blob: 4475b3bb494d5d126dd844b0159629cd672afbdc [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 Dumazetf54b3112013-12-05 22:36:05 -0800288int sysctl_tcp_autocorking __read_mostly = 1;
289
Eric Dumazetdd24c002008-11-25 21:17:14 -0800290struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700291EXPORT_SYMBOL_GPL(tcp_orphan_count);
292
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700293long sysctl_tcp_mem[3] __read_mostly;
David S. Millerb8059ea2006-03-25 01:36:56 -0800294int sysctl_tcp_wmem[3] __read_mostly;
295int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700297EXPORT_SYMBOL(sysctl_tcp_mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298EXPORT_SYMBOL(sysctl_tcp_rmem);
299EXPORT_SYMBOL(sysctl_tcp_wmem);
300
Eric Dumazet8d987e52010-11-09 23:24:26 +0000301atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800303
304/*
305 * Current number of TCP sockets.
306 */
307struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308EXPORT_SYMBOL(tcp_sockets_allocated);
309
310/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800311 * TCP splice context
312 */
313struct tcp_splice_state {
314 struct pipe_inode_info *pipe;
315 size_t len;
316 unsigned int flags;
317};
318
319/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * Pressure flag: try to collapse.
321 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800322 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * is strict, actions are advisory and have some latency.
324 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700325int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326EXPORT_SYMBOL(tcp_memory_pressure);
327
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700328void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
330 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700331 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 tcp_memory_pressure = 1;
333 }
334}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335EXPORT_SYMBOL(tcp_enter_memory_pressure);
336
Julian Anastasovb103cf32009-10-19 10:10:40 +0000337/* Convert seconds to retransmits based on initial and max timeout */
338static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
339{
340 u8 res = 0;
341
342 if (seconds > 0) {
343 int period = timeout;
344
345 res = 1;
346 while (seconds > period && res < 255) {
347 res++;
348 timeout <<= 1;
349 if (timeout > rto_max)
350 timeout = rto_max;
351 period += timeout;
352 }
353 }
354 return res;
355}
356
357/* Convert retransmits to seconds based on initial and max timeout */
358static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
359{
360 int period = 0;
361
362 if (retrans > 0) {
363 period = timeout;
364 while (--retrans) {
365 timeout <<= 1;
366 if (timeout > rto_max)
367 timeout = rto_max;
368 period += timeout;
369 }
370 }
371 return period;
372}
373
Neal Cardwell900f65d2012-04-19 09:55:21 +0000374/* Address-family independent initialization for a tcp_sock.
375 *
376 * NOTE: A lot of things set to zero explicitly by call to
377 * sk_alloc() so need not be done here.
378 */
379void tcp_init_sock(struct sock *sk)
380{
381 struct inet_connection_sock *icsk = inet_csk(sk);
382 struct tcp_sock *tp = tcp_sk(sk);
383
Eric Dumazet996b1752014-01-06 09:36:12 -0800384 __skb_queue_head_init(&tp->out_of_order_queue);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000385 tcp_init_xmit_timers(sk);
386 tcp_prequeue_init(tp);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000387 INIT_LIST_HEAD(&tp->tsq_node);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000388
389 icsk->icsk_rto = TCP_TIMEOUT_INIT;
390 tp->mdev = TCP_TIMEOUT_INIT;
391
392 /* So many TCP implementations out there (incorrectly) count the
393 * initial SYN frame in their delayed-ACK and congestion control
394 * algorithms that we must have the following bandaid to talk
395 * efficiently to them. -DaveM
396 */
397 tp->snd_cwnd = TCP_INIT_CWND;
398
399 /* See draft-stevens-tcpca-spec-01 for discussion of the
400 * initialization of these values.
401 */
402 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
403 tp->snd_cwnd_clamp = ~0;
404 tp->mss_cache = TCP_MSS_DEFAULT;
405
406 tp->reordering = sysctl_tcp_reordering;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000407 tcp_enable_early_retrans(tp);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000408 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
409
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000410 tp->tsoffset = 0;
411
Neal Cardwell900f65d2012-04-19 09:55:21 +0000412 sk->sk_state = TCP_CLOSE;
413
414 sk->sk_write_space = sk_stream_write_space;
415 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
416
417 icsk->icsk_sync_mss = tcp_sync_mss;
418
Neal Cardwell900f65d2012-04-19 09:55:21 +0000419 sk->sk_sndbuf = sysctl_tcp_wmem[1];
420 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
421
422 local_bh_disable();
423 sock_update_memcg(sk);
424 sk_sockets_allocated_inc(sk);
425 local_bh_enable();
426}
427EXPORT_SYMBOL(tcp_init_sock);
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 * Wait for a TCP event.
431 *
432 * Note that we don't need to lock the socket, as the upper poll layers
433 * take care of normal races (between the test and the event) and we don't
434 * go look at any of the socket buffers directly.
435 */
436unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
437{
438 unsigned int mask;
439 struct sock *sk = sock->sk;
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400440 const struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
David Majnemerc3f1dba2013-05-31 13:15:38 +0000442 sock_rps_record_flow(sk);
443
Eric Dumazetaa395142010-04-20 13:03:51 +0000444 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700446 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700449 * by poll logic and correct handling of state changes
450 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 */
452
453 mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 /*
456 * POLLHUP is certainly not done right. But poll() doesn't
457 * have a notion of HUP in just one direction, and for a
458 * socket the read side is more interesting.
459 *
460 * Some poll() documentation says that POLLHUP is incompatible
461 * with the POLLOUT/POLLWR flags, so somebody should check this
462 * all. But careful, it tends to be safer to return too many
463 * bits than too few, and you can easily break real applications
464 * if you don't tell them that something has hung up!
465 *
466 * Check-me.
467 *
468 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
469 * our fs/select.c). It means that after we received EOF,
470 * poll always returns immediately, making impossible poll() on write()
471 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
472 * if and only if shutdown has been made in both directions.
473 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700474 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 * then we could set it on SND_SHUTDOWN. BTW examples given
476 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700477 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 *
479 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
480 * blocking on fresh not-connected or disconnected socket. --ANK
481 */
482 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
483 mask |= POLLHUP;
484 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800485 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Jerry Chu83368862012-08-31 12:29:12 +0000487 /* Connected or passive Fast Open socket? */
488 if (sk->sk_state != TCP_SYN_SENT &&
489 (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) {
David S. Millerc7004482008-10-06 10:43:54 -0700490 int target = sock_rcvlowat(sk, 0, INT_MAX);
491
492 if (tp->urg_seq == tp->copied_seq &&
493 !sock_flag(sk, SOCK_URGINLINE) &&
494 tp->urg_data)
Alexandra Kossovskyb634f872010-03-18 20:29:24 -0700495 target++;
David S. Millerc7004482008-10-06 10:43:54 -0700496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 /* Potential race condition. If read of tp below will
498 * escape above sk->sk_state, we can be illegally awaken
499 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700500 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 mask |= POLLIN | POLLRDNORM;
502
503 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
Eric Dumazet64dc6132013-07-22 20:26:31 -0700504 if (sk_stream_is_writeable(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 mask |= POLLOUT | POLLWRNORM;
506 } else { /* send SIGIO later */
507 set_bit(SOCK_ASYNC_NOSPACE,
508 &sk->sk_socket->flags);
509 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
510
511 /* Race breaker. If space is freed after
512 * wspace test but before the flags are set,
513 * IO signal will be lost.
514 */
Eric Dumazet64dc6132013-07-22 20:26:31 -0700515 if (sk_stream_is_writeable(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 mask |= POLLOUT | POLLWRNORM;
517 }
KOSAKI Motohirod84ba632010-08-24 16:05:48 +0000518 } else
519 mask |= POLLOUT | POLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521 if (tp->urg_data & TCP_URG_VALID)
522 mask |= POLLPRI;
523 }
Tom Marshalla4d25802010-09-20 15:42:05 -0700524 /* This barrier is coupled with smp_wmb() in tcp_reset() */
525 smp_rmb();
526 if (sk->sk_err)
527 mask |= POLLERR;
528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 return mask;
530}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000531EXPORT_SYMBOL(tcp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
534{
535 struct tcp_sock *tp = tcp_sk(sk);
536 int answ;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000537 bool slow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
539 switch (cmd) {
540 case SIOCINQ:
541 if (sk->sk_state == TCP_LISTEN)
542 return -EINVAL;
543
Eric Dumazet0e71c552012-10-21 20:06:56 +0000544 slow = lock_sock_fast(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
546 answ = 0;
547 else if (sock_flag(sk, SOCK_URGINLINE) ||
548 !tp->urg_data ||
549 before(tp->urg_seq, tp->copied_seq) ||
550 !before(tp->urg_seq, tp->rcv_nxt)) {
David S. Miller91521942009-05-28 21:35:47 -0700551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 answ = tp->rcv_nxt - tp->copied_seq;
553
Eric Dumazeta3374c42012-10-18 09:14:12 +0000554 /* Subtract 1, if FIN was received */
555 if (answ && sock_flag(sk, SOCK_DONE))
556 answ--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 } else
558 answ = tp->urg_seq - tp->copied_seq;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000559 unlock_sock_fast(sk, slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 break;
561 case SIOCATMARK:
562 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
563 break;
564 case SIOCOUTQ:
565 if (sk->sk_state == TCP_LISTEN)
566 return -EINVAL;
567
568 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
569 answ = 0;
570 else
571 answ = tp->write_seq - tp->snd_una;
572 break;
Mario Schuknecht2f4e1b32011-03-09 14:08:09 -0800573 case SIOCOUTQNSD:
574 if (sk->sk_state == TCP_LISTEN)
575 return -EINVAL;
576
577 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
578 answ = 0;
579 else
580 answ = tp->write_seq - tp->snd_nxt;
581 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 default:
583 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 return put_user(answ, (int __user *)arg);
587}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000588EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
591{
Eric Dumazet4de075e2011-09-27 13:25:05 -0400592 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 tp->pushed_seq = tp->write_seq;
594}
595
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000596static inline bool forced_push(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
599}
600
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700601static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700603 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200604 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
605
606 skb->csum = 0;
607 tcb->seq = tcb->end_seq = tp->write_seq;
Eric Dumazet4de075e2011-09-27 13:25:05 -0400608 tcb->tcp_flags = TCPHDR_ACK;
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200609 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800611 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800612 sk->sk_wmem_queued += skb->truesize;
613 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700614 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900615 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
Krishna Kumarafeca342009-12-10 07:16:52 +0000618static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700620 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
Eric Dumazetf54b3112013-12-05 22:36:05 -0800624/* If a not yet filled skb is pushed, do not send it if
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800625 * we have data packets in Qdisc or NIC queues :
Eric Dumazetf54b3112013-12-05 22:36:05 -0800626 * Because TX completion will happen shortly, it gives a chance
627 * to coalesce future sendmsg() payload into this skb, without
628 * need for a timer, and with no latency trade off.
629 * As packets containing data payload have a bigger truesize
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800630 * than pure acks (dataless) packets, the last checks prevent
631 * autocorking if we only have an ACK in Qdisc/NIC queues,
632 * or if TX completion was delayed after we processed ACK packet.
Eric Dumazetf54b3112013-12-05 22:36:05 -0800633 */
634static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
635 int size_goal)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Eric Dumazetf54b3112013-12-05 22:36:05 -0800637 return skb->len < size_goal &&
638 sysctl_tcp_autocorking &&
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800639 skb != tcp_write_queue_head(sk) &&
Eric Dumazetf54b3112013-12-05 22:36:05 -0800640 atomic_read(&sk->sk_wmem_alloc) > skb->truesize;
641}
Krishna Kumarafeca342009-12-10 07:16:52 +0000642
Eric Dumazetf54b3112013-12-05 22:36:05 -0800643static void tcp_push(struct sock *sk, int flags, int mss_now,
644 int nonagle, int size_goal)
645{
646 struct tcp_sock *tp = tcp_sk(sk);
647 struct sk_buff *skb;
Krishna Kumarafeca342009-12-10 07:16:52 +0000648
Eric Dumazetf54b3112013-12-05 22:36:05 -0800649 if (!tcp_send_head(sk))
650 return;
651
652 skb = tcp_write_queue_tail(sk);
653 if (!(flags & MSG_MORE) || forced_push(tp))
654 tcp_mark_push(tp, skb);
655
656 tcp_mark_urg(tp, flags);
657
658 if (tcp_should_autocork(sk, skb, size_goal)) {
659
660 /* avoid atomic op if TSQ_THROTTLED bit is already set */
661 if (!test_bit(TSQ_THROTTLED, &tp->tsq_flags)) {
662 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
663 set_bit(TSQ_THROTTLED, &tp->tsq_flags);
664 }
Eric Dumazeta181ceb2013-12-17 09:58:30 -0800665 /* It is possible TX completion already happened
666 * before we set TSQ_THROTTLED.
667 */
668 if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize)
669 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
Eric Dumazetf54b3112013-12-05 22:36:05 -0800671
672 if (flags & MSG_MORE)
673 nonagle = TCP_NAGLE_CORK;
674
675 __tcp_push_pending_frames(sk, mss_now, nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Adrian Bunk6ff77512007-11-06 23:32:26 -0800678static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
679 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800680{
681 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd2009-01-13 16:04:36 -0800682 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800683
Dimitris Michailidis9fa5fdf2009-01-26 22:15:31 -0800684 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
685 tss->flags);
Willy Tarreau33966dd2009-01-13 16:04:36 -0800686 if (ret > 0)
687 rd_desc->count -= ret;
688 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800689}
690
691static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
692{
693 /* Store TCP splice context information in read_descriptor_t. */
694 read_descriptor_t rd_desc = {
695 .arg.data = tss,
Willy Tarreau33966dd2009-01-13 16:04:36 -0800696 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800697 };
698
699 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
700}
701
702/**
703 * tcp_splice_read - splice data from TCP socket to a pipe
704 * @sock: socket to splice from
705 * @ppos: position (not valid)
706 * @pipe: pipe to splice to
707 * @len: number of bytes to splice
708 * @flags: splice modifier flags
709 *
710 * Description:
711 * Will read pages from given socket and fill them into a pipe.
712 *
713 **/
714ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
715 struct pipe_inode_info *pipe, size_t len,
716 unsigned int flags)
717{
718 struct sock *sk = sock->sk;
719 struct tcp_splice_state tss = {
720 .pipe = pipe,
721 .len = len,
722 .flags = flags,
723 };
724 long timeo;
725 ssize_t spliced;
726 int ret;
727
Changli Gao3a047bf2010-07-12 21:00:12 +0000728 sock_rps_record_flow(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -0800729 /*
730 * We can't seek on a socket input
731 */
732 if (unlikely(*ppos))
733 return -ESPIPE;
734
735 ret = spliced = 0;
736
737 lock_sock(sk);
738
Eric Dumazet42324c62009-10-01 15:26:00 -0700739 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
Jens Axboe9c55e012007-11-06 23:30:13 -0800740 while (tss.len) {
741 ret = __tcp_splice_read(sk, &tss);
742 if (ret < 0)
743 break;
744 else if (!ret) {
745 if (spliced)
746 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800747 if (sock_flag(sk, SOCK_DONE))
748 break;
749 if (sk->sk_err) {
750 ret = sock_error(sk);
751 break;
752 }
753 if (sk->sk_shutdown & RCV_SHUTDOWN)
754 break;
755 if (sk->sk_state == TCP_CLOSE) {
756 /*
757 * This occurs when user tries to read
758 * from never connected socket.
759 */
760 if (!sock_flag(sk, SOCK_DONE))
761 ret = -ENOTCONN;
762 break;
763 }
764 if (!timeo) {
765 ret = -EAGAIN;
766 break;
767 }
768 sk_wait_data(sk, &timeo);
769 if (signal_pending(current)) {
770 ret = sock_intr_errno(timeo);
771 break;
772 }
773 continue;
774 }
775 tss.len -= ret;
776 spliced += ret;
777
Willy Tarreau33966dd2009-01-13 16:04:36 -0800778 if (!timeo)
779 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800780 release_sock(sk);
781 lock_sock(sk);
782
783 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd2009-01-13 16:04:36 -0800784 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800785 signal_pending(current))
786 break;
787 }
788
789 release_sock(sk);
790
791 if (spliced)
792 return spliced;
793
794 return ret;
795}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000796EXPORT_SYMBOL(tcp_splice_read);
Jens Axboe9c55e012007-11-06 23:30:13 -0800797
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100798struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100799{
800 struct sk_buff *skb;
801
802 /* The TCP header must be at least 32-bit aligned. */
803 size = ALIGN(size, 4);
804
805 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
806 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800807 if (sk_wmem_schedule(sk, skb->truesize)) {
Eric Dumazeta21d4572012-04-10 20:30:48 +0000808 skb_reserve(skb, sk->sk_prot->max_header);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100809 /*
810 * Make sure that we have exactly size bytes
811 * available to the caller, no more, no less.
812 */
Eric Dumazet16fad692013-03-14 05:40:32 +0000813 skb->reserved_tailroom = skb->end - skb->tail - size;
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100814 return skb;
815 }
816 __kfree_skb(skb);
817 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700818 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100819 sk_stream_moderate_sndbuf(sk);
820 }
821 return NULL;
822}
823
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000824static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
825 int large_allowed)
826{
827 struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000828 u32 xmit_size_goal, old_size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000829
830 xmit_size_goal = mss_now;
831
832 if (large_allowed && sk_can_gso(sk)) {
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700833 u32 gso_size, hlen;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000834
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700835 /* Maybe we should/could use sk->sk_prot->max_header here ? */
836 hlen = inet_csk(sk)->icsk_af_ops->net_header_len +
837 inet_csk(sk)->icsk_ext_hdr_len +
838 tp->tcp_header_len;
839
840 /* Goal is to send at least one packet per ms,
841 * not one big TSO packet every 100 ms.
842 * This preserves ACK clocking and is consistent
843 * with tcp_tso_should_defer() heuristic.
844 */
845 gso_size = sk->sk_pacing_rate / (2 * MSEC_PER_SEC);
846 gso_size = max_t(u32, gso_size,
847 sysctl_tcp_min_tso_segs * mss_now);
848
849 xmit_size_goal = min_t(u32, gso_size,
850 sk->sk_gso_max_size - 1 - hlen);
851
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000852 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000853
854 /* We try hard to avoid divides here */
855 old_size_goal = tp->xmit_size_goal_segs * mss_now;
856
857 if (likely(old_size_goal <= xmit_size_goal &&
858 old_size_goal + mss_now > xmit_size_goal)) {
859 xmit_size_goal = old_size_goal;
860 } else {
Ben Hutchings14853482012-07-30 16:11:42 +0000861 tp->xmit_size_goal_segs =
862 min_t(u16, xmit_size_goal / mss_now,
863 sk->sk_gso_max_segs);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000864 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
865 }
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000866 }
867
Ilpo Järvinenafece1c2009-03-14 14:23:07 +0000868 return max(xmit_size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000869}
870
871static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
872{
873 int mss_now;
874
875 mss_now = tcp_current_mss(sk);
876 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
877
878 return mss_now;
879}
880
Eric Dumazet64022d02012-12-01 13:07:02 +0000881static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
882 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
884 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700885 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 int err;
887 ssize_t copied;
888 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
889
Jerry Chu83368862012-08-31 12:29:12 +0000890 /* Wait for a connection to finish. One exception is TCP Fast Open
891 * (passive side) where data is allowed to be sent before a connection
892 * is fully established.
893 */
894 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
895 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
897 goto out_err;
Jerry Chu83368862012-08-31 12:29:12 +0000898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
901
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000902 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 copied = 0;
904
905 err = -EPIPE;
906 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000907 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Eric Dumazet64022d02012-12-01 13:07:02 +0000909 while (size > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800910 struct sk_buff *skb = tcp_write_queue_tail(sk);
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000911 int copy, i;
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000912 bool can_coalesce;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
David S. Millerfe067e82007-03-07 12:12:44 -0800914 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915new_segment:
916 if (!sk_stream_memory_free(sk))
917 goto wait_for_sndbuf;
918
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100919 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 if (!skb)
921 goto wait_for_memory;
922
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700923 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700924 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 }
926
927 if (copy > size)
928 copy = size;
929
930 i = skb_shinfo(skb)->nr_frags;
931 can_coalesce = skb_can_coalesce(skb, i, page, offset);
932 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
933 tcp_mark_push(tp, skb);
934 goto new_segment;
935 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800936 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (can_coalesce) {
Eric Dumazet9e903e02011-10-18 21:00:24 +0000940 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 } else {
942 get_page(page);
943 skb_fill_page_desc(skb, i, page, offset, copy);
944 }
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000945 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +0000946
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 skb->len += copy;
948 skb->data_len += copy;
949 skb->truesize += copy;
950 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800951 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700952 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 tp->write_seq += copy;
954 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700955 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -0400958 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 copied += copy;
Eric Dumazet64022d02012-12-01 13:07:02 +0000961 offset += copy;
962 if (!(size -= copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 goto out;
964
Herbert Xu69d15062008-03-22 15:47:05 -0700965 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 continue;
967
968 if (forced_push(tp)) {
969 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700970 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800971 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 tcp_push_one(sk, mss_now);
973 continue;
974
975wait_for_sndbuf:
976 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
977wait_for_memory:
Eric Dumazetf54b3112013-12-05 22:36:05 -0800978 tcp_push(sk, flags & ~MSG_MORE, mss_now,
979 TCP_NAGLE_PUSH, size_goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
982 goto do_error;
983
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000984 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 }
986
987out:
Eric Dumazet35f9c092012-04-05 03:05:35 +0000988 if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
Eric Dumazetf54b3112013-12-05 22:36:05 -0800989 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return copied;
991
992do_error:
993 if (copied)
994 goto out;
995out_err:
996 return sk_stream_error(sk, flags, err);
997}
998
Changli Gao7ba42912010-07-10 20:41:55 +0000999int tcp_sendpage(struct sock *sk, struct page *page, int offset,
1000 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
1002 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -07001005 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Changli Gao7ba42912010-07-10 20:41:55 +00001006 return sock_no_sendpage(sk->sk_socket, page, offset, size,
1007 flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 lock_sock(sk);
Eric Dumazet64022d02012-12-01 13:07:02 +00001010 res = do_tcp_sendpages(sk, page, offset, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 release_sock(sk);
1012 return res;
1013}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001014EXPORT_SYMBOL(tcp_sendpage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Eric Dumazet690e99c2011-11-28 00:27:47 +00001016static inline int select_size(const struct sock *sk, bool sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001018 const struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001019 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Krishna Kumardef87cf2009-12-10 07:16:59 +00001021 if (sg) {
Eric Dumazetf07d9602011-11-28 22:41:47 +00001022 if (sk_can_gso(sk)) {
1023 /* Small frames wont use a full page:
1024 * Payload will immediately follow tcp header.
1025 */
1026 tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
1027 } else {
David S. Millerb4e26f52005-07-05 15:20:27 -07001028 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
1029
1030 if (tmp >= pgbreak &&
1031 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
1032 tmp = pgbreak;
1033 }
1034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return tmp;
1037}
1038
Yuchung Chengcf60af02012-07-19 06:43:09 +00001039void tcp_free_fastopen_req(struct tcp_sock *tp)
1040{
1041 if (tp->fastopen_req != NULL) {
1042 kfree(tp->fastopen_req);
1043 tp->fastopen_req = NULL;
1044 }
1045}
1046
1047static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size)
1048{
1049 struct tcp_sock *tp = tcp_sk(sk);
1050 int err, flags;
1051
1052 if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
1053 return -EOPNOTSUPP;
1054 if (tp->fastopen_req != NULL)
1055 return -EALREADY; /* Another Fast Open is in progress */
1056
1057 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1058 sk->sk_allocation);
1059 if (unlikely(tp->fastopen_req == NULL))
1060 return -ENOBUFS;
1061 tp->fastopen_req->data = msg;
1062
1063 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1064 err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1065 msg->msg_namelen, flags);
1066 *size = tp->fastopen_req->copied;
1067 tcp_free_fastopen_req(tp);
1068 return err;
1069}
1070
Changli Gao7ba42912010-07-10 20:41:55 +00001071int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 size_t size)
1073{
1074 struct iovec *iov;
1075 struct tcp_sock *tp = tcp_sk(sk);
1076 struct sk_buff *skb;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001077 int iovlen, flags, err, copied = 0;
1078 int mss_now = 0, size_goal, copied_syn = 0, offset = 0;
Eric Dumazet690e99c2011-11-28 00:27:47 +00001079 bool sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 long timeo;
1081
1082 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084 flags = msg->msg_flags;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001085 if (flags & MSG_FASTOPEN) {
1086 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn);
1087 if (err == -EINPROGRESS && copied_syn > 0)
1088 goto out;
1089 else if (err)
1090 goto out_err;
1091 offset = copied_syn;
1092 }
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1095
Jerry Chu83368862012-08-31 12:29:12 +00001096 /* Wait for a connection to finish. One exception is TCP Fast Open
1097 * (passive side) where data is allowed to be sent before a connection
1098 * is fully established.
1099 */
1100 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1101 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001103 goto do_error;
Jerry Chu83368862012-08-31 12:29:12 +00001104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001106 if (unlikely(tp->repair)) {
1107 if (tp->repair_queue == TCP_RECV_QUEUE) {
1108 copied = tcp_send_rcvq(sk, msg, size);
1109 goto out;
1110 }
1111
1112 err = -EINVAL;
1113 if (tp->repair_queue == TCP_NO_QUEUE)
1114 goto out_err;
1115
1116 /* 'common' sending to sendq */
1117 }
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 /* This should be in poll */
1120 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1121
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001122 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124 /* Ok commence sending. */
1125 iovlen = msg->msg_iovlen;
1126 iov = msg->msg_iov;
1127 copied = 0;
1128
1129 err = -EPIPE;
1130 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +00001131 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Eric Dumazet690e99c2011-11-28 00:27:47 +00001133 sg = !!(sk->sk_route_caps & NETIF_F_SG);
Krishna Kumardef87cf2009-12-10 07:16:59 +00001134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 while (--iovlen >= 0) {
David S. Miller01db4032010-09-27 20:24:54 -07001136 size_t seglen = iov->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 unsigned char __user *from = iov->iov_base;
1138
1139 iov++;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001140 if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */
1141 if (offset >= seglen) {
1142 offset -= seglen;
1143 continue;
1144 }
1145 seglen -= offset;
1146 from += offset;
1147 offset = 0;
1148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
1150 while (seglen > 0) {
Herbert Xu6828b922009-06-28 18:06:41 +00001151 int copy = 0;
1152 int max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
David S. Millerfe067e82007-03-07 12:12:44 -08001154 skb = tcp_write_queue_tail(sk);
Herbert Xu6828b922009-06-28 18:06:41 +00001155 if (tcp_send_head(sk)) {
1156 if (skb->ip_summed == CHECKSUM_NONE)
1157 max = mss_now;
1158 copy = max - skb->len;
1159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Herbert Xu6828b922009-06-28 18:06:41 +00001161 if (copy <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162new_segment:
1163 /* Allocate new segment. If the interface is SG,
1164 * allocate skb fitting to single page.
1165 */
1166 if (!sk_stream_memory_free(sk))
1167 goto wait_for_sndbuf;
1168
Krishna Kumardef87cf2009-12-10 07:16:59 +00001169 skb = sk_stream_alloc_skb(sk,
1170 select_size(sk, sg),
1171 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (!skb)
1173 goto wait_for_memory;
1174
1175 /*
Andrey Vagin7ed5c5a2013-08-16 19:04:36 +04001176 * All packets are restored as if they have
1177 * already been sent.
1178 */
1179 if (tp->repair)
1180 TCP_SKB_CB(skb)->when = tcp_time_stamp;
1181
1182 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 * Check whether we can use HW checksum.
1184 */
Herbert Xu8648b302006-06-17 22:06:05 -07001185 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -07001186 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001188 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001189 copy = size_goal;
Herbert Xu6828b922009-06-28 18:06:41 +00001190 max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
1192
1193 /* Try to append data to the end of skb. */
1194 if (copy > seglen)
1195 copy = seglen;
1196
1197 /* Where to copy to? */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001198 if (skb_availroom(skb) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 /* We have some space in skb head. Superb! */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001200 copy = min_t(int, copy, skb_availroom(skb));
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001201 err = skb_add_data_nocache(sk, skb, from, copy);
1202 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 goto do_fault;
1204 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001205 bool merge = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 int i = skb_shinfo(skb)->nr_frags;
Eric Dumazet5640f762012-09-23 23:04:42 +00001207 struct page_frag *pfrag = sk_page_frag(sk);
Eric Dumazet761965e2011-12-04 07:05:17 +00001208
Eric Dumazet5640f762012-09-23 23:04:42 +00001209 if (!sk_page_frag_refill(sk, pfrag))
1210 goto wait_for_memory;
Eric Dumazet761965e2011-12-04 07:05:17 +00001211
Eric Dumazet5640f762012-09-23 23:04:42 +00001212 if (!skb_can_coalesce(skb, i, pfrag->page,
1213 pfrag->offset)) {
1214 if (i == MAX_SKB_FRAGS || !sg) {
1215 tcp_mark_push(tp, skb);
1216 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001218 merge = false;
1219 }
Herbert Xuef015782005-09-01 17:48:59 -07001220
Eric Dumazet5640f762012-09-23 23:04:42 +00001221 copy = min_t(int, copy, pfrag->size - pfrag->offset);
Herbert Xuef015782005-09-01 17:48:59 -07001222
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001223 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -07001224 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001226 err = skb_copy_to_page_nocache(sk, from, skb,
Eric Dumazet5640f762012-09-23 23:04:42 +00001227 pfrag->page,
1228 pfrag->offset,
1229 copy);
1230 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 goto do_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 /* Update the skb. */
1234 if (merge) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001235 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001237 skb_fill_page_desc(skb, i, pfrag->page,
1238 pfrag->offset, copy);
1239 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001241 pfrag->offset += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
1243
1244 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -04001245 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 tp->write_seq += copy;
1248 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -07001249 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
1251 from += copy;
1252 copied += copy;
1253 if ((seglen -= copy) == 0 && iovlen == 0)
1254 goto out;
1255
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001256 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 continue;
1258
1259 if (forced_push(tp)) {
1260 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001261 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001262 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 tcp_push_one(sk, mss_now);
1264 continue;
1265
1266wait_for_sndbuf:
1267 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1268wait_for_memory:
Andrew Vaginec342322012-11-15 04:03:17 +00001269 if (copied)
Eric Dumazetf54b3112013-12-05 22:36:05 -08001270 tcp_push(sk, flags & ~MSG_MORE, mss_now,
1271 TCP_NAGLE_PUSH, size_goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1274 goto do_error;
1275
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001276 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 }
1278 }
1279
1280out:
Andrew Vaginec342322012-11-15 04:03:17 +00001281 if (copied)
Eric Dumazetf54b3112013-12-05 22:36:05 -08001282 tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 release_sock(sk);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001284 return copied + copied_syn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286do_fault:
1287 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001288 tcp_unlink_write_queue(skb, sk);
1289 /* It is the one place in all of TCP, except connection
1290 * reset, where we can be unlinking the send_head.
1291 */
1292 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001293 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295
1296do_error:
Yuchung Chengcf60af02012-07-19 06:43:09 +00001297 if (copied + copied_syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 goto out;
1299out_err:
1300 err = sk_stream_error(sk, flags, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 release_sock(sk);
1302 return err;
1303}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001304EXPORT_SYMBOL(tcp_sendmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306/*
1307 * Handle reading urgent data. BSD has very simple semantics for
1308 * this, no blocking and very strange errors 8)
1309 */
1310
Rami Rosen377f0a02009-03-31 14:43:17 -07001311static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
1313 struct tcp_sock *tp = tcp_sk(sk);
1314
1315 /* No URG data to read. */
1316 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1317 tp->urg_data == TCP_URG_READ)
1318 return -EINVAL; /* Yes this is right ! */
1319
1320 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1321 return -ENOTCONN;
1322
1323 if (tp->urg_data & TCP_URG_VALID) {
1324 int err = 0;
1325 char c = tp->urg_data;
1326
1327 if (!(flags & MSG_PEEK))
1328 tp->urg_data = TCP_URG_READ;
1329
1330 /* Read urgent data. */
1331 msg->msg_flags |= MSG_OOB;
1332
1333 if (len > 0) {
1334 if (!(flags & MSG_TRUNC))
1335 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1336 len = 1;
1337 } else
1338 msg->msg_flags |= MSG_TRUNC;
1339
1340 return err ? -EFAULT : len;
1341 }
1342
1343 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1344 return 0;
1345
1346 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1347 * the available implementations agree in this case:
1348 * this call should never block, independent of the
1349 * blocking state of the socket.
1350 * Mike <pall@rz.uni-karlsruhe.de>
1351 */
1352 return -EAGAIN;
1353}
1354
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001355static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1356{
1357 struct sk_buff *skb;
1358 int copied = 0, err = 0;
1359
1360 /* XXX -- need to support SO_PEEK_OFF */
1361
1362 skb_queue_walk(&sk->sk_write_queue, skb) {
1363 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len);
1364 if (err)
1365 break;
1366
1367 copied += skb->len;
1368 }
1369
1370 return err ?: copied;
1371}
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373/* Clean up the receive buffer for full frames taken by the user,
1374 * then send an ACK if necessary. COPIED is the number of bytes
1375 * tcp_recvmsg has given to the user so far, it speeds up the
1376 * calculation of whether or not we must ACK for the sake of
1377 * a window update.
1378 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001379void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
1381 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001382 bool time_to_ack = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1385
Ilpo Järvinend792c102009-11-13 13:56:33 -08001386 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001387 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
Ilpo Järvinend792c102009-11-13 13:56:33 -08001388 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001390 if (inet_csk_ack_scheduled(sk)) {
1391 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 /* Delayed ACKs frequently hit locked sockets during bulk
1393 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001394 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001396 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 /*
1398 * If this read emptied read buffer, we send ACK, if
1399 * connection is not bidirectional, user drained
1400 * receive buffer and there was a small segment
1401 * in queue.
1402 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001403 (copied > 0 &&
1404 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1405 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1406 !icsk->icsk_ack.pingpong)) &&
1407 !atomic_read(&sk->sk_rmem_alloc)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001408 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 }
1410
1411 /* We send an ACK if we can now advertise a non-zero window
1412 * which has been raised "significantly".
1413 *
1414 * Even if window raised up to infinity, do not send window open ACK
1415 * in states, where we will not receive more. It is useless.
1416 */
1417 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1418 __u32 rcv_window_now = tcp_receive_window(tp);
1419
1420 /* Optimize, __tcp_select_window() is not cheap. */
1421 if (2*rcv_window_now <= tp->window_clamp) {
1422 __u32 new_window = __tcp_select_window(sk);
1423
1424 /* Send ACK now, if this read freed lots of space
1425 * in our buffer. Certainly, new_window is new window.
1426 * We can advertise it now, if it is not less than current one.
1427 * "Lots" means "at least twice" here.
1428 */
1429 if (new_window && new_window >= 2 * rcv_window_now)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001430 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
1432 }
1433 if (time_to_ack)
1434 tcp_send_ack(sk);
1435}
1436
1437static void tcp_prequeue_process(struct sock *sk)
1438{
1439 struct sk_buff *skb;
1440 struct tcp_sock *tp = tcp_sk(sk);
1441
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001442 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444 /* RX process wants to run with disabled BHs, though it is not
1445 * necessary */
1446 local_bh_disable();
1447 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001448 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 local_bh_enable();
1450
1451 /* Clear memory counter. */
1452 tp->ucopy.memory = 0;
1453}
1454
Steven J. Magnani73852e82010-03-16 05:22:44 +00001455#ifdef CONFIG_NET_DMA
1456static void tcp_service_net_dma(struct sock *sk, bool wait)
1457{
1458 dma_cookie_t done, used;
1459 dma_cookie_t last_issued;
1460 struct tcp_sock *tp = tcp_sk(sk);
1461
1462 if (!tp->ucopy.dma_chan)
1463 return;
1464
1465 last_issued = tp->ucopy.dma_cookie;
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001466 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001467
1468 do {
Bartlomiej Zolnierkiewicze239345f2012-11-08 10:01:01 +00001469 if (dma_async_is_tx_complete(tp->ucopy.dma_chan,
Steven J. Magnani73852e82010-03-16 05:22:44 +00001470 last_issued, &done,
Vinod Koul27bf6972013-10-16 21:07:39 +05301471 &used) == DMA_COMPLETE) {
Steven J. Magnani73852e82010-03-16 05:22:44 +00001472 /* Safe to free early-copied skbs now */
1473 __skb_queue_purge(&sk->sk_async_wait_queue);
1474 break;
1475 } else {
1476 struct sk_buff *skb;
1477 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1478 (dma_async_is_complete(skb->dma_cookie, done,
Vinod Koul27bf6972013-10-16 21:07:39 +05301479 used) == DMA_COMPLETE)) {
Steven J. Magnani73852e82010-03-16 05:22:44 +00001480 __skb_dequeue(&sk->sk_async_wait_queue);
1481 kfree_skb(skb);
1482 }
1483 }
1484 } while (wait);
1485}
1486#endif
1487
Eric Dumazetf26845b2013-01-09 20:59:09 +00001488static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
1490 struct sk_buff *skb;
1491 u32 offset;
1492
Eric Dumazetf26845b2013-01-09 20:59:09 +00001493 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001495 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001497 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 *off = offset;
1499 return skb;
1500 }
Eric Dumazetf26845b2013-01-09 20:59:09 +00001501 /* This looks weird, but this can happen if TCP collapsing
1502 * splitted a fat GRO packet, while we released socket lock
1503 * in skb_splice_bits()
1504 */
1505 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
1507 return NULL;
1508}
1509
1510/*
1511 * This routine provides an alternative to tcp_recvmsg() for routines
1512 * that would like to handle copying from skbuffs directly in 'sendfile'
1513 * fashion.
1514 * Note:
1515 * - It is assumed that the socket was locked by the caller.
1516 * - The routine does not block.
1517 * - At present, there is no support for reading OOB data
1518 * or for 'peeking' the socket using this routine
1519 * (although both would be easy to implement).
1520 */
1521int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1522 sk_read_actor_t recv_actor)
1523{
1524 struct sk_buff *skb;
1525 struct tcp_sock *tp = tcp_sk(sk);
1526 u32 seq = tp->copied_seq;
1527 u32 offset;
1528 int copied = 0;
1529
1530 if (sk->sk_state == TCP_LISTEN)
1531 return -ENOTCONN;
1532 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1533 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001534 int used;
1535 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
1537 len = skb->len - offset;
1538 /* Stop reading if we hit a patch of urgent data */
1539 if (tp->urg_data) {
1540 u32 urg_offset = tp->urg_seq - seq;
1541 if (urg_offset < len)
1542 len = urg_offset;
1543 if (!len)
1544 break;
1545 }
1546 used = recv_actor(desc, skb, offset, len);
Eric Dumazetff905b12013-01-10 07:06:10 +00001547 if (used <= 0) {
Jens Axboeddb61a52007-06-23 23:07:50 -07001548 if (!copied)
1549 copied = used;
1550 break;
1551 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 seq += used;
1553 copied += used;
1554 offset += used;
1555 }
Willy Tarreau02275a22012-12-02 11:49:27 +00001556 /* If recv_actor drops the lock (e.g. TCP splice
Octavian Purdila293ad602008-06-04 15:45:58 -07001557 * receive) the skb pointer might be invalid when
1558 * getting here: tcp_collapse might have deleted it
1559 * while aggregating skbs from the socket queue.
1560 */
Willy Tarreau02275a22012-12-02 11:49:27 +00001561 skb = tcp_recv_skb(sk, seq - 1, &offset);
1562 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 break;
Willy Tarreau02275a22012-12-02 11:49:27 +00001564 /* TCP coalescing might have appended data to the skb.
1565 * Try to splice more frags
1566 */
1567 if (offset + 1 != skb->len)
1568 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001570 if (tcp_hdr(skb)->fin) {
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001571 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 ++seq;
1573 break;
1574 }
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001575 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 if (!desc->count)
1577 break;
Steven J. Magnanibaff42a2010-03-30 13:56:01 -07001578 tp->copied_seq = seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 }
1580 tp->copied_seq = seq;
1581
1582 tcp_rcv_space_adjust(sk);
1583
1584 /* Clean up data we have read: This will do ACK frames. */
Eric Dumazetf26845b2013-01-09 20:59:09 +00001585 if (copied > 0) {
1586 tcp_recv_skb(sk, seq, &offset);
Chris Leech0e4b4992006-05-23 18:00:16 -07001587 tcp_cleanup_rbuf(sk, copied);
Eric Dumazetf26845b2013-01-09 20:59:09 +00001588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 return copied;
1590}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001591EXPORT_SYMBOL(tcp_read_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593/*
1594 * This routine copies from a sock struct into the user buffer.
1595 *
1596 * Technical note: in 2.3 we work on _locked_ socket, so that
1597 * tricks with *seq access order and skb->users are not required.
1598 * Probably, code can be easily improved even more.
1599 */
1600
1601int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1602 size_t len, int nonblock, int flags, int *addr_len)
1603{
1604 struct tcp_sock *tp = tcp_sk(sk);
1605 int copied = 0;
1606 u32 peek_seq;
1607 u32 *seq;
1608 unsigned long used;
1609 int err;
1610 int target; /* Read at least this many bytes */
1611 long timeo;
1612 struct task_struct *user_recv = NULL;
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001613 bool copied_early = false;
Chris Leech2b1244a2007-03-08 09:57:36 -08001614 struct sk_buff *skb;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001615 u32 urg_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
Eliezer Tamircbf55002013-07-08 16:20:34 +03001617 if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
1618 (sk->sk_state == TCP_ESTABLISHED))
1619 sk_busy_loop(sk, nonblock);
Eliezer Tamird30e3832013-06-10 11:40:10 +03001620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 lock_sock(sk);
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 err = -ENOTCONN;
1624 if (sk->sk_state == TCP_LISTEN)
1625 goto out;
1626
1627 timeo = sock_rcvtimeo(sk, nonblock);
1628
1629 /* Urgent data needs to be handled specially. */
1630 if (flags & MSG_OOB)
1631 goto recv_urg;
1632
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001633 if (unlikely(tp->repair)) {
1634 err = -EPERM;
1635 if (!(flags & MSG_PEEK))
1636 goto out;
1637
1638 if (tp->repair_queue == TCP_SEND_QUEUE)
1639 goto recv_sndq;
1640
1641 err = -EINVAL;
1642 if (tp->repair_queue == TCP_NO_QUEUE)
1643 goto out;
1644
1645 /* 'common' recv queue MSG_PEEK-ing */
1646 }
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 seq = &tp->copied_seq;
1649 if (flags & MSG_PEEK) {
1650 peek_seq = tp->copied_seq;
1651 seq = &peek_seq;
1652 }
1653
1654 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1655
Chris Leech1a2449a2006-05-23 18:05:53 -07001656#ifdef CONFIG_NET_DMA
1657 tp->ucopy.dma_chan = NULL;
1658 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001659 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001660 {
1661 int available = 0;
1662
1663 if (skb)
1664 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1665 if ((available < target) &&
1666 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1667 !sysctl_tcp_low_latency &&
Dave Jianga2bd1142012-04-04 16:10:46 -07001668 net_dma_find_channel()) {
Peter Zijlstra1774e9f2013-11-19 16:13:38 +01001669 preempt_enable();
Andrew Mortone00c5d82007-03-08 09:57:36 -08001670 tp->ucopy.pinned_list =
1671 dma_pin_iovec_pages(msg->msg_iov, len);
1672 } else {
Peter Zijlstra1774e9f2013-11-19 16:13:38 +01001673 preempt_enable();
Andrew Mortone00c5d82007-03-08 09:57:36 -08001674 }
1675 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001676#endif
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 u32 offset;
1680
1681 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1682 if (tp->urg_data && tp->urg_seq == *seq) {
1683 if (copied)
1684 break;
1685 if (signal_pending(current)) {
1686 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1687 break;
1688 }
1689 }
1690
1691 /* Next get a buffer. */
1692
David S. Miller91521942009-05-28 21:35:47 -07001693 skb_queue_walk(&sk->sk_receive_queue, skb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 /* Now that we have two receive queues this
1695 * shouldn't happen.
1696 */
Ilpo Järvinend792c102009-11-13 13:56:33 -08001697 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001698 "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1699 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1700 flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 break;
Ilpo Järvinend792c102009-11-13 13:56:33 -08001702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001704 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 offset--;
1706 if (offset < skb->len)
1707 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001708 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 goto found_fin_ok;
Joe Perches2af6fd82010-10-30 11:08:53 +00001710 WARN(!(flags & MSG_PEEK),
1711 "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1712 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
David S. Miller91521942009-05-28 21:35:47 -07001713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715 /* Well, if we have backlog, try to process it now yet. */
1716
1717 if (copied >= target && !sk->sk_backlog.tail)
1718 break;
1719
1720 if (copied) {
1721 if (sk->sk_err ||
1722 sk->sk_state == TCP_CLOSE ||
1723 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1724 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08001725 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 break;
1727 } else {
1728 if (sock_flag(sk, SOCK_DONE))
1729 break;
1730
1731 if (sk->sk_err) {
1732 copied = sock_error(sk);
1733 break;
1734 }
1735
1736 if (sk->sk_shutdown & RCV_SHUTDOWN)
1737 break;
1738
1739 if (sk->sk_state == TCP_CLOSE) {
1740 if (!sock_flag(sk, SOCK_DONE)) {
1741 /* This occurs when user tries to read
1742 * from never connected socket.
1743 */
1744 copied = -ENOTCONN;
1745 break;
1746 }
1747 break;
1748 }
1749
1750 if (!timeo) {
1751 copied = -EAGAIN;
1752 break;
1753 }
1754
1755 if (signal_pending(current)) {
1756 copied = sock_intr_errno(timeo);
1757 break;
1758 }
1759 }
1760
Chris Leech0e4b4992006-05-23 18:00:16 -07001761 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
David S. Miller7df55122005-06-18 23:01:10 -07001763 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 /* Install new reader */
1765 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1766 user_recv = current;
1767 tp->ucopy.task = user_recv;
1768 tp->ucopy.iov = msg->msg_iov;
1769 }
1770
1771 tp->ucopy.len = len;
1772
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001773 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1774 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 /* Ugly... If prequeue is not empty, we have to
1777 * process it before releasing socket, otherwise
1778 * order will be broken at second iteration.
1779 * More elegant solution is required!!!
1780 *
1781 * Look: we have the following (pseudo)queues:
1782 *
1783 * 1. packets in flight
1784 * 2. backlog
1785 * 3. prequeue
1786 * 4. receive_queue
1787 *
1788 * Each queue can be processed only if the next ones
1789 * are empty. At this point we have empty receive_queue.
1790 * But prequeue _can_ be not empty after 2nd iteration,
1791 * when we jumped to start of loop because backlog
1792 * processing added something to receive_queue.
1793 * We cannot release_sock(), because backlog contains
1794 * packets arrived _after_ prequeued ones.
1795 *
1796 * Shortly, algorithm is clear --- to process all
1797 * the queues in order. We could make it more directly,
1798 * requeueing packets from backlog to prequeue, if
1799 * is not empty. It is more elegant, but eats cycles,
1800 * unfortunately.
1801 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001802 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 goto do_prequeue;
1804
1805 /* __ Set realtime policy in scheduler __ */
1806 }
1807
Steven J. Magnani73852e82010-03-16 05:22:44 +00001808#ifdef CONFIG_NET_DMA
Michal Kubeček15c04172012-09-14 04:59:52 +00001809 if (tp->ucopy.dma_chan) {
1810 if (tp->rcv_wnd == 0 &&
1811 !skb_queue_empty(&sk->sk_async_wait_queue)) {
1812 tcp_service_net_dma(sk, true);
1813 tcp_cleanup_rbuf(sk, copied);
1814 } else
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001815 dma_async_issue_pending(tp->ucopy.dma_chan);
Michal Kubeček15c04172012-09-14 04:59:52 +00001816 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001817#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 if (copied >= target) {
1819 /* Do not sleep, just process backlog. */
1820 release_sock(sk);
1821 lock_sock(sk);
1822 } else
1823 sk_wait_data(sk, &timeo);
1824
Chris Leech1a2449a2006-05-23 18:05:53 -07001825#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001826 tcp_service_net_dma(sk, false); /* Don't block */
Chris Leech1a2449a2006-05-23 18:05:53 -07001827 tp->ucopy.wakeup = 0;
1828#endif
1829
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 if (user_recv) {
1831 int chunk;
1832
1833 /* __ Restore normal policy in scheduler __ */
1834
1835 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001836 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 len -= chunk;
1838 copied += chunk;
1839 }
1840
1841 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001842 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843do_prequeue:
1844 tcp_prequeue_process(sk);
1845
1846 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001847 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 len -= chunk;
1849 copied += chunk;
1850 }
1851 }
1852 }
Ilpo Järvinen77527312009-05-10 20:32:34 +00001853 if ((flags & MSG_PEEK) &&
1854 (peek_seq - copied - urg_hole != tp->copied_seq)) {
Joe Perchese87cc472012-05-13 21:56:26 +00001855 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1856 current->comm,
1857 task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 peek_seq = tp->copied_seq;
1859 }
1860 continue;
1861
1862 found_ok_skb:
1863 /* Ok so how much can we use? */
1864 used = skb->len - offset;
1865 if (len < used)
1866 used = len;
1867
1868 /* Do we have urgent data here? */
1869 if (tp->urg_data) {
1870 u32 urg_offset = tp->urg_seq - *seq;
1871 if (urg_offset < used) {
1872 if (!urg_offset) {
1873 if (!sock_flag(sk, SOCK_URGINLINE)) {
1874 ++*seq;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001875 urg_hole++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 offset++;
1877 used--;
1878 if (!used)
1879 goto skip_copy;
1880 }
1881 } else
1882 used = urg_offset;
1883 }
1884 }
1885
1886 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001887#ifdef CONFIG_NET_DMA
1888 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dave Jianga2bd1142012-04-04 16:10:46 -07001889 tp->ucopy.dma_chan = net_dma_find_channel();
Chris Leech1a2449a2006-05-23 18:05:53 -07001890
1891 if (tp->ucopy.dma_chan) {
1892 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1893 tp->ucopy.dma_chan, skb, offset,
1894 msg->msg_iov, used,
1895 tp->ucopy.pinned_list);
1896
1897 if (tp->ucopy.dma_cookie < 0) {
1898
Joe Perchesafd465032012-03-12 07:03:32 +00001899 pr_alert("%s: dma_cookie < 0\n",
1900 __func__);
Chris Leech1a2449a2006-05-23 18:05:53 -07001901
1902 /* Exception. Bailout! */
1903 if (!copied)
1904 copied = -EFAULT;
1905 break;
1906 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001907
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001908 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001909
Chris Leech1a2449a2006-05-23 18:05:53 -07001910 if ((offset + used) == skb->len)
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001911 copied_early = true;
Chris Leech1a2449a2006-05-23 18:05:53 -07001912
1913 } else
1914#endif
1915 {
1916 err = skb_copy_datagram_iovec(skb, offset,
1917 msg->msg_iov, used);
1918 if (err) {
1919 /* Exception. Bailout! */
1920 if (!copied)
1921 copied = -EFAULT;
1922 break;
1923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925 }
1926
1927 *seq += used;
1928 copied += used;
1929 len -= used;
1930
1931 tcp_rcv_space_adjust(sk);
1932
1933skip_copy:
1934 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1935 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001936 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 }
1938 if (used + offset < skb->len)
1939 continue;
1940
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001941 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001943 if (!(flags & MSG_PEEK)) {
1944 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001945 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 continue;
1948
1949 found_fin_ok:
1950 /* Process the FIN. */
1951 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001952 if (!(flags & MSG_PEEK)) {
1953 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001954 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 break;
1957 } while (len > 0);
1958
1959 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001960 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 int chunk;
1962
1963 tp->ucopy.len = copied > 0 ? len : 0;
1964
1965 tcp_prequeue_process(sk);
1966
1967 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001968 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 len -= chunk;
1970 copied += chunk;
1971 }
1972 }
1973
1974 tp->ucopy.task = NULL;
1975 tp->ucopy.len = 0;
1976 }
1977
Chris Leech1a2449a2006-05-23 18:05:53 -07001978#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001979 tcp_service_net_dma(sk, true); /* Wait for queue to drain */
1980 tp->ucopy.dma_chan = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001981
Chris Leech1a2449a2006-05-23 18:05:53 -07001982 if (tp->ucopy.pinned_list) {
1983 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1984 tp->ucopy.pinned_list = NULL;
1985 }
1986#endif
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 /* According to UNIX98, msg_name/msg_namelen are ignored
1989 * on connected socket. I was just happy when found this 8) --ANK
1990 */
1991
1992 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001993 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 release_sock(sk);
1996 return copied;
1997
1998out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 release_sock(sk);
2000 return err;
2001
2002recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07002003 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 goto out;
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00002005
2006recv_sndq:
2007 err = tcp_peek_sndq(sk, msg, len);
2008 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002010EXPORT_SYMBOL(tcp_recvmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002012void tcp_set_state(struct sock *sk, int state)
2013{
2014 int oldstate = sk->sk_state;
2015
2016 switch (state) {
2017 case TCP_ESTABLISHED:
2018 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07002019 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002020 break;
2021
2022 case TCP_CLOSE:
2023 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07002024 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002025
2026 sk->sk_prot->unhash(sk);
2027 if (inet_csk(sk)->icsk_bind_hash &&
2028 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08002029 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002030 /* fall through */
2031 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08002032 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07002033 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002034 }
2035
2036 /* Change state AFTER socket is unhashed to avoid closed
2037 * socket sitting in hash tables.
2038 */
2039 sk->sk_state = state;
2040
2041#ifdef STATE_TRACE
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08002042 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08002043#endif
2044}
2045EXPORT_SYMBOL_GPL(tcp_set_state);
2046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047/*
2048 * State processing on a close. This implements the state shift for
2049 * sending our FIN frame. Note that we only send a FIN for some
2050 * states. A shutdown() may have already sent the FIN, or we may be
2051 * closed.
2052 */
2053
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08002054static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 /* current state: new state: action: */
2056 /* (Invalid) */ TCP_CLOSE,
2057 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2058 /* TCP_SYN_SENT */ TCP_CLOSE,
2059 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2060 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
2061 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
2062 /* TCP_TIME_WAIT */ TCP_CLOSE,
2063 /* TCP_CLOSE */ TCP_CLOSE,
2064 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
2065 /* TCP_LAST_ACK */ TCP_LAST_ACK,
2066 /* TCP_LISTEN */ TCP_CLOSE,
2067 /* TCP_CLOSING */ TCP_CLOSING,
2068};
2069
2070static int tcp_close_state(struct sock *sk)
2071{
2072 int next = (int)new_state[sk->sk_state];
2073 int ns = next & TCP_STATE_MASK;
2074
2075 tcp_set_state(sk, ns);
2076
2077 return next & TCP_ACTION_FIN;
2078}
2079
2080/*
2081 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07002082 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 */
2084
2085void tcp_shutdown(struct sock *sk, int how)
2086{
2087 /* We need to grab some memory, and put together a FIN,
2088 * and then put it into the queue to be sent.
2089 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2090 */
2091 if (!(how & SEND_SHUTDOWN))
2092 return;
2093
2094 /* If we've already sent a FIN, or it's a closed state, skip this. */
2095 if ((1 << sk->sk_state) &
2096 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2097 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2098 /* Clear out any half completed packets. FIN if needed. */
2099 if (tcp_close_state(sk))
2100 tcp_send_fin(sk);
2101 }
2102}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002103EXPORT_SYMBOL(tcp_shutdown);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002105bool tcp_check_oom(struct sock *sk, int shift)
2106{
2107 bool too_many_orphans, out_of_socket_memory;
2108
2109 too_many_orphans = tcp_too_many_orphans(sk, shift);
2110 out_of_socket_memory = tcp_out_of_memory(sk);
2111
Joe Perchese87cc472012-05-13 21:56:26 +00002112 if (too_many_orphans)
2113 net_info_ratelimited("too many orphaned sockets\n");
2114 if (out_of_socket_memory)
2115 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002116 return too_many_orphans || out_of_socket_memory;
2117}
2118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119void tcp_close(struct sock *sk, long timeout)
2120{
2121 struct sk_buff *skb;
2122 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07002123 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125 lock_sock(sk);
2126 sk->sk_shutdown = SHUTDOWN_MASK;
2127
2128 if (sk->sk_state == TCP_LISTEN) {
2129 tcp_set_state(sk, TCP_CLOSE);
2130
2131 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002132 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
2134 goto adjudge_to_death;
2135 }
2136
2137 /* We need to flush the recv. buffs. We do this only on the
2138 * descriptor close, not protocol-sourced closes, because the
2139 * reader process may not have drained the data yet!
2140 */
2141 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2142 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002143 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 data_was_unread += len;
2145 __kfree_skb(skb);
2146 }
2147
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002148 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
Konstantin Khorenko565b7b22010-06-24 21:54:58 -07002150 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2151 if (sk->sk_state == TCP_CLOSE)
2152 goto adjudge_to_death;
2153
Gerrit Renker65bb7232007-04-28 21:21:46 -07002154 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2155 * data was lost. To witness the awful effects of the old behavior of
2156 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2157 * GET in an FTP client, suspend the process, wait for the client to
2158 * advertise a zero window, then kill -9 the FTP client, wheee...
2159 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 */
Pavel Emelyanovee995282012-04-19 03:40:39 +00002161 if (unlikely(tcp_sk(sk)->repair)) {
2162 sk->sk_prot->disconnect(sk, 0);
2163 } else if (data_was_unread) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002165 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 tcp_set_state(sk, TCP_CLOSE);
Wu Fengguangaa133072009-09-02 23:45:45 -07002167 tcp_send_active_reset(sk, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2169 /* Check zero linger _after_ checking for unread data. */
2170 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002171 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 } else if (tcp_close_state(sk)) {
2173 /* We FIN if the application ate all the data before
2174 * zapping the connection.
2175 */
2176
2177 /* RED-PEN. Formally speaking, we have broken TCP state
2178 * machine. State transitions:
2179 *
2180 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2181 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2182 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2183 *
2184 * are legal only when FIN has been sent (i.e. in window),
2185 * rather than queued out of window. Purists blame.
2186 *
2187 * F.e. "RFC state" is ESTABLISHED,
2188 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2189 *
2190 * The visible declinations are that sometimes
2191 * we enter time-wait state, when it is not required really
2192 * (harmless), do not send active resets, when they are
2193 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2194 * they look as CLOSING or LAST_ACK for Linux)
2195 * Probably, I missed some more holelets.
2196 * --ANK
Jerry Chu83368862012-08-31 12:29:12 +00002197 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2198 * in a single packet! (May consider it later but will
2199 * probably need API support or TCP_CORK SYN-ACK until
2200 * data is written and socket is closed.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 */
2202 tcp_send_fin(sk);
2203 }
2204
2205 sk_stream_wait_close(sk, timeout);
2206
2207adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07002208 state = sk->sk_state;
2209 sock_hold(sk);
2210 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07002211
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 /* It is the last release_sock in its life. It will remove backlog. */
2213 release_sock(sk);
2214
2215
2216 /* Now socket is owned by kernel and we acquire BH lock
2217 to finish close. No need to check for user refs.
2218 */
2219 local_bh_disable();
2220 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002221 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Herbert Xueb4dea52008-12-29 23:04:08 -08002223 percpu_counter_inc(sk->sk_prot->orphan_count);
2224
Herbert Xu75c2d9072006-05-03 23:31:35 -07002225 /* Have we already been destroyed by a softirq or backlog? */
2226 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2227 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 /* This is a (useful) BSD violating of the RFC. There is a
2230 * problem with TCP as specified in that the other end could
2231 * keep a socket open forever with no application left this end.
2232 * We use a 3 minute timeout (about the same as BSD) then kill
2233 * our end. If they send after that then tough - BUT: long enough
2234 * that we won't make the old 4*rto = almost no time - whoops
2235 * reset mistake.
2236 *
2237 * Nope, it was not mistake. It is really desired behaviour
2238 * f.e. on http servers, when such sockets are useless, but
2239 * consume significant resources. Let's do it with special
2240 * linger2 option. --ANK
2241 */
2242
2243 if (sk->sk_state == TCP_FIN_WAIT2) {
2244 struct tcp_sock *tp = tcp_sk(sk);
2245 if (tp->linger2 < 0) {
2246 tcp_set_state(sk, TCP_CLOSE);
2247 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002248 NET_INC_STATS_BH(sock_net(sk),
2249 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002251 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07002254 inet_csk_reset_keepalive_timer(sk,
2255 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2258 goto out;
2259 }
2260 }
2261 }
2262 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002263 sk_mem_reclaim(sk);
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002264 if (tcp_check_oom(sk, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 tcp_set_state(sk, TCP_CLOSE);
2266 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002267 NET_INC_STATS_BH(sock_net(sk),
2268 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 }
2270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
Jerry Chu83368862012-08-31 12:29:12 +00002272 if (sk->sk_state == TCP_CLOSE) {
2273 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2274 /* We could get here with a non-NULL req if the socket is
2275 * aborted (e.g., closed with unread data) before 3WHS
2276 * finishes.
2277 */
2278 if (req != NULL)
2279 reqsk_fastopen_remove(sk, req, false);
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002280 inet_csk_destroy_sock(sk);
Jerry Chu83368862012-08-31 12:29:12 +00002281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 /* Otherwise, socket is reprieved until protocol close. */
2283
2284out:
2285 bh_unlock_sock(sk);
2286 local_bh_enable();
2287 sock_put(sk);
2288}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002289EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291/* These states need RST on ABORT according to RFC793 */
2292
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002293static inline bool tcp_need_reset(int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294{
2295 return (1 << state) &
2296 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2297 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2298}
2299
2300int tcp_disconnect(struct sock *sk, int flags)
2301{
2302 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002303 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 struct tcp_sock *tp = tcp_sk(sk);
2305 int err = 0;
2306 int old_state = sk->sk_state;
2307
2308 if (old_state != TCP_CLOSE)
2309 tcp_set_state(sk, TCP_CLOSE);
2310
2311 /* ABORT function of RFC793 */
2312 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002313 inet_csk_listen_stop(sk);
Pavel Emelyanovee995282012-04-19 03:40:39 +00002314 } else if (unlikely(tp->repair)) {
2315 sk->sk_err = ECONNABORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 } else if (tcp_need_reset(old_state) ||
2317 (tp->snd_nxt != tp->write_seq &&
2318 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08002319 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 * states
2321 */
2322 tcp_send_active_reset(sk, gfp_any());
2323 sk->sk_err = ECONNRESET;
2324 } else if (old_state == TCP_SYN_SENT)
2325 sk->sk_err = ECONNRESET;
2326
2327 tcp_clear_xmit_timers(sk);
2328 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08002329 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07002331#ifdef CONFIG_NET_DMA
2332 __skb_queue_purge(&sk->sk_async_wait_queue);
2333#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002335 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
2337 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2338 inet_reset_saddr(sk);
2339
2340 sk->sk_shutdown = 0;
2341 sock_reset_flag(sk, SOCK_DONE);
2342 tp->srtt = 0;
2343 if ((tp->write_seq += tp->max_window + 2) == 0)
2344 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002345 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002347 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 tp->packets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002349 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 tp->snd_cwnd_cnt = 0;
Eric Dumazet1fdf4752009-11-30 12:53:30 -08002351 tp->window_clamp = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002352 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002354 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08002355 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002356 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 __sk_dst_reset(sk);
2358
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002359 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
2361 sk->sk_error_report(sk);
2362 return err;
2363}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002364EXPORT_SYMBOL(tcp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Christoph Paaschbb68b642012-09-18 14:19:23 +00002366void tcp_sock_destruct(struct sock *sk)
2367{
2368 inet_sock_destruct(sk);
2369
2370 kfree(inet_csk(sk)->icsk_accept_queue.fastopenq);
2371}
2372
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002373static inline bool tcp_can_repair_sock(const struct sock *sk)
Pavel Emelyanovee995282012-04-19 03:40:39 +00002374{
Eric W. Biederman52e804c2012-11-16 03:03:05 +00002375 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
Pavel Emelyanovee995282012-04-19 03:40:39 +00002376 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
2377}
2378
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002379static int tcp_repair_options_est(struct tcp_sock *tp,
2380 struct tcp_repair_opt __user *optbuf, unsigned int len)
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002381{
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002382 struct tcp_repair_opt opt;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002383
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002384 while (len >= sizeof(opt)) {
2385 if (copy_from_user(&opt, optbuf, sizeof(opt)))
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002386 return -EFAULT;
2387
2388 optbuf++;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002389 len -= sizeof(opt);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002390
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002391 switch (opt.opt_code) {
2392 case TCPOPT_MSS:
2393 tp->rx_opt.mss_clamp = opt.opt_val;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002394 break;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002395 case TCPOPT_WINDOW:
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002396 {
2397 u16 snd_wscale = opt.opt_val & 0xFFFF;
2398 u16 rcv_wscale = opt.opt_val >> 16;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002399
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002400 if (snd_wscale > 14 || rcv_wscale > 14)
2401 return -EFBIG;
2402
2403 tp->rx_opt.snd_wscale = snd_wscale;
2404 tp->rx_opt.rcv_wscale = rcv_wscale;
2405 tp->rx_opt.wscale_ok = 1;
2406 }
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002407 break;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002408 case TCPOPT_SACK_PERM:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002409 if (opt.opt_val != 0)
2410 return -EINVAL;
2411
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002412 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2413 if (sysctl_tcp_fack)
2414 tcp_enable_fack(tp);
2415 break;
2416 case TCPOPT_TIMESTAMP:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002417 if (opt.opt_val != 0)
2418 return -EINVAL;
2419
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002420 tp->rx_opt.tstamp_ok = 1;
2421 break;
2422 }
2423 }
2424
2425 return 0;
2426}
2427
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 * Socket option code for TCP.
2430 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002431static int do_tcp_setsockopt(struct sock *sk, int level,
David S. Millerb7058842009-09-30 16:12:20 -07002432 int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433{
2434 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002435 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 int val;
2437 int err = 0;
2438
William Allen Simpsone56fb502009-12-02 18:19:30 +00002439 /* These are data/string values, all the others are ints */
2440 switch (optname) {
2441 case TCP_CONGESTION: {
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002442 char name[TCP_CA_NAME_MAX];
2443
2444 if (optlen < 1)
2445 return -EINVAL;
2446
2447 val = strncpy_from_user(name, optval,
Andrew Morton4fdb78d2009-10-01 15:02:20 -07002448 min_t(long, TCP_CA_NAME_MAX-1, optlen));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002449 if (val < 0)
2450 return -EFAULT;
2451 name[val] = 0;
2452
2453 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002454 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002455 release_sock(sk);
2456 return err;
2457 }
William Allen Simpsone56fb502009-12-02 18:19:30 +00002458 default:
2459 /* fallthru */
2460 break;
Joe Perchesccbd6a52010-05-14 10:58:26 +00002461 }
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002462
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 if (optlen < sizeof(int))
2464 return -EINVAL;
2465
2466 if (get_user(val, (int __user *)optval))
2467 return -EFAULT;
2468
2469 lock_sock(sk);
2470
2471 switch (optname) {
2472 case TCP_MAXSEG:
2473 /* Values greater than interface MTU won't take effect. However
2474 * at the point when this call is done we typically don't yet
2475 * know which interface is going to be used */
David S. Millerc39508d2010-11-24 11:47:22 -08002476 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 err = -EINVAL;
2478 break;
2479 }
2480 tp->rx_opt.user_mss = val;
2481 break;
2482
2483 case TCP_NODELAY:
2484 if (val) {
2485 /* TCP_NODELAY is weaker than TCP_CORK, so that
2486 * this option on corked socket is remembered, but
2487 * it is not activated until cork is cleared.
2488 *
2489 * However, when TCP_NODELAY is set we make
2490 * an explicit push, which overrides even TCP_CORK
2491 * for currently queued segments.
2492 */
2493 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002494 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 } else {
2496 tp->nonagle &= ~TCP_NAGLE_OFF;
2497 }
2498 break;
2499
Andreas Petlund36e31b0a2010-02-18 02:47:01 +00002500 case TCP_THIN_LINEAR_TIMEOUTS:
2501 if (val < 0 || val > 1)
2502 err = -EINVAL;
2503 else
2504 tp->thin_lto = val;
2505 break;
2506
Andreas Petlund7e380172010-02-18 04:48:19 +00002507 case TCP_THIN_DUPACK:
2508 if (val < 0 || val > 1)
2509 err = -EINVAL;
Dave Jonese2e5c4c2013-09-05 13:43:34 -04002510 else {
Andreas Petlund7e380172010-02-18 04:48:19 +00002511 tp->thin_dupack = val;
Yuchung Chengeed530b2012-05-02 13:30:03 +00002512 if (tp->thin_dupack)
2513 tcp_disable_early_retrans(tp);
Dave Jonese2e5c4c2013-09-05 13:43:34 -04002514 }
Andreas Petlund7e380172010-02-18 04:48:19 +00002515 break;
2516
Pavel Emelyanovee995282012-04-19 03:40:39 +00002517 case TCP_REPAIR:
2518 if (!tcp_can_repair_sock(sk))
2519 err = -EPERM;
2520 else if (val == 1) {
2521 tp->repair = 1;
2522 sk->sk_reuse = SK_FORCE_REUSE;
2523 tp->repair_queue = TCP_NO_QUEUE;
2524 } else if (val == 0) {
2525 tp->repair = 0;
2526 sk->sk_reuse = SK_NO_REUSE;
2527 tcp_send_window_probe(sk);
2528 } else
2529 err = -EINVAL;
2530
2531 break;
2532
2533 case TCP_REPAIR_QUEUE:
2534 if (!tp->repair)
2535 err = -EPERM;
2536 else if (val < TCP_QUEUES_NR)
2537 tp->repair_queue = val;
2538 else
2539 err = -EINVAL;
2540 break;
2541
2542 case TCP_QUEUE_SEQ:
2543 if (sk->sk_state != TCP_CLOSE)
2544 err = -EPERM;
2545 else if (tp->repair_queue == TCP_SEND_QUEUE)
2546 tp->write_seq = val;
2547 else if (tp->repair_queue == TCP_RECV_QUEUE)
2548 tp->rcv_nxt = val;
2549 else
2550 err = -EINVAL;
2551 break;
2552
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002553 case TCP_REPAIR_OPTIONS:
2554 if (!tp->repair)
2555 err = -EINVAL;
2556 else if (sk->sk_state == TCP_ESTABLISHED)
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002557 err = tcp_repair_options_est(tp,
2558 (struct tcp_repair_opt __user *)optval,
2559 optlen);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002560 else
2561 err = -EPERM;
2562 break;
2563
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 case TCP_CORK:
2565 /* When set indicates to always queue non-full frames.
2566 * Later the user clears this option and we transmit
2567 * any pending partial frames in the queue. This is
2568 * meant to be used alongside sendfile() to get properly
2569 * filled frames when the user (for example) must write
2570 * out headers with a write() call first and then use
2571 * sendfile to send out the data parts.
2572 *
2573 * TCP_CORK can be set together with TCP_NODELAY and it is
2574 * stronger than TCP_NODELAY.
2575 */
2576 if (val) {
2577 tp->nonagle |= TCP_NAGLE_CORK;
2578 } else {
2579 tp->nonagle &= ~TCP_NAGLE_CORK;
2580 if (tp->nonagle&TCP_NAGLE_OFF)
2581 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002582 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 }
2584 break;
2585
2586 case TCP_KEEPIDLE:
2587 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2588 err = -EINVAL;
2589 else {
2590 tp->keepalive_time = val * HZ;
2591 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2592 !((1 << sk->sk_state) &
2593 (TCPF_CLOSE | TCPF_LISTEN))) {
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00002594 u32 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 if (tp->keepalive_time > elapsed)
2596 elapsed = tp->keepalive_time - elapsed;
2597 else
2598 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002599 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 }
2601 }
2602 break;
2603 case TCP_KEEPINTVL:
2604 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2605 err = -EINVAL;
2606 else
2607 tp->keepalive_intvl = val * HZ;
2608 break;
2609 case TCP_KEEPCNT:
2610 if (val < 1 || val > MAX_TCP_KEEPCNT)
2611 err = -EINVAL;
2612 else
2613 tp->keepalive_probes = val;
2614 break;
2615 case TCP_SYNCNT:
2616 if (val < 1 || val > MAX_TCP_SYNCNT)
2617 err = -EINVAL;
2618 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002619 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 break;
2621
2622 case TCP_LINGER2:
2623 if (val < 0)
2624 tp->linger2 = -1;
2625 else if (val > sysctl_tcp_fin_timeout / HZ)
2626 tp->linger2 = 0;
2627 else
2628 tp->linger2 = val * HZ;
2629 break;
2630
2631 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002632 /* Translate value in seconds to number of retransmits */
2633 icsk->icsk_accept_queue.rskq_defer_accept =
2634 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2635 TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 break;
2637
2638 case TCP_WINDOW_CLAMP:
2639 if (!val) {
2640 if (sk->sk_state != TCP_CLOSE) {
2641 err = -EINVAL;
2642 break;
2643 }
2644 tp->window_clamp = 0;
2645 } else
2646 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2647 SOCK_MIN_RCVBUF / 2 : val;
2648 break;
2649
2650 case TCP_QUICKACK:
2651 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002652 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002654 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if ((1 << sk->sk_state) &
2656 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002657 inet_csk_ack_scheduled(sk)) {
2658 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002659 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002661 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 }
2663 }
2664 break;
2665
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002666#ifdef CONFIG_TCP_MD5SIG
2667 case TCP_MD5SIG:
2668 /* Read the IP->Key mappings from userspace */
2669 err = tp->af_specific->md5_parse(sk, optval, optlen);
2670 break;
2671#endif
Jerry Chudca43c72010-08-27 19:13:28 +00002672 case TCP_USER_TIMEOUT:
2673 /* Cap the max timeout in ms TCP will retry/retrans
2674 * before giving up and aborting (ETIMEDOUT) a connection.
2675 */
Hangbin Liu42493572012-07-26 22:52:21 +00002676 if (val < 0)
2677 err = -EINVAL;
2678 else
2679 icsk->icsk_user_timeout = msecs_to_jiffies(val);
Jerry Chudca43c72010-08-27 19:13:28 +00002680 break;
Jerry Chu83368862012-08-31 12:29:12 +00002681
2682 case TCP_FASTOPEN:
2683 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
2684 TCPF_LISTEN)))
2685 err = fastopen_init_queue(sk, val);
2686 else
2687 err = -EINVAL;
2688 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002689 case TCP_TIMESTAMP:
2690 if (!tp->repair)
2691 err = -EPERM;
2692 else
2693 tp->tsoffset = val - tcp_time_stamp;
2694 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002695 case TCP_NOTSENT_LOWAT:
2696 tp->notsent_lowat = val;
2697 sk->sk_write_space(sk);
2698 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 default:
2700 err = -ENOPROTOOPT;
2701 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002702 }
2703
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 release_sock(sk);
2705 return err;
2706}
2707
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002708int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
David S. Millerb7058842009-09-30 16:12:20 -07002709 unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002710{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002711 const struct inet_connection_sock *icsk = inet_csk(sk);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002712
2713 if (level != SOL_TCP)
2714 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2715 optval, optlen);
2716 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2717}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002718EXPORT_SYMBOL(tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002719
2720#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002721int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002722 char __user *optval, unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002723{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002724 if (level != SOL_TCP)
2725 return inet_csk_compat_setsockopt(sk, level, optname,
2726 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002727 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2728}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002729EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002730#endif
2731
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732/* Return information about state of tcp endpoint in API format. */
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002733void tcp_get_info(const struct sock *sk, struct tcp_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002735 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002736 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 u32 now = tcp_time_stamp;
2738
2739 memset(info, 0, sizeof(*info));
2740
2741 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002742 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002743 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002744 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002745 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
2747 if (tp->rx_opt.tstamp_ok)
2748 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002749 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 info->tcpi_options |= TCPI_OPT_SACK;
2751 if (tp->rx_opt.wscale_ok) {
2752 info->tcpi_options |= TCPI_OPT_WSCALE;
2753 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2754 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756
Eric Dumazetb5c56932011-10-03 14:01:21 -04002757 if (tp->ecn_flags & TCP_ECN_OK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 info->tcpi_options |= TCPI_OPT_ECN;
Eric Dumazetb5c56932011-10-03 14:01:21 -04002759 if (tp->ecn_flags & TCP_ECN_SEEN)
2760 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
Yuchung Cheng6f736012012-10-19 15:14:44 +00002761 if (tp->syn_data_acked)
2762 info->tcpi_options |= TCPI_OPT_SYN_DATA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002764 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2765 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002766 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002767 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768
Rick Jones5ee3afb2007-09-18 13:26:31 -07002769 if (sk->sk_state == TCP_LISTEN) {
2770 info->tcpi_unacked = sk->sk_ack_backlog;
2771 info->tcpi_sacked = sk->sk_max_ack_backlog;
2772 } else {
2773 info->tcpi_unacked = tp->packets_out;
2774 info->tcpi_sacked = tp->sacked_out;
2775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 info->tcpi_lost = tp->lost_out;
2777 info->tcpi_retrans = tp->retrans_out;
2778 info->tcpi_fackets = tp->fackets_out;
2779
2780 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002781 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2783
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002784 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2786 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2787 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2788 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2789 info->tcpi_snd_cwnd = tp->snd_cwnd;
2790 info->tcpi_advmss = tp->advmss;
2791 info->tcpi_reordering = tp->reordering;
2792
2793 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2794 info->tcpi_rcv_space = tp->rcvq_space.space;
2795
2796 info->tcpi_total_retrans = tp->total_retrans;
2797}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798EXPORT_SYMBOL_GPL(tcp_get_info);
2799
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002800static int do_tcp_getsockopt(struct sock *sk, int level,
2801 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002803 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 struct tcp_sock *tp = tcp_sk(sk);
2805 int val, len;
2806
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 if (get_user(len, optlen))
2808 return -EFAULT;
2809
2810 len = min_t(unsigned int, len, sizeof(int));
2811
2812 if (len < 0)
2813 return -EINVAL;
2814
2815 switch (optname) {
2816 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002817 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2819 val = tp->rx_opt.user_mss;
Pavel Emelyanov5e6a3ce2012-04-19 03:41:32 +00002820 if (tp->repair)
2821 val = tp->rx_opt.mss_clamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 break;
2823 case TCP_NODELAY:
2824 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2825 break;
2826 case TCP_CORK:
2827 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2828 break;
2829 case TCP_KEEPIDLE:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002830 val = keepalive_time_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 break;
2832 case TCP_KEEPINTVL:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002833 val = keepalive_intvl_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 break;
2835 case TCP_KEEPCNT:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002836 val = keepalive_probes(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 break;
2838 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002839 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 break;
2841 case TCP_LINGER2:
2842 val = tp->linger2;
2843 if (val >= 0)
2844 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2845 break;
2846 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002847 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2848 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 break;
2850 case TCP_WINDOW_CLAMP:
2851 val = tp->window_clamp;
2852 break;
2853 case TCP_INFO: {
2854 struct tcp_info info;
2855
2856 if (get_user(len, optlen))
2857 return -EFAULT;
2858
2859 tcp_get_info(sk, &info);
2860
2861 len = min_t(unsigned int, len, sizeof(info));
2862 if (put_user(len, optlen))
2863 return -EFAULT;
2864 if (copy_to_user(optval, &info, len))
2865 return -EFAULT;
2866 return 0;
2867 }
2868 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002869 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002871
2872 case TCP_CONGESTION:
2873 if (get_user(len, optlen))
2874 return -EFAULT;
2875 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2876 if (put_user(len, optlen))
2877 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002878 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002879 return -EFAULT;
2880 return 0;
William Allen Simpsone56fb502009-12-02 18:19:30 +00002881
Josh Hunt3c0fef02010-07-30 13:49:35 +00002882 case TCP_THIN_LINEAR_TIMEOUTS:
2883 val = tp->thin_lto;
2884 break;
2885 case TCP_THIN_DUPACK:
2886 val = tp->thin_dupack;
2887 break;
Jerry Chudca43c72010-08-27 19:13:28 +00002888
Pavel Emelyanovee995282012-04-19 03:40:39 +00002889 case TCP_REPAIR:
2890 val = tp->repair;
2891 break;
2892
2893 case TCP_REPAIR_QUEUE:
2894 if (tp->repair)
2895 val = tp->repair_queue;
2896 else
2897 return -EINVAL;
2898 break;
2899
2900 case TCP_QUEUE_SEQ:
2901 if (tp->repair_queue == TCP_SEND_QUEUE)
2902 val = tp->write_seq;
2903 else if (tp->repair_queue == TCP_RECV_QUEUE)
2904 val = tp->rcv_nxt;
2905 else
2906 return -EINVAL;
2907 break;
2908
Jerry Chudca43c72010-08-27 19:13:28 +00002909 case TCP_USER_TIMEOUT:
2910 val = jiffies_to_msecs(icsk->icsk_user_timeout);
2911 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002912 case TCP_TIMESTAMP:
2913 val = tcp_time_stamp + tp->tsoffset;
2914 break;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002915 case TCP_NOTSENT_LOWAT:
2916 val = tp->notsent_lowat;
2917 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 default:
2919 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002920 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921
2922 if (put_user(len, optlen))
2923 return -EFAULT;
2924 if (copy_to_user(optval, &val, len))
2925 return -EFAULT;
2926 return 0;
2927}
2928
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002929int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2930 int __user *optlen)
2931{
2932 struct inet_connection_sock *icsk = inet_csk(sk);
2933
2934 if (level != SOL_TCP)
2935 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2936 optval, optlen);
2937 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2938}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002939EXPORT_SYMBOL(tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002940
2941#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002942int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2943 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002944{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002945 if (level != SOL_TCP)
2946 return inet_csk_compat_getsockopt(sk, level, optname,
2947 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002948 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2949}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002950EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002951#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002953#ifdef CONFIG_TCP_MD5SIG
Eric Dumazet71cea172013-05-20 06:52:26 +00002954static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool __read_mostly;
2955static DEFINE_MUTEX(tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002956
Eric Dumazet765cf992011-09-12 20:28:37 +00002957static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002958{
2959 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00002960
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002961 for_each_possible_cpu(cpu) {
Eric Dumazet765cf992011-09-12 20:28:37 +00002962 struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu);
2963
2964 if (p->md5_desc.tfm)
2965 crypto_free_hash(p->md5_desc.tfm);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002966 }
2967 free_percpu(pool);
2968}
2969
Eric Dumazet71cea172013-05-20 06:52:26 +00002970static void __tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002971{
2972 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00002973 struct tcp_md5sig_pool __percpu *pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002974
Eric Dumazet765cf992011-09-12 20:28:37 +00002975 pool = alloc_percpu(struct tcp_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002976 if (!pool)
Eric Dumazet71cea172013-05-20 06:52:26 +00002977 return;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002978
2979 for_each_possible_cpu(cpu) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002980 struct crypto_hash *hash;
2981
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002982 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
YOSHIFUJI Hideaki / 吉藤英明50c3a482013-01-22 06:32:49 +00002983 if (IS_ERR_OR_NULL(hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002984 goto out_free;
2985
Eric Dumazet765cf992011-09-12 20:28:37 +00002986 per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002987 }
Eric Dumazet71cea172013-05-20 06:52:26 +00002988 /* before setting tcp_md5sig_pool, we must commit all writes
2989 * to memory. See ACCESS_ONCE() in tcp_get_md5sig_pool()
2990 */
2991 smp_wmb();
2992 tcp_md5sig_pool = pool;
2993 return;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002994out_free:
2995 __tcp_free_md5sig_pool(pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002996}
2997
Eric Dumazet71cea172013-05-20 06:52:26 +00002998bool tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002999{
Eric Dumazet71cea172013-05-20 06:52:26 +00003000 if (unlikely(!tcp_md5sig_pool)) {
3001 mutex_lock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003002
Eric Dumazet71cea172013-05-20 06:52:26 +00003003 if (!tcp_md5sig_pool)
3004 __tcp_alloc_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003005
Eric Dumazet71cea172013-05-20 06:52:26 +00003006 mutex_unlock(&tcp_md5sig_mutex);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003007 }
Eric Dumazet71cea172013-05-20 06:52:26 +00003008 return tcp_md5sig_pool != NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003009}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003010EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3011
Eric Dumazet35790c02010-05-16 00:34:04 -07003012
3013/**
3014 * tcp_get_md5sig_pool - get md5sig_pool for this user
3015 *
3016 * We use percpu structure, so if we succeed, we exit with preemption
3017 * and BH disabled, to make sure another thread or softirq handling
3018 * wont try to get same context.
3019 */
3020struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003021{
Eric Dumazet765cf992011-09-12 20:28:37 +00003022 struct tcp_md5sig_pool __percpu *p;
Eric Dumazet35790c02010-05-16 00:34:04 -07003023
3024 local_bh_disable();
Eric Dumazet71cea172013-05-20 06:52:26 +00003025 p = ACCESS_ONCE(tcp_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003026 if (p)
Eric Dumazet71cea172013-05-20 06:52:26 +00003027 return __this_cpu_ptr(p);
Eric Dumazet35790c02010-05-16 00:34:04 -07003028
3029 local_bh_enable();
3030 return NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003031}
Eric Dumazet35790c02010-05-16 00:34:04 -07003032EXPORT_SYMBOL(tcp_get_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003033
Adam Langley49a72df2008-07-19 00:01:42 -07003034int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003035 const struct tcphdr *th)
Adam Langley49a72df2008-07-19 00:01:42 -07003036{
3037 struct scatterlist sg;
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003038 struct tcphdr hdr;
Adam Langley49a72df2008-07-19 00:01:42 -07003039 int err;
3040
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003041 /* We are not allowed to change tcphdr, make a local copy */
3042 memcpy(&hdr, th, sizeof(hdr));
3043 hdr.check = 0;
3044
Adam Langley49a72df2008-07-19 00:01:42 -07003045 /* options aren't included in the hash */
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003046 sg_init_one(&sg, &hdr, sizeof(hdr));
3047 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr));
Adam Langley49a72df2008-07-19 00:01:42 -07003048 return err;
3049}
Adam Langley49a72df2008-07-19 00:01:42 -07003050EXPORT_SYMBOL(tcp_md5_hash_header);
3051
3052int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003053 const struct sk_buff *skb, unsigned int header_len)
Adam Langley49a72df2008-07-19 00:01:42 -07003054{
3055 struct scatterlist sg;
3056 const struct tcphdr *tp = tcp_hdr(skb);
3057 struct hash_desc *desc = &hp->md5_desc;
Eric Dumazet95c96172012-04-15 05:58:06 +00003058 unsigned int i;
3059 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3060 skb_headlen(skb) - header_len : 0;
Adam Langley49a72df2008-07-19 00:01:42 -07003061 const struct skb_shared_info *shi = skb_shinfo(skb);
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003062 struct sk_buff *frag_iter;
Adam Langley49a72df2008-07-19 00:01:42 -07003063
3064 sg_init_table(&sg, 1);
3065
3066 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3067 if (crypto_hash_update(desc, &sg, head_data_len))
3068 return 1;
3069
3070 for (i = 0; i < shi->nr_frags; ++i) {
3071 const struct skb_frag_struct *f = &shi->frags[i];
Eric Dumazet54d27fc2013-05-13 21:25:52 +00003072 unsigned int offset = f->page_offset;
3073 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3074
3075 sg_set_page(&sg, page, skb_frag_size(f),
3076 offset_in_page(offset));
Eric Dumazet9e903e02011-10-18 21:00:24 +00003077 if (crypto_hash_update(desc, &sg, skb_frag_size(f)))
Adam Langley49a72df2008-07-19 00:01:42 -07003078 return 1;
3079 }
3080
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003081 skb_walk_frags(skb, frag_iter)
3082 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3083 return 1;
3084
Adam Langley49a72df2008-07-19 00:01:42 -07003085 return 0;
3086}
Adam Langley49a72df2008-07-19 00:01:42 -07003087EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3088
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003089int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
Adam Langley49a72df2008-07-19 00:01:42 -07003090{
3091 struct scatterlist sg;
3092
3093 sg_init_one(&sg, key->key, key->keylen);
3094 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
3095}
Adam Langley49a72df2008-07-19 00:01:42 -07003096EXPORT_SYMBOL(tcp_md5_hash_key);
3097
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003098#endif
3099
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003100void tcp_done(struct sock *sk)
3101{
Jerry Chu83368862012-08-31 12:29:12 +00003102 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3103
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08003104 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07003105 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003106
3107 tcp_set_state(sk, TCP_CLOSE);
3108 tcp_clear_xmit_timers(sk);
Jerry Chu83368862012-08-31 12:29:12 +00003109 if (req != NULL)
3110 reqsk_fastopen_remove(sk, req, false);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003111
3112 sk->sk_shutdown = SHUTDOWN_MASK;
3113
3114 if (!sock_flag(sk, SOCK_DEAD))
3115 sk->sk_state_change(sk);
3116 else
3117 inet_csk_destroy_sock(sk);
3118}
3119EXPORT_SYMBOL_GPL(tcp_done);
3120
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003121extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122
3123static __initdata unsigned long thash_entries;
3124static int __init set_thash_entries(char *str)
3125{
Eldad Zack413c27d2012-05-19 14:13:18 +00003126 ssize_t ret;
3127
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 if (!str)
3129 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00003130
3131 ret = kstrtoul(str, 0, &thash_entries);
3132 if (ret)
3133 return 0;
3134
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 return 1;
3136}
3137__setup("thash_entries=", set_thash_entries);
3138
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -07003139static void tcp_init_mem(void)
Glauber Costa4acb4192012-01-30 01:20:17 +00003140{
Glauber Costa4acb4192012-01-30 01:20:17 +00003141 unsigned long limit = nr_free_buffer_pages() / 8;
3142 limit = max(limit, 128UL);
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -07003143 sysctl_tcp_mem[0] = limit / 4 * 3;
3144 sysctl_tcp_mem[1] = limit;
3145 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
Glauber Costa4acb4192012-01-30 01:20:17 +00003146}
3147
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148void __init tcp_init(void)
3149{
3150 struct sk_buff *skb = NULL;
Eric Dumazetf03d78d2011-07-07 00:27:05 -07003151 unsigned long limit;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003152 int max_rshare, max_wshare, cnt;
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003153 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08003155 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156
Eric Dumazet17483762008-11-25 21:16:35 -08003157 percpu_counter_init(&tcp_sockets_allocated, 0);
Eric Dumazetdd24c002008-11-25 21:17:14 -08003158 percpu_counter_init(&tcp_orphan_count, 0);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003159 tcp_hashinfo.bind_bucket_cachep =
3160 kmem_cache_create("tcp_bind_bucket",
3161 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003162 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 /* Size and allocate the main established and bind bucket
3165 * hash tables.
3166 *
3167 * The methodology is similar to that of the buffer cache.
3168 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003169 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003171 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 thash_entries,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003173 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003174 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 NULL,
Eric Dumazetf373b532009-10-09 00:16:19 +00003176 &tcp_hashinfo.ehash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003177 0,
Jean Delvare0ccfe612007-10-30 00:59:25 -07003178 thash_entries ? 0 : 512 * 1024);
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07003179 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08003180 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07003181
Eric Dumazet230140c2007-11-07 02:40:20 -08003182 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3183 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003184 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003186 sizeof(struct inet_bind_hashbucket),
Eric Dumazetf373b532009-10-09 00:16:19 +00003187 tcp_hashinfo.ehash_mask + 1,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003188 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003189 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003190 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003192 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 64 * 1024);
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003194 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003195 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3196 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3197 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 }
3199
Eric Dumazetc5ed63d2010-08-25 23:02:17 -07003200
3201 cnt = tcp_hashinfo.ehash_mask + 1;
3202
3203 tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
3204 sysctl_tcp_max_orphans = cnt / 2;
3205 sysctl_max_syn_backlog = max(128, cnt / 256);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -07003207 tcp_init_mem();
Jason Wangc43b8742012-02-02 00:07:00 +00003208 /* Set per-socket limits to no more than 1/128 the pressure threshold */
Eric Dumazet5fb84b12012-04-10 00:56:42 +00003209 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
Eric Dumazetb49960a2012-05-02 02:28:41 +00003210 max_wshare = min(4UL*1024*1024, limit);
3211 max_rshare = min(6UL*1024*1024, limit);
John Heffner7b4f4b52006-03-25 01:34:07 -08003212
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003213 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003214 sysctl_tcp_wmem[1] = 16*1024;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003215 sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
John Heffner7b4f4b52006-03-25 01:34:07 -08003216
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003217 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003218 sysctl_tcp_rmem[1] = 87380;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003219 sysctl_tcp_rmem[2] = max(87380, max_rshare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
Joe Perchesafd465032012-03-12 07:03:32 +00003221 pr_info("Hash tables configured (established %u bind %u)\n",
Joe Perches058bd4d2012-03-11 18:36:11 +00003222 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003223
David S. Miller51c5d0c2012-07-10 00:49:14 -07003224 tcp_metrics_init();
3225
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003226 tcp_register_congestion_control(&tcp_reno);
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003227
Eric Dumazet46d3cea2012-07-11 05:50:31 +00003228 tcp_tasklet_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229}