blob: 6e5617b9f9db6f3c2aeacc7923ea763660761d99 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * 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
Joe Perchesafd465032012-03-12 07:03:32 +0000248#define pr_fmt(fmt) "TCP: " fmt
249
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700250#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251#include <linux/module.h>
252#include <linux/types.h>
253#include <linux/fcntl.h>
254#include <linux/poll.h>
255#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256#include <linux/fs.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800257#include <linux/skbuff.h>
Andrew Morton81b23b42008-07-03 03:22:02 -0700258#include <linux/scatterlist.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800259#include <linux/splice.h>
260#include <linux/net.h>
261#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262#include <linux/random.h>
263#include <linux/bootmem.h>
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -0700264#include <linux/highmem.h>
265#include <linux/swap.h>
David S. Millerb8059ea2006-03-25 01:36:56 -0800266#include <linux/cache.h>
Herbert Xuf4c50d92006-06-22 03:02:40 -0700267#include <linux/err.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800268#include <linux/crypto.h>
William Allen Simpsonda5c78c2009-12-02 18:12:09 +0000269#include <linux/time.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +0900270#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
272#include <net/icmp.h>
Yuchung Chengcf60af02012-07-19 06:43:09 +0000273#include <net/inet_common.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274#include <net/tcp.h>
275#include <net/xfrm.h>
276#include <net/ip.h>
Chris Leech1a2449a2006-05-23 18:05:53 -0700277#include <net/netdma.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800278#include <net/sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280#include <asm/uaccess.h>
281#include <asm/ioctls.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +0300282#include <net/busy_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
Brian Haleyab32ea52006-09-22 14:15:41 -0700284int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700286int sysctl_tcp_min_tso_segs __read_mostly = 2;
287
Eric Dumazetdd24c002008-11-25 21:17:14 -0800288struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700289EXPORT_SYMBOL_GPL(tcp_orphan_count);
290
David S. Millerb8059ea2006-03-25 01:36:56 -0800291int sysctl_tcp_wmem[3] __read_mostly;
292int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294EXPORT_SYMBOL(sysctl_tcp_rmem);
295EXPORT_SYMBOL(sysctl_tcp_wmem);
296
Eric Dumazet8d987e52010-11-09 23:24:26 +0000297atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800299
300/*
301 * Current number of TCP sockets.
302 */
303struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304EXPORT_SYMBOL(tcp_sockets_allocated);
305
306/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800307 * TCP splice context
308 */
309struct tcp_splice_state {
310 struct pipe_inode_info *pipe;
311 size_t len;
312 unsigned int flags;
313};
314
315/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Pressure flag: try to collapse.
317 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800318 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * is strict, actions are advisory and have some latency.
320 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700321int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322EXPORT_SYMBOL(tcp_memory_pressure);
323
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700324void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700327 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 tcp_memory_pressure = 1;
329 }
330}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331EXPORT_SYMBOL(tcp_enter_memory_pressure);
332
Julian Anastasovb103cf32009-10-19 10:10:40 +0000333/* Convert seconds to retransmits based on initial and max timeout */
334static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
335{
336 u8 res = 0;
337
338 if (seconds > 0) {
339 int period = timeout;
340
341 res = 1;
342 while (seconds > period && res < 255) {
343 res++;
344 timeout <<= 1;
345 if (timeout > rto_max)
346 timeout = rto_max;
347 period += timeout;
348 }
349 }
350 return res;
351}
352
353/* Convert retransmits to seconds based on initial and max timeout */
354static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
355{
356 int period = 0;
357
358 if (retrans > 0) {
359 period = timeout;
360 while (--retrans) {
361 timeout <<= 1;
362 if (timeout > rto_max)
363 timeout = rto_max;
364 period += timeout;
365 }
366 }
367 return period;
368}
369
Neal Cardwell900f65d2012-04-19 09:55:21 +0000370/* Address-family independent initialization for a tcp_sock.
371 *
372 * NOTE: A lot of things set to zero explicitly by call to
373 * sk_alloc() so need not be done here.
374 */
375void tcp_init_sock(struct sock *sk)
376{
377 struct inet_connection_sock *icsk = inet_csk(sk);
378 struct tcp_sock *tp = tcp_sk(sk);
379
380 skb_queue_head_init(&tp->out_of_order_queue);
381 tcp_init_xmit_timers(sk);
382 tcp_prequeue_init(tp);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000383 INIT_LIST_HEAD(&tp->tsq_node);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000384
385 icsk->icsk_rto = TCP_TIMEOUT_INIT;
386 tp->mdev = TCP_TIMEOUT_INIT;
387
388 /* So many TCP implementations out there (incorrectly) count the
389 * initial SYN frame in their delayed-ACK and congestion control
390 * algorithms that we must have the following bandaid to talk
391 * efficiently to them. -DaveM
392 */
393 tp->snd_cwnd = TCP_INIT_CWND;
394
395 /* See draft-stevens-tcpca-spec-01 for discussion of the
396 * initialization of these values.
397 */
398 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
399 tp->snd_cwnd_clamp = ~0;
400 tp->mss_cache = TCP_MSS_DEFAULT;
401
402 tp->reordering = sysctl_tcp_reordering;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000403 tcp_enable_early_retrans(tp);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000404 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
405
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000406 tp->tsoffset = 0;
407
Neal Cardwell900f65d2012-04-19 09:55:21 +0000408 sk->sk_state = TCP_CLOSE;
409
410 sk->sk_write_space = sk_stream_write_space;
411 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
412
413 icsk->icsk_sync_mss = tcp_sync_mss;
414
Neal Cardwell900f65d2012-04-19 09:55:21 +0000415 sk->sk_sndbuf = sysctl_tcp_wmem[1];
416 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
417
418 local_bh_disable();
419 sock_update_memcg(sk);
420 sk_sockets_allocated_inc(sk);
421 local_bh_enable();
422}
423EXPORT_SYMBOL(tcp_init_sock);
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 * Wait for a TCP event.
427 *
428 * Note that we don't need to lock the socket, as the upper poll layers
429 * take care of normal races (between the test and the event) and we don't
430 * go look at any of the socket buffers directly.
431 */
432unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
433{
434 unsigned int mask;
435 struct sock *sk = sock->sk;
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400436 const struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
David Majnemerc3f1dba2013-05-31 13:15:38 +0000438 sock_rps_record_flow(sk);
439
Eric Dumazetaa395142010-04-20 13:03:51 +0000440 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700442 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700445 * by poll logic and correct handling of state changes
446 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 */
448
449 mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 /*
452 * POLLHUP is certainly not done right. But poll() doesn't
453 * have a notion of HUP in just one direction, and for a
454 * socket the read side is more interesting.
455 *
456 * Some poll() documentation says that POLLHUP is incompatible
457 * with the POLLOUT/POLLWR flags, so somebody should check this
458 * all. But careful, it tends to be safer to return too many
459 * bits than too few, and you can easily break real applications
460 * if you don't tell them that something has hung up!
461 *
462 * Check-me.
463 *
464 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
465 * our fs/select.c). It means that after we received EOF,
466 * poll always returns immediately, making impossible poll() on write()
467 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
468 * if and only if shutdown has been made in both directions.
469 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700470 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 * then we could set it on SND_SHUTDOWN. BTW examples given
472 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700473 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 *
475 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
476 * blocking on fresh not-connected or disconnected socket. --ANK
477 */
478 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
479 mask |= POLLHUP;
480 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800481 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Jerry Chu83368862012-08-31 12:29:12 +0000483 /* Connected or passive Fast Open socket? */
484 if (sk->sk_state != TCP_SYN_SENT &&
485 (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) {
David S. Millerc7004482008-10-06 10:43:54 -0700486 int target = sock_rcvlowat(sk, 0, INT_MAX);
487
488 if (tp->urg_seq == tp->copied_seq &&
489 !sock_flag(sk, SOCK_URGINLINE) &&
490 tp->urg_data)
Alexandra Kossovskyb634f872010-03-18 20:29:24 -0700491 target++;
David S. Millerc7004482008-10-06 10:43:54 -0700492
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 /* Potential race condition. If read of tp below will
494 * escape above sk->sk_state, we can be illegally awaken
495 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700496 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 mask |= POLLIN | POLLRDNORM;
498
499 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700500 if (sk_stream_is_writeable(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 mask |= POLLOUT | POLLWRNORM;
502 } else { /* send SIGIO later */
503 set_bit(SOCK_ASYNC_NOSPACE,
504 &sk->sk_socket->flags);
505 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
506
507 /* Race breaker. If space is freed after
508 * wspace test but before the flags are set,
509 * IO signal will be lost.
510 */
Eric Dumazet64dc6132013-07-22 20:26:31 -0700511 if (sk_stream_is_writeable(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 mask |= POLLOUT | POLLWRNORM;
513 }
KOSAKI Motohirod84ba632010-08-24 16:05:48 +0000514 } else
515 mask |= POLLOUT | POLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
517 if (tp->urg_data & TCP_URG_VALID)
518 mask |= POLLPRI;
519 }
Tom Marshalla4d25802010-09-20 15:42:05 -0700520 /* This barrier is coupled with smp_wmb() in tcp_reset() */
521 smp_rmb();
522 if (sk->sk_err)
523 mask |= POLLERR;
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return mask;
526}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000527EXPORT_SYMBOL(tcp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
529int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
530{
531 struct tcp_sock *tp = tcp_sk(sk);
532 int answ;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000533 bool slow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 switch (cmd) {
536 case SIOCINQ:
537 if (sk->sk_state == TCP_LISTEN)
538 return -EINVAL;
539
Eric Dumazet0e71c552012-10-21 20:06:56 +0000540 slow = lock_sock_fast(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
542 answ = 0;
543 else if (sock_flag(sk, SOCK_URGINLINE) ||
544 !tp->urg_data ||
545 before(tp->urg_seq, tp->copied_seq) ||
546 !before(tp->urg_seq, tp->rcv_nxt)) {
David S. Miller91521942009-05-28 21:35:47 -0700547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 answ = tp->rcv_nxt - tp->copied_seq;
549
Eric Dumazeta3374c42012-10-18 09:14:12 +0000550 /* Subtract 1, if FIN was received */
551 if (answ && sock_flag(sk, SOCK_DONE))
552 answ--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 } else
554 answ = tp->urg_seq - tp->copied_seq;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000555 unlock_sock_fast(sk, slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 break;
557 case SIOCATMARK:
558 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
559 break;
560 case SIOCOUTQ:
561 if (sk->sk_state == TCP_LISTEN)
562 return -EINVAL;
563
564 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
565 answ = 0;
566 else
567 answ = tp->write_seq - tp->snd_una;
568 break;
Mario Schuknecht2f4e1b32011-03-09 14:08:09 -0800569 case SIOCOUTQNSD:
570 if (sk->sk_state == TCP_LISTEN)
571 return -EINVAL;
572
573 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
574 answ = 0;
575 else
576 answ = tp->write_seq - tp->snd_nxt;
577 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 default:
579 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582 return put_user(answ, (int __user *)arg);
583}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000584EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
587{
Eric Dumazet4de075e2011-09-27 13:25:05 -0400588 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 tp->pushed_seq = tp->write_seq;
590}
591
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000592static inline bool forced_push(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
594 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
595}
596
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700597static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700599 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200600 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
601
602 skb->csum = 0;
603 tcb->seq = tcb->end_seq = tp->write_seq;
Eric Dumazet4de075e2011-09-27 13:25:05 -0400604 tcb->tcp_flags = TCPHDR_ACK;
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200605 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800607 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800608 sk->sk_wmem_queued += skb->truesize;
609 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700610 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900611 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
Krishna Kumarafeca342009-12-10 07:16:52 +0000614static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700616 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700620static inline void tcp_push(struct sock *sk, int flags, int mss_now,
621 int nonagle)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
David S. Millerfe067e82007-03-07 12:12:44 -0800623 if (tcp_send_head(sk)) {
Krishna Kumarafeca342009-12-10 07:16:52 +0000624 struct tcp_sock *tp = tcp_sk(sk);
625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (!(flags & MSG_MORE) || forced_push(tp))
Krishna Kumarafeca342009-12-10 07:16:52 +0000627 tcp_mark_push(tp, tcp_write_queue_tail(sk));
628
629 tcp_mark_urg(tp, flags);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700630 __tcp_push_pending_frames(sk, mss_now,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
632 }
633}
634
Adrian Bunk6ff77512007-11-06 23:32:26 -0800635static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
636 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800637{
638 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd2009-01-13 16:04:36 -0800639 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800640
Dimitris Michailidis9fa5fdf2009-01-26 22:15:31 -0800641 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
642 tss->flags);
Willy Tarreau33966dd2009-01-13 16:04:36 -0800643 if (ret > 0)
644 rd_desc->count -= ret;
645 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800646}
647
648static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
649{
650 /* Store TCP splice context information in read_descriptor_t. */
651 read_descriptor_t rd_desc = {
652 .arg.data = tss,
Willy Tarreau33966dd2009-01-13 16:04:36 -0800653 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800654 };
655
656 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
657}
658
659/**
660 * tcp_splice_read - splice data from TCP socket to a pipe
661 * @sock: socket to splice from
662 * @ppos: position (not valid)
663 * @pipe: pipe to splice to
664 * @len: number of bytes to splice
665 * @flags: splice modifier flags
666 *
667 * Description:
668 * Will read pages from given socket and fill them into a pipe.
669 *
670 **/
671ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
672 struct pipe_inode_info *pipe, size_t len,
673 unsigned int flags)
674{
675 struct sock *sk = sock->sk;
676 struct tcp_splice_state tss = {
677 .pipe = pipe,
678 .len = len,
679 .flags = flags,
680 };
681 long timeo;
682 ssize_t spliced;
683 int ret;
684
Changli Gao3a047bf2010-07-12 21:00:12 +0000685 sock_rps_record_flow(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -0800686 /*
687 * We can't seek on a socket input
688 */
689 if (unlikely(*ppos))
690 return -ESPIPE;
691
692 ret = spliced = 0;
693
694 lock_sock(sk);
695
Eric Dumazet42324c62009-10-01 15:26:00 -0700696 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
Jens Axboe9c55e012007-11-06 23:30:13 -0800697 while (tss.len) {
698 ret = __tcp_splice_read(sk, &tss);
699 if (ret < 0)
700 break;
701 else if (!ret) {
702 if (spliced)
703 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800704 if (sock_flag(sk, SOCK_DONE))
705 break;
706 if (sk->sk_err) {
707 ret = sock_error(sk);
708 break;
709 }
710 if (sk->sk_shutdown & RCV_SHUTDOWN)
711 break;
712 if (sk->sk_state == TCP_CLOSE) {
713 /*
714 * This occurs when user tries to read
715 * from never connected socket.
716 */
717 if (!sock_flag(sk, SOCK_DONE))
718 ret = -ENOTCONN;
719 break;
720 }
721 if (!timeo) {
722 ret = -EAGAIN;
723 break;
724 }
725 sk_wait_data(sk, &timeo);
726 if (signal_pending(current)) {
727 ret = sock_intr_errno(timeo);
728 break;
729 }
730 continue;
731 }
732 tss.len -= ret;
733 spliced += ret;
734
Willy Tarreau33966dd2009-01-13 16:04:36 -0800735 if (!timeo)
736 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800737 release_sock(sk);
738 lock_sock(sk);
739
740 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd2009-01-13 16:04:36 -0800741 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800742 signal_pending(current))
743 break;
744 }
745
746 release_sock(sk);
747
748 if (spliced)
749 return spliced;
750
751 return ret;
752}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000753EXPORT_SYMBOL(tcp_splice_read);
Jens Axboe9c55e012007-11-06 23:30:13 -0800754
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100755struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100756{
757 struct sk_buff *skb;
758
759 /* The TCP header must be at least 32-bit aligned. */
760 size = ALIGN(size, 4);
761
762 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
763 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800764 if (sk_wmem_schedule(sk, skb->truesize)) {
Eric Dumazeta21d4572012-04-10 20:30:48 +0000765 skb_reserve(skb, sk->sk_prot->max_header);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100766 /*
767 * Make sure that we have exactly size bytes
768 * available to the caller, no more, no less.
769 */
Eric Dumazet16fad692013-03-14 05:40:32 +0000770 skb->reserved_tailroom = skb->end - skb->tail - size;
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100771 return skb;
772 }
773 __kfree_skb(skb);
774 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700775 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100776 sk_stream_moderate_sndbuf(sk);
777 }
778 return NULL;
779}
780
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000781static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
782 int large_allowed)
783{
784 struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000785 u32 xmit_size_goal, old_size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000786
787 xmit_size_goal = mss_now;
788
789 if (large_allowed && sk_can_gso(sk)) {
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700790 u32 gso_size, hlen;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000791
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700792 /* Maybe we should/could use sk->sk_prot->max_header here ? */
793 hlen = inet_csk(sk)->icsk_af_ops->net_header_len +
794 inet_csk(sk)->icsk_ext_hdr_len +
795 tp->tcp_header_len;
796
797 /* Goal is to send at least one packet per ms,
798 * not one big TSO packet every 100 ms.
799 * This preserves ACK clocking and is consistent
800 * with tcp_tso_should_defer() heuristic.
801 */
802 gso_size = sk->sk_pacing_rate / (2 * MSEC_PER_SEC);
803 gso_size = max_t(u32, gso_size,
804 sysctl_tcp_min_tso_segs * mss_now);
805
806 xmit_size_goal = min_t(u32, gso_size,
807 sk->sk_gso_max_size - 1 - hlen);
808
809 /* TSQ : try to have at least two segments in flight
810 * (one in NIC TX ring, another in Qdisc)
811 */
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000812 xmit_size_goal = min_t(u32, xmit_size_goal,
813 sysctl_tcp_limit_output_bytes >> 1);
814
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000815 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000816
817 /* We try hard to avoid divides here */
818 old_size_goal = tp->xmit_size_goal_segs * mss_now;
819
820 if (likely(old_size_goal <= xmit_size_goal &&
821 old_size_goal + mss_now > xmit_size_goal)) {
822 xmit_size_goal = old_size_goal;
823 } else {
Ben Hutchings14853482012-07-30 16:11:42 +0000824 tp->xmit_size_goal_segs =
825 min_t(u16, xmit_size_goal / mss_now,
826 sk->sk_gso_max_segs);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000827 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
828 }
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000829 }
830
Ilpo Järvinenafece1c2009-03-14 14:23:07 +0000831 return max(xmit_size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000832}
833
834static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
835{
836 int mss_now;
837
838 mss_now = tcp_current_mss(sk);
839 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
840
841 return mss_now;
842}
843
Eric Dumazet64022d02012-12-01 13:07:02 +0000844static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
845 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
847 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700848 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 int err;
850 ssize_t copied;
851 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
852
Jerry Chu83368862012-08-31 12:29:12 +0000853 /* Wait for a connection to finish. One exception is TCP Fast Open
854 * (passive side) where data is allowed to be sent before a connection
855 * is fully established.
856 */
857 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
858 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
860 goto out_err;
Jerry Chu83368862012-08-31 12:29:12 +0000861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
864
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000865 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 copied = 0;
867
868 err = -EPIPE;
869 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000870 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Eric Dumazet64022d02012-12-01 13:07:02 +0000872 while (size > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800873 struct sk_buff *skb = tcp_write_queue_tail(sk);
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000874 int copy, i;
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000875 bool can_coalesce;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
David S. Millerfe067e82007-03-07 12:12:44 -0800877 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878new_segment:
879 if (!sk_stream_memory_free(sk))
880 goto wait_for_sndbuf;
881
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100882 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 if (!skb)
884 goto wait_for_memory;
885
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700886 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700887 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889
890 if (copy > size)
891 copy = size;
892
893 i = skb_shinfo(skb)->nr_frags;
894 can_coalesce = skb_can_coalesce(skb, i, page, offset);
895 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
896 tcp_mark_push(tp, skb);
897 goto new_segment;
898 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800899 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (can_coalesce) {
Eric Dumazet9e903e02011-10-18 21:00:24 +0000903 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 } else {
905 get_page(page);
906 skb_fill_page_desc(skb, i, page, offset, copy);
907 }
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000908 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +0000909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 skb->len += copy;
911 skb->data_len += copy;
912 skb->truesize += copy;
913 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800914 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700915 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 tp->write_seq += copy;
917 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700918 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -0400921 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 copied += copy;
Eric Dumazet64022d02012-12-01 13:07:02 +0000924 offset += copy;
925 if (!(size -= copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 goto out;
927
Herbert Xu69d15062008-03-22 15:47:05 -0700928 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 continue;
930
931 if (forced_push(tp)) {
932 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700933 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800934 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 tcp_push_one(sk, mss_now);
936 continue;
937
938wait_for_sndbuf:
939 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
940wait_for_memory:
Willy Tarreaubad115c2012-05-17 11:14:14 +0000941 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
944 goto do_error;
945
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000946 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 }
948
949out:
Eric Dumazet35f9c092012-04-05 03:05:35 +0000950 if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700951 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 return copied;
953
954do_error:
955 if (copied)
956 goto out;
957out_err:
958 return sk_stream_error(sk, flags, err);
959}
960
Changli Gao7ba42912010-07-10 20:41:55 +0000961int tcp_sendpage(struct sock *sk, struct page *page, int offset,
962 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
964 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -0700967 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Changli Gao7ba42912010-07-10 20:41:55 +0000968 return sock_no_sendpage(sk->sk_socket, page, offset, size,
969 flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 lock_sock(sk);
Eric Dumazet64022d02012-12-01 13:07:02 +0000972 res = do_tcp_sendpages(sk, page, offset, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 release_sock(sk);
974 return res;
975}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000976EXPORT_SYMBOL(tcp_sendpage);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Eric Dumazet690e99c2011-11-28 00:27:47 +0000978static inline int select_size(const struct sock *sk, bool sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400980 const struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700981 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Krishna Kumardef87cf2009-12-10 07:16:59 +0000983 if (sg) {
Eric Dumazetf07d9602011-11-28 22:41:47 +0000984 if (sk_can_gso(sk)) {
985 /* Small frames wont use a full page:
986 * Payload will immediately follow tcp header.
987 */
988 tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
989 } else {
David S. Millerb4e26f52005-07-05 15:20:27 -0700990 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
991
992 if (tmp >= pgbreak &&
993 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
994 tmp = pgbreak;
995 }
996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 return tmp;
999}
1000
Yuchung Chengcf60af02012-07-19 06:43:09 +00001001void tcp_free_fastopen_req(struct tcp_sock *tp)
1002{
1003 if (tp->fastopen_req != NULL) {
1004 kfree(tp->fastopen_req);
1005 tp->fastopen_req = NULL;
1006 }
1007}
1008
1009static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size)
1010{
1011 struct tcp_sock *tp = tcp_sk(sk);
1012 int err, flags;
1013
1014 if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
1015 return -EOPNOTSUPP;
1016 if (tp->fastopen_req != NULL)
1017 return -EALREADY; /* Another Fast Open is in progress */
1018
1019 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1020 sk->sk_allocation);
1021 if (unlikely(tp->fastopen_req == NULL))
1022 return -ENOBUFS;
1023 tp->fastopen_req->data = msg;
1024
1025 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1026 err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1027 msg->msg_namelen, flags);
1028 *size = tp->fastopen_req->copied;
1029 tcp_free_fastopen_req(tp);
1030 return err;
1031}
1032
Changli Gao7ba42912010-07-10 20:41:55 +00001033int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 size_t size)
1035{
1036 struct iovec *iov;
1037 struct tcp_sock *tp = tcp_sk(sk);
1038 struct sk_buff *skb;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001039 int iovlen, flags, err, copied = 0;
1040 int mss_now = 0, size_goal, copied_syn = 0, offset = 0;
Eric Dumazet690e99c2011-11-28 00:27:47 +00001041 bool sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 long timeo;
1043
1044 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 flags = msg->msg_flags;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001047 if (flags & MSG_FASTOPEN) {
1048 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn);
1049 if (err == -EINPROGRESS && copied_syn > 0)
1050 goto out;
1051 else if (err)
1052 goto out_err;
1053 offset = copied_syn;
1054 }
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1057
Jerry Chu83368862012-08-31 12:29:12 +00001058 /* Wait for a connection to finish. One exception is TCP Fast Open
1059 * (passive side) where data is allowed to be sent before a connection
1060 * is fully established.
1061 */
1062 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1063 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001065 goto do_error;
Jerry Chu83368862012-08-31 12:29:12 +00001066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001068 if (unlikely(tp->repair)) {
1069 if (tp->repair_queue == TCP_RECV_QUEUE) {
1070 copied = tcp_send_rcvq(sk, msg, size);
1071 goto out;
1072 }
1073
1074 err = -EINVAL;
1075 if (tp->repair_queue == TCP_NO_QUEUE)
1076 goto out_err;
1077
1078 /* 'common' sending to sendq */
1079 }
1080
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /* This should be in poll */
1082 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1083
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001084 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 /* Ok commence sending. */
1087 iovlen = msg->msg_iovlen;
1088 iov = msg->msg_iov;
1089 copied = 0;
1090
1091 err = -EPIPE;
1092 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +00001093 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Eric Dumazet690e99c2011-11-28 00:27:47 +00001095 sg = !!(sk->sk_route_caps & NETIF_F_SG);
Krishna Kumardef87cf2009-12-10 07:16:59 +00001096
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 while (--iovlen >= 0) {
David S. Miller01db4032010-09-27 20:24:54 -07001098 size_t seglen = iov->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 unsigned char __user *from = iov->iov_base;
1100
1101 iov++;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001102 if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */
1103 if (offset >= seglen) {
1104 offset -= seglen;
1105 continue;
1106 }
1107 seglen -= offset;
1108 from += offset;
1109 offset = 0;
1110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 while (seglen > 0) {
Herbert Xu6828b922009-06-28 18:06:41 +00001113 int copy = 0;
1114 int max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
David S. Millerfe067e82007-03-07 12:12:44 -08001116 skb = tcp_write_queue_tail(sk);
Herbert Xu6828b922009-06-28 18:06:41 +00001117 if (tcp_send_head(sk)) {
1118 if (skb->ip_summed == CHECKSUM_NONE)
1119 max = mss_now;
1120 copy = max - skb->len;
1121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Herbert Xu6828b922009-06-28 18:06:41 +00001123 if (copy <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124new_segment:
1125 /* Allocate new segment. If the interface is SG,
1126 * allocate skb fitting to single page.
1127 */
1128 if (!sk_stream_memory_free(sk))
1129 goto wait_for_sndbuf;
1130
Krishna Kumardef87cf2009-12-10 07:16:59 +00001131 skb = sk_stream_alloc_skb(sk,
1132 select_size(sk, sg),
1133 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (!skb)
1135 goto wait_for_memory;
1136
1137 /*
Andrey Vagin7ed5c5a2013-08-16 19:04:36 +04001138 * All packets are restored as if they have
1139 * already been sent.
1140 */
1141 if (tp->repair)
1142 TCP_SKB_CB(skb)->when = tcp_time_stamp;
1143
1144 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 * Check whether we can use HW checksum.
1146 */
Herbert Xu8648b302006-06-17 22:06:05 -07001147 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -07001148 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001150 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001151 copy = size_goal;
Herbert Xu6828b922009-06-28 18:06:41 +00001152 max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
1154
1155 /* Try to append data to the end of skb. */
1156 if (copy > seglen)
1157 copy = seglen;
1158
1159 /* Where to copy to? */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001160 if (skb_availroom(skb) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 /* We have some space in skb head. Superb! */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001162 copy = min_t(int, copy, skb_availroom(skb));
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001163 err = skb_add_data_nocache(sk, skb, from, copy);
1164 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 goto do_fault;
1166 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001167 bool merge = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 int i = skb_shinfo(skb)->nr_frags;
Eric Dumazet5640f762012-09-23 23:04:42 +00001169 struct page_frag *pfrag = sk_page_frag(sk);
Eric Dumazet761965e2011-12-04 07:05:17 +00001170
Eric Dumazet5640f762012-09-23 23:04:42 +00001171 if (!sk_page_frag_refill(sk, pfrag))
1172 goto wait_for_memory;
Eric Dumazet761965e2011-12-04 07:05:17 +00001173
Eric Dumazet5640f762012-09-23 23:04:42 +00001174 if (!skb_can_coalesce(skb, i, pfrag->page,
1175 pfrag->offset)) {
1176 if (i == MAX_SKB_FRAGS || !sg) {
1177 tcp_mark_push(tp, skb);
1178 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001180 merge = false;
1181 }
Herbert Xuef015782005-09-01 17:48:59 -07001182
Eric Dumazet5640f762012-09-23 23:04:42 +00001183 copy = min_t(int, copy, pfrag->size - pfrag->offset);
Herbert Xuef015782005-09-01 17:48:59 -07001184
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001185 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -07001186 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001188 err = skb_copy_to_page_nocache(sk, from, skb,
Eric Dumazet5640f762012-09-23 23:04:42 +00001189 pfrag->page,
1190 pfrag->offset,
1191 copy);
1192 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 goto do_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 /* Update the skb. */
1196 if (merge) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001197 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001199 skb_fill_page_desc(skb, i, pfrag->page,
1200 pfrag->offset, copy);
1201 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001203 pfrag->offset += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205
1206 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -04001207 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
1209 tp->write_seq += copy;
1210 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -07001211 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 from += copy;
1214 copied += copy;
1215 if ((seglen -= copy) == 0 && iovlen == 0)
1216 goto out;
1217
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001218 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 continue;
1220
1221 if (forced_push(tp)) {
1222 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001223 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001224 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 tcp_push_one(sk, mss_now);
1226 continue;
1227
1228wait_for_sndbuf:
1229 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1230wait_for_memory:
Andrew Vaginec342322012-11-15 04:03:17 +00001231 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001232 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1235 goto do_error;
1236
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001237 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 }
1239 }
1240
1241out:
Andrew Vaginec342322012-11-15 04:03:17 +00001242 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001243 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 release_sock(sk);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001245 return copied + copied_syn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247do_fault:
1248 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001249 tcp_unlink_write_queue(skb, sk);
1250 /* It is the one place in all of TCP, except connection
1251 * reset, where we can be unlinking the send_head.
1252 */
1253 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001254 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256
1257do_error:
Yuchung Chengcf60af02012-07-19 06:43:09 +00001258 if (copied + copied_syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 goto out;
1260out_err:
1261 err = sk_stream_error(sk, flags, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 release_sock(sk);
1263 return err;
1264}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001265EXPORT_SYMBOL(tcp_sendmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267/*
1268 * Handle reading urgent data. BSD has very simple semantics for
1269 * this, no blocking and very strange errors 8)
1270 */
1271
Rami Rosen377f0a02009-03-31 14:43:17 -07001272static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
1274 struct tcp_sock *tp = tcp_sk(sk);
1275
1276 /* No URG data to read. */
1277 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1278 tp->urg_data == TCP_URG_READ)
1279 return -EINVAL; /* Yes this is right ! */
1280
1281 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1282 return -ENOTCONN;
1283
1284 if (tp->urg_data & TCP_URG_VALID) {
1285 int err = 0;
1286 char c = tp->urg_data;
1287
1288 if (!(flags & MSG_PEEK))
1289 tp->urg_data = TCP_URG_READ;
1290
1291 /* Read urgent data. */
1292 msg->msg_flags |= MSG_OOB;
1293
1294 if (len > 0) {
1295 if (!(flags & MSG_TRUNC))
1296 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1297 len = 1;
1298 } else
1299 msg->msg_flags |= MSG_TRUNC;
1300
1301 return err ? -EFAULT : len;
1302 }
1303
1304 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1305 return 0;
1306
1307 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1308 * the available implementations agree in this case:
1309 * this call should never block, independent of the
1310 * blocking state of the socket.
1311 * Mike <pall@rz.uni-karlsruhe.de>
1312 */
1313 return -EAGAIN;
1314}
1315
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001316static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1317{
1318 struct sk_buff *skb;
1319 int copied = 0, err = 0;
1320
1321 /* XXX -- need to support SO_PEEK_OFF */
1322
1323 skb_queue_walk(&sk->sk_write_queue, skb) {
1324 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len);
1325 if (err)
1326 break;
1327
1328 copied += skb->len;
1329 }
1330
1331 return err ?: copied;
1332}
1333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334/* Clean up the receive buffer for full frames taken by the user,
1335 * then send an ACK if necessary. COPIED is the number of bytes
1336 * tcp_recvmsg has given to the user so far, it speeds up the
1337 * calculation of whether or not we must ACK for the sake of
1338 * a window update.
1339 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001340void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001343 bool time_to_ack = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1346
Ilpo Järvinend792c102009-11-13 13:56:33 -08001347 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001348 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
Ilpo Järvinend792c102009-11-13 13:56:33 -08001349 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001351 if (inet_csk_ack_scheduled(sk)) {
1352 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 /* Delayed ACKs frequently hit locked sockets during bulk
1354 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001355 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001357 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 /*
1359 * If this read emptied read buffer, we send ACK, if
1360 * connection is not bidirectional, user drained
1361 * receive buffer and there was a small segment
1362 * in queue.
1363 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001364 (copied > 0 &&
1365 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1366 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1367 !icsk->icsk_ack.pingpong)) &&
1368 !atomic_read(&sk->sk_rmem_alloc)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001369 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371
1372 /* We send an ACK if we can now advertise a non-zero window
1373 * which has been raised "significantly".
1374 *
1375 * Even if window raised up to infinity, do not send window open ACK
1376 * in states, where we will not receive more. It is useless.
1377 */
1378 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1379 __u32 rcv_window_now = tcp_receive_window(tp);
1380
1381 /* Optimize, __tcp_select_window() is not cheap. */
1382 if (2*rcv_window_now <= tp->window_clamp) {
1383 __u32 new_window = __tcp_select_window(sk);
1384
1385 /* Send ACK now, if this read freed lots of space
1386 * in our buffer. Certainly, new_window is new window.
1387 * We can advertise it now, if it is not less than current one.
1388 * "Lots" means "at least twice" here.
1389 */
1390 if (new_window && new_window >= 2 * rcv_window_now)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001391 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 }
1393 }
1394 if (time_to_ack)
1395 tcp_send_ack(sk);
1396}
1397
1398static void tcp_prequeue_process(struct sock *sk)
1399{
1400 struct sk_buff *skb;
1401 struct tcp_sock *tp = tcp_sk(sk);
1402
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001403 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 /* RX process wants to run with disabled BHs, though it is not
1406 * necessary */
1407 local_bh_disable();
1408 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001409 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 local_bh_enable();
1411
1412 /* Clear memory counter. */
1413 tp->ucopy.memory = 0;
1414}
1415
Steven J. Magnani73852e82010-03-16 05:22:44 +00001416#ifdef CONFIG_NET_DMA
1417static void tcp_service_net_dma(struct sock *sk, bool wait)
1418{
1419 dma_cookie_t done, used;
1420 dma_cookie_t last_issued;
1421 struct tcp_sock *tp = tcp_sk(sk);
1422
1423 if (!tp->ucopy.dma_chan)
1424 return;
1425
1426 last_issued = tp->ucopy.dma_cookie;
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001427 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001428
1429 do {
Bartlomiej Zolnierkiewicze239345f2012-11-08 10:01:01 +00001430 if (dma_async_is_tx_complete(tp->ucopy.dma_chan,
Steven J. Magnani73852e82010-03-16 05:22:44 +00001431 last_issued, &done,
1432 &used) == DMA_SUCCESS) {
1433 /* Safe to free early-copied skbs now */
1434 __skb_queue_purge(&sk->sk_async_wait_queue);
1435 break;
1436 } else {
1437 struct sk_buff *skb;
1438 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1439 (dma_async_is_complete(skb->dma_cookie, done,
1440 used) == DMA_SUCCESS)) {
1441 __skb_dequeue(&sk->sk_async_wait_queue);
1442 kfree_skb(skb);
1443 }
1444 }
1445 } while (wait);
1446}
1447#endif
1448
Eric Dumazetf26845b2013-01-09 20:59:09 +00001449static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 struct sk_buff *skb;
1452 u32 offset;
1453
Eric Dumazetf26845b2013-01-09 20:59:09 +00001454 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001456 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001458 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 *off = offset;
1460 return skb;
1461 }
Eric Dumazetf26845b2013-01-09 20:59:09 +00001462 /* This looks weird, but this can happen if TCP collapsing
1463 * splitted a fat GRO packet, while we released socket lock
1464 * in skb_splice_bits()
1465 */
1466 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
1468 return NULL;
1469}
1470
1471/*
1472 * This routine provides an alternative to tcp_recvmsg() for routines
1473 * that would like to handle copying from skbuffs directly in 'sendfile'
1474 * fashion.
1475 * Note:
1476 * - It is assumed that the socket was locked by the caller.
1477 * - The routine does not block.
1478 * - At present, there is no support for reading OOB data
1479 * or for 'peeking' the socket using this routine
1480 * (although both would be easy to implement).
1481 */
1482int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1483 sk_read_actor_t recv_actor)
1484{
1485 struct sk_buff *skb;
1486 struct tcp_sock *tp = tcp_sk(sk);
1487 u32 seq = tp->copied_seq;
1488 u32 offset;
1489 int copied = 0;
1490
1491 if (sk->sk_state == TCP_LISTEN)
1492 return -ENOTCONN;
1493 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1494 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001495 int used;
1496 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 len = skb->len - offset;
1499 /* Stop reading if we hit a patch of urgent data */
1500 if (tp->urg_data) {
1501 u32 urg_offset = tp->urg_seq - seq;
1502 if (urg_offset < len)
1503 len = urg_offset;
1504 if (!len)
1505 break;
1506 }
1507 used = recv_actor(desc, skb, offset, len);
Eric Dumazetff905b12013-01-10 07:06:10 +00001508 if (used <= 0) {
Jens Axboeddb61a52007-06-23 23:07:50 -07001509 if (!copied)
1510 copied = used;
1511 break;
1512 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 seq += used;
1514 copied += used;
1515 offset += used;
1516 }
Willy Tarreau02275a22012-12-02 11:49:27 +00001517 /* If recv_actor drops the lock (e.g. TCP splice
Octavian Purdila293ad602008-06-04 15:45:58 -07001518 * receive) the skb pointer might be invalid when
1519 * getting here: tcp_collapse might have deleted it
1520 * while aggregating skbs from the socket queue.
1521 */
Willy Tarreau02275a22012-12-02 11:49:27 +00001522 skb = tcp_recv_skb(sk, seq - 1, &offset);
1523 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 break;
Willy Tarreau02275a22012-12-02 11:49:27 +00001525 /* TCP coalescing might have appended data to the skb.
1526 * Try to splice more frags
1527 */
1528 if (offset + 1 != skb->len)
1529 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001531 if (tcp_hdr(skb)->fin) {
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001532 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 ++seq;
1534 break;
1535 }
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001536 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 if (!desc->count)
1538 break;
Steven J. Magnanibaff42a2010-03-30 13:56:01 -07001539 tp->copied_seq = seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 }
1541 tp->copied_seq = seq;
1542
1543 tcp_rcv_space_adjust(sk);
1544
1545 /* Clean up data we have read: This will do ACK frames. */
Eric Dumazetf26845b2013-01-09 20:59:09 +00001546 if (copied > 0) {
1547 tcp_recv_skb(sk, seq, &offset);
Chris Leech0e4b4992006-05-23 18:00:16 -07001548 tcp_cleanup_rbuf(sk, copied);
Eric Dumazetf26845b2013-01-09 20:59:09 +00001549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return copied;
1551}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001552EXPORT_SYMBOL(tcp_read_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554/*
1555 * This routine copies from a sock struct into the user buffer.
1556 *
1557 * Technical note: in 2.3 we work on _locked_ socket, so that
1558 * tricks with *seq access order and skb->users are not required.
1559 * Probably, code can be easily improved even more.
1560 */
1561
1562int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1563 size_t len, int nonblock, int flags, int *addr_len)
1564{
1565 struct tcp_sock *tp = tcp_sk(sk);
1566 int copied = 0;
1567 u32 peek_seq;
1568 u32 *seq;
1569 unsigned long used;
1570 int err;
1571 int target; /* Read at least this many bytes */
1572 long timeo;
1573 struct task_struct *user_recv = NULL;
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001574 bool copied_early = false;
Chris Leech2b1244a2007-03-08 09:57:36 -08001575 struct sk_buff *skb;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001576 u32 urg_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Eliezer Tamircbf55002013-07-08 16:20:34 +03001578 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1579 (sk->sk_state == TCP_ESTABLISHED))
1580 sk_busy_loop(sk, nonblock);
Eliezer Tamird30e3832013-06-10 11:40:10 +03001581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 lock_sock(sk);
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 err = -ENOTCONN;
1585 if (sk->sk_state == TCP_LISTEN)
1586 goto out;
1587
1588 timeo = sock_rcvtimeo(sk, nonblock);
1589
1590 /* Urgent data needs to be handled specially. */
1591 if (flags & MSG_OOB)
1592 goto recv_urg;
1593
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001594 if (unlikely(tp->repair)) {
1595 err = -EPERM;
1596 if (!(flags & MSG_PEEK))
1597 goto out;
1598
1599 if (tp->repair_queue == TCP_SEND_QUEUE)
1600 goto recv_sndq;
1601
1602 err = -EINVAL;
1603 if (tp->repair_queue == TCP_NO_QUEUE)
1604 goto out;
1605
1606 /* 'common' recv queue MSG_PEEK-ing */
1607 }
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 seq = &tp->copied_seq;
1610 if (flags & MSG_PEEK) {
1611 peek_seq = tp->copied_seq;
1612 seq = &peek_seq;
1613 }
1614
1615 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1616
Chris Leech1a2449a2006-05-23 18:05:53 -07001617#ifdef CONFIG_NET_DMA
1618 tp->ucopy.dma_chan = NULL;
1619 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001620 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001621 {
1622 int available = 0;
1623
1624 if (skb)
1625 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1626 if ((available < target) &&
1627 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1628 !sysctl_tcp_low_latency &&
Dave Jianga2bd1142012-04-04 16:10:46 -07001629 net_dma_find_channel()) {
Andrew Mortone00c5d82007-03-08 09:57:36 -08001630 preempt_enable_no_resched();
1631 tp->ucopy.pinned_list =
1632 dma_pin_iovec_pages(msg->msg_iov, len);
1633 } else {
1634 preempt_enable_no_resched();
1635 }
1636 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001637#endif
1638
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 u32 offset;
1641
1642 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1643 if (tp->urg_data && tp->urg_seq == *seq) {
1644 if (copied)
1645 break;
1646 if (signal_pending(current)) {
1647 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1648 break;
1649 }
1650 }
1651
1652 /* Next get a buffer. */
1653
David S. Miller91521942009-05-28 21:35:47 -07001654 skb_queue_walk(&sk->sk_receive_queue, skb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 /* Now that we have two receive queues this
1656 * shouldn't happen.
1657 */
Ilpo Järvinend792c102009-11-13 13:56:33 -08001658 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001659 "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1660 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1661 flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 break;
Ilpo Järvinend792c102009-11-13 13:56:33 -08001663
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001665 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 offset--;
1667 if (offset < skb->len)
1668 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001669 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 goto found_fin_ok;
Joe Perches2af6fd82010-10-30 11:08:53 +00001671 WARN(!(flags & MSG_PEEK),
1672 "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1673 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
David S. Miller91521942009-05-28 21:35:47 -07001674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 /* Well, if we have backlog, try to process it now yet. */
1677
1678 if (copied >= target && !sk->sk_backlog.tail)
1679 break;
1680
1681 if (copied) {
1682 if (sk->sk_err ||
1683 sk->sk_state == TCP_CLOSE ||
1684 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1685 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08001686 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 break;
1688 } else {
1689 if (sock_flag(sk, SOCK_DONE))
1690 break;
1691
1692 if (sk->sk_err) {
1693 copied = sock_error(sk);
1694 break;
1695 }
1696
1697 if (sk->sk_shutdown & RCV_SHUTDOWN)
1698 break;
1699
1700 if (sk->sk_state == TCP_CLOSE) {
1701 if (!sock_flag(sk, SOCK_DONE)) {
1702 /* This occurs when user tries to read
1703 * from never connected socket.
1704 */
1705 copied = -ENOTCONN;
1706 break;
1707 }
1708 break;
1709 }
1710
1711 if (!timeo) {
1712 copied = -EAGAIN;
1713 break;
1714 }
1715
1716 if (signal_pending(current)) {
1717 copied = sock_intr_errno(timeo);
1718 break;
1719 }
1720 }
1721
Chris Leech0e4b4992006-05-23 18:00:16 -07001722 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
David S. Miller7df55122005-06-18 23:01:10 -07001724 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 /* Install new reader */
1726 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1727 user_recv = current;
1728 tp->ucopy.task = user_recv;
1729 tp->ucopy.iov = msg->msg_iov;
1730 }
1731
1732 tp->ucopy.len = len;
1733
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001734 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1735 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
1737 /* Ugly... If prequeue is not empty, we have to
1738 * process it before releasing socket, otherwise
1739 * order will be broken at second iteration.
1740 * More elegant solution is required!!!
1741 *
1742 * Look: we have the following (pseudo)queues:
1743 *
1744 * 1. packets in flight
1745 * 2. backlog
1746 * 3. prequeue
1747 * 4. receive_queue
1748 *
1749 * Each queue can be processed only if the next ones
1750 * are empty. At this point we have empty receive_queue.
1751 * But prequeue _can_ be not empty after 2nd iteration,
1752 * when we jumped to start of loop because backlog
1753 * processing added something to receive_queue.
1754 * We cannot release_sock(), because backlog contains
1755 * packets arrived _after_ prequeued ones.
1756 *
1757 * Shortly, algorithm is clear --- to process all
1758 * the queues in order. We could make it more directly,
1759 * requeueing packets from backlog to prequeue, if
1760 * is not empty. It is more elegant, but eats cycles,
1761 * unfortunately.
1762 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001763 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 goto do_prequeue;
1765
1766 /* __ Set realtime policy in scheduler __ */
1767 }
1768
Steven J. Magnani73852e82010-03-16 05:22:44 +00001769#ifdef CONFIG_NET_DMA
Michal Kubeček15c04172012-09-14 04:59:52 +00001770 if (tp->ucopy.dma_chan) {
1771 if (tp->rcv_wnd == 0 &&
1772 !skb_queue_empty(&sk->sk_async_wait_queue)) {
1773 tcp_service_net_dma(sk, true);
1774 tcp_cleanup_rbuf(sk, copied);
1775 } else
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001776 dma_async_issue_pending(tp->ucopy.dma_chan);
Michal Kubeček15c04172012-09-14 04:59:52 +00001777 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001778#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 if (copied >= target) {
1780 /* Do not sleep, just process backlog. */
1781 release_sock(sk);
1782 lock_sock(sk);
1783 } else
1784 sk_wait_data(sk, &timeo);
1785
Chris Leech1a2449a2006-05-23 18:05:53 -07001786#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001787 tcp_service_net_dma(sk, false); /* Don't block */
Chris Leech1a2449a2006-05-23 18:05:53 -07001788 tp->ucopy.wakeup = 0;
1789#endif
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 if (user_recv) {
1792 int chunk;
1793
1794 /* __ Restore normal policy in scheduler __ */
1795
1796 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001797 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 len -= chunk;
1799 copied += chunk;
1800 }
1801
1802 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001803 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804do_prequeue:
1805 tcp_prequeue_process(sk);
1806
1807 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001808 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 len -= chunk;
1810 copied += chunk;
1811 }
1812 }
1813 }
Ilpo Järvinen77527312009-05-10 20:32:34 +00001814 if ((flags & MSG_PEEK) &&
1815 (peek_seq - copied - urg_hole != tp->copied_seq)) {
Joe Perchese87cc472012-05-13 21:56:26 +00001816 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1817 current->comm,
1818 task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 peek_seq = tp->copied_seq;
1820 }
1821 continue;
1822
1823 found_ok_skb:
1824 /* Ok so how much can we use? */
1825 used = skb->len - offset;
1826 if (len < used)
1827 used = len;
1828
1829 /* Do we have urgent data here? */
1830 if (tp->urg_data) {
1831 u32 urg_offset = tp->urg_seq - *seq;
1832 if (urg_offset < used) {
1833 if (!urg_offset) {
1834 if (!sock_flag(sk, SOCK_URGINLINE)) {
1835 ++*seq;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001836 urg_hole++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 offset++;
1838 used--;
1839 if (!used)
1840 goto skip_copy;
1841 }
1842 } else
1843 used = urg_offset;
1844 }
1845 }
1846
1847 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001848#ifdef CONFIG_NET_DMA
1849 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dave Jianga2bd1142012-04-04 16:10:46 -07001850 tp->ucopy.dma_chan = net_dma_find_channel();
Chris Leech1a2449a2006-05-23 18:05:53 -07001851
1852 if (tp->ucopy.dma_chan) {
1853 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1854 tp->ucopy.dma_chan, skb, offset,
1855 msg->msg_iov, used,
1856 tp->ucopy.pinned_list);
1857
1858 if (tp->ucopy.dma_cookie < 0) {
1859
Joe Perchesafd465032012-03-12 07:03:32 +00001860 pr_alert("%s: dma_cookie < 0\n",
1861 __func__);
Chris Leech1a2449a2006-05-23 18:05:53 -07001862
1863 /* Exception. Bailout! */
1864 if (!copied)
1865 copied = -EFAULT;
1866 break;
1867 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001868
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001869 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001870
Chris Leech1a2449a2006-05-23 18:05:53 -07001871 if ((offset + used) == skb->len)
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001872 copied_early = true;
Chris Leech1a2449a2006-05-23 18:05:53 -07001873
1874 } else
1875#endif
1876 {
1877 err = skb_copy_datagram_iovec(skb, offset,
1878 msg->msg_iov, used);
1879 if (err) {
1880 /* Exception. Bailout! */
1881 if (!copied)
1882 copied = -EFAULT;
1883 break;
1884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 }
1886 }
1887
1888 *seq += used;
1889 copied += used;
1890 len -= used;
1891
1892 tcp_rcv_space_adjust(sk);
1893
1894skip_copy:
1895 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1896 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001897 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 }
1899 if (used + offset < skb->len)
1900 continue;
1901
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001902 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001904 if (!(flags & MSG_PEEK)) {
1905 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001906 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 continue;
1909
1910 found_fin_ok:
1911 /* Process the FIN. */
1912 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001913 if (!(flags & MSG_PEEK)) {
1914 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001915 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 break;
1918 } while (len > 0);
1919
1920 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001921 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 int chunk;
1923
1924 tp->ucopy.len = copied > 0 ? len : 0;
1925
1926 tcp_prequeue_process(sk);
1927
1928 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001929 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 len -= chunk;
1931 copied += chunk;
1932 }
1933 }
1934
1935 tp->ucopy.task = NULL;
1936 tp->ucopy.len = 0;
1937 }
1938
Chris Leech1a2449a2006-05-23 18:05:53 -07001939#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001940 tcp_service_net_dma(sk, true); /* Wait for queue to drain */
1941 tp->ucopy.dma_chan = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001942
Chris Leech1a2449a2006-05-23 18:05:53 -07001943 if (tp->ucopy.pinned_list) {
1944 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1945 tp->ucopy.pinned_list = NULL;
1946 }
1947#endif
1948
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 /* According to UNIX98, msg_name/msg_namelen are ignored
1950 * on connected socket. I was just happy when found this 8) --ANK
1951 */
1952
1953 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001954 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 release_sock(sk);
1957 return copied;
1958
1959out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 release_sock(sk);
1961 return err;
1962
1963recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07001964 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 goto out;
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001966
1967recv_sndq:
1968 err = tcp_peek_sndq(sk, msg, len);
1969 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001971EXPORT_SYMBOL(tcp_recvmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001973void tcp_set_state(struct sock *sk, int state)
1974{
1975 int oldstate = sk->sk_state;
1976
1977 switch (state) {
1978 case TCP_ESTABLISHED:
1979 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001980 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001981 break;
1982
1983 case TCP_CLOSE:
1984 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001985 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001986
1987 sk->sk_prot->unhash(sk);
1988 if (inet_csk(sk)->icsk_bind_hash &&
1989 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001990 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001991 /* fall through */
1992 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001993 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07001994 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001995 }
1996
1997 /* Change state AFTER socket is unhashed to avoid closed
1998 * socket sitting in hash tables.
1999 */
2000 sk->sk_state = state;
2001
2002#ifdef STATE_TRACE
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08002003 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002004#endif
2005}
2006EXPORT_SYMBOL_GPL(tcp_set_state);
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008/*
2009 * State processing on a close. This implements the state shift for
2010 * sending our FIN frame. Note that we only send a FIN for some
2011 * states. A shutdown() may have already sent the FIN, or we may be
2012 * closed.
2013 */
2014
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08002015static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 /* current state: new state: action: */
2017 /* (Invalid) */ TCP_CLOSE,
2018 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2019 /* TCP_SYN_SENT */ TCP_CLOSE,
2020 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2021 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
2022 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
2023 /* TCP_TIME_WAIT */ TCP_CLOSE,
2024 /* TCP_CLOSE */ TCP_CLOSE,
2025 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
2026 /* TCP_LAST_ACK */ TCP_LAST_ACK,
2027 /* TCP_LISTEN */ TCP_CLOSE,
2028 /* TCP_CLOSING */ TCP_CLOSING,
2029};
2030
2031static int tcp_close_state(struct sock *sk)
2032{
2033 int next = (int)new_state[sk->sk_state];
2034 int ns = next & TCP_STATE_MASK;
2035
2036 tcp_set_state(sk, ns);
2037
2038 return next & TCP_ACTION_FIN;
2039}
2040
2041/*
2042 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07002043 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 */
2045
2046void tcp_shutdown(struct sock *sk, int how)
2047{
2048 /* We need to grab some memory, and put together a FIN,
2049 * and then put it into the queue to be sent.
2050 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2051 */
2052 if (!(how & SEND_SHUTDOWN))
2053 return;
2054
2055 /* If we've already sent a FIN, or it's a closed state, skip this. */
2056 if ((1 << sk->sk_state) &
2057 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2058 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2059 /* Clear out any half completed packets. FIN if needed. */
2060 if (tcp_close_state(sk))
2061 tcp_send_fin(sk);
2062 }
2063}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002064EXPORT_SYMBOL(tcp_shutdown);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002066bool tcp_check_oom(struct sock *sk, int shift)
2067{
2068 bool too_many_orphans, out_of_socket_memory;
2069
2070 too_many_orphans = tcp_too_many_orphans(sk, shift);
2071 out_of_socket_memory = tcp_out_of_memory(sk);
2072
Joe Perchese87cc472012-05-13 21:56:26 +00002073 if (too_many_orphans)
2074 net_info_ratelimited("too many orphaned sockets\n");
2075 if (out_of_socket_memory)
2076 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002077 return too_many_orphans || out_of_socket_memory;
2078}
2079
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080void tcp_close(struct sock *sk, long timeout)
2081{
2082 struct sk_buff *skb;
2083 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07002084 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 lock_sock(sk);
2087 sk->sk_shutdown = SHUTDOWN_MASK;
2088
2089 if (sk->sk_state == TCP_LISTEN) {
2090 tcp_set_state(sk, TCP_CLOSE);
2091
2092 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002093 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
2095 goto adjudge_to_death;
2096 }
2097
2098 /* We need to flush the recv. buffs. We do this only on the
2099 * descriptor close, not protocol-sourced closes, because the
2100 * reader process may not have drained the data yet!
2101 */
2102 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2103 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002104 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 data_was_unread += len;
2106 __kfree_skb(skb);
2107 }
2108
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002109 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Konstantin Khorenko565b7b22010-06-24 21:54:58 -07002111 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2112 if (sk->sk_state == TCP_CLOSE)
2113 goto adjudge_to_death;
2114
Gerrit Renker65bb7232007-04-28 21:21:46 -07002115 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2116 * data was lost. To witness the awful effects of the old behavior of
2117 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2118 * GET in an FTP client, suspend the process, wait for the client to
2119 * advertise a zero window, then kill -9 the FTP client, wheee...
2120 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 */
Pavel Emelyanovee995282012-04-19 03:40:39 +00002122 if (unlikely(tcp_sk(sk)->repair)) {
2123 sk->sk_prot->disconnect(sk, 0);
2124 } else if (data_was_unread) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002126 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 tcp_set_state(sk, TCP_CLOSE);
Wu Fengguangaa133072009-09-02 23:45:45 -07002128 tcp_send_active_reset(sk, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2130 /* Check zero linger _after_ checking for unread data. */
2131 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002132 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 } else if (tcp_close_state(sk)) {
2134 /* We FIN if the application ate all the data before
2135 * zapping the connection.
2136 */
2137
2138 /* RED-PEN. Formally speaking, we have broken TCP state
2139 * machine. State transitions:
2140 *
2141 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2142 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2143 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2144 *
2145 * are legal only when FIN has been sent (i.e. in window),
2146 * rather than queued out of window. Purists blame.
2147 *
2148 * F.e. "RFC state" is ESTABLISHED,
2149 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2150 *
2151 * The visible declinations are that sometimes
2152 * we enter time-wait state, when it is not required really
2153 * (harmless), do not send active resets, when they are
2154 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2155 * they look as CLOSING or LAST_ACK for Linux)
2156 * Probably, I missed some more holelets.
2157 * --ANK
Jerry Chu83368862012-08-31 12:29:12 +00002158 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2159 * in a single packet! (May consider it later but will
2160 * probably need API support or TCP_CORK SYN-ACK until
2161 * data is written and socket is closed.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 */
2163 tcp_send_fin(sk);
2164 }
2165
2166 sk_stream_wait_close(sk, timeout);
2167
2168adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07002169 state = sk->sk_state;
2170 sock_hold(sk);
2171 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07002172
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 /* It is the last release_sock in its life. It will remove backlog. */
2174 release_sock(sk);
2175
2176
2177 /* Now socket is owned by kernel and we acquire BH lock
2178 to finish close. No need to check for user refs.
2179 */
2180 local_bh_disable();
2181 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002182 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Herbert Xueb4dea52008-12-29 23:04:08 -08002184 percpu_counter_inc(sk->sk_prot->orphan_count);
2185
Herbert Xu75c2d9072006-05-03 23:31:35 -07002186 /* Have we already been destroyed by a softirq or backlog? */
2187 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2188 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190 /* This is a (useful) BSD violating of the RFC. There is a
2191 * problem with TCP as specified in that the other end could
2192 * keep a socket open forever with no application left this end.
2193 * We use a 3 minute timeout (about the same as BSD) then kill
2194 * our end. If they send after that then tough - BUT: long enough
2195 * that we won't make the old 4*rto = almost no time - whoops
2196 * reset mistake.
2197 *
2198 * Nope, it was not mistake. It is really desired behaviour
2199 * f.e. on http servers, when such sockets are useless, but
2200 * consume significant resources. Let's do it with special
2201 * linger2 option. --ANK
2202 */
2203
2204 if (sk->sk_state == TCP_FIN_WAIT2) {
2205 struct tcp_sock *tp = tcp_sk(sk);
2206 if (tp->linger2 < 0) {
2207 tcp_set_state(sk, TCP_CLOSE);
2208 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002209 NET_INC_STATS_BH(sock_net(sk),
2210 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002212 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
2214 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07002215 inet_csk_reset_keepalive_timer(sk,
2216 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2219 goto out;
2220 }
2221 }
2222 }
2223 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002224 sk_mem_reclaim(sk);
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002225 if (tcp_check_oom(sk, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 tcp_set_state(sk, TCP_CLOSE);
2227 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002228 NET_INC_STATS_BH(sock_net(sk),
2229 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 }
2231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
Jerry Chu83368862012-08-31 12:29:12 +00002233 if (sk->sk_state == TCP_CLOSE) {
2234 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2235 /* We could get here with a non-NULL req if the socket is
2236 * aborted (e.g., closed with unread data) before 3WHS
2237 * finishes.
2238 */
2239 if (req != NULL)
2240 reqsk_fastopen_remove(sk, req, false);
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002241 inet_csk_destroy_sock(sk);
Jerry Chu83368862012-08-31 12:29:12 +00002242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 /* Otherwise, socket is reprieved until protocol close. */
2244
2245out:
2246 bh_unlock_sock(sk);
2247 local_bh_enable();
2248 sock_put(sk);
2249}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002250EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252/* These states need RST on ABORT according to RFC793 */
2253
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002254static inline bool tcp_need_reset(int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
2256 return (1 << state) &
2257 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2258 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2259}
2260
2261int tcp_disconnect(struct sock *sk, int flags)
2262{
2263 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002264 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 struct tcp_sock *tp = tcp_sk(sk);
2266 int err = 0;
2267 int old_state = sk->sk_state;
2268
2269 if (old_state != TCP_CLOSE)
2270 tcp_set_state(sk, TCP_CLOSE);
2271
2272 /* ABORT function of RFC793 */
2273 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002274 inet_csk_listen_stop(sk);
Pavel Emelyanovee995282012-04-19 03:40:39 +00002275 } else if (unlikely(tp->repair)) {
2276 sk->sk_err = ECONNABORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 } else if (tcp_need_reset(old_state) ||
2278 (tp->snd_nxt != tp->write_seq &&
2279 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08002280 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 * states
2282 */
2283 tcp_send_active_reset(sk, gfp_any());
2284 sk->sk_err = ECONNRESET;
2285 } else if (old_state == TCP_SYN_SENT)
2286 sk->sk_err = ECONNRESET;
2287
2288 tcp_clear_xmit_timers(sk);
2289 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08002290 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07002292#ifdef CONFIG_NET_DMA
2293 __skb_queue_purge(&sk->sk_async_wait_queue);
2294#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002296 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
2298 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2299 inet_reset_saddr(sk);
2300
2301 sk->sk_shutdown = 0;
2302 sock_reset_flag(sk, SOCK_DONE);
2303 tp->srtt = 0;
2304 if ((tp->write_seq += tp->max_window + 2) == 0)
2305 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002306 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002308 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 tp->packets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002310 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 tp->snd_cwnd_cnt = 0;
Eric Dumazet1fdf4752009-11-30 12:53:30 -08002312 tp->window_clamp = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002313 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002315 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08002316 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002317 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 __sk_dst_reset(sk);
2319
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002320 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
2322 sk->sk_error_report(sk);
2323 return err;
2324}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002325EXPORT_SYMBOL(tcp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Christoph Paaschbb68b642012-09-18 14:19:23 +00002327void tcp_sock_destruct(struct sock *sk)
2328{
2329 inet_sock_destruct(sk);
2330
2331 kfree(inet_csk(sk)->icsk_accept_queue.fastopenq);
2332}
2333
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002334static inline bool tcp_can_repair_sock(const struct sock *sk)
Pavel Emelyanovee995282012-04-19 03:40:39 +00002335{
Eric W. Biederman52e804c2012-11-16 03:03:05 +00002336 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
Pavel Emelyanovee995282012-04-19 03:40:39 +00002337 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
2338}
2339
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002340static int tcp_repair_options_est(struct tcp_sock *tp,
2341 struct tcp_repair_opt __user *optbuf, unsigned int len)
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002342{
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002343 struct tcp_repair_opt opt;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002344
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002345 while (len >= sizeof(opt)) {
2346 if (copy_from_user(&opt, optbuf, sizeof(opt)))
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002347 return -EFAULT;
2348
2349 optbuf++;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002350 len -= sizeof(opt);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002351
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002352 switch (opt.opt_code) {
2353 case TCPOPT_MSS:
2354 tp->rx_opt.mss_clamp = opt.opt_val;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002355 break;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002356 case TCPOPT_WINDOW:
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002357 {
2358 u16 snd_wscale = opt.opt_val & 0xFFFF;
2359 u16 rcv_wscale = opt.opt_val >> 16;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002360
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002361 if (snd_wscale > 14 || rcv_wscale > 14)
2362 return -EFBIG;
2363
2364 tp->rx_opt.snd_wscale = snd_wscale;
2365 tp->rx_opt.rcv_wscale = rcv_wscale;
2366 tp->rx_opt.wscale_ok = 1;
2367 }
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002368 break;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002369 case TCPOPT_SACK_PERM:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002370 if (opt.opt_val != 0)
2371 return -EINVAL;
2372
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002373 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2374 if (sysctl_tcp_fack)
2375 tcp_enable_fack(tp);
2376 break;
2377 case TCPOPT_TIMESTAMP:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002378 if (opt.opt_val != 0)
2379 return -EINVAL;
2380
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002381 tp->rx_opt.tstamp_ok = 1;
2382 break;
2383 }
2384 }
2385
2386 return 0;
2387}
2388
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 * Socket option code for TCP.
2391 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002392static int do_tcp_setsockopt(struct sock *sk, int level,
David S. Millerb7058842009-09-30 16:12:20 -07002393 int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394{
2395 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002396 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 int val;
2398 int err = 0;
2399
William Allen Simpsone56fb502009-12-02 18:19:30 +00002400 /* These are data/string values, all the others are ints */
2401 switch (optname) {
2402 case TCP_CONGESTION: {
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002403 char name[TCP_CA_NAME_MAX];
2404
2405 if (optlen < 1)
2406 return -EINVAL;
2407
2408 val = strncpy_from_user(name, optval,
Andrew Morton4fdb78d2009-10-01 15:02:20 -07002409 min_t(long, TCP_CA_NAME_MAX-1, optlen));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002410 if (val < 0)
2411 return -EFAULT;
2412 name[val] = 0;
2413
2414 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002415 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002416 release_sock(sk);
2417 return err;
2418 }
William Allen Simpsone56fb502009-12-02 18:19:30 +00002419 default:
2420 /* fallthru */
2421 break;
Joe Perchesccbd6a52010-05-14 10:58:26 +00002422 }
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 if (optlen < sizeof(int))
2425 return -EINVAL;
2426
2427 if (get_user(val, (int __user *)optval))
2428 return -EFAULT;
2429
2430 lock_sock(sk);
2431
2432 switch (optname) {
2433 case TCP_MAXSEG:
2434 /* Values greater than interface MTU won't take effect. However
2435 * at the point when this call is done we typically don't yet
2436 * know which interface is going to be used */
David S. Millerc39508d2010-11-24 11:47:22 -08002437 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 err = -EINVAL;
2439 break;
2440 }
2441 tp->rx_opt.user_mss = val;
2442 break;
2443
2444 case TCP_NODELAY:
2445 if (val) {
2446 /* TCP_NODELAY is weaker than TCP_CORK, so that
2447 * this option on corked socket is remembered, but
2448 * it is not activated until cork is cleared.
2449 *
2450 * However, when TCP_NODELAY is set we make
2451 * an explicit push, which overrides even TCP_CORK
2452 * for currently queued segments.
2453 */
2454 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002455 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 } else {
2457 tp->nonagle &= ~TCP_NAGLE_OFF;
2458 }
2459 break;
2460
Andreas Petlund36e31b0a2010-02-18 02:47:01 +00002461 case TCP_THIN_LINEAR_TIMEOUTS:
2462 if (val < 0 || val > 1)
2463 err = -EINVAL;
2464 else
2465 tp->thin_lto = val;
2466 break;
2467
Andreas Petlund7e380172010-02-18 04:48:19 +00002468 case TCP_THIN_DUPACK:
2469 if (val < 0 || val > 1)
2470 err = -EINVAL;
Dave Jonese2e5c4c2013-09-05 13:43:34 -04002471 else {
Andreas Petlund7e380172010-02-18 04:48:19 +00002472 tp->thin_dupack = val;
Yuchung Chengeed530b2012-05-02 13:30:03 +00002473 if (tp->thin_dupack)
2474 tcp_disable_early_retrans(tp);
Dave Jonese2e5c4c2013-09-05 13:43:34 -04002475 }
Andreas Petlund7e380172010-02-18 04:48:19 +00002476 break;
2477
Pavel Emelyanovee995282012-04-19 03:40:39 +00002478 case TCP_REPAIR:
2479 if (!tcp_can_repair_sock(sk))
2480 err = -EPERM;
2481 else if (val == 1) {
2482 tp->repair = 1;
2483 sk->sk_reuse = SK_FORCE_REUSE;
2484 tp->repair_queue = TCP_NO_QUEUE;
2485 } else if (val == 0) {
2486 tp->repair = 0;
2487 sk->sk_reuse = SK_NO_REUSE;
2488 tcp_send_window_probe(sk);
2489 } else
2490 err = -EINVAL;
2491
2492 break;
2493
2494 case TCP_REPAIR_QUEUE:
2495 if (!tp->repair)
2496 err = -EPERM;
2497 else if (val < TCP_QUEUES_NR)
2498 tp->repair_queue = val;
2499 else
2500 err = -EINVAL;
2501 break;
2502
2503 case TCP_QUEUE_SEQ:
2504 if (sk->sk_state != TCP_CLOSE)
2505 err = -EPERM;
2506 else if (tp->repair_queue == TCP_SEND_QUEUE)
2507 tp->write_seq = val;
2508 else if (tp->repair_queue == TCP_RECV_QUEUE)
2509 tp->rcv_nxt = val;
2510 else
2511 err = -EINVAL;
2512 break;
2513
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002514 case TCP_REPAIR_OPTIONS:
2515 if (!tp->repair)
2516 err = -EINVAL;
2517 else if (sk->sk_state == TCP_ESTABLISHED)
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002518 err = tcp_repair_options_est(tp,
2519 (struct tcp_repair_opt __user *)optval,
2520 optlen);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002521 else
2522 err = -EPERM;
2523 break;
2524
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 case TCP_CORK:
2526 /* When set indicates to always queue non-full frames.
2527 * Later the user clears this option and we transmit
2528 * any pending partial frames in the queue. This is
2529 * meant to be used alongside sendfile() to get properly
2530 * filled frames when the user (for example) must write
2531 * out headers with a write() call first and then use
2532 * sendfile to send out the data parts.
2533 *
2534 * TCP_CORK can be set together with TCP_NODELAY and it is
2535 * stronger than TCP_NODELAY.
2536 */
2537 if (val) {
2538 tp->nonagle |= TCP_NAGLE_CORK;
2539 } else {
2540 tp->nonagle &= ~TCP_NAGLE_CORK;
2541 if (tp->nonagle&TCP_NAGLE_OFF)
2542 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002543 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 }
2545 break;
2546
2547 case TCP_KEEPIDLE:
2548 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2549 err = -EINVAL;
2550 else {
2551 tp->keepalive_time = val * HZ;
2552 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2553 !((1 << sk->sk_state) &
2554 (TCPF_CLOSE | TCPF_LISTEN))) {
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00002555 u32 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 if (tp->keepalive_time > elapsed)
2557 elapsed = tp->keepalive_time - elapsed;
2558 else
2559 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002560 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 }
2562 }
2563 break;
2564 case TCP_KEEPINTVL:
2565 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2566 err = -EINVAL;
2567 else
2568 tp->keepalive_intvl = val * HZ;
2569 break;
2570 case TCP_KEEPCNT:
2571 if (val < 1 || val > MAX_TCP_KEEPCNT)
2572 err = -EINVAL;
2573 else
2574 tp->keepalive_probes = val;
2575 break;
2576 case TCP_SYNCNT:
2577 if (val < 1 || val > MAX_TCP_SYNCNT)
2578 err = -EINVAL;
2579 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002580 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 break;
2582
2583 case TCP_LINGER2:
2584 if (val < 0)
2585 tp->linger2 = -1;
2586 else if (val > sysctl_tcp_fin_timeout / HZ)
2587 tp->linger2 = 0;
2588 else
2589 tp->linger2 = val * HZ;
2590 break;
2591
2592 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002593 /* Translate value in seconds to number of retransmits */
2594 icsk->icsk_accept_queue.rskq_defer_accept =
2595 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2596 TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 break;
2598
2599 case TCP_WINDOW_CLAMP:
2600 if (!val) {
2601 if (sk->sk_state != TCP_CLOSE) {
2602 err = -EINVAL;
2603 break;
2604 }
2605 tp->window_clamp = 0;
2606 } else
2607 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2608 SOCK_MIN_RCVBUF / 2 : val;
2609 break;
2610
2611 case TCP_QUICKACK:
2612 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002613 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002615 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 if ((1 << sk->sk_state) &
2617 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002618 inet_csk_ack_scheduled(sk)) {
2619 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002620 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002622 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 }
2624 }
2625 break;
2626
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002627#ifdef CONFIG_TCP_MD5SIG
2628 case TCP_MD5SIG:
2629 /* Read the IP->Key mappings from userspace */
2630 err = tp->af_specific->md5_parse(sk, optval, optlen);
2631 break;
2632#endif
Jerry Chudca43c72010-08-27 19:13:28 +00002633 case TCP_USER_TIMEOUT:
2634 /* Cap the max timeout in ms TCP will retry/retrans
2635 * before giving up and aborting (ETIMEDOUT) a connection.
2636 */
Hangbin Liu42493572012-07-26 22:52:21 +00002637 if (val < 0)
2638 err = -EINVAL;
2639 else
2640 icsk->icsk_user_timeout = msecs_to_jiffies(val);
Jerry Chudca43c72010-08-27 19:13:28 +00002641 break;
Jerry Chu83368862012-08-31 12:29:12 +00002642
2643 case TCP_FASTOPEN:
2644 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
2645 TCPF_LISTEN)))
2646 err = fastopen_init_queue(sk, val);
2647 else
2648 err = -EINVAL;
2649 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002650 case TCP_TIMESTAMP:
2651 if (!tp->repair)
2652 err = -EPERM;
2653 else
2654 tp->tsoffset = val - tcp_time_stamp;
2655 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002656 case TCP_NOTSENT_LOWAT:
2657 tp->notsent_lowat = val;
2658 sk->sk_write_space(sk);
2659 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 default:
2661 err = -ENOPROTOOPT;
2662 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002663 }
2664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 release_sock(sk);
2666 return err;
2667}
2668
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002669int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
David S. Millerb7058842009-09-30 16:12:20 -07002670 unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002671{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002672 const struct inet_connection_sock *icsk = inet_csk(sk);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002673
2674 if (level != SOL_TCP)
2675 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2676 optval, optlen);
2677 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2678}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002679EXPORT_SYMBOL(tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002680
2681#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002682int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002683 char __user *optval, unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002684{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002685 if (level != SOL_TCP)
2686 return inet_csk_compat_setsockopt(sk, level, optname,
2687 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002688 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2689}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002690EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002691#endif
2692
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693/* Return information about state of tcp endpoint in API format. */
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002694void tcp_get_info(const struct sock *sk, struct tcp_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002696 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002697 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 u32 now = tcp_time_stamp;
2699
2700 memset(info, 0, sizeof(*info));
2701
2702 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002703 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002704 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002705 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002706 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
2708 if (tp->rx_opt.tstamp_ok)
2709 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002710 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 info->tcpi_options |= TCPI_OPT_SACK;
2712 if (tp->rx_opt.wscale_ok) {
2713 info->tcpi_options |= TCPI_OPT_WSCALE;
2714 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2715 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
Eric Dumazetb5c56932011-10-03 14:01:21 -04002718 if (tp->ecn_flags & TCP_ECN_OK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 info->tcpi_options |= TCPI_OPT_ECN;
Eric Dumazetb5c56932011-10-03 14:01:21 -04002720 if (tp->ecn_flags & TCP_ECN_SEEN)
2721 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
Yuchung Cheng6f736012012-10-19 15:14:44 +00002722 if (tp->syn_data_acked)
2723 info->tcpi_options |= TCPI_OPT_SYN_DATA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002725 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2726 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002727 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002728 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
Rick Jones5ee3afb2007-09-18 13:26:31 -07002730 if (sk->sk_state == TCP_LISTEN) {
2731 info->tcpi_unacked = sk->sk_ack_backlog;
2732 info->tcpi_sacked = sk->sk_max_ack_backlog;
2733 } else {
2734 info->tcpi_unacked = tp->packets_out;
2735 info->tcpi_sacked = tp->sacked_out;
2736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 info->tcpi_lost = tp->lost_out;
2738 info->tcpi_retrans = tp->retrans_out;
2739 info->tcpi_fackets = tp->fackets_out;
2740
2741 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002742 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2744
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002745 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2747 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2748 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2749 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2750 info->tcpi_snd_cwnd = tp->snd_cwnd;
2751 info->tcpi_advmss = tp->advmss;
2752 info->tcpi_reordering = tp->reordering;
2753
2754 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2755 info->tcpi_rcv_space = tp->rcvq_space.space;
2756
2757 info->tcpi_total_retrans = tp->total_retrans;
2758}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759EXPORT_SYMBOL_GPL(tcp_get_info);
2760
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002761static int do_tcp_getsockopt(struct sock *sk, int level,
2762 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002764 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 struct tcp_sock *tp = tcp_sk(sk);
2766 int val, len;
2767
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 if (get_user(len, optlen))
2769 return -EFAULT;
2770
2771 len = min_t(unsigned int, len, sizeof(int));
2772
2773 if (len < 0)
2774 return -EINVAL;
2775
2776 switch (optname) {
2777 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002778 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2780 val = tp->rx_opt.user_mss;
Pavel Emelyanov5e6a3ce2012-04-19 03:41:32 +00002781 if (tp->repair)
2782 val = tp->rx_opt.mss_clamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 break;
2784 case TCP_NODELAY:
2785 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2786 break;
2787 case TCP_CORK:
2788 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2789 break;
2790 case TCP_KEEPIDLE:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002791 val = keepalive_time_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 break;
2793 case TCP_KEEPINTVL:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002794 val = keepalive_intvl_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 break;
2796 case TCP_KEEPCNT:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002797 val = keepalive_probes(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 break;
2799 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002800 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 break;
2802 case TCP_LINGER2:
2803 val = tp->linger2;
2804 if (val >= 0)
2805 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2806 break;
2807 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002808 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2809 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 break;
2811 case TCP_WINDOW_CLAMP:
2812 val = tp->window_clamp;
2813 break;
2814 case TCP_INFO: {
2815 struct tcp_info info;
2816
2817 if (get_user(len, optlen))
2818 return -EFAULT;
2819
2820 tcp_get_info(sk, &info);
2821
2822 len = min_t(unsigned int, len, sizeof(info));
2823 if (put_user(len, optlen))
2824 return -EFAULT;
2825 if (copy_to_user(optval, &info, len))
2826 return -EFAULT;
2827 return 0;
2828 }
2829 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002830 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002832
2833 case TCP_CONGESTION:
2834 if (get_user(len, optlen))
2835 return -EFAULT;
2836 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2837 if (put_user(len, optlen))
2838 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002839 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002840 return -EFAULT;
2841 return 0;
William Allen Simpsone56fb502009-12-02 18:19:30 +00002842
Josh Hunt3c0fef02010-07-30 13:49:35 +00002843 case TCP_THIN_LINEAR_TIMEOUTS:
2844 val = tp->thin_lto;
2845 break;
2846 case TCP_THIN_DUPACK:
2847 val = tp->thin_dupack;
2848 break;
Jerry Chudca43c72010-08-27 19:13:28 +00002849
Pavel Emelyanovee995282012-04-19 03:40:39 +00002850 case TCP_REPAIR:
2851 val = tp->repair;
2852 break;
2853
2854 case TCP_REPAIR_QUEUE:
2855 if (tp->repair)
2856 val = tp->repair_queue;
2857 else
2858 return -EINVAL;
2859 break;
2860
2861 case TCP_QUEUE_SEQ:
2862 if (tp->repair_queue == TCP_SEND_QUEUE)
2863 val = tp->write_seq;
2864 else if (tp->repair_queue == TCP_RECV_QUEUE)
2865 val = tp->rcv_nxt;
2866 else
2867 return -EINVAL;
2868 break;
2869
Jerry Chudca43c72010-08-27 19:13:28 +00002870 case TCP_USER_TIMEOUT:
2871 val = jiffies_to_msecs(icsk->icsk_user_timeout);
2872 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002873 case TCP_TIMESTAMP:
2874 val = tcp_time_stamp + tp->tsoffset;
2875 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002876 case TCP_NOTSENT_LOWAT:
2877 val = tp->notsent_lowat;
2878 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 default:
2880 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
2883 if (put_user(len, optlen))
2884 return -EFAULT;
2885 if (copy_to_user(optval, &val, len))
2886 return -EFAULT;
2887 return 0;
2888}
2889
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002890int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2891 int __user *optlen)
2892{
2893 struct inet_connection_sock *icsk = inet_csk(sk);
2894
2895 if (level != SOL_TCP)
2896 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2897 optval, optlen);
2898 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2899}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002900EXPORT_SYMBOL(tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002901
2902#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002903int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2904 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002905{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002906 if (level != SOL_TCP)
2907 return inet_csk_compat_getsockopt(sk, level, optname,
2908 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002909 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2910}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002911EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002912#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002914#ifdef CONFIG_TCP_MD5SIG
Eric Dumazet71cea172013-05-20 06:52:26 +00002915static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool __read_mostly;
2916static DEFINE_MUTEX(tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002917
Eric Dumazet765cf992011-09-12 20:28:37 +00002918static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002919{
2920 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00002921
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002922 for_each_possible_cpu(cpu) {
Eric Dumazet765cf992011-09-12 20:28:37 +00002923 struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu);
2924
2925 if (p->md5_desc.tfm)
2926 crypto_free_hash(p->md5_desc.tfm);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002927 }
2928 free_percpu(pool);
2929}
2930
Eric Dumazet71cea172013-05-20 06:52:26 +00002931static void __tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002932{
2933 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00002934 struct tcp_md5sig_pool __percpu *pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002935
Eric Dumazet765cf992011-09-12 20:28:37 +00002936 pool = alloc_percpu(struct tcp_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002937 if (!pool)
Eric Dumazet71cea172013-05-20 06:52:26 +00002938 return;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002939
2940 for_each_possible_cpu(cpu) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002941 struct crypto_hash *hash;
2942
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002943 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
YOSHIFUJI Hideaki / 吉藤英明50c3a482013-01-22 06:32:49 +00002944 if (IS_ERR_OR_NULL(hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002945 goto out_free;
2946
Eric Dumazet765cf992011-09-12 20:28:37 +00002947 per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002948 }
Eric Dumazet71cea172013-05-20 06:52:26 +00002949 /* before setting tcp_md5sig_pool, we must commit all writes
2950 * to memory. See ACCESS_ONCE() in tcp_get_md5sig_pool()
2951 */
2952 smp_wmb();
2953 tcp_md5sig_pool = pool;
2954 return;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002955out_free:
2956 __tcp_free_md5sig_pool(pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002957}
2958
Eric Dumazet71cea172013-05-20 06:52:26 +00002959bool tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002960{
Eric Dumazet71cea172013-05-20 06:52:26 +00002961 if (unlikely(!tcp_md5sig_pool)) {
2962 mutex_lock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002963
Eric Dumazet71cea172013-05-20 06:52:26 +00002964 if (!tcp_md5sig_pool)
2965 __tcp_alloc_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002966
Eric Dumazet71cea172013-05-20 06:52:26 +00002967 mutex_unlock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002968 }
Eric Dumazet71cea172013-05-20 06:52:26 +00002969 return tcp_md5sig_pool != NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002970}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002971EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2972
Eric Dumazet35790c02010-05-16 00:34:04 -07002973
2974/**
2975 * tcp_get_md5sig_pool - get md5sig_pool for this user
2976 *
2977 * We use percpu structure, so if we succeed, we exit with preemption
2978 * and BH disabled, to make sure another thread or softirq handling
2979 * wont try to get same context.
2980 */
2981struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002982{
Eric Dumazet765cf992011-09-12 20:28:37 +00002983 struct tcp_md5sig_pool __percpu *p;
Eric Dumazet35790c02010-05-16 00:34:04 -07002984
2985 local_bh_disable();
Eric Dumazet71cea172013-05-20 06:52:26 +00002986 p = ACCESS_ONCE(tcp_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002987 if (p)
Eric Dumazet71cea172013-05-20 06:52:26 +00002988 return __this_cpu_ptr(p);
Eric Dumazet35790c02010-05-16 00:34:04 -07002989
2990 local_bh_enable();
2991 return NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002992}
Eric Dumazet35790c02010-05-16 00:34:04 -07002993EXPORT_SYMBOL(tcp_get_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002994
Adam Langley49a72df2008-07-19 00:01:42 -07002995int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
Eric Dumazetca35a0e2011-10-24 01:52:35 -04002996 const struct tcphdr *th)
Adam Langley49a72df2008-07-19 00:01:42 -07002997{
2998 struct scatterlist sg;
Eric Dumazetca35a0e2011-10-24 01:52:35 -04002999 struct tcphdr hdr;
Adam Langley49a72df2008-07-19 00:01:42 -07003000 int err;
3001
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003002 /* We are not allowed to change tcphdr, make a local copy */
3003 memcpy(&hdr, th, sizeof(hdr));
3004 hdr.check = 0;
3005
Adam Langley49a72df2008-07-19 00:01:42 -07003006 /* options aren't included in the hash */
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003007 sg_init_one(&sg, &hdr, sizeof(hdr));
3008 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr));
Adam Langley49a72df2008-07-19 00:01:42 -07003009 return err;
3010}
Adam Langley49a72df2008-07-19 00:01:42 -07003011EXPORT_SYMBOL(tcp_md5_hash_header);
3012
3013int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003014 const struct sk_buff *skb, unsigned int header_len)
Adam Langley49a72df2008-07-19 00:01:42 -07003015{
3016 struct scatterlist sg;
3017 const struct tcphdr *tp = tcp_hdr(skb);
3018 struct hash_desc *desc = &hp->md5_desc;
Eric Dumazet95c96172012-04-15 05:58:06 +00003019 unsigned int i;
3020 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3021 skb_headlen(skb) - header_len : 0;
Adam Langley49a72df2008-07-19 00:01:42 -07003022 const struct skb_shared_info *shi = skb_shinfo(skb);
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003023 struct sk_buff *frag_iter;
Adam Langley49a72df2008-07-19 00:01:42 -07003024
3025 sg_init_table(&sg, 1);
3026
3027 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3028 if (crypto_hash_update(desc, &sg, head_data_len))
3029 return 1;
3030
3031 for (i = 0; i < shi->nr_frags; ++i) {
3032 const struct skb_frag_struct *f = &shi->frags[i];
Eric Dumazet54d27fc2013-05-13 21:25:52 +00003033 unsigned int offset = f->page_offset;
3034 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3035
3036 sg_set_page(&sg, page, skb_frag_size(f),
3037 offset_in_page(offset));
Eric Dumazet9e903e02011-10-18 21:00:24 +00003038 if (crypto_hash_update(desc, &sg, skb_frag_size(f)))
Adam Langley49a72df2008-07-19 00:01:42 -07003039 return 1;
3040 }
3041
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003042 skb_walk_frags(skb, frag_iter)
3043 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3044 return 1;
3045
Adam Langley49a72df2008-07-19 00:01:42 -07003046 return 0;
3047}
Adam Langley49a72df2008-07-19 00:01:42 -07003048EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3049
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003050int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
Adam Langley49a72df2008-07-19 00:01:42 -07003051{
3052 struct scatterlist sg;
3053
3054 sg_init_one(&sg, key->key, key->keylen);
3055 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
3056}
Adam Langley49a72df2008-07-19 00:01:42 -07003057EXPORT_SYMBOL(tcp_md5_hash_key);
3058
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003059#endif
3060
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003061void tcp_done(struct sock *sk)
3062{
Jerry Chu83368862012-08-31 12:29:12 +00003063 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3064
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08003065 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07003066 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003067
3068 tcp_set_state(sk, TCP_CLOSE);
3069 tcp_clear_xmit_timers(sk);
Jerry Chu83368862012-08-31 12:29:12 +00003070 if (req != NULL)
3071 reqsk_fastopen_remove(sk, req, false);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003072
3073 sk->sk_shutdown = SHUTDOWN_MASK;
3074
3075 if (!sock_flag(sk, SOCK_DEAD))
3076 sk->sk_state_change(sk);
3077 else
3078 inet_csk_destroy_sock(sk);
3079}
3080EXPORT_SYMBOL_GPL(tcp_done);
3081
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003082extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
3084static __initdata unsigned long thash_entries;
3085static int __init set_thash_entries(char *str)
3086{
Eldad Zack413c27d2012-05-19 14:13:18 +00003087 ssize_t ret;
3088
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 if (!str)
3090 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00003091
3092 ret = kstrtoul(str, 0, &thash_entries);
3093 if (ret)
3094 return 0;
3095
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 return 1;
3097}
3098__setup("thash_entries=", set_thash_entries);
3099
Glauber Costa4acb4192012-01-30 01:20:17 +00003100void tcp_init_mem(struct net *net)
3101{
Glauber Costa4acb4192012-01-30 01:20:17 +00003102 unsigned long limit = nr_free_buffer_pages() / 8;
3103 limit = max(limit, 128UL);
3104 net->ipv4.sysctl_tcp_mem[0] = limit / 4 * 3;
3105 net->ipv4.sysctl_tcp_mem[1] = limit;
3106 net->ipv4.sysctl_tcp_mem[2] = net->ipv4.sysctl_tcp_mem[0] * 2;
3107}
3108
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109void __init tcp_init(void)
3110{
3111 struct sk_buff *skb = NULL;
Eric Dumazetf03d78d2011-07-07 00:27:05 -07003112 unsigned long limit;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003113 int max_rshare, max_wshare, cnt;
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003114 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08003116 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117
Eric Dumazet17483762008-11-25 21:16:35 -08003118 percpu_counter_init(&tcp_sockets_allocated, 0);
Eric Dumazetdd24c002008-11-25 21:17:14 -08003119 percpu_counter_init(&tcp_orphan_count, 0);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003120 tcp_hashinfo.bind_bucket_cachep =
3121 kmem_cache_create("tcp_bind_bucket",
3122 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003123 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 /* Size and allocate the main established and bind bucket
3126 * hash tables.
3127 *
3128 * The methodology is similar to that of the buffer cache.
3129 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003130 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003132 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 thash_entries,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003134 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003135 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 NULL,
Eric Dumazetf373b532009-10-09 00:16:19 +00003137 &tcp_hashinfo.ehash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003138 0,
Jean Delvare0ccfe612007-10-30 00:59:25 -07003139 thash_entries ? 0 : 512 * 1024);
Eric Dumazetf373b532009-10-09 00:16:19 +00003140 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) {
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08003141 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3142 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 }
Eric Dumazet230140c2007-11-07 02:40:20 -08003144 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3145 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003146 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003148 sizeof(struct inet_bind_hashbucket),
Eric Dumazetf373b532009-10-09 00:16:19 +00003149 tcp_hashinfo.ehash_mask + 1,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003150 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003151 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003152 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003154 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 64 * 1024);
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003156 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003157 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3158 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3159 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 }
3161
Eric Dumazetc5ed63d2010-08-25 23:02:17 -07003162
3163 cnt = tcp_hashinfo.ehash_mask + 1;
3164
3165 tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
3166 sysctl_tcp_max_orphans = cnt / 2;
3167 sysctl_max_syn_backlog = max(128, cnt / 256);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
Glauber Costa4acb4192012-01-30 01:20:17 +00003169 tcp_init_mem(&init_net);
Jason Wangc43b8742012-02-02 00:07:00 +00003170 /* Set per-socket limits to no more than 1/128 the pressure threshold */
Eric Dumazet5fb84b12012-04-10 00:56:42 +00003171 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
Eric Dumazetb49960a2012-05-02 02:28:41 +00003172 max_wshare = min(4UL*1024*1024, limit);
3173 max_rshare = min(6UL*1024*1024, limit);
John Heffner7b4f4b52006-03-25 01:34:07 -08003174
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003175 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003176 sysctl_tcp_wmem[1] = 16*1024;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003177 sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
John Heffner7b4f4b52006-03-25 01:34:07 -08003178
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003179 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003180 sysctl_tcp_rmem[1] = 87380;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003181 sysctl_tcp_rmem[2] = max(87380, max_rshare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182
Joe Perchesafd465032012-03-12 07:03:32 +00003183 pr_info("Hash tables configured (established %u bind %u)\n",
Joe Perches058bd4d2012-03-11 18:36:11 +00003184 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003185
David S. Miller51c5d0c2012-07-10 00:49:14 -07003186 tcp_metrics_init();
3187
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003188 tcp_register_congestion_control(&tcp_reno);
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003189
Eric Dumazet46d3cea2012-07-11 05:50:31 +00003190 tcp_tasklet_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191}