                 mss_now = tcp_send_mss(sk, &size_goal, flags);
994         }
995 
996 out:
997         if (copied) {
998                 tcp_tx_timestamp(sk, sk->sk_tsflags, tcp_write_queue_tail(sk));
999                 if (!(flags & MSG_SENDPAGE_NOTLAST))
1000                         tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1001         }
1002         return copied;
1003 
1004 do_error:
1005         if (copied)
1006                 goto out;
1007 out_err:
1008         /* make sure we wake any epoll edge trigger waiter */
1009         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1010                      err == -EAGAIN)) {
1011                 sk->sk_write_space(sk);
1012                 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1013         }
1014         return sk_stream_error(sk, flags, err);
1015 }
1016 
1017 int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1018                  size_t size, int flags)
1019 {
1020         ssize_t res;
1021 
1022         if (!(sk->sk_route_caps & NETIF_F_SG) ||
1023             !sk_check_csum_caps(sk))
1024                 return sock_no_sendpage(sk->sk_socket, page, offset, size,
1025                                         flags);
1026 
1027         lock_sock(sk);
1028 
1029         tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1030 
1031         res = do_tcp_sendpages(sk, page, offset, size, flags);
1032         release_sock(sk);
1033         return res;
1034 }
1035 EXPORT_SYMBOL(tcp_sendpage);
1036 
1037 /* Do not bother using a page frag for very small frames.
1038  * But use this heuristic only for the first skb in write queue.
1039  *
1040  * Having no payload in skb->head allows better SACK shifting
1041  * in tcp_shift_skb_data(), reducing sack/rack overhead, because
1042  * write queue has less skbs.
1043  * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB.
1044  * This also speeds up tso_fragment(), since it wont fallback
1045  * to tcp_fragment().
1046  */
1047 static int linear_payload_sz(bool first_skb)
1048 {
1049         if (first_skb)
1050                 return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1051         return 0;
1052 }
1053 
1054 static int select_size(const struct sock *sk, bool sg, bool first_skb)
1055 {
1056         const struct tcp_sock *tp = tcp_sk(sk);
1057         int tmp = tp->mss_cache;
1058 
1059         if (sg) {
1060                 if (sk_can_gso(sk)) {
1061                         tmp = linear_payload_sz(first_skb);
1062                 } else {
1063                         int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
1064 
1065                         if (tmp >= pgbreak &&
1066                             tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
1067                                 tmp = pgbreak;
1068                 }
1069         }
1070 
1071         return tmp;
1072 }
1073 
1074 void tcp_free_fastopen_req(struct tcp_sock *tp)
1075 {
1076         if (tp->fastopen_req) {
1077                 kfree(tp->fastopen_req);
1078                 tp->fastopen_req = NULL;
1079         }
1080 }
1081 
1082 static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1083                                 int *copied, size_t size)
1084 {
1085         struct tcp_sock *tp = tcp_sk(sk);
1086         struct inet_sock *inet = inet_sk(sk);
1087         struct sockaddr *uaddr = msg->msg_name;
1088         int err, flags;
1089 
1090         if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) ||
1091             (uaddr && msg->msg_namelen >= sizeof(uaddr->sa_family) &&
1092              uaddr->sa_family == AF_UNSPEC))
1093                 return -EOPNOTSUPP;
1094         if (tp->fastopen_req)
1095                 return -EALREADY; /* Another Fast Open is in progress */
1096 
1097         tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1098                                    sk->sk_allocation);
1099         if (unlikely(!tp->fastopen_req))
1100                 return -ENOBUFS;
1101         tp->fastopen_req->data = msg;
1102         tp->fastopen_req->size = size;
1103 
1104         if (inet->defer_connect) {
1105                 err = tcp_connect(sk);
1106                 /* Same failure procedure as in tcp_v4/6_connect */
1107                 if (err) {
1108                         tcp_set_state(sk, TCP_CLOSE);
1109                         inet->inet_dport = 0;
1110                         sk->sk_route_caps = 0;
1111                 }
1112         }
1113         flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1114         err = __inet_stream_connect(sk->sk_socket, uaddr,
1115                                     msg->msg_namelen, flags, 1);
1116         /* fastopen_req could already be freed in __inet_stream_connect
1117          * if the connection times out or gets rst
1118          */
1119         if (tp->fastopen_req) {
1120                 *copied = tp->fastopen_req->copied;
1121                 tcp_free_fastopen_req(tp);
1122                 inet->defer_connect = 0;
1123         }
1124         return err;
1125 }
1126 
1127 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
1128 {
1129         struct tcp_sock *tp = tcp_sk(sk);
1130         struct sk_buff *skb;
1131         struct sockcm_cookie sockc;
1132         int flags, err, copied = 0;
1133         int mss_now = 0, size_goal, copied_syn = 0;
1134         bool process_backlog = false;
1135         bool sg;
1136         long timeo;
1137 
1138         lock_sock(sk);
1139 
1140         flags = msg->msg_flags;
1141         if (unlikely(flags & MSG_FASTOPEN || inet_sk(sk)->defer_connect)) {
1142                 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size);
1143                 if (err == -EINPROGRESS && copied_syn > 0)
1144                         goto out;
1145                 else if (err)
1146                         goto out_err;
1147         }
1148 
1149         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1150 
1151         tcp_rate_check_app_limited(sk);  /* is sending application-limited? */
1152 
1153         /* Wait for a connection to finish. One exception is TCP Fast Open
1154          * (passive side) where data is allowed to be sent before a connection
1155          * is fully established.
1156          */
1157         if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1158             !tcp_passive_fastopen(sk)) {
1159                 err = sk_stream_wait_connect(sk, &timeo);
1160                 if (err != 0)
1161                         goto do_error;
1162         }
1163 
1164         if (unlikely(tp->repair)) {
1165                 if (tp->repair_queue == TCP_RECV_QUEUE) {
1166                         copied = tcp_send_rcvq(sk, msg, size);
1167                         goto out_nopush;
1168                 }
1169 
1170                 err = -EINVAL;
1171                 if (tp->repair_queue == TCP_NO_QUEUE)
1172                         goto out_err;
1173 
1174                 /* 'common' sending to sendq */
1175         }
1176 
1177         sockc.tsflags = sk->sk_tsflags;
1178         if (msg->msg_controllen) {
1179                 err = sock_cmsg_send(sk, msg, &sockc);
1180                 if (unlikely(err)) {
1181                         err = -EINVAL;
1182                         goto out_err;
1183                 }
1184         }
1185 
1186         /* This should be in poll */
1187         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1188 
1189         /* Ok commence sending. */
1190         copied = 0;
1191 
1192 restart:
1193         mss_now = tcp_send_mss(sk, &size_goal, flags);
1194 
1195         err = -EPIPE;
1196         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
1197                 goto do_error;
1198 
1199         sg = !!(sk->sk_route_caps & NETIF_F_SG);
1200 
1201         while (msg_data_left(msg)) {
1202                 int copy = 0;
1203                 int max = size_goal;
1204 
1205                 skb = tcp_write_queue_tail(sk);
1206                 if (tcp_send_head(sk)) {
1207                         if (skb->ip_summed == CHECKSUM_NONE)
1208                                 max = mss_now;
1209                         copy = max - skb->len;
1210                 }
1211 
1212                 if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
1213                         bool first_skb;
1214 
1215 new_segment:
1216                         /* Allocate new segment. If the interface is SG,
1217                          * allocate skb fitting to single page.
1218                          */
1219                         if (!sk_stream_memory_free(sk))
1220                                 goto wait_for_sndbuf;
1221 
1222                         if (process_backlog && sk_flush_backlog(sk)) {
1223                                 process_backlog = false;
1224                                 goto restart;
1225                         }
1226                         first_skb = skb_queue_empty(&sk->sk_write_queue);
1227                         skb = sk_stream_alloc_skb(sk,
1228                                                   select_size(sk, sg, first_skb),
1229                                                   sk->sk_allocation,
1230                                                   first_skb);
1231                         if (!skb)
1232                                 goto wait_for_memory;
1233 
1234                         process_backlog = true;
1235                         /*
1236                          * Check whether we can use HW checksum.
1237                          */
1238                         if (sk_check_csum_caps(sk))
1239                                 skb->ip_summed = CHECKSUM_PARTIAL;
1240 
1241                         skb_entail(sk, skb);
1242                         copy = size_goal;
1243                         max = size_goal;
1244 
1245                         /* All packets are restored as if they have
1246                          * already been sent. skb_mstamp isn't set to
1247                          * avoid wrong rtt estimation.
1248                          */
1249                         if (tp->repair)
1250                                 TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
1251                 }
1252 
1253                 /* Try to append data to the end of skb. */
1254                 if (copy > msg_data_left(msg))
1255                         copy = msg_data_left(msg);
1256 
1257                 /* Where to copy to? */
1258                 if (skb_availroom(skb) > 0) {
1259                         /* We have some space in skb head. Superb! */
1260                         copy = min_t(int, copy, skb_availroom(skb));
1261                         err = skb_add_data_nocache(sk, skb, &msg->msg_iter, copy);
1262                         if (err)
1263                                 goto do_fault;
1264                 } else {
1265                         bool merge = true;
1266                         int i = skb_shinfo(skb)->nr_frags;
1267                         struct page_frag *pfrag = sk_page_frag(sk);
1268 
1269                         if (!sk_page_frag_refill(sk, pfrag))
1270                                 goto wait_for_memory;
1271 
1272                         if (!skb_can_coalesce(skb, i, pfrag->page,
1273                                               pfrag->offset)) {
1274                                 if (i >= sysctl_max_skb_frags || !sg) {
1275                                         tcp_mark_push(tp, skb);
1276                                         goto new_segment;
1277                                 }
1278                                 merge = false;
1279                         }
1280 
1281                         copy = min_t(int, copy, pfrag->size - pfrag->offset);
1282 
1283                         if (!sk_wmem_schedule(sk, copy))
1284                                 goto wait_for_memory;
1285 
1286                         err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
1287                                                        pfrag->page,
1288                                                        pfrag->offset,
1289                                                        copy);
1290                         if (err)
1291                                 goto do_error;
1292 
1293                         /* Update the skb. */
1294                         if (merge) {
1295                                 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1296                         } else {
1297                                 skb_fill_page_desc(skb, i, pfrag->page,
1298                                                    pfrag->offset, copy);
1299                                 page_ref_inc(pfrag->page);
1300                         }
1301                         pfrag->offset += copy;
1302                 }
1303 
1304                 if (!copied)
1305                         TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1306 
1307                 tp->write_seq += copy;
1308                 TCP_SKB_CB(skb)->end_seq += copy;
1309                 tcp_skb_pcount_set(skb, 0);
1310 
1311                 copied += copy;
1312                 if (!msg_data_left(msg)) {
1313                         if (unlikely(flags & MSG_EOR))
1314                                 TCP_SKB_CB(skb)->eor = 1;
1315                         goto out;
1316                 }
1317 
1318                 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
1319                         continue;
1320 
1321                 if (forced_push(tp)) {
1322                         tcp_mark_push(tp, skb);
1323                         __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
1324                 } else if (skb == tcp_send_head(sk))
1325                         tcp_push_one(sk, mss_now);
1326                 continue;
1327 
1328 wait_for_sndbuf:
1329                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1330 wait_for_memory:
1331                 if (copied)
1332                         tcp_push(sk, flags & ~MSG_MORE, mss_now,
1333                                  TCP_NAGLE_PUSH, size_goal);
1334 
1335                 err = sk_stream_wait_memory(sk, &timeo);
1336                 if (err != 0)
1337                         goto do_error;
1338 
1339                 mss_now = tcp_send_mss(sk, &size_goal, flags);
1340         }
1341 
1342 out:
1343         if (copied) {
1344                 tcp_tx_timestamp(sk, sockc.tsflags, tcp_write_queue_tail(sk));
1345                 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
1346         }
1347 out_nopush:
1348         release_sock(sk);
1349         return copied + copied_syn;
1350 
1351 do_fault:
1352         if (!skb->len) {
1353                 tcp_unlink_write_queue(skb, sk);
1354                 /* It is the one place in all of TCP, except connection
1355                  * reset, where we can be unlinking the send_head.
1356                  */
1357                 tcp_check_send_head(sk, skb);
1358                 sk_wmem_free_skb(sk, skb);
1359         }
1360 
1361 do_error:
1362         if (copied + copied_syn)
1363                 goto out;
1364 out_err:
1365         err = sk_stream_error(sk, flags, err);
1366         /* make sure we wake any epoll edge trigger waiter */
1367         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&
1368                      err == -EAGAIN)) {
1369                 sk->sk_write_space(sk);
1370                 tcp_chrono_stop(sk, TCP_CHRONO_SNDBUF_LIMITED);
1371         }
1372         release_sock(sk);
1373         return err;
1374 }
1375 EXPORT_SYMBOL(tcp_sendmsg);
1376 
1377 /*
1378  *      Handle reading urgent data. BSD has very simple semantics for
1379  *      this, no blocking and very strange errors 8)
1380  */
1381 
1382 static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
1383 {
1384         struct tcp_sock *tp = tcp_sk(sk);
1385 
1386         /* No URG data to read. */
1387         if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1388             tp->urg_data == TCP_URG_READ)
1389                 return -EINVAL; /* Yes this is right ! */
1390 
1391         if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1392                 return -ENOTCONN;
1393 
1394         if (tp->urg_data & TCP_URG_VALID) {
1395                 int err = 0;
1396                 char c = tp->urg_data;
1397 
1398                 if (!(flags & MSG_PEEK))
1399                         tp->urg_data = TCP_URG_READ;
1400 
1401                 /* Read urgent data. */
1402                 msg->msg_flags |= MSG_OOB;
1403 
1404                 if (len > 0) {
1405                         if (!(flags & MSG_TRUNC))
1406                                 err = memcpy_to_msg(msg, &c, 1);
1407                         len = 1;
1408                 } else
1409                         msg->msg_flags |= MSG_TRUNC;
1410 
1411                 return err ? -EFAULT : len;
1412         }
1413 
1414         if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1415                 return 0;
1416 
1417         /* Fixed the recv(..., MSG_OOB) behaviour.  BSD docs and
1418          * the available implementations agree in this case:
1419          * this call should never block, independent of the
1420          * blocking state of the socket.
1421          * Mike <pall@rz.uni-karlsruhe.de>
1422          */
1423         return -EAGAIN;
1424 }
1425 
1426 static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1427 {
1428         struct sk_buff *skb;
1429         int copied = 0, err = 0;
1430 
1431         /* XXX -- need to support SO_PEEK_OFF */
1432 
1433         skb_queue_walk(&sk->sk_write_queue, skb) {
1434                 err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
1435                 if (err)
1436                         break;
1437 
1438                 copied += skb->len;
1439         }
1440 
1441         return err ?: copied;
1442 }
1443 
1444 /* Clean up the receive buffer for full frames taken by the user,
1445  * then send an ACK if necessary.  COPIED is the number of bytes
1446  * tcp_recvmsg has given to the user so far, it speeds up the
1447  * calculation of whether or not we must ACK for the sake of
1448  * a window update.
1449  */
1450 static void tcp_cleanup_rbuf(struct sock *sk, int copied)
1451 {
1452         struct tcp_sock *tp = tcp_sk(sk);
1453         bool time_to_ack = false;
1454 
1455         struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1456 
1457         WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1458              "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1459              tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
1460 
1461         if (inet_csk_ack_scheduled(sk)) {
1462                 const struct inet_connection_sock *icsk = inet_csk(sk);
1463                    /* Delayed ACKs frequently hit locked sockets during bulk
1464                     * receive. */
1465                 if (icsk->icsk_ack.blocked ||
1466                     /* Once-per-two-segments ACK was not sent by tcp_input.c */
1467                     tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
1468                     /*
1469                      * If this read emptied read buffer, we send ACK, if
1470                      * connection is not bidirectional, user drained
1471                      * receive buffer and there was a small segment
1472                      * in queue.
1473                      */
1474                     (copied > 0 &&
1475                      ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1476                       ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1477                        !icsk->icsk_ack.pingpong)) &&
1478                       !atomic_read(&sk->sk_rmem_alloc)))
1479                         time_to_ack = true;
1480         }
1481 
1482         /* We send an ACK if we can now advertise a non-zero window
1483          * which has been raised "significantly".
1484          *
1485          * Even if window raised up to infinity, do not send window open ACK
1486          * in states, where we will not receive more. It is useless.
1487          */
1488         if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1489                 __u32 rcv_window_now = tcp_receive_window(tp);
1490 
1491                 /* Optimize, __tcp_select_window() is not cheap. */
1492                 if (2*rcv_window_now <= tp->window_clamp) {
1493                         __u32 new_window = __tcp_select_window(sk);
1494 
1495                         /* Send ACK now, if this read freed lots of space
1496                          * in our buffer. Certainly, new_window is new window.
1497                          * We can advertise it now, if it is not less than current one.
1498                          * "Lots" means "at least twice" here.
1499                          */
1500                         if (new_window && new_window >= 2 * rcv_window_now)
1501                                 time_to_ack = true;
1502                 }
1503         }
1504         if (time_to_ack)
1505                 tcp_send_ack(sk);
1506 }
1507 
1508 static void tcp_prequeue_process(struct sock *sk)
1509 {
1510         struct sk_buff *skb;
1511         struct tcp_sock *tp = tcp_sk(sk);
1512 
1513         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
1514 
1515         while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1516                 sk_backlog_rcv(sk, skb);
1517 
1518         /* Clear memory counter. */
1519         tp->ucopy.memory = 0;
1520 }
1521 
1522 static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1523 {
1524         struct sk_buff *skb;
1525         u32 offset;
1526 
1527         while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
1528                 offset = seq - TCP_SKB_CB(skb)->seq;
1529                 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1530                         pr_err_once("%s: found a SYN, please report !\n", __func__);
1531                         offset--;
1532                 }
1533                 if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
1534                         *off = offset;
1535                         return skb;
1536                 }
1537                 /* This looks weird, but this can happen if TCP collapsing
1538                  * splitted a fat GRO packet, while we released socket lock
1539                  * in skb_splice_bits()
1540                  */
1541                 sk_eat_skb(sk, skb);
1542         }
1543         return NULL;
1544 }
1545 
1546 /*
1547  * This routine provides an alternative to tcp_recvmsg() for routines
1548  * that would like to handle copying from skbuffs directly in 'sendfile'
1549  * fashion.
1550  * Note:
1551  *      - It is assumed that the socket was locked by the caller.
1552  *      - The routine does not block.
1553  *      - At present, there is no support for reading OOB data
1554  *        or for 'peeking' the socket using this routine
1555  *        (although both would be easy to implement).
1556  */
1557 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1558                   sk_read_actor_t recv_actor)
1559 {
1560         struct sk_buff *skb;
1561         struct tcp_sock *tp = tcp_sk(sk);
1562         u32 seq = tp->copied_seq;
1563         u32 offset;
1564         int copied = 0;
1565 
1566         if (sk->sk_state == TCP_LISTEN)
1567                 return -ENOTCONN;
1568         while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1569                 if (offset < skb->len) {
1570                         int used;
1571                         size_t len;
1572 
1573                         len = skb->len - offset;
1574                         /* Stop reading if we hit a patch of urgent data */
1575                         if (tp->urg_data) {
1576                                 u32 urg_offset = tp->urg_seq - seq;
1577                                 if (urg_offset < len)
1578                                         len = urg_offset;
1579                                 if (!len)
1580                                         break;
1581                         }
1582                         used = recv_actor(desc, skb, offset, len);
1583                         if (used <= 0) {
1584                                 if (!copied)
1585                                         copied = used;
1586                                 break;
1587                         } else if (used <= len) {
1588                                 seq += used;
1589                                 copied += used;
1590                                 offset += used;
1591                         }
1592                         /* If recv_actor drops the lock (e.g. TCP splice
1593                          * receive) the skb pointer might be invalid when
1594                          * getting here: tcp_collapse might have deleted it
1595                          * while aggregating skbs from the socket queue.
1596                          */
1597                         skb = tcp_recv_skb(sk, seq - 1, &offset);
1598                         if (!skb)
1599                                 break;
1600                         /* TCP coalescing might have appended data to the skb.
1601                          * Try to splice more frags
1602                          */
1603                         if (offset + 1 != skb->len)
1604                                 continue;
1605                 }
1606                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
1607                         sk_eat_skb(sk, skb);
1608                         ++seq;
1609                         break;
1610                 }
1611                 sk_eat_skb(sk, skb);
1612                 if (!desc->count)
1613                         break;
1614                 tp->copied_seq = seq;
1615         }
1616         tp->copied_seq = seq;
1617 
1618         tcp_rcv_space_adjust(sk);
1619 
1620         /* Clean up data we have read: This will do ACK frames. */
1621         if (copied > 0) {
1622                 tcp_recv_skb(sk, seq, &offset);
1623                 tcp_cleanup_rbuf(sk, copied);
1624         }
1625         return copied;
1626 }
1627 EXPORT_SYMBOL(tcp_read_sock);
1628 
1629 int tcp_peek_len(struct socket *sock)
1630 {
1631         return tcp_inq(sock->sk);
1632 }
1633 EXPORT_SYMBOL(tcp_peek_len);
1634 
1635 /*
1636  *      This routine copies from a sock struct into the user buffer.
1637  *
1638  *      Technical note: in 2.3 we work on _locked_ socket, so that
1639  *      tricks with *seq access order and skb->users are not required.
1640  *      Probably, code can be easily improved even more.
1641  */
1642 
1643 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
1644                 int flags, int *addr_len)
1645 {
1646         struct tcp_sock *tp = tcp_sk(sk);
1647         int copied = 0;
1648         u32 peek_seq;
1649         u32 *seq;
1650         unsigned long used;
1651         int err;
1652         int target;             /* Read at least this many bytes */
1653         long timeo;
1654         struct task_struct *user_recv = NULL;
1655         struct sk_buff *skb, *last;
1656         u32 urg_hole = 0;
1657 
1658         if (unlikely(flags & MSG_ERRQUEUE))
1659                 return inet_recv_error(sk, msg, len, addr_len);
1660 
1661         if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1662             (sk->sk_state == TCP_ESTABLISHED))
1663                 sk_busy_loop(sk, nonblock);
1664 
1665         lock_sock(sk);
1666 
1667         err = -ENOTCONN;
1668         if (sk->sk_state == TCP_LISTEN)
1669                 goto out;
1670 
1671         timeo = sock_rcvtimeo(sk, nonblock);
1672 
1673         /* Urgent data needs to be handled specially. */
1674         if (flags & MSG_OOB)
1675                 goto recv_urg;
1676 
1677         if (unlikely(tp->repair)) {
1678                 err = -EPERM;
1679                 if (!(flags & MSG_PEEK))
1680                         goto out;
1681 
1682                 if (tp->repair_queue == TCP_SEND_QUEUE)
1683                         goto recv_sndq;
1684 
1685                 err = -EINVAL;
1686                 if (tp->repair_queue == TCP_NO_QUEUE)
1687                         goto out;
1688 
1689                 /* 'common' recv queue MSG_PEEK-ing */
1690         }
1691 
1692         seq = &tp->copied_seq;
1693         if (flags & MSG_PEEK) {
1694                 peek_seq = tp->copied_seq;
1695                 seq = &peek_seq;
1696         }
1697 
1698         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1699 
1700         do {
1701                 u32 offset;
1702 
1703                 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1704                 if (tp->urg_data && tp->urg_seq == *seq) {
1705                         if (copied)
1706                                 break;
1707                         if (signal_pending(current)) {
1708                                 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1709                                 break;
1710                         }
1711                 }
1712 
1713                 /* Next get a buffer. */
1714 
1715                 last = skb_peek_tail(&sk->sk_receive_queue);
1716                 skb_queue_walk(&sk->sk_receive_queue, skb) {
1717                         last = skb;
1718                         /* Now that we have two receive queues this
1719                          * shouldn't happen.
1720                          */
1721                         if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
1722                                  "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1723                                  *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1724                                  flags))
1725                                 break;
1726 
1727                         offset = *seq - TCP_SKB_CB(skb)->seq;
1728                         if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
1729                                 pr_err_once("%s: found a SYN, please report !\n", __func__);
1730                                 offset--;
1731                         }
1732                         if (offset < skb->len)
1733                                 goto found_ok_skb;
1734                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1735                                 goto found_fin_ok;
1736                         WARN(!(flags & MSG_PEEK),
1737                              "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1738                              *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
1739                 }
1740 
1741                 /* Well, if we have backlog, try to process it now yet. */
1742 
1743                 if (copied >= target && !sk->sk_backlog.tail)
1744                         break;
1745 
1746                 if (copied) {
1747                         if (sk->sk_err ||
1748                             sk->sk_state == TCP_CLOSE ||
1749                             (sk->sk_shutdown & RCV_SHUTDOWN) ||
1750                             !timeo ||
1751                             signal_pending(current))
1752                                 break;
1753                 } else {
1754                         if (sock_flag(sk, SOCK_DONE))
1755                                 break;
1756 
1757                         if (sk->sk_err) {
1758                                 copied = sock_error(sk);
1759                                 break;
1760                         }
1761 
1762                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1763                                 break;
1764 
1765                         if (sk->sk_state == TCP_CLOSE) {
1766                                 if (!sock_flag(sk, SOCK_DONE)) {
1767                                         /* This occurs when user tries to read
1768                                          * from never connected socket.
1769                                          */
1770                                         copied = -ENOTCONN;
1771                                         break;
1772                                 }
1773                                 break;
1774                         }
1775 
1776                         if (!timeo) {
1777                                 copied = -EAGAIN;
1778                                 break;
1779                         }
1780 
1781                         if (signal_pending(current)) {
1782                                 copied = sock_intr_errno(timeo);
1783                                 break;
1784                         }
1785                 }
1786 
1787                 tcp_cleanup_rbuf(sk, copied);
1788 
1789                 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
1790                         /* Install new reader */
1791                         if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1792                                 user_recv = current;
1793                                 tp->ucopy.task = user_recv;
1794                                 tp->ucopy.msg = msg;
1795                         }
1796 
1797                         tp->ucopy.len = len;
1798 
1799                         WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1800                                 !(flags & (MSG_PEEK | MSG_TRUNC)));
1801 
1802                         /* Ugly... If prequeue is not empty, we have to
1803                          * process it before releasing socket, otherwise
1804                          * order will be broken at second iteration.
1805                          * More elegant solution is required!!!
1806                          *
1807                          * Look: we have the following (pseudo)queues:
1808                          *
1809                          * 1. packets in flight
1810                          * 2. backlog
1811                          * 3. prequeue
1812                          * 4. receive_queue
1813                          *
1814                          * Each queue can be processed only if the next ones
1815                          * are empty. At this point we have empty receive_queue.
1816                          * But prequeue _can_ be not empty after 2nd iteration,
1817                          * when we jumped to start of loop because backlog
1818                          * processing added something to receive_queue.
1819                          * We cannot release_sock(), because backlog contains
1820                          * packets arrived _after_ prequeued ones.
1821                          *
1822                          * Shortly, algorithm is clear --- to process all
1823                          * the queues in order. We could make it more directly,
1824                          * requeueing packets from backlog to prequeue, if
1825                          * is not empty. It is more elegant, but eats cycles,
1826                          * unfortunately.
1827                          */
1828                         if (!skb_queue_empty(&tp->ucopy.prequeue))
1829                                 goto do_prequeue;
1830 
1831                         /* __ Set realtime policy in scheduler __ */
1832                 }
1833 
1834                 if (copied >= target) {
1835                         /* Do not sleep, just process backlog. */
1836                         release_sock(sk);
1837                         lock_sock(sk);
1838                 } else {
1839                         sk_wait_data(sk, &timeo, last);
1840                 }
1841 
1842                 if (user_recv) {
1843                         int chunk;
1844 
1845                         /* __ Restore normal policy in scheduler __ */
1846 
1847                         chunk = len - tp->ucopy.len;
1848                         if (chunk != 0) {
1849                                 NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
1850                                 len -= chunk;
1851                                 copied += chunk;
1852                         }
1853 
1854                         if (tp->rcv_nxt == tp->copied_seq &&
1855                             !skb_queue_empty(&tp->ucopy.prequeue)) {
1856 do_prequeue:
1857                                 tcp_prequeue_process(sk);
1858 
1859                                 chunk = len - tp->ucopy.len;
1860                                 if (chunk != 0) {
1861                                         NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1862                                         len -= chunk;
1863                                         copied += chunk;
1864                                 }
1865                         }
1866                 }
1867                 if ((flags & MSG_PEEK) &&
1868                     (peek_seq - copied - urg_hole != tp->copied_seq)) {
1869                         net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1870                                             current->comm,
1871                                             task_pid_nr(current));
1872                         peek_seq = tp->copied_seq;
1873                 }
1874                 continue;
1875 
1876         found_ok_skb:
1877                 /* Ok so how much can we use? */
1878                 used = skb->len - offset;
1879                 if (len < used)
1880                         used = len;
1881 
1882                 /* Do we have urgent data here? */
1883                 if (tp->urg_data) {
1884                         u32 urg_offset = tp->urg_seq - *seq;
1885                         if (urg_offset < used) {
1886                                 if (!urg_offset) {
1887                                         if (!sock_flag(sk, SOCK_URGINLINE)) {
1888                                                 ++*seq;
1889                                                 urg_hole++;
1890                                                 offset++;
1891                                                 used--;
1892                                                 if (!used)
1893                                                         goto skip_copy;
1894                                         }
1895                                 } else
1896                                         used = urg_offset;
1897                         }
1898                 }
1899 
1900                 if (!(flags & MSG_TRUNC)) {
1901                         err = skb_copy_datagram_msg(skb, offset, msg, used);
1902                         if (err) {
1903                                 /* Exception. Bailout! */
1904                                 if (!copied)
1905                                         copied = -EFAULT;
1906                                 break;
1907                         }
1908                 }
1909 
1910                 *seq += used;
1911                 copied += used;
1912                 len -= used;
1913 
1914                 tcp_rcv_space_adjust(sk);
1915 
1916 skip_copy:
1917                 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1918                         tp->urg_data = 0;
1919                         tcp_fast_path_check(sk);
1920                 }
1921                 if (used + offset < skb->len)
1922                         continue;
1923 
1924                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
1925                         goto found_fin_ok;
1926                 if (!(flags & MSG_PEEK))
1927                         sk_eat_skb(sk, skb);
1928                 continue;
1929 
1930         found_fin_ok:
1931                 /* Process the FIN. */
1932                 ++*seq;
1933                 if (!(flags & MSG_PEEK))
1934                         sk_eat_skb(sk, skb);
1935                 break;
1936         } while (len > 0);
1937 
1938         if (user_recv) {
1939                 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
1940                         int chunk;
1941 
1942                         tp->ucopy.len = copied > 0 ? len : 0;
1943 
1944                         tcp_prequeue_process(sk);
1945 
1946                         if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
1947                                 NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
1948                                 len -= chunk;
1949                                 copied += chunk;
1950                         }
1951                 }
1952 
1953                 tp->ucopy.task = NULL;
1954                 tp->ucopy.len = 0;
1955         }
1956 
1957         /* According to UNIX98, msg_name/msg_namelen are ignored
1958          * on connected socket. I was just happy when found this 8) --ANK
1959          */
1960 
1961         /* Clean up data we have read: This will do ACK frames. */
1962         tcp_cleanup_rbuf(sk, copied);
1963 
1964         release_sock(sk);
1965         return copied;
1966 
1967 out:
1968         release_sock(sk);
1969         return err;
1970 
1971 recv_urg:
1972         err = tcp_recv_urg(sk, msg, len, flags);
1973         goto out;
1974 
1975 recv_sndq:
1976         err = tcp_peek_sndq(sk, msg, len);
1977         goto out;
1978 }
1979 EXPORT_SYMBOL(tcp_recvmsg);
1980 
1981 void tcp_set_state(struct sock *sk, int state)
1982 {
1983         int oldstate = sk->sk_state;
1984 
1985         switch (state) {
1986         case TCP_ESTABLISHED:
1987                 if (oldstate != TCP_ESTABLISHED)
1988                         TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
1989                 break;
1990 
1991         case TCP_CLOSE:
1992                 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1993                         TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
1994 
1995                 sk->sk_prot->unhash(sk);
1996                 if (inet_csk(sk)->icsk_bind_hash &&
1997                     !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1998                         inet_put_port(sk);
1999                 /* fall through */
2000         default:
2001                 if (oldstate == TCP_ESTABLISHED)
2002                         TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
2003         }
2004 
2005         /* Change state AFTER socket is unhashed to avoid closed
2006          * socket sitting in hash tables.
2007          */
2008         sk_state_store(sk, state);
2009 
2010 #ifdef STATE_TRACE
2011         SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
2012 #endif
2013 }
2014 EXPORT_SYMBOL_GPL(tcp_set_state);
2015 
2016 /*
2017  *      State processing on a close. This implements the state shift for
2018  *      sending our FIN frame. Note that we only send a FIN for some
2019  *      states. A shutdown() may have already sent the FIN, or we may be
2020  *      closed.
2021  */
2022 
2023 static const unsigned char new_state[16] = {
2024   /* current state:        new state:      action:      */
2025   [0 /* (Invalid) */]   = TCP_CLOSE,
2026   [TCP_ESTABLISHED]     = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2027   [TCP_SYN_SENT]        = TCP_CLOSE,
2028   [TCP_SYN_RECV]        = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2029   [TCP_FIN_WAIT1]       = TCP_FIN_WAIT1,
2030   [TCP_FIN_WAIT2]       = TCP_FIN_WAIT2,
2031   [TCP_TIME_WAIT]       = TCP_CLOSE,
2032   [TCP_CLOSE]           = TCP_CLOSE,
2033   [TCP_CLOSE_WAIT]      = TCP_LAST_ACK  | TCP_ACTION_FIN,
2034   [TCP_LAST_ACK]        = TCP_LAST_ACK,
2035   [TCP_LISTEN]          = TCP_CLOSE,
2036   [TCP_CLOSING]         = TCP_CLOSING,
2037   [TCP_NEW_SYN_RECV]    = TCP_CLOSE,    /* should not happen ! */
2038 };
2039 
2040 static int tcp_close_state(struct sock *sk)
2041 {
2042         int next = (int)new_state[sk->sk_state];
2043         int ns = next & TCP_STATE_MASK;
2044 
2045         tcp_set_state(sk, ns);
2046 
2047         return next & TCP_ACTION_FIN;
2048 }
2049 
2050 /*
2051  *      Shutdown the sending side of a connection. Much like close except
2052  *      that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
2053  */
2054 
2055 void tcp_shutdown(struct sock *sk, int how)
2056 {
2057         /*      We need to grab some memory, and put together a FIN,
2058          *      and then put it into the queue to be sent.
2059          *              Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2060          */
2061         if (!(how & SEND_SHUTDOWN))
2062                 return;
2063 
2064         /* If we've already sent a FIN, or it's a closed state, skip this. */
2065         if ((1 << sk->sk_state) &
2066             (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2067              TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2068                 /* Clear out any half completed packets.  FIN if needed. */
2069                 if (tcp_close_state(sk))
2070                         tcp_send_fin(sk);
2071         }
2072 }
2073 EXPORT_SYMBOL(tcp_shutdown);
2074 
2075 bool tcp_check_oom(struct sock *sk, int shift)
2076 {
2077         bool too_many_orphans, out_of_socket_memory;
2078 
2079         too_many_orphans = tcp_too_many_orphans(sk, shift);
2080         out_of_socket_memory = tcp_out_of_memory(sk);
2081 
2082         if (too_many_orphans)
2083                 net_info_ratelimited("too many orphaned sockets\n");
2084         if (out_of_socket_memory)
2085                 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
2086         return too_many_orphans || out_of_socket_memory;
2087 }
2088 
2089 void tcp_close(struct sock *sk, long timeout)
2090 {
2091         struct sk_buff *skb;
2092         int data_was_unread = 0;
2093         int state;
2094 
2095         lock_sock(sk);
2096         sk->sk_shutdown = SHUTDOWN_MASK;
2097 
2098         if (sk->sk_state == TCP_LISTEN) {
2099                 tcp_set_state(sk, TCP_CLOSE);
2100 
2101                 /* Special case. */
2102                 inet_csk_listen_stop(sk);
2103 
2104                 goto adjudge_to_death;
2105         }
2106 
2107         /*  We need to flush the recv. buffs.  We do this only on the
2108          *  descriptor close, not protocol-sourced closes, because the
2109          *  reader process may not have drained the data yet!
2110          */
2111         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2112                 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
2113 
2114                 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
2115                         len--;
2116                 data_was_unread += len;
2117                 __kfree_skb(skb);
2118         }
2119 
2120         sk_mem_reclaim(sk);
2121 
2122         /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2123         if (sk->sk_state == TCP_CLOSE)
2124                 goto adjudge_to_death;
2125 
2126         /* As outlined in RFC 2525, section 2.17, we send a RST here because
2127          * data was lost. To witness the awful effects of the old behavior of
2128          * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2129          * GET in an FTP client, suspend the process, wait for the client to
2130          * advertise a zero window, then kill -9 the FTP client, wheee...
2131          * Note: timeout is always zero in such a case.
2132          */
2133         if (unlikely(tcp_sk(sk)->repair)) {
2134                 sk->sk_prot->disconnect(sk, 0);
2135         } else if (data_was_unread) {
2136                 /* Unread data was tossed, zap the connection. */
2137                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
2138                 tcp_set_state(sk, TCP_CLOSE);
2139                 tcp_send_active_reset(sk, sk->sk_allocation);
2140         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2141                 /* Check zero linger _after_ checking for unread data. */
2142                 sk->sk_prot->disconnect(sk, 0);
2143                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
2144         } else if (tcp_close_state(sk)) {
2145                 /* We FIN if the application ate all the data before
2146                  * zapping the connection.
2147                  */
2148 
2149                 /* RED-PEN. Formally speaking, we have broken TCP state
2150                  * machine. State transitions:
2151                  *
2152                  * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2153                  * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2154                  * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2155                  *
2156                  * are legal only when FIN has been sent (i.e. in window),
2157                  * rather than queued out of window. Purists blame.
2158                  *
2159                  * F.e. "RFC state" is ESTABLISHED,
2160                  * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2161                  *
2162                  * The visible declinations are that sometimes
2163                  * we enter time-wait state, when it is not required really
2164                  * (harmless), do not send active resets, when they are
2165                  * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2166                  * they look as CLOSING or LAST_ACK for Linux)
2167                  * Probably, I missed some more holelets.
2168                  *                                              --ANK
2169                  * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2170                  * in a single packet! (May consider it later but will
2171                  * probably need API support or TCP_CORK SYN-ACK until
2172                  * data is written and socket is closed.)
2173                  */
2174                 tcp_send_fin(sk);
2175         }
2176 
2177         sk_stream_wait_close(sk, timeout);
2178 
2179 adjudge_to_death:
2180         state = sk->sk_state;
2181         sock_hold(sk);
2182         sock_orphan(sk);
2183 
2184         /* It is the last release_sock in its life. It will remove backlog. */
2185         release_sock(sk);
2186 
2187 
2188         /* Now socket is owned by kernel and we acquire BH lock
2189            to finish close. No need to check for user refs.
2190          */
2191         local_bh_disable();
2192         bh_lock_sock(sk);
2193         WARN_ON(sock_owned_by_user(sk));
2194 
2195         percpu_counter_inc(sk->sk_prot->orphan_count);
2196 
2197         /* Have we already been destroyed by a softirq or backlog? */
2198         if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2199                 goto out;
2200 
2201         /*      This is a (useful) BSD violating of the RFC. There is a
2202          *      problem with TCP as specified in that the other end could
2203          *      keep a socket open forever with no application left this end.
2204          *      We use a 1 minute timeout (about the same as BSD) then kill
2205          *      our end. If they send after that then tough - BUT: long enough
2206          *      that we won't make the old 4*rto = almost no time - whoops
2207          *      reset mistake.
2208          *
2209          *      Nope, it was not mistake. It is really desired behaviour
2210          *      f.e. on http servers, when such sockets are useless, but
2211          *      consume significant resources. Let's do it with special
2212          *      linger2 option.                                 --ANK
2213          */
2214 
2215         if (sk->sk_state == TCP_FIN_WAIT2) {
2216                 struct tcp_sock *tp = tcp_sk(sk);
2217                 if (tp->linger2 < 0) {
2218                         tcp_set_state(sk, TCP_CLOSE);
2219                         tcp_send_active_reset(sk, GFP_ATOMIC);
2220                         __NET_INC_STATS(sock_net(sk),
2221                                         LINUX_MIB_TCPABORTONLINGER);
2222                 } else {
2223                         const int tmo = tcp_fin_time(sk);
2224 
2225                         if (tmo > TCP_TIMEWAIT_LEN) {
2226                                 inet_csk_reset_keepalive_timer(sk,
2227                                                 tmo - TCP_TIMEWAIT_LEN);
2228                         } else {
2229                                 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2230                                 goto out;
2231                         }
2232                 }
2233         }
2234         if (sk->sk_state != TCP_CLOSE) {
2235                 sk_mem_reclaim(sk);
2236                 if (tcp_check_oom(sk, 0)) {
2237                         tcp_set_state(sk, TCP_CLOSE);
2238                         tcp_send_active_reset(sk, GFP_ATOMIC);
2239                         __NET_INC_STATS(sock_net(sk),
2240                                         LINUX_MIB_TCPABORTONMEMORY);
2241                 }
2242         }
2243 
2244         if (sk->sk_state == TCP_CLOSE) {
2245                 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2246                 /* We could get here with a non-NULL req if the socket is
2247                  * aborted (e.g., closed with unread data) before 3WHS
2248                  * finishes.
2249                  */
2250                 if (req)
2251                         reqsk_fastopen_remove(sk, req, false);
2252                 inet_csk_destroy_sock(sk);
2253         }
2254         /* Otherwise, socket is reprieved until protocol close. */
2255 
2256 out:
2257         bh_unlock_sock(sk);
2258         local_bh_enable();
2259         sock_put(sk);
2260 }
2261 EXPORT_SYMBOL(tcp_close);
2262 
2263 /* These states need RST on ABORT according to RFC793 */
2264 
2265 static inline bool tcp_need_reset(int state)
2266 {
2267         return (1 << state) &
2268                (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2269                 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2270 }
2271 
2272 int tcp_disconnect(struct sock *sk, int flags)
2273 {
2274         struct inet_sock *inet = inet_sk(sk);
2275         struct inet_connection_sock *icsk = inet_csk(sk);
2276         struct tcp_sock *tp = tcp_sk(sk);
2277         int err = 0;
2278         int old_state = sk->sk_state;
2279 
2280         if (old_state != TCP_CLOSE)
2281                 tcp_set_state(sk, TCP_CLOSE);
2282 
2283         /* ABORT function of RFC793 */
2284         if (old_state == TCP_LISTEN) {
2285                 inet_csk_listen_stop(sk);
2286         } else if (unlikely(tp->repair)) {
2287                 sk->sk_err = ECONNABORTED;
2288         } else if (tcp_need_reset(old_state) ||
2289                    (tp->snd_nxt != tp->write_seq &&
2290                     (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
2291                 /* The last check adjusts for discrepancy of Linux wrt. RFC
2292                  * states
2293                  */
2294                 tcp_send_active_reset(sk, gfp_any());
2295                 sk->sk_err = ECONNRESET;
2296         } else if (old_state == TCP_SYN_SENT)
2297                 sk->sk_err = ECONNRESET;
2298 
2299         tcp_clear_xmit_timers(sk);
2300         __skb_queue_purge(&sk->sk_receive_queue);
2301         tcp_write_queue_purge(sk);
2302         skb_rbtree_purge(&tp->out_of_order_queue);
2303 
2304         inet->inet_dport = 0;
2305 
2306         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2307                 inet_reset_saddr(sk);
2308 
2309         sk->sk_shutdown = 0;
2310         sock_reset_flag(sk, SOCK_DONE);
2311         tp->srtt_us = 0;
2312         tp->write_seq += tp->max_window + 2;
2313         if (tp->write_seq == 0)
2314                 tp->write_seq = 1;
2315         icsk->icsk_backoff = 0;
2316         tp->snd_cwnd = 2;
2317         icsk->icsk_probes_out = 0;
2318         tp->packets_out = 0;
2319         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
2320         tp->snd_cwnd_cnt = 0;
2321         tp->window_clamp = 0;
2322         tcp_set_ca_state(sk, TCP_CA_Open);
2323         tcp_clear_retrans(tp);
2324         inet_csk_delack_init(sk);
2325         tcp_init_send_head(sk);
2326         memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
2327         __sk_dst_reset(sk);
2328         dst_release(sk->sk_rx_dst);
2329         sk->sk_rx_dst = NULL;
2330         tcp_saved_syn_free(tp);
2331 
2332         /* Clean up fastopen related fields */
2333         tcp_free_fastopen_req(tp);
2334         inet->defer_connect = 0;
2335 
2336         WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
2337 
2338         sk->sk_error_report(sk);
2339         return err;
2340 }
2341 EXPORT_SYMBOL(tcp_disconnect);
2342 
2343 static inline bool tcp_can_repair_sock(const struct sock *sk)
2344 {
2345         return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
2346                 (sk->sk_state != TCP_LISTEN);
2347 }
2348 
2349 static int tcp_repair_set_window(struct tcp_sock *tp, char __user *optbuf, int len)
2350 {
2351         struct tcp_repair_window opt;
2352 
2353         if (!tp->repair)
2354                 return -EPERM;
2355 
2356         if (len != sizeof(opt))
2357                 return -EINVAL;
2358 
2359         if (copy_from_user(&opt, optbuf, sizeof(opt)))
2360                 return -EFAULT;
2361 
2362         if (opt.max_window < opt.snd_wnd)
2363                 return -EINVAL;
2364 
2365         if (after(opt.snd_wl1, tp->rcv_nxt + opt.rcv_wnd))
2366                 return -EINVAL;
2367 
2368         if (after(opt.rcv_wup, tp->rcv_nxt))
2369                 return -EINVAL;
2370 
2371         tp->snd_wl1     = opt.snd_wl1;
2372         tp->snd_wnd     = opt.snd_wnd;
2373         tp->max_window  = opt.max_window;
2374 
2375         tp->rcv_wnd     = opt.rcv_wnd;
2376         tp->rcv_wup     = opt.rcv_wup;
2377 
2378         return 0;
2379 }
2380 
2381 static int tcp_repair_options_est(struct tcp_sock *tp,
2382                 struct tcp_repair_opt __user *optbuf, unsigned int len)
2383 {
2384         struct tcp_repair_opt opt;
2385 
2386         while (len >= sizeof(opt)) {
2387                 if (copy_from_user(&opt, optbuf, sizeof(opt)))
2388                         return -EFAULT;
2389 
2390                 optbuf++;
2391                 len -= sizeof(opt);
2392 
2393                 switch (opt.opt_code) {
2394                 case TCPOPT_MSS:
2395                         tp->rx_opt.mss_clamp = opt.opt_val;
2396                         break;
2397                 case TCPOPT_WINDOW:
2398                         {
2399                                 u16 snd_wscale = opt.opt_val & 0xFFFF;
2400                                 u16 rcv_wscale = opt.opt_val >> 16;
2401 
2402                                 if (snd_wscale > 14 || rcv_wscale > 14)
2403                                         return -EFBIG;
2404 
2405                                 tp->rx_opt.snd_wscale = snd_wscale;
2406                                 tp->rx_opt.rcv_wscale = rcv_wscale;
2407                                 tp->rx_opt.wscale_ok = 1;
2408                         }
2409                         break;
2410                 case TCPOPT_SACK_PERM:
2411                         if (opt.opt_val != 0)
2412                                 return -EINVAL;
2413 
2414                         tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2415                         if (sysctl_tcp_fack)
2416                                 tcp_enable_fack(tp);
2417                         break;
2418                 case TCPOPT_TIMESTAMP:
2419                         if (opt.opt_val != 0)
2420                                 return -EINVAL;
2421 
2422                         tp->rx_opt.tstamp_ok = 1;
2423                         break;
2424                 }
2425         }
2426 
2427         return 0;
2428 }
2429 
2430 /*
2431  *      Socket option code for TCP.
2432  */
2433 static int do_tcp_setsockopt(struct sock *sk, int level,
2434                 int optname, char __user *optval, unsigned int optlen)
2435 {
2436         struct tcp_sock *tp = tcp_sk(sk);
2437         struct inet_connection_sock *icsk = inet_csk(sk);
2438         struct net *net = sock_net(sk);
2439         int val;
2440         int err = 0;
2441 
2442         /* These are data/string values, all the others are ints */
2443         switch (optname) {
2444         case TCP_CONGESTION: {
2445                 char name[TCP_CA_NAME_MAX];
2446 
2447                 if (optlen < 1)
2448                         return -EINVAL;
2449 
2450                 val = strncpy_from_user(name, optval,
2451                                         min_t(long, TCP_CA_NAME_MAX-1, optlen));
2452                 if (val < 0)
2453                         return -EFAULT;
2454                 name[val] = 0;
2455 
2456                 lock_sock(sk);
2457                 err = tcp_set_congestion_control(sk, name);
2458                 release_sock(sk);
2459                 return err;
2460         }
2461         default:
2462                 /* fallthru */
2463                 break;
2464         }
2465 
2466         if (optlen < sizeof(int))
2467                 return -EINVAL;
2468 
2469         if (get_user(val, (int __user *)optval))
2470                 return -EFAULT;
2471 
2472         lock_sock(sk);
2473 
2474         switch (optname) {
2475         case TCP_MAXSEG:
2476                 /* Values greater than interface MTU won't take effect. However
2477                  * at the point when this call is done we typically don't yet
2478                  * know which interface is going to be used */
2479                 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
2480                         err = -EINVAL;
2481                         break;
2482                 }
2483                 tp->rx_opt.user_mss = val;
2484                 break;
2485 
2486         case TCP_NODELAY:
2487                 if (val) {
2488                         /* TCP_NODELAY is weaker than TCP_CORK, so that
2489                          * this option on corked socket is remembered, but
2490                          * it is not activated until cork is cleared.
2491                          *
2492                          * However, when TCP_NODELAY is set we make
2493                          * an explicit push, which overrides even TCP_CORK
2494                          * for currently queued segments.
2495                          */
2496                         tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
2497                         tcp_push_pending_frames(sk);
2498                 } else {
2499                         tp->nonagle &= ~TCP_NAGLE_OFF;
2500                 }
2501                 break;
2502 
2503         case TCP_THIN_LINEAR_TIMEOUTS:
2504                 if (val < 0 || val > 1)
2505                         err = -EINVAL;
2506                 else
2507                         tp->thin_lto = val;
2508                 break;
2509 
2510         case TCP_THIN_DUPACK:
2511                 if (val < 0 || val > 1)
2512                         err = -EINVAL;
2513                 break;
2514 
2515         case TCP_REPAIR:
2516                 if (!tcp_can_repair_sock(sk))
2517                         err = -EPERM;
2518                 else if (val == 1) {
2519                         tp->repair = 1;
2520                         sk->sk_reuse = SK_FORCE_REUSE;
2521                         tp->repair_queue = TCP_NO_QUEUE;
2522                 } else if (val == 0) {
2523                         tp->repair = 0;
2524                         sk->sk_reuse = SK_NO_REUSE;
2525                         tcp_send_window_probe(sk);
2526                 } else
2527                         err = -EINVAL;
2528 
2529                 break;
2530 
2531         case TCP_REPAIR_QUEUE:
2532                 if (!tp->repair)
2533                         err = -EPERM;
2534                 else if (val < TCP_QUEUES_NR)
2535                         tp->repair_queue = val;
2536                 else
2537                         err = -EINVAL;
2538                 break;
2539 
2540         case TCP_QUEUE_SEQ:
2541                 if (sk->sk_state != TCP_CLOSE)
2542                         err = -EPERM;
2543                 else if (tp->repair_queue == TCP_SEND_QUEUE)
2544                         tp->write_seq = val;
2545                 else if (tp->repair_queue == TCP_RECV_QUEUE)
2546                         tp->rcv_nxt = val;
2547                 else
2548                         err = -EINVAL;
2549                 break;
2550 
2551         case TCP_REPAIR_OPTIONS:
2552                 if (!tp->repair)
2553                         err = -EINVAL;
2554                 else if (sk->sk_state == TCP_ESTABLISHED)
2555                         err = tcp_repair_options_est(tp,
2556                                         (struct tcp_repair_opt __user *)optval,
2557                                         optlen);
2558                 else
2559                         err = -EPERM;
2560                 break;
2561 
2562         case TCP_CORK:
2563                 /* When set indicates to always queue non-full frames.
2564                  * Later the user clears this option and we transmit
2565                  * any pending partial frames in the queue.  This is
2566                  * meant to be used alongside sendfile() to get properly
2567                  * filled frames when the user (for example) must write
2568                  * out headers with a write() call first and then use
2569                  * sendfile to send out the data parts.
2570                  *
2571                  * TCP_CORK can be set together with TCP_NODELAY and it is
2572                  * stronger than TCP_NODELAY.
2573                  */
2574                 if (val) {
2575                         tp->nonagle |= TCP_NAGLE_CORK;
2576                 } else {
2577                         tp->nonagle &= ~TCP_NAGLE_CORK;
2578                         if (tp->nonagle&TCP_NAGLE_OFF)
2579                                 tp->nonagle |= TCP_NAGLE_PUSH;
2580                         tcp_push_pending_frames(sk);
2581                 }
2582                 break;
2583 
2584         case TCP_KEEPIDLE:
2585                 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2586                         err = -EINVAL;
2587                 else {
2588                         tp->keepalive_time = val * HZ;
2589                         if (sock_flag(sk, SOCK_KEEPOPEN) &&
2590                             !((1 << sk->sk_state) &
2591                               (TCPF_CLOSE | TCPF_LISTEN))) {
2592                                 u32 elapsed = keepalive_time_elapsed(tp);
2593                                 if (tp->keepalive_time > elapsed)
2594                                         elapsed = tp->keepalive_time - elapsed;
2595                                 else
2596                                         elapsed = 0;
2597                                 inet_csk_reset_keepalive_timer(sk, elapsed);
2598                         }
2599                 }
2600                 break;
2601         case TCP_KEEPINTVL:
2602                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2603                         err = -EINVAL;
2604                 else
2605                         tp->keepalive_intvl = val * HZ;
2606                 break;
2607         case TCP_KEEPCNT:
2608                 if (val < 1 || val > MAX_TCP_KEEPCNT)
2609                         err = -EINVAL;
2610                 else
2611                         tp->keepalive_probes = val;
2612                 break;
2613         case TCP_SYNCNT:
2614                 if (val < 1 || val > MAX_TCP_SYNCNT)
2615                         err = -EINVAL;
2616                 else
2617                         icsk->icsk_syn_retries = val;
2618                 break;
2619 
2620         case TCP_SAVE_SYN:
2621                 if (val < 0 || val > 1)
2622                         err = -EINVAL;
2623                 else
2624                         tp->save_syn = val;
2625                 break;
2626 
2627         case TCP_LINGER2:
2628                 if (val < 0)
2629                         tp->linger2 = -1;
2630                 else if (val > net->ipv4.sysctl_tcp_fin_timeout / HZ)
2631                         tp->linger2 = 0;
2632                 else
2633                         tp->linger2 = val * HZ;
2634                 break;
2635 
2636         case TCP_DEFER_ACCEPT:
2637                 /* Translate value in seconds to number of retransmits */
2638                 icsk->icsk_accept_queue.rskq_defer_accept =
2639                         secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2640                                         TCP_RTO_MAX / HZ);
2641                 break;
2642 
2643         case TCP_WINDOW_CLAMP:
2644                 if (!val) {
2645                         if (sk->sk_state != TCP_CLOSE) {
2646                                 err = -EINVAL;
2647                                 break;
2648                         }
2649                         tp->window_clamp = 0;
2650                 } else
2651                         tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2652                                                 SOCK_MIN_RCVBUF / 2 : val;
2653                 break;
2654 
2655         case TCP_QUICKACK:
2656                 if (!val) {
2657                         icsk->icsk_ack.pingpong = 1;
2658                 } else {
2659                         icsk->icsk_ack.pingpong = 0;
2660                         if ((1 << sk->sk_state) &
2661                             (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
2662                             inet_csk_ack_scheduled(sk)) {
2663                                 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
2664                                 tcp_cleanup_rbuf(sk, 1);
2665                                 if (!(val & 1))
2666                                         icsk->icsk_ack.pingpong = 1;
2667                         }
2668                 }
2669                 break;
2670 
2671 #ifdef CONFIG_TCP_MD5SIG
2672         case TCP_MD5SIG:
2673                 /* Read the IP->Key mappings from userspace */
2674                 err = tp->af_specific->md5_parse(sk, optval, optlen);
2675                 break;
2676 #endif
2677         case TCP_USER_TIMEOUT:
2678                 /* Cap the max time in ms TCP will retry or probe the window
2679                  * before giving up and aborting (ETIMEDOUT) a connection.
2680                  */
2681                 if (val < 0)
2682                         err = -EINVAL;
2683                 else
2684                         icsk->icsk_user_timeout = msecs_to_jiffies(val);
2685                 break;
2686 
2687         case TCP_FASTOPEN:
2688                 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
2689                     TCPF_LISTEN))) {
2690                         tcp_fastopen_init_key_once(true);
2691 
2692                         fastopen_queue_tune(sk, val);
2693                 } else {
2694                         err = -EINVAL;
2695                 }
2696                 break;
2697         case TCP_FASTOPEN_CONNECT:
2698                 if (val > 1 || val < 0) {
2699                         err = -EINVAL;
2700                 } else if (sysctl_tcp_fastopen & TFO_CLIENT_ENABLE) {
2701                         if (sk->sk_state == TCP_CLOSE)
2702                                 tp->fastopen_connect = val;
2703                         else
2704                                 err = -EINVAL;
2705                 } else {
2706                         err = -EOPNOTSUPP;
2707                 }
2708                 break;
2709         case TCP_TIMESTAMP:
2710                 if (!tp->repair)
2711                         err = -EPERM;
2712                 else
2713                         tp->tsoffset = val - tcp_time_stamp;
2714                 break;
2715         case TCP_REPAIR_WINDOW:
2716                 err = tcp_repair_set_window(tp, optval, optlen);
2717                 break;
2718         case TCP_NOTSENT_LOWAT:
2719                 tp->notsent_lowat = val;
2720                 sk->sk_write_space(sk);
2721                 break;
2722         default:
2723                 err = -ENOPROTOOPT;
2724                 break;
2725         }
2726 
2727         release_sock(sk);
2728         return err;
2729 }
2730 
2731 int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
2732                    unsigned int optlen)
2733 {
2734         const struct inet_connection_sock *icsk = inet_csk(sk);
2735 
2736         if (level != SOL_TCP)
2737                 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2738                                                      optval, optlen);
2739         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2740 }
2741 EXPORT_SYMBOL(tcp_setsockopt);
2742 
2743 #ifdef CONFIG_COMPAT
2744 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
2745                           char __user *optval, unsigned int optlen)
2746 {
2747         if (level != SOL_TCP)
2748                 return inet_csk_compat_setsockopt(sk, level, optname,
2749                                                   optval, optlen);
2750         return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2751 }
2752 EXPORT_SYMBOL(compat_tcp_setsockopt);
2753 #endif
2754 
2755 static void tcp_get_info_chrono_stats(const struct tcp_sock *tp,
2756                                       struct tcp_info *info)
2757 {
2758         u64 stats[__TCP_CHRONO_MAX], total = 0;
2759         enum tcp_chrono i;
2760 
2761         for (i = TCP_CHRONO_BUSY; i < __TCP_CHRONO_MAX; ++i) {
2762                 stats[i] = tp->chrono_stat[i - 1];
2763                 if (i == tp->chrono_type)
2764                         stats[i] += tcp_time_stamp - tp->chrono_start;
2765                 stats[i] *= USEC_PER_SEC / HZ;
2766                 total += stats[i];
2767         }
2768 
2769         info->tcpi_busy_time = total;
2770         info->tcpi_rwnd_limited = stats[TCP_CHRONO_RWND_LIMITED];
2771         info->tcpi_sndbuf_limited = stats[TCP_CHRONO_SNDBUF_LIMITED];
2772 }
2773 
2774 /* Return information about state of tcp endpoint in API format. */
2775 void tcp_get_info(struct sock *sk, struct tcp_info *info)
2776 {
2777         const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
2778         const struct inet_connection_sock *icsk = inet_csk(sk);
2779         u32 now, intv;
2780         u64 rate64;
2781         bool slow;
2782         u32 rate;
2783 
2784         memset(info, 0, sizeof(*info));
2785         if (sk->sk_type != SOCK_STREAM)
2786                 return;
2787 
2788         info->tcpi_state = sk_state_load(sk);
2789 
2790         /* Report meaningful fields for all TCP states, including listeners */
2791         rate = READ_ONCE(sk->sk_pacing_rate);
2792         rate64 = rate != ~0U ? rate : ~0ULL;
2793         info->tcpi_pacing_rate = rate64;
2794 
2795         rate = READ_ONCE(sk->sk_max_pacing_rate);
2796         rate64 = rate != ~0U ? rate : ~0ULL;
2797         info->tcpi_max_pacing_rate = rate64;
2798 
2799         info->tcpi_reordering = tp->reordering;
2800         info->tcpi_snd_cwnd = tp->snd_cwnd;
2801 
2802         if (info->tcpi_state == TCP_LISTEN) {
2803                 /* listeners aliased fields :
2804                  * tcpi_unacked -> Number of children ready for accept()
2805                  * tcpi_sacked  -> max backlog
2806                  */
2807                 info->tcpi_unacked = sk->sk_ack_backlog;
2808                 info->tcpi_sacked = sk->sk_max_ack_backlog;
2809                 return;
2810         }
2811 
2812         slow = lock_sock_fast(sk);
2813 
2814         info->tcpi_ca_state = icsk->icsk_ca_state;
2815         info->tcpi_retransmits = icsk->icsk_retransmits;
2816         info->tcpi_probes = icsk->icsk_probes_out;
2817         info->tcpi_backoff = icsk->icsk_backoff;
2818 
2819         if (tp->rx_opt.tstamp_ok)
2820                 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
2821         if (tcp_is_sack(tp))
2822                 info->tcpi_options |= TCPI_OPT_SACK;
2823         if (tp->rx_opt.wscale_ok) {
2824                 info->tcpi_options |= TCPI_OPT_WSCALE;
2825                 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2826                 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
2827         }
2828 
2829         if (tp->ecn_flags & TCP_ECN_OK)
2830                 info->tcpi_options |= TCPI_OPT_ECN;
2831         if (tp->ecn_flags & TCP_ECN_SEEN)
2832                 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
2833         if (tp->syn_data_acked)
2834                 info->tcpi_options |= TCPI_OPT_SYN_DATA;
2835 
2836         info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2837         info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
2838         info->tcpi_snd_mss = tp->mss_cache;
2839         info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
2840 
2841         info->tcpi_unacked = tp->packets_out;
2842         info->tcpi_sacked = tp->sacked_out;
2843 
2844         info->tcpi_lost = tp->lost_out;
2845         info->tcpi_retrans = tp->retrans_out;
2846         info->tcpi_fackets = tp->fackets_out;
2847 
2848         now = tcp_time_stamp;
2849         info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
2850         info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
2851         info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2852 
2853         info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
2854         info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2855         info->tcpi_rtt = tp->srtt_us >> 3;
2856         info->tcpi_rttvar = tp->mdev_us >> 2;
2857         info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2858         info->tcpi_advmss = tp->advmss;
2859 
2860         info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2861         info->tcpi_rcv_space = tp->rcvq_space.space;
2862 
2863         info->tcpi_total_retrans = tp->total_retrans;
2864 
2865         info->tcpi_bytes_acked = tp->bytes_acked;
2866         info->tcpi_bytes_received = tp->bytes_received;
2867         info->tcpi_notsent_bytes = max_t(int, 0, tp->write_seq - tp->snd_nxt);
2868         tcp_get_info_chrono_stats(tp, info);
2869 
2870         info->tcpi_segs_out = tp->segs_out;
2871         info->tcpi_segs_in = tp->segs_in;
2872 
2873         info->tcpi_min_rtt = tcp_min_rtt(tp);
2874         info->tcpi_data_segs_in = tp->data_segs_in;
2875         info->tcpi_data_segs_out = tp->data_segs_out;
2876 
2877         info->tcpi_delivery_rate_app_limited = tp->rate_app_limited ? 1 : 0;
2878         rate = READ_ONCE(tp->rate_delivered);
2879         intv = READ_ONCE(tp->rate_interval_us);
2880         if (rate && intv) {
2881                 rate64 = (u64)rate * tp->mss_cache * USEC_PER_SEC;
2882                 do_div(rate64, intv);
2883                 info->tcpi_delivery_rate = rate64;
2884         }
2885         unlock_sock_fast(sk, slow);
2886 }
2887 EXPORT_SYMBOL_GPL(tcp_get_info);
2888 
2889 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk)
2890 {
2891         const struct tcp_sock *tp = tcp_sk(sk);
2892         struct sk_buff *stats;
2893         struct tcp_info info;
2894 
2895         stats = alloc_skb(5 * nla_total_size_64bit(sizeof(u64)), GFP_ATOMIC);
2896         if (!stats)
2897                 return NULL;
2898 
2899         tcp_get_info_chrono_stats(tp, &info);
2900         nla_put_u64_64bit(stats, TCP_NLA_BUSY,
2901                           info.tcpi_busy_time, TCP_NLA_PAD);
2902         nla_put_u64_64bit(stats, TCP_NLA_RWND_LIMITED,
2903                           info.tcpi_rwnd_limited, TCP_NLA_PAD);
2904         nla_put_u64_64bit(stats, TCP_NLA_SNDBUF_LIMITED,
2905                           info.tcpi_sndbuf_limited, TCP_NLA_PAD);
2906         nla_put_u64_64bit(stats, TCP_NLA_DATA_SEGS_OUT,
2907                           tp->data_segs_out, TCP_NLA_PAD);
2908         nla_put_u64_64bit(stats, TCP_NLA_TOTAL_RETRANS,
2909                           tp->total_retrans, TCP_NLA_PAD);
2910         return stats;
2911 }
2912 
2913 static int do_tcp_getsockopt(struct sock *sk, int level,
2914                 int optname, char __user *optval, int __user *optlen)
2915 {
2916         struct inet_connection_sock *icsk = inet_csk(sk);
2917         struct tcp_sock *tp = tcp_sk(sk);
2918         struct net *net = sock_net(sk);
2919         int val, len;
2920 
2921         if (get_user(len, optlen))
2922                 return -EFAULT;
2923 
2924         len = min_t(unsigned int, len, sizeof(int));
2925 
2926         if (len < 0)
2927                 return -EINVAL;
2928 
2929         switch (optname) {
2930         case TCP_MAXSEG:
2931                 val = tp->mss_cache;
2932                 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2933                         val = tp->rx_opt.user_mss;
2934                 if (tp->repair)
2935                         val = tp->rx_opt.mss_clamp;
2936                 break;
2937         case TCP_NODELAY:
2938                 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2939                 break;
2940         case TCP_CORK:
2941                 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2942                 break;
2943         case TCP_KEEPIDLE:
2944                 val = keepalive_time_when(tp) / HZ;
2945                 break;
2946         case TCP_KEEPINTVL:
2947                 val = keepalive_intvl_when(tp) / HZ;
2948                 break;
2949         case TCP_KEEPCNT:
2950                 val = keepalive_probes(tp);
2951                 break;
2952         case TCP_SYNCNT:
2953                 val = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
2954                 break;
2955         case TCP_LINGER2:
2956                 val = tp->linger2;
2957                 if (val >= 0)
2958                         val = (val ? : net->ipv4.sysctl_tcp_fin_timeout) / HZ;
2959                 break;
2960         case TCP_DEFER_ACCEPT:
2961                 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2962                                       TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
2963                 break;
2964         case TCP_WINDOW_CLAMP:
2965                 val = tp->window_clamp;
2966                 break;
2967         case TCP_INFO: {
2968                 struct tcp_info info;
2969 
2970                 if (get_user(len, optlen))
2971                         return -EFAULT;
2972 
2973                 tcp_get_info(sk, &info);
2974 
2975                 len = min_t(unsigned int, len, sizeof(info));
2976                 if (put_user(len, optlen))
2977                         return -EFAULT;
2978                 if (copy_to_user(optval, &info, len))
2979                         return -EFAULT;
2980                 return 0;
2981         }
2982         case TCP_CC_INFO: {
2983                 const struct tcp_congestion_ops *ca_ops;
2984                 union tcp_cc_info info;
2985                 size_t sz = 0;
2986                 int attr;
2987 
2988                 if (get_user(len, optlen))
2989                         return -EFAULT;
2990 
2991                 ca_ops = icsk->icsk_ca_ops;
2992                 if (ca_ops && ca_ops->get_info)
2993                         sz = ca_ops->get_info(sk, ~0U, &attr, &info);
2994 
2995                 len = min_t(unsigned int, len, sz);
2996                 if (put_user(len, optlen))
2997                         return -EFAULT;
2998                 if (copy_to_user(optval, &info, len))
2999                         return -EFAULT;
3000                 return 0;
3001         }
3002         case TCP_QUICKACK:
3003                 val = !icsk->icsk_ack.pingpong;
3004                 break;
3005 
3006         case TCP_CONGESTION:
3007                 if (get_user(len, optlen))
3008                         return -EFAULT;
3009                 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
3010                 if (put_user(len, optlen))
3011                         return -EFAULT;
3012                 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
3013                         return -EFAULT;
3014                 return 0;
3015 
3016         case TCP_THIN_LINEAR_TIMEOUTS:
3017                 val = tp->thin_lto;
3018                 break;
3019 
3020         case TCP_THIN_DUPACK:
3021                 val = 0;
3022                 break;
3023 
3024         case TCP_REPAIR:
3025                 val = tp->repair;
3026                 break;
3027 
3028         case TCP_REPAIR_QUEUE:
3029                 if (tp->repair)
3030                         val = tp->repair_queue;
3031                 else
3032                         return -EINVAL;
3033                 break;
3034 
3035         case TCP_REPAIR_WINDOW: {
3036                 struct tcp_repair_window opt;
3037 
3038                 if (get_user(len, optlen))
3039                         return -EFAULT;
3040 
3041                 if (len != sizeof(opt))
3042                         return -EINVAL;
3043 
3044                 if (!tp->repair)
3045                         return -EPERM;
3046 
3047                 opt.snd_wl1     = tp->snd_wl1;
3048                 opt.snd_wnd     = tp->snd_wnd;
3049                 opt.max_window  = tp->max_window;
3050                 opt.rcv_wnd     = tp->rcv_wnd;
3051                 opt.rcv_wup     = tp->rcv_wup;
3052 
3053                 if (copy_to_user(optval, &opt, len))
3054                         return -EFAULT;
3055                 return 0;
3056         }
3057         case TCP_QUEUE_SEQ:
3058                 if (tp->repair_queue == TCP_SEND_QUEUE)
3059                         val = tp->write_seq;
3060                 else if (tp->repair_queue == TCP_RECV_QUEUE)
3061                         val = tp->rcv_nxt;
3062                 else
3063                         return -EINVAL;
3064                 break;
3065 
3066         case TCP_USER_TIMEOUT:
3067                 val = jiffies_to_msecs(icsk->icsk_user_timeout);
3068                 break;
3069 
3070         case TCP_FASTOPEN:
3071                 val = icsk->icsk_accept_queue.fastopenq.max_qlen;
3072                 break;
3073 
3074         case TCP_FASTOPEN_CONNECT:
3075                 val = tp->fastopen_connect;
3076                 break;
3077 
3078         case TCP_TIMESTAMP:
3079                 val = tcp_time_stamp + tp->tsoffset;
3080                 break;
3081         case TCP_NOTSENT_LOWAT:
3082                 val = tp->notsent_lowat;
3083                 break;
3084         case TCP_SAVE_SYN:
3085                 val = tp->save_syn;
3086                 break;
3087         case TCP_SAVED_SYN: {
3088                 if (get_user(len, optlen))
3089                         return -EFAULT;
3090 
3091                 lock_sock(sk);
3092                 if (tp->saved_syn) {
3093                         if (len < tp->saved_syn[0]) {
3094                                 if (put_user(tp->saved_syn[0], optlen)) {
3095                                         release_sock(sk);
3096                                         return -EFAULT;
3097                                 }
3098                                 release_sock(sk);
3099                                 return -EINVAL;
3100                         }
3101                         len = tp->saved_syn[0];
3102                         if (put_user(len, optlen)) {
3103                                 release_sock(sk);
3104                                 return -EFAULT;
3105                         }
3106                         if (copy_to_user(optval, tp->saved_syn + 1, len)) {
3107                                 release_sock(sk);
3108                                 return -EFAULT;
3109                         }
3110                         tcp_saved_syn_free(tp);
3111                         release_sock(sk);
3112                 } else {
3113                         release_sock(sk);
3114                         len = 0;
3115                         if (put_user(len, optlen))
3116                                 return -EFAULT;
3117                 }
3118                 return 0;
3119         }
3120         default:
3121                 return -ENOPROTOOPT;
3122         }
3123 
3124         if (put_user(len, optlen))
3125                 return -EFAULT;
3126         if (copy_to_user(optval, &val, len))
3127                 return -EFAULT;
3128         return 0;
3129 }
3130 
3131 int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
3132                    int __user *optlen)
3133 {
3134         struct inet_connection_sock *icsk = inet_csk(sk);
3135 
3136         if (level != SOL_TCP)
3137                 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
3138                                                      optval, optlen);
3139         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3140 }
3141 EXPORT_SYMBOL(tcp_getsockopt);
3142 
3143 #ifdef CONFIG_COMPAT
3144 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
3145                           char __user *optval, int __user *optlen)
3146 {
3147         if (level != SOL_TCP)
3148                 return inet_csk_compat_getsockopt(sk, level, optname,
3149                                                   optval, optlen);
3150         return do_tcp_getsockopt(sk, level, optname, optval, optlen);
3151 }
3152 EXPORT_SYMBOL(compat_tcp_getsockopt);
3153 #endif
3154 
3155 #ifdef CONFIG_TCP_MD5SIG
3156 static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
3157 static DEFINE_MUTEX(tcp_md5sig_mutex);
3158 static bool tcp_md5sig_pool_populated = false;
3159 
3160 static void __tcp_alloc_md5sig_pool(void)
3161 {
3162         struct crypto_ahash *hash;
3163         int cpu;
3164 
3165         hash = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
3166         if (IS_ERR(hash))
3167                 return;
3168 
3169         for_each_possible_cpu(cpu) {
3170                 void *scratch = per_cpu(tcp_md5sig_pool, cpu).scratch;
3171                 struct ahash_request *req;
3172 
3173                 if (!scratch) {
3174                         scratch = kmalloc_node(sizeof(union tcp_md5sum_block) +
3175                                                sizeof(struct tcphdr),
3176                                                GFP_KERNEL,
3177                                                cpu_to_node(cpu));
3178                         if (!scratch)
3179                                 return;
3180                         per_cpu(tcp_md5sig_pool, cpu).scratch = scratch;
3181                 }
3182                 if (per_cpu(tcp_md5sig_pool, cpu).md5_req)
3183                         continue;
3184 
3185                 req = ahash_request_alloc(hash, GFP_KERNEL);
3186                 if (!req)
3187                         return;
3188 
3189                 ahash_request_set_callback(req, 0, NULL, NULL);
3190 
3191                 per_cpu(tcp_md5sig_pool, cpu).md5_req = req;
3192         }
3193         /* before setting tcp_md5sig_pool_populated, we must commit all writes
3194          * to memory. See smp_rmb() in tcp_get_md5sig_pool()
3195          */
3196         smp_wmb();
3197         tcp_md5sig_pool_populated = true;
3198 }
3199 
3200 bool tcp_alloc_md5sig_pool(void)
3201 {
3202         if (unlikely(!tcp_md5sig_pool_populated)) {
3203                 mutex_lock(&tcp_md5sig_mutex);
3204 
3205                 if (!tcp_md5sig_pool_populated)
3206                         __tcp_alloc_md5sig_pool();
3207 
3208                 mutex_unlock(&tcp_md5sig_mutex);
3209         }
3210         return tcp_md5sig_pool_populated;
3211 }
3212 EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3213 
3214 
3215 /**
3216  *      tcp_get_md5sig_pool - get md5sig_pool for this user
3217  *
3218  *      We use percpu structure, so if we succeed, we exit with preemption
3219  *      and BH disabled, to make sure another thread or softirq handling
3220  *      wont try to get same context.
3221  */
3222 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
3223 {
3224         local_bh_disable();
3225 
3226         if (tcp_md5sig_pool_populated) {
3227                 /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
3228                 smp_rmb();
3229                 return this_cpu_ptr(&tcp_md5sig_pool);
3230         }
3231         local_bh_enable();
3232         return NULL;
3233 }
3234 EXPORT_SYMBOL(tcp_get_md5sig_pool);
3235 
3236 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
3237                           const struct sk_buff *skb, unsigned int header_len)
3238 {
3239         struct scatterlist sg;
3240         const struct tcphdr *tp = tcp_hdr(skb);
3241         struct ahash_request *req = hp->md5_req;
3242         unsigned int i;
3243         const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3244                                            skb_headlen(skb) - header_len : 0;
3245         const struct skb_shared_info *shi = skb_shinfo(skb);
3246         struct sk_buff *frag_iter;
3247 
3248         sg_init_table(&sg, 1);
3249 
3250         sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3251         ahash_request_set_crypt(req, &sg, NULL, head_data_len);
3252         if (crypto_ahash_update(req))
3253                 return 1;
3254 
3255         for (i = 0; i < shi->nr_frags; ++i) {
3256                 const struct skb_frag_struct *f = &shi->frags[i];
3257                 unsigned int offset = f->page_offset;
3258                 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3259 
3260                 sg_set_page(&sg, page, skb_frag_size(f),
3261                             offset_in_page(offset));
3262                 ahash_request_set_crypt(req, &sg, NULL, skb_frag_size(f));
3263                 if (crypto_ahash_update(req))
3264                         return 1;
3265         }
3266 
3267         skb_walk_frags(skb, frag_iter)
3268                 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3269                         return 1;
3270 
3271         return 0;
3272 }
3273 EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3274 
3275 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
3276 {
3277         struct scatterlist sg;
3278 
3279         sg_init_one(&sg, key->key, key->keylen);
3280         ahash_request_set_crypt(hp->md5_req, &sg, NULL, key->keylen);
3281         return crypto_ahash_update(hp->md5_req);
3282 }
3283 EXPORT_SYMBOL(tcp_md5_hash_key);
3284 
3285 #endif
3286 
3287 void tcp_done(struct sock *sk)
3288 {
3289         struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3290 
3291         if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
3292                 TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
3293 
3294         tcp_set_state(sk, TCP_CLOSE);
3295         tcp_clear_xmit_timers(sk);
3296         if (req)
3297                 reqsk_fastopen_remove(sk, req, false);
3298 
3299         sk->sk_shutdown = SHUTDOWN_MASK;
3300 
3301         if (!sock_flag(sk, SOCK_DEAD))
3302                 sk->sk_state_change(sk);
3303         else
3304                 inet_csk_destroy_sock(sk);
3305 }
3306 EXPORT_SYMBOL_GPL(tcp_done);
3307 
3308 int tcp_abort(struct sock *sk, int err)
3309 {
3310         if (!sk_fullsock(sk)) {
3311                 if (sk->sk_state == TCP_NEW_SYN_RECV) {
3312                         struct request_sock *req = inet_reqsk(sk);
3313 
3314                         local_bh_disable();
3315                         inet_csk_reqsk_queue_drop_and_put(req->rsk_listener,
3316                                                           req);
3317                         local_bh_enable();
3318                         return 0;
3319                 }
3320                 return -EOPNOTSUPP;
3321         }
3322 
3323         /* Don't race with userspace socket closes such as tcp_close. */
3324         lock_sock(sk);
3325 
3326         if (sk->sk_state == TCP_LISTEN) {
3327                 tcp_set_state(sk, TCP_CLOSE);
3328                 inet_csk_listen_stop(sk);
3329         }
3330 
3331         /* Don't race with BH socket closes such as inet_csk_listen_stop. */
3332         local_bh_disable();
3333         bh_lock_sock(sk);
3334 
3335         if (!sock_flag(sk, SOCK_DEAD)) {
3336                 sk->sk_err = err;
3337                 /* This barrier is coupled with smp_rmb() in tcp_poll() */
3338                 smp_wmb();
3339                 sk->sk_error_report(sk);
3340                 if (tcp_need_reset(sk->sk_state))
3341                         tcp_send_active_reset(sk, GFP_ATOMIC);
3342                 tcp_done(sk);
3343         }
3344 
3345         bh_unlock_sock(sk);
3346         local_bh_enable();
3347         release_sock(sk);
3348         return 0;
3349 }
3350 EXPORT_SYMBOL_GPL(tcp_abort);
3351 
3352 extern struct tcp_congestion_ops tcp_reno;
3353 
3354 static __initdata unsigned long thash_entries;
3355 static int __init set_thash_entries(char *str)
3356 {
3357         ssize_t ret;
3358 
3359         if (!str)
3360                 return 0;
3361 
3362         ret = kstrtoul(str, 0, &thash_entries);
3363         if (ret)
3364                 return 0;
3365 
3366         return 1;
3367 }
3368 __setup("thash_entries=", set_thash_entries);
3369 
3370 static void __init tcp_init_mem(void)
3371 {
3372         unsigned long limit = nr_free_buffer_pages() / 16;
3373 
3374         limit = max(limit, 128UL);
3375         sysctl_tcp_mem[0] = limit / 4 * 3;              /* 4.68 % */
3376         sysctl_tcp_mem[1] = limit;                      /* 6.25 % */
3377         sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;      /* 9.37 % */
3378 }
3379 
3380 void __init tcp_init(void)
3381 {
3382         int max_rshare, max_wshare, cnt;
3383         unsigned long limit;
3384         unsigned int i;
3385 
3386         BUILD_BUG_ON(sizeof(struct tcp_skb_cb) >
3387                      FIELD_SIZEOF(struct sk_buff, cb));
3388 
3389         percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
3390         percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
3391         inet_hashinfo_init(&tcp_hashinfo);
3392         tcp_hashinfo.bind_bucket_cachep =
3393                 kmem_cache_create("tcp_bind_bucket",
3394                                   sizeof(struct inet_bind_bucket), 0,
3395                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3396 
3397         /* Size and allocate the main established and bind bucket
3398          * hash tables.
3399          *
3400          * The methodology is similar to that of the buffer cache.
3401          */
3402         tcp_hashinfo.ehash =
3403                 alloc_large_system_hash("TCP established",
3404                                         sizeof(struct inet_ehash_bucket),
3405                                         thash_entries,
3406                                         17, /* one slot per 128 KB of memory */
3407                                         0,
3408                                         NULL,
3409                                         &tcp_hashinfo.ehash_mask,
3410                                         0,
3411                                         thash_entries ? 0 : 512 * 1024);
3412         for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
3413                 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3414 
3415         if (inet_ehash_locks_alloc(&tcp_hashinfo))
3416                 panic("TCP: failed to alloc ehash_locks");
3417         tcp_hashinfo.bhash =
3418                 alloc_large_system_hash("TCP bind",
3419                                         sizeof(struct inet_bind_hashbucket),
3420                                         tcp_hashinfo.ehash_mask + 1,
3421                                         17, /* one slot per 128 KB of memory */
3422                                         0,
3423                                         &tcp_hashinfo.bhash_size,
3424                                         NULL,
3425                                         0,
3426                                         64 * 1024);
3427         tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
3428         for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3429                 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3430                 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
3431         }
3432 
3433 
3434         cnt = tcp_hashinfo.ehash_mask + 1;
3435         sysctl_tcp_max_orphans = cnt / 2;
3436 
3437         tcp_init_mem();
3438         /* Set per-socket limits to no more than 1/128 the pressure threshold */
3439         limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
3440         max_wshare = min(4UL*1024*1024, limit);
3441         max_rshare = min(6UL*1024*1024, limit);
3442 
3443         sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
3444         sysctl_tcp_wmem[1] = 16*1024;
3445         sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
3446 
3447         sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
3448         sysctl_tcp_rmem[1] = 87380;
3449         sysctl_tcp_rmem[2] = max(87380, max_rshare);
3450 
3451         pr_info("Hash tables configured (established %u bind %u)\n",
3452                 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
3453 
3454         tcp_v4_init();
3455         tcp_metrics_init();
3456         BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
3457         tcp_tasklet_init();
3458 }
3459 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp