blob: 0ce860b93c188b176f3d0eae199849009151b47e [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>
Robert Love4b015882008-05-12 17:08:29 -0400277#include <net/ip6_route.h>
278#include <net/ipv6.h>
279#include <net/transp_v6.h>
Chris Leech1a2449a2006-05-23 18:05:53 -0700280#include <net/netdma.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800281#include <net/sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283#include <asm/uaccess.h>
284#include <asm/ioctls.h>
285
Brian Haleyab32ea52006-09-22 14:15:41 -0700286int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Eric Dumazetdd24c002008-11-25 21:17:14 -0800288struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700289EXPORT_SYMBOL_GPL(tcp_orphan_count);
290
David S. Millerb8059ea2006-03-25 01:36:56 -0800291int sysctl_tcp_wmem[3] __read_mostly;
292int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294EXPORT_SYMBOL(sysctl_tcp_rmem);
295EXPORT_SYMBOL(sysctl_tcp_wmem);
296
Eric Dumazet8d987e52010-11-09 23:24:26 +0000297atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800299
300/*
301 * Current number of TCP sockets.
302 */
303struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304EXPORT_SYMBOL(tcp_sockets_allocated);
305
306/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800307 * TCP splice context
308 */
309struct tcp_splice_state {
310 struct pipe_inode_info *pipe;
311 size_t len;
312 unsigned int flags;
313};
314
315/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Pressure flag: try to collapse.
317 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800318 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * is strict, actions are advisory and have some latency.
320 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700321int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322EXPORT_SYMBOL(tcp_memory_pressure);
323
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700324void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700327 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 tcp_memory_pressure = 1;
329 }
330}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331EXPORT_SYMBOL(tcp_enter_memory_pressure);
332
Julian Anastasovb103cf32009-10-19 10:10:40 +0000333/* Convert seconds to retransmits based on initial and max timeout */
334static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
335{
336 u8 res = 0;
337
338 if (seconds > 0) {
339 int period = timeout;
340
341 res = 1;
342 while (seconds > period && res < 255) {
343 res++;
344 timeout <<= 1;
345 if (timeout > rto_max)
346 timeout = rto_max;
347 period += timeout;
348 }
349 }
350 return res;
351}
352
353/* Convert retransmits to seconds based on initial and max timeout */
354static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
355{
356 int period = 0;
357
358 if (retrans > 0) {
359 period = timeout;
360 while (--retrans) {
361 timeout <<= 1;
362 if (timeout > rto_max)
363 timeout = rto_max;
364 period += timeout;
365 }
366 }
367 return period;
368}
369
Neal Cardwell900f65d2012-04-19 09:55:21 +0000370/* Address-family independent initialization for a tcp_sock.
371 *
372 * NOTE: A lot of things set to zero explicitly by call to
373 * sk_alloc() so need not be done here.
374 */
375void tcp_init_sock(struct sock *sk)
376{
377 struct inet_connection_sock *icsk = inet_csk(sk);
378 struct tcp_sock *tp = tcp_sk(sk);
379
380 skb_queue_head_init(&tp->out_of_order_queue);
381 tcp_init_xmit_timers(sk);
382 tcp_prequeue_init(tp);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000383 INIT_LIST_HEAD(&tp->tsq_node);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000384
385 icsk->icsk_rto = TCP_TIMEOUT_INIT;
386 tp->mdev = TCP_TIMEOUT_INIT;
387
388 /* So many TCP implementations out there (incorrectly) count the
389 * initial SYN frame in their delayed-ACK and congestion control
390 * algorithms that we must have the following bandaid to talk
391 * efficiently to them. -DaveM
392 */
393 tp->snd_cwnd = TCP_INIT_CWND;
394
395 /* See draft-stevens-tcpca-spec-01 for discussion of the
396 * initialization of these values.
397 */
398 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
399 tp->snd_cwnd_clamp = ~0;
400 tp->mss_cache = TCP_MSS_DEFAULT;
401
402 tp->reordering = sysctl_tcp_reordering;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000403 tcp_enable_early_retrans(tp);
Neal Cardwell900f65d2012-04-19 09:55:21 +0000404 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
405
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000406 tp->tsoffset = 0;
407
Neal Cardwell900f65d2012-04-19 09:55:21 +0000408 sk->sk_state = TCP_CLOSE;
409
410 sk->sk_write_space = sk_stream_write_space;
411 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
412
413 icsk->icsk_sync_mss = tcp_sync_mss;
414
Neal Cardwell900f65d2012-04-19 09:55:21 +0000415 /* Presumed zeroed, in order of appearance:
416 * cookie_in_always, cookie_out_never,
417 * s_data_constant, s_data_in, s_data_out
418 */
419 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
Eric Dumazetaa395142010-04-20 13:03:51 +0000442 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700444 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700447 * by poll logic and correct handling of state changes
448 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
450
451 mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 /*
454 * POLLHUP is certainly not done right. But poll() doesn't
455 * have a notion of HUP in just one direction, and for a
456 * socket the read side is more interesting.
457 *
458 * Some poll() documentation says that POLLHUP is incompatible
459 * with the POLLOUT/POLLWR flags, so somebody should check this
460 * all. But careful, it tends to be safer to return too many
461 * bits than too few, and you can easily break real applications
462 * if you don't tell them that something has hung up!
463 *
464 * Check-me.
465 *
466 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
467 * our fs/select.c). It means that after we received EOF,
468 * poll always returns immediately, making impossible poll() on write()
469 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
470 * if and only if shutdown has been made in both directions.
471 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700472 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 * then we could set it on SND_SHUTDOWN. BTW examples given
474 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700475 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 *
477 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
478 * blocking on fresh not-connected or disconnected socket. --ANK
479 */
480 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
481 mask |= POLLHUP;
482 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800483 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Jerry Chu83368862012-08-31 12:29:12 +0000485 /* Connected or passive Fast Open socket? */
486 if (sk->sk_state != TCP_SYN_SENT &&
487 (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) {
David S. Millerc7004482008-10-06 10:43:54 -0700488 int target = sock_rcvlowat(sk, 0, INT_MAX);
489
490 if (tp->urg_seq == tp->copied_seq &&
491 !sock_flag(sk, SOCK_URGINLINE) &&
492 tp->urg_data)
Alexandra Kossovskyb634f872010-03-18 20:29:24 -0700493 target++;
David S. Millerc7004482008-10-06 10:43:54 -0700494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 /* Potential race condition. If read of tp below will
496 * escape above sk->sk_state, we can be illegally awaken
497 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700498 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 mask |= POLLIN | POLLRDNORM;
500
501 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
502 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
503 mask |= POLLOUT | POLLWRNORM;
504 } else { /* send SIGIO later */
505 set_bit(SOCK_ASYNC_NOSPACE,
506 &sk->sk_socket->flags);
507 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
508
509 /* Race breaker. If space is freed after
510 * wspace test but before the flags are set,
511 * IO signal will be lost.
512 */
513 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
514 mask |= POLLOUT | POLLWRNORM;
515 }
KOSAKI Motohirod84ba632010-08-24 16:05:48 +0000516 } else
517 mask |= POLLOUT | POLLWRNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
519 if (tp->urg_data & TCP_URG_VALID)
520 mask |= POLLPRI;
521 }
Tom Marshalla4d25802010-09-20 15:42:05 -0700522 /* This barrier is coupled with smp_wmb() in tcp_reset() */
523 smp_rmb();
524 if (sk->sk_err)
525 mask |= POLLERR;
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return mask;
528}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000529EXPORT_SYMBOL(tcp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
532{
533 struct tcp_sock *tp = tcp_sk(sk);
534 int answ;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000535 bool slow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
537 switch (cmd) {
538 case SIOCINQ:
539 if (sk->sk_state == TCP_LISTEN)
540 return -EINVAL;
541
Eric Dumazet0e71c552012-10-21 20:06:56 +0000542 slow = lock_sock_fast(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
544 answ = 0;
545 else if (sock_flag(sk, SOCK_URGINLINE) ||
546 !tp->urg_data ||
547 before(tp->urg_seq, tp->copied_seq) ||
548 !before(tp->urg_seq, tp->rcv_nxt)) {
David S. Miller91521942009-05-28 21:35:47 -0700549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 answ = tp->rcv_nxt - tp->copied_seq;
551
Eric Dumazeta3374c42012-10-18 09:14:12 +0000552 /* Subtract 1, if FIN was received */
553 if (answ && sock_flag(sk, SOCK_DONE))
554 answ--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 } else
556 answ = tp->urg_seq - tp->copied_seq;
Eric Dumazet0e71c552012-10-21 20:06:56 +0000557 unlock_sock_fast(sk, slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 break;
559 case SIOCATMARK:
560 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
561 break;
562 case SIOCOUTQ:
563 if (sk->sk_state == TCP_LISTEN)
564 return -EINVAL;
565
566 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
567 answ = 0;
568 else
569 answ = tp->write_seq - tp->snd_una;
570 break;
Mario Schuknecht2f4e1b32011-03-09 14:08:09 -0800571 case SIOCOUTQNSD:
572 if (sk->sk_state == TCP_LISTEN)
573 return -EINVAL;
574
575 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
576 answ = 0;
577 else
578 answ = tp->write_seq - tp->snd_nxt;
579 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 default:
581 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 return put_user(answ, (int __user *)arg);
585}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000586EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
589{
Eric Dumazet4de075e2011-09-27 13:25:05 -0400590 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 tp->pushed_seq = tp->write_seq;
592}
593
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000594static inline bool forced_push(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595{
596 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
597}
598
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700599static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700601 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200602 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
603
604 skb->csum = 0;
605 tcb->seq = tcb->end_seq = tp->write_seq;
Eric Dumazet4de075e2011-09-27 13:25:05 -0400606 tcb->tcp_flags = TCPHDR_ACK;
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200607 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800609 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800610 sk->sk_wmem_queued += skb->truesize;
611 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700612 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900613 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614}
615
Krishna Kumarafeca342009-12-10 07:16:52 +0000616static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700618 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700622static inline void tcp_push(struct sock *sk, int flags, int mss_now,
623 int nonagle)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
David S. Millerfe067e82007-03-07 12:12:44 -0800625 if (tcp_send_head(sk)) {
Krishna Kumarafeca342009-12-10 07:16:52 +0000626 struct tcp_sock *tp = tcp_sk(sk);
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (!(flags & MSG_MORE) || forced_push(tp))
Krishna Kumarafeca342009-12-10 07:16:52 +0000629 tcp_mark_push(tp, tcp_write_queue_tail(sk));
630
631 tcp_mark_urg(tp, flags);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700632 __tcp_push_pending_frames(sk, mss_now,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
634 }
635}
636
Adrian Bunk6ff77512007-11-06 23:32:26 -0800637static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
638 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800639{
640 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd02009-01-13 16:04:36 -0800641 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800642
Dimitris Michailidis9fa5fdf2009-01-26 22:15:31 -0800643 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
644 tss->flags);
Willy Tarreau33966dd02009-01-13 16:04:36 -0800645 if (ret > 0)
646 rd_desc->count -= ret;
647 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800648}
649
650static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
651{
652 /* Store TCP splice context information in read_descriptor_t. */
653 read_descriptor_t rd_desc = {
654 .arg.data = tss,
Willy Tarreau33966dd02009-01-13 16:04:36 -0800655 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800656 };
657
658 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
659}
660
661/**
662 * tcp_splice_read - splice data from TCP socket to a pipe
663 * @sock: socket to splice from
664 * @ppos: position (not valid)
665 * @pipe: pipe to splice to
666 * @len: number of bytes to splice
667 * @flags: splice modifier flags
668 *
669 * Description:
670 * Will read pages from given socket and fill them into a pipe.
671 *
672 **/
673ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
674 struct pipe_inode_info *pipe, size_t len,
675 unsigned int flags)
676{
677 struct sock *sk = sock->sk;
678 struct tcp_splice_state tss = {
679 .pipe = pipe,
680 .len = len,
681 .flags = flags,
682 };
683 long timeo;
684 ssize_t spliced;
685 int ret;
686
Changli Gao3a047bf2010-07-12 21:00:12 +0000687 sock_rps_record_flow(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -0800688 /*
689 * We can't seek on a socket input
690 */
691 if (unlikely(*ppos))
692 return -ESPIPE;
693
694 ret = spliced = 0;
695
696 lock_sock(sk);
697
Eric Dumazet42324c62009-10-01 15:26:00 -0700698 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
Jens Axboe9c55e012007-11-06 23:30:13 -0800699 while (tss.len) {
700 ret = __tcp_splice_read(sk, &tss);
701 if (ret < 0)
702 break;
703 else if (!ret) {
704 if (spliced)
705 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800706 if (sock_flag(sk, SOCK_DONE))
707 break;
708 if (sk->sk_err) {
709 ret = sock_error(sk);
710 break;
711 }
712 if (sk->sk_shutdown & RCV_SHUTDOWN)
713 break;
714 if (sk->sk_state == TCP_CLOSE) {
715 /*
716 * This occurs when user tries to read
717 * from never connected socket.
718 */
719 if (!sock_flag(sk, SOCK_DONE))
720 ret = -ENOTCONN;
721 break;
722 }
723 if (!timeo) {
724 ret = -EAGAIN;
725 break;
726 }
727 sk_wait_data(sk, &timeo);
728 if (signal_pending(current)) {
729 ret = sock_intr_errno(timeo);
730 break;
731 }
732 continue;
733 }
734 tss.len -= ret;
735 spliced += ret;
736
Willy Tarreau33966dd02009-01-13 16:04:36 -0800737 if (!timeo)
738 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800739 release_sock(sk);
740 lock_sock(sk);
741
742 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd02009-01-13 16:04:36 -0800743 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800744 signal_pending(current))
745 break;
746 }
747
748 release_sock(sk);
749
750 if (spliced)
751 return spliced;
752
753 return ret;
754}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000755EXPORT_SYMBOL(tcp_splice_read);
Jens Axboe9c55e012007-11-06 23:30:13 -0800756
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100757struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100758{
759 struct sk_buff *skb;
760
761 /* The TCP header must be at least 32-bit aligned. */
762 size = ALIGN(size, 4);
763
764 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
765 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800766 if (sk_wmem_schedule(sk, skb->truesize)) {
Eric Dumazeta21d4572012-04-10 20:30:48 +0000767 skb_reserve(skb, sk->sk_prot->max_header);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100768 /*
769 * Make sure that we have exactly size bytes
770 * available to the caller, no more, no less.
771 */
Eric Dumazet16fad692013-03-14 05:40:32 +0000772 skb->reserved_tailroom = skb->end - skb->tail - size;
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100773 return skb;
774 }
775 __kfree_skb(skb);
776 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700777 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100778 sk_stream_moderate_sndbuf(sk);
779 }
780 return NULL;
781}
782
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000783static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
784 int large_allowed)
785{
786 struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000787 u32 xmit_size_goal, old_size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000788
789 xmit_size_goal = mss_now;
790
791 if (large_allowed && sk_can_gso(sk)) {
792 xmit_size_goal = ((sk->sk_gso_max_size - 1) -
793 inet_csk(sk)->icsk_af_ops->net_header_len -
794 inet_csk(sk)->icsk_ext_hdr_len -
795 tp->tcp_header_len);
796
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000797 /* TSQ : try to have two TSO segments in flight */
798 xmit_size_goal = min_t(u32, xmit_size_goal,
799 sysctl_tcp_limit_output_bytes >> 1);
800
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000801 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000802
803 /* We try hard to avoid divides here */
804 old_size_goal = tp->xmit_size_goal_segs * mss_now;
805
806 if (likely(old_size_goal <= xmit_size_goal &&
807 old_size_goal + mss_now > xmit_size_goal)) {
808 xmit_size_goal = old_size_goal;
809 } else {
Ben Hutchings14853482012-07-30 16:11:42 +0000810 tp->xmit_size_goal_segs =
811 min_t(u16, xmit_size_goal / mss_now,
812 sk->sk_gso_max_segs);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000813 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
814 }
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000815 }
816
Ilpo Järvinenafece1c2009-03-14 14:23:07 +0000817 return max(xmit_size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000818}
819
820static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
821{
822 int mss_now;
823
824 mss_now = tcp_current_mss(sk);
825 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
826
827 return mss_now;
828}
829
Eric Dumazet64022d02012-12-01 13:07:02 +0000830static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
831 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700834 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 int err;
836 ssize_t copied;
837 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
838
Jerry Chu83368862012-08-31 12:29:12 +0000839 /* Wait for a connection to finish. One exception is TCP Fast Open
840 * (passive side) where data is allowed to be sent before a connection
841 * is fully established.
842 */
843 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
844 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
846 goto out_err;
Jerry Chu83368862012-08-31 12:29:12 +0000847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
850
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000851 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 copied = 0;
853
854 err = -EPIPE;
855 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000856 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Eric Dumazet64022d02012-12-01 13:07:02 +0000858 while (size > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800859 struct sk_buff *skb = tcp_write_queue_tail(sk);
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000860 int copy, i;
Eric Dumazet38ba0a62012-04-23 17:48:27 +0000861 bool can_coalesce;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
David S. Millerfe067e82007-03-07 12:12:44 -0800863 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864new_segment:
865 if (!sk_stream_memory_free(sk))
866 goto wait_for_sndbuf;
867
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100868 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (!skb)
870 goto wait_for_memory;
871
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700872 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700873 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
875
876 if (copy > size)
877 copy = size;
878
879 i = skb_shinfo(skb)->nr_frags;
880 can_coalesce = skb_can_coalesce(skb, i, page, offset);
881 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
882 tcp_mark_push(tp, skb);
883 goto new_segment;
884 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800885 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (can_coalesce) {
Eric Dumazet9e903e02011-10-18 21:00:24 +0000889 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 } else {
891 get_page(page);
892 skb_fill_page_desc(skb, i, page, offset, copy);
893 }
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000894 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +0000895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 skb->len += copy;
897 skb->data_len += copy;
898 skb->truesize += copy;
899 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800900 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700901 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 tp->write_seq += copy;
903 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700904 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -0400907 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 copied += copy;
Eric Dumazet64022d02012-12-01 13:07:02 +0000910 offset += copy;
911 if (!(size -= copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 goto out;
913
Herbert Xu69d15062008-03-22 15:47:05 -0700914 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 continue;
916
917 if (forced_push(tp)) {
918 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700919 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800920 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 tcp_push_one(sk, mss_now);
922 continue;
923
924wait_for_sndbuf:
925 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
926wait_for_memory:
Willy Tarreaubad115c2012-05-17 11:14:14 +0000927 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
930 goto do_error;
931
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000932 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
934
935out:
Eric Dumazet35f9c092012-04-05 03:05:35 +0000936 if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700937 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 return copied;
939
940do_error:
941 if (copied)
942 goto out;
943out_err:
944 return sk_stream_error(sk, flags, err);
945}
946
Changli Gao7ba42912010-07-10 20:41:55 +0000947int tcp_sendpage(struct sock *sk, struct page *page, int offset,
948 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
950 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -0700953 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Changli Gao7ba42912010-07-10 20:41:55 +0000954 return sock_no_sendpage(sk->sk_socket, page, offset, size,
955 flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 lock_sock(sk);
Eric Dumazet64022d02012-12-01 13:07:02 +0000958 res = do_tcp_sendpages(sk, page, offset, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 release_sock(sk);
960 return res;
961}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000962EXPORT_SYMBOL(tcp_sendpage);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Eric Dumazet690e99c2011-11-28 00:27:47 +0000964static inline int select_size(const struct sock *sk, bool sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400966 const struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700967 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Krishna Kumardef87cf2009-12-10 07:16:59 +0000969 if (sg) {
Eric Dumazetf07d9602011-11-28 22:41:47 +0000970 if (sk_can_gso(sk)) {
971 /* Small frames wont use a full page:
972 * Payload will immediately follow tcp header.
973 */
974 tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
975 } else {
David S. Millerb4e26f52005-07-05 15:20:27 -0700976 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
977
978 if (tmp >= pgbreak &&
979 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
980 tmp = pgbreak;
981 }
982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 return tmp;
985}
986
Yuchung Chengcf60af02012-07-19 06:43:09 +0000987void tcp_free_fastopen_req(struct tcp_sock *tp)
988{
989 if (tp->fastopen_req != NULL) {
990 kfree(tp->fastopen_req);
991 tp->fastopen_req = NULL;
992 }
993}
994
995static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *size)
996{
997 struct tcp_sock *tp = tcp_sk(sk);
998 int err, flags;
999
1000 if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
1001 return -EOPNOTSUPP;
1002 if (tp->fastopen_req != NULL)
1003 return -EALREADY; /* Another Fast Open is in progress */
1004
1005 tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
1006 sk->sk_allocation);
1007 if (unlikely(tp->fastopen_req == NULL))
1008 return -ENOBUFS;
1009 tp->fastopen_req->data = msg;
1010
1011 flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
1012 err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1013 msg->msg_namelen, flags);
1014 *size = tp->fastopen_req->copied;
1015 tcp_free_fastopen_req(tp);
1016 return err;
1017}
1018
Changli Gao7ba42912010-07-10 20:41:55 +00001019int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 size_t size)
1021{
1022 struct iovec *iov;
1023 struct tcp_sock *tp = tcp_sk(sk);
1024 struct sk_buff *skb;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001025 int iovlen, flags, err, copied = 0;
1026 int mss_now = 0, size_goal, copied_syn = 0, offset = 0;
Eric Dumazet690e99c2011-11-28 00:27:47 +00001027 bool sg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 long timeo;
1029
1030 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 flags = msg->msg_flags;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001033 if (flags & MSG_FASTOPEN) {
1034 err = tcp_sendmsg_fastopen(sk, msg, &copied_syn);
1035 if (err == -EINPROGRESS && copied_syn > 0)
1036 goto out;
1037 else if (err)
1038 goto out_err;
1039 offset = copied_syn;
1040 }
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1043
Jerry Chu83368862012-08-31 12:29:12 +00001044 /* Wait for a connection to finish. One exception is TCP Fast Open
1045 * (passive side) where data is allowed to be sent before a connection
1046 * is fully established.
1047 */
1048 if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
1049 !tcp_passive_fastopen(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
Yuchung Chengcf60af02012-07-19 06:43:09 +00001051 goto do_error;
Jerry Chu83368862012-08-31 12:29:12 +00001052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001054 if (unlikely(tp->repair)) {
1055 if (tp->repair_queue == TCP_RECV_QUEUE) {
1056 copied = tcp_send_rcvq(sk, msg, size);
1057 goto out;
1058 }
1059
1060 err = -EINVAL;
1061 if (tp->repair_queue == TCP_NO_QUEUE)
1062 goto out_err;
1063
1064 /* 'common' sending to sendq */
1065 }
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 /* This should be in poll */
1068 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1069
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001070 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072 /* Ok commence sending. */
1073 iovlen = msg->msg_iovlen;
1074 iov = msg->msg_iov;
1075 copied = 0;
1076
1077 err = -EPIPE;
1078 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +00001079 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Eric Dumazet690e99c2011-11-28 00:27:47 +00001081 sg = !!(sk->sk_route_caps & NETIF_F_SG);
Krishna Kumardef87cf2009-12-10 07:16:59 +00001082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 while (--iovlen >= 0) {
David S. Miller01db4032010-09-27 20:24:54 -07001084 size_t seglen = iov->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 unsigned char __user *from = iov->iov_base;
1086
1087 iov++;
Yuchung Chengcf60af02012-07-19 06:43:09 +00001088 if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */
1089 if (offset >= seglen) {
1090 offset -= seglen;
1091 continue;
1092 }
1093 seglen -= offset;
1094 from += offset;
1095 offset = 0;
1096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 while (seglen > 0) {
Herbert Xu6828b922009-06-28 18:06:41 +00001099 int copy = 0;
1100 int max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
David S. Millerfe067e82007-03-07 12:12:44 -08001102 skb = tcp_write_queue_tail(sk);
Herbert Xu6828b922009-06-28 18:06:41 +00001103 if (tcp_send_head(sk)) {
1104 if (skb->ip_summed == CHECKSUM_NONE)
1105 max = mss_now;
1106 copy = max - skb->len;
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Herbert Xu6828b922009-06-28 18:06:41 +00001109 if (copy <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110new_segment:
1111 /* Allocate new segment. If the interface is SG,
1112 * allocate skb fitting to single page.
1113 */
1114 if (!sk_stream_memory_free(sk))
1115 goto wait_for_sndbuf;
1116
Krishna Kumardef87cf2009-12-10 07:16:59 +00001117 skb = sk_stream_alloc_skb(sk,
1118 select_size(sk, sg),
1119 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if (!skb)
1121 goto wait_for_memory;
1122
1123 /*
1124 * Check whether we can use HW checksum.
1125 */
Herbert Xu8648b302006-06-17 22:06:05 -07001126 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -07001127 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001129 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001130 copy = size_goal;
Herbert Xu6828b922009-06-28 18:06:41 +00001131 max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 }
1133
1134 /* Try to append data to the end of skb. */
1135 if (copy > seglen)
1136 copy = seglen;
1137
1138 /* Where to copy to? */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001139 if (skb_availroom(skb) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 /* We have some space in skb head. Superb! */
Eric Dumazeta21d4572012-04-10 20:30:48 +00001141 copy = min_t(int, copy, skb_availroom(skb));
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001142 err = skb_add_data_nocache(sk, skb, from, copy);
1143 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 goto do_fault;
1145 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001146 bool merge = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 int i = skb_shinfo(skb)->nr_frags;
Eric Dumazet5640f762012-09-23 23:04:42 +00001148 struct page_frag *pfrag = sk_page_frag(sk);
Eric Dumazet761965e2011-12-04 07:05:17 +00001149
Eric Dumazet5640f762012-09-23 23:04:42 +00001150 if (!sk_page_frag_refill(sk, pfrag))
1151 goto wait_for_memory;
Eric Dumazet761965e2011-12-04 07:05:17 +00001152
Eric Dumazet5640f762012-09-23 23:04:42 +00001153 if (!skb_can_coalesce(skb, i, pfrag->page,
1154 pfrag->offset)) {
1155 if (i == MAX_SKB_FRAGS || !sg) {
1156 tcp_mark_push(tp, skb);
1157 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001159 merge = false;
1160 }
Herbert Xuef015782005-09-01 17:48:59 -07001161
Eric Dumazet5640f762012-09-23 23:04:42 +00001162 copy = min_t(int, copy, pfrag->size - pfrag->offset);
Herbert Xuef015782005-09-01 17:48:59 -07001163
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001164 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -07001165 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Tom Herbertc6e1a0d2011-04-04 22:30:30 -07001167 err = skb_copy_to_page_nocache(sk, from, skb,
Eric Dumazet5640f762012-09-23 23:04:42 +00001168 pfrag->page,
1169 pfrag->offset,
1170 copy);
1171 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 goto do_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 /* Update the skb. */
1175 if (merge) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001176 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 } else {
Eric Dumazet5640f762012-09-23 23:04:42 +00001178 skb_fill_page_desc(skb, i, pfrag->page,
1179 pfrag->offset, copy);
1180 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001182 pfrag->offset += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 }
1184
1185 if (!copied)
Eric Dumazet4de075e2011-09-27 13:25:05 -04001186 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
1188 tp->write_seq += copy;
1189 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -07001190 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 from += copy;
1193 copied += copy;
1194 if ((seglen -= copy) == 0 && iovlen == 0)
1195 goto out;
1196
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001197 if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 continue;
1199
1200 if (forced_push(tp)) {
1201 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001202 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001203 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 tcp_push_one(sk, mss_now);
1205 continue;
1206
1207wait_for_sndbuf:
1208 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1209wait_for_memory:
Andrew Vaginec342322012-11-15 04:03:17 +00001210 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001211 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1214 goto do_error;
1215
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001216 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
1218 }
1219
1220out:
Andrew Vaginec342322012-11-15 04:03:17 +00001221 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001222 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 release_sock(sk);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001224 return copied + copied_syn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
1226do_fault:
1227 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001228 tcp_unlink_write_queue(skb, sk);
1229 /* It is the one place in all of TCP, except connection
1230 * reset, where we can be unlinking the send_head.
1231 */
1232 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001233 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235
1236do_error:
Yuchung Chengcf60af02012-07-19 06:43:09 +00001237 if (copied + copied_syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 goto out;
1239out_err:
1240 err = sk_stream_error(sk, flags, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 release_sock(sk);
1242 return err;
1243}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001244EXPORT_SYMBOL(tcp_sendmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
1246/*
1247 * Handle reading urgent data. BSD has very simple semantics for
1248 * this, no blocking and very strange errors 8)
1249 */
1250
Rami Rosen377f0a02009-03-31 14:43:17 -07001251static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252{
1253 struct tcp_sock *tp = tcp_sk(sk);
1254
1255 /* No URG data to read. */
1256 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1257 tp->urg_data == TCP_URG_READ)
1258 return -EINVAL; /* Yes this is right ! */
1259
1260 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1261 return -ENOTCONN;
1262
1263 if (tp->urg_data & TCP_URG_VALID) {
1264 int err = 0;
1265 char c = tp->urg_data;
1266
1267 if (!(flags & MSG_PEEK))
1268 tp->urg_data = TCP_URG_READ;
1269
1270 /* Read urgent data. */
1271 msg->msg_flags |= MSG_OOB;
1272
1273 if (len > 0) {
1274 if (!(flags & MSG_TRUNC))
1275 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1276 len = 1;
1277 } else
1278 msg->msg_flags |= MSG_TRUNC;
1279
1280 return err ? -EFAULT : len;
1281 }
1282
1283 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1284 return 0;
1285
1286 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1287 * the available implementations agree in this case:
1288 * this call should never block, independent of the
1289 * blocking state of the socket.
1290 * Mike <pall@rz.uni-karlsruhe.de>
1291 */
1292 return -EAGAIN;
1293}
1294
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001295static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
1296{
1297 struct sk_buff *skb;
1298 int copied = 0, err = 0;
1299
1300 /* XXX -- need to support SO_PEEK_OFF */
1301
1302 skb_queue_walk(&sk->sk_write_queue, skb) {
1303 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, skb->len);
1304 if (err)
1305 break;
1306
1307 copied += skb->len;
1308 }
1309
1310 return err ?: copied;
1311}
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313/* Clean up the receive buffer for full frames taken by the user,
1314 * then send an ACK if necessary. COPIED is the number of bytes
1315 * tcp_recvmsg has given to the user so far, it speeds up the
1316 * calculation of whether or not we must ACK for the sake of
1317 * a window update.
1318 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001319void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
1321 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001322 bool time_to_ack = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1325
Ilpo Järvinend792c102009-11-13 13:56:33 -08001326 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001327 "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
Ilpo Järvinend792c102009-11-13 13:56:33 -08001328 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001330 if (inet_csk_ack_scheduled(sk)) {
1331 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 /* Delayed ACKs frequently hit locked sockets during bulk
1333 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001334 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001336 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 /*
1338 * If this read emptied read buffer, we send ACK, if
1339 * connection is not bidirectional, user drained
1340 * receive buffer and there was a small segment
1341 * in queue.
1342 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001343 (copied > 0 &&
1344 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1345 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1346 !icsk->icsk_ack.pingpong)) &&
1347 !atomic_read(&sk->sk_rmem_alloc)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001348 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
1350
1351 /* We send an ACK if we can now advertise a non-zero window
1352 * which has been raised "significantly".
1353 *
1354 * Even if window raised up to infinity, do not send window open ACK
1355 * in states, where we will not receive more. It is useless.
1356 */
1357 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1358 __u32 rcv_window_now = tcp_receive_window(tp);
1359
1360 /* Optimize, __tcp_select_window() is not cheap. */
1361 if (2*rcv_window_now <= tp->window_clamp) {
1362 __u32 new_window = __tcp_select_window(sk);
1363
1364 /* Send ACK now, if this read freed lots of space
1365 * in our buffer. Certainly, new_window is new window.
1366 * We can advertise it now, if it is not less than current one.
1367 * "Lots" means "at least twice" here.
1368 */
1369 if (new_window && new_window >= 2 * rcv_window_now)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001370 time_to_ack = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372 }
1373 if (time_to_ack)
1374 tcp_send_ack(sk);
1375}
1376
1377static void tcp_prequeue_process(struct sock *sk)
1378{
1379 struct sk_buff *skb;
1380 struct tcp_sock *tp = tcp_sk(sk);
1381
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001382 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 /* RX process wants to run with disabled BHs, though it is not
1385 * necessary */
1386 local_bh_disable();
1387 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001388 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 local_bh_enable();
1390
1391 /* Clear memory counter. */
1392 tp->ucopy.memory = 0;
1393}
1394
Steven J. Magnani73852e82010-03-16 05:22:44 +00001395#ifdef CONFIG_NET_DMA
1396static void tcp_service_net_dma(struct sock *sk, bool wait)
1397{
1398 dma_cookie_t done, used;
1399 dma_cookie_t last_issued;
1400 struct tcp_sock *tp = tcp_sk(sk);
1401
1402 if (!tp->ucopy.dma_chan)
1403 return;
1404
1405 last_issued = tp->ucopy.dma_cookie;
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001406 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001407
1408 do {
Bartlomiej Zolnierkiewicze239345f2012-11-08 10:01:01 +00001409 if (dma_async_is_tx_complete(tp->ucopy.dma_chan,
Steven J. Magnani73852e82010-03-16 05:22:44 +00001410 last_issued, &done,
1411 &used) == DMA_SUCCESS) {
1412 /* Safe to free early-copied skbs now */
1413 __skb_queue_purge(&sk->sk_async_wait_queue);
1414 break;
1415 } else {
1416 struct sk_buff *skb;
1417 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1418 (dma_async_is_complete(skb->dma_cookie, done,
1419 used) == DMA_SUCCESS)) {
1420 __skb_dequeue(&sk->sk_async_wait_queue);
1421 kfree_skb(skb);
1422 }
1423 }
1424 } while (wait);
1425}
1426#endif
1427
Eric Dumazetf26845b2013-01-09 20:59:09 +00001428static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
1430 struct sk_buff *skb;
1431 u32 offset;
1432
Eric Dumazetf26845b2013-01-09 20:59:09 +00001433 while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001435 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001437 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 *off = offset;
1439 return skb;
1440 }
Eric Dumazetf26845b2013-01-09 20:59:09 +00001441 /* This looks weird, but this can happen if TCP collapsing
1442 * splitted a fat GRO packet, while we released socket lock
1443 * in skb_splice_bits()
1444 */
1445 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447 return NULL;
1448}
1449
1450/*
1451 * This routine provides an alternative to tcp_recvmsg() for routines
1452 * that would like to handle copying from skbuffs directly in 'sendfile'
1453 * fashion.
1454 * Note:
1455 * - It is assumed that the socket was locked by the caller.
1456 * - The routine does not block.
1457 * - At present, there is no support for reading OOB data
1458 * or for 'peeking' the socket using this routine
1459 * (although both would be easy to implement).
1460 */
1461int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1462 sk_read_actor_t recv_actor)
1463{
1464 struct sk_buff *skb;
1465 struct tcp_sock *tp = tcp_sk(sk);
1466 u32 seq = tp->copied_seq;
1467 u32 offset;
1468 int copied = 0;
1469
1470 if (sk->sk_state == TCP_LISTEN)
1471 return -ENOTCONN;
1472 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1473 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001474 int used;
1475 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
1477 len = skb->len - offset;
1478 /* Stop reading if we hit a patch of urgent data */
1479 if (tp->urg_data) {
1480 u32 urg_offset = tp->urg_seq - seq;
1481 if (urg_offset < len)
1482 len = urg_offset;
1483 if (!len)
1484 break;
1485 }
1486 used = recv_actor(desc, skb, offset, len);
Eric Dumazetff905b12013-01-10 07:06:10 +00001487 if (used <= 0) {
Jens Axboeddb61a52007-06-23 23:07:50 -07001488 if (!copied)
1489 copied = used;
1490 break;
1491 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 seq += used;
1493 copied += used;
1494 offset += used;
1495 }
Willy Tarreau02275a22012-12-02 11:49:27 +00001496 /* If recv_actor drops the lock (e.g. TCP splice
Octavian Purdila293ad602008-06-04 15:45:58 -07001497 * receive) the skb pointer might be invalid when
1498 * getting here: tcp_collapse might have deleted it
1499 * while aggregating skbs from the socket queue.
1500 */
Willy Tarreau02275a22012-12-02 11:49:27 +00001501 skb = tcp_recv_skb(sk, seq - 1, &offset);
1502 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 break;
Willy Tarreau02275a22012-12-02 11:49:27 +00001504 /* TCP coalescing might have appended data to the skb.
1505 * Try to splice more frags
1506 */
1507 if (offset + 1 != skb->len)
1508 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001510 if (tcp_hdr(skb)->fin) {
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001511 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 ++seq;
1513 break;
1514 }
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001515 sk_eat_skb(sk, skb, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 if (!desc->count)
1517 break;
Steven J. Magnanibaff42a2010-03-30 13:56:01 -07001518 tp->copied_seq = seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 }
1520 tp->copied_seq = seq;
1521
1522 tcp_rcv_space_adjust(sk);
1523
1524 /* Clean up data we have read: This will do ACK frames. */
Eric Dumazetf26845b2013-01-09 20:59:09 +00001525 if (copied > 0) {
1526 tcp_recv_skb(sk, seq, &offset);
Chris Leech0e4b4992006-05-23 18:00:16 -07001527 tcp_cleanup_rbuf(sk, copied);
Eric Dumazetf26845b2013-01-09 20:59:09 +00001528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return copied;
1530}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001531EXPORT_SYMBOL(tcp_read_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533/*
1534 * This routine copies from a sock struct into the user buffer.
1535 *
1536 * Technical note: in 2.3 we work on _locked_ socket, so that
1537 * tricks with *seq access order and skb->users are not required.
1538 * Probably, code can be easily improved even more.
1539 */
1540
1541int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1542 size_t len, int nonblock, int flags, int *addr_len)
1543{
1544 struct tcp_sock *tp = tcp_sk(sk);
1545 int copied = 0;
1546 u32 peek_seq;
1547 u32 *seq;
1548 unsigned long used;
1549 int err;
1550 int target; /* Read at least this many bytes */
1551 long timeo;
1552 struct task_struct *user_recv = NULL;
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001553 bool copied_early = false;
Chris Leech2b1244a2007-03-08 09:57:36 -08001554 struct sk_buff *skb;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001555 u32 urg_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 lock_sock(sk);
1558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 err = -ENOTCONN;
1560 if (sk->sk_state == TCP_LISTEN)
1561 goto out;
1562
1563 timeo = sock_rcvtimeo(sk, nonblock);
1564
1565 /* Urgent data needs to be handled specially. */
1566 if (flags & MSG_OOB)
1567 goto recv_urg;
1568
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001569 if (unlikely(tp->repair)) {
1570 err = -EPERM;
1571 if (!(flags & MSG_PEEK))
1572 goto out;
1573
1574 if (tp->repair_queue == TCP_SEND_QUEUE)
1575 goto recv_sndq;
1576
1577 err = -EINVAL;
1578 if (tp->repair_queue == TCP_NO_QUEUE)
1579 goto out;
1580
1581 /* 'common' recv queue MSG_PEEK-ing */
1582 }
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 seq = &tp->copied_seq;
1585 if (flags & MSG_PEEK) {
1586 peek_seq = tp->copied_seq;
1587 seq = &peek_seq;
1588 }
1589
1590 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1591
Chris Leech1a2449a2006-05-23 18:05:53 -07001592#ifdef CONFIG_NET_DMA
1593 tp->ucopy.dma_chan = NULL;
1594 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001595 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001596 {
1597 int available = 0;
1598
1599 if (skb)
1600 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1601 if ((available < target) &&
1602 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1603 !sysctl_tcp_low_latency &&
Dave Jianga2bd1142012-04-04 16:10:46 -07001604 net_dma_find_channel()) {
Andrew Mortone00c5d82007-03-08 09:57:36 -08001605 preempt_enable_no_resched();
1606 tp->ucopy.pinned_list =
1607 dma_pin_iovec_pages(msg->msg_iov, len);
1608 } else {
1609 preempt_enable_no_resched();
1610 }
1611 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001612#endif
1613
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 u32 offset;
1616
1617 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1618 if (tp->urg_data && tp->urg_seq == *seq) {
1619 if (copied)
1620 break;
1621 if (signal_pending(current)) {
1622 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1623 break;
1624 }
1625 }
1626
1627 /* Next get a buffer. */
1628
David S. Miller91521942009-05-28 21:35:47 -07001629 skb_queue_walk(&sk->sk_receive_queue, skb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 /* Now that we have two receive queues this
1631 * shouldn't happen.
1632 */
Ilpo Järvinend792c102009-11-13 13:56:33 -08001633 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
Joe Perches2af6fd82010-10-30 11:08:53 +00001634 "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
1635 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1636 flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 break;
Ilpo Järvinend792c102009-11-13 13:56:33 -08001638
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001640 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 offset--;
1642 if (offset < skb->len)
1643 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001644 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 goto found_fin_ok;
Joe Perches2af6fd82010-10-30 11:08:53 +00001646 WARN(!(flags & MSG_PEEK),
1647 "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
1648 *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
David S. Miller91521942009-05-28 21:35:47 -07001649 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 /* Well, if we have backlog, try to process it now yet. */
1652
1653 if (copied >= target && !sk->sk_backlog.tail)
1654 break;
1655
1656 if (copied) {
1657 if (sk->sk_err ||
1658 sk->sk_state == TCP_CLOSE ||
1659 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1660 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08001661 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 break;
1663 } else {
1664 if (sock_flag(sk, SOCK_DONE))
1665 break;
1666
1667 if (sk->sk_err) {
1668 copied = sock_error(sk);
1669 break;
1670 }
1671
1672 if (sk->sk_shutdown & RCV_SHUTDOWN)
1673 break;
1674
1675 if (sk->sk_state == TCP_CLOSE) {
1676 if (!sock_flag(sk, SOCK_DONE)) {
1677 /* This occurs when user tries to read
1678 * from never connected socket.
1679 */
1680 copied = -ENOTCONN;
1681 break;
1682 }
1683 break;
1684 }
1685
1686 if (!timeo) {
1687 copied = -EAGAIN;
1688 break;
1689 }
1690
1691 if (signal_pending(current)) {
1692 copied = sock_intr_errno(timeo);
1693 break;
1694 }
1695 }
1696
Chris Leech0e4b4992006-05-23 18:00:16 -07001697 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
David S. Miller7df55122005-06-18 23:01:10 -07001699 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 /* Install new reader */
1701 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1702 user_recv = current;
1703 tp->ucopy.task = user_recv;
1704 tp->ucopy.iov = msg->msg_iov;
1705 }
1706
1707 tp->ucopy.len = len;
1708
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001709 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1710 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 /* Ugly... If prequeue is not empty, we have to
1713 * process it before releasing socket, otherwise
1714 * order will be broken at second iteration.
1715 * More elegant solution is required!!!
1716 *
1717 * Look: we have the following (pseudo)queues:
1718 *
1719 * 1. packets in flight
1720 * 2. backlog
1721 * 3. prequeue
1722 * 4. receive_queue
1723 *
1724 * Each queue can be processed only if the next ones
1725 * are empty. At this point we have empty receive_queue.
1726 * But prequeue _can_ be not empty after 2nd iteration,
1727 * when we jumped to start of loop because backlog
1728 * processing added something to receive_queue.
1729 * We cannot release_sock(), because backlog contains
1730 * packets arrived _after_ prequeued ones.
1731 *
1732 * Shortly, algorithm is clear --- to process all
1733 * the queues in order. We could make it more directly,
1734 * requeueing packets from backlog to prequeue, if
1735 * is not empty. It is more elegant, but eats cycles,
1736 * unfortunately.
1737 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001738 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 goto do_prequeue;
1740
1741 /* __ Set realtime policy in scheduler __ */
1742 }
1743
Steven J. Magnani73852e82010-03-16 05:22:44 +00001744#ifdef CONFIG_NET_DMA
Michal Kubeček15c04172012-09-14 04:59:52 +00001745 if (tp->ucopy.dma_chan) {
1746 if (tp->rcv_wnd == 0 &&
1747 !skb_queue_empty(&sk->sk_async_wait_queue)) {
1748 tcp_service_net_dma(sk, true);
1749 tcp_cleanup_rbuf(sk, copied);
1750 } else
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001751 dma_async_issue_pending(tp->ucopy.dma_chan);
Michal Kubeček15c04172012-09-14 04:59:52 +00001752 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001753#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 if (copied >= target) {
1755 /* Do not sleep, just process backlog. */
1756 release_sock(sk);
1757 lock_sock(sk);
1758 } else
1759 sk_wait_data(sk, &timeo);
1760
Chris Leech1a2449a2006-05-23 18:05:53 -07001761#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001762 tcp_service_net_dma(sk, false); /* Don't block */
Chris Leech1a2449a2006-05-23 18:05:53 -07001763 tp->ucopy.wakeup = 0;
1764#endif
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (user_recv) {
1767 int chunk;
1768
1769 /* __ Restore normal policy in scheduler __ */
1770
1771 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001772 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 len -= chunk;
1774 copied += chunk;
1775 }
1776
1777 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001778 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779do_prequeue:
1780 tcp_prequeue_process(sk);
1781
1782 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001783 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 len -= chunk;
1785 copied += chunk;
1786 }
1787 }
1788 }
Ilpo Järvinen77527312009-05-10 20:32:34 +00001789 if ((flags & MSG_PEEK) &&
1790 (peek_seq - copied - urg_hole != tp->copied_seq)) {
Joe Perchese87cc472012-05-13 21:56:26 +00001791 net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
1792 current->comm,
1793 task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 peek_seq = tp->copied_seq;
1795 }
1796 continue;
1797
1798 found_ok_skb:
1799 /* Ok so how much can we use? */
1800 used = skb->len - offset;
1801 if (len < used)
1802 used = len;
1803
1804 /* Do we have urgent data here? */
1805 if (tp->urg_data) {
1806 u32 urg_offset = tp->urg_seq - *seq;
1807 if (urg_offset < used) {
1808 if (!urg_offset) {
1809 if (!sock_flag(sk, SOCK_URGINLINE)) {
1810 ++*seq;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001811 urg_hole++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 offset++;
1813 used--;
1814 if (!used)
1815 goto skip_copy;
1816 }
1817 } else
1818 used = urg_offset;
1819 }
1820 }
1821
1822 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001823#ifdef CONFIG_NET_DMA
1824 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dave Jianga2bd1142012-04-04 16:10:46 -07001825 tp->ucopy.dma_chan = net_dma_find_channel();
Chris Leech1a2449a2006-05-23 18:05:53 -07001826
1827 if (tp->ucopy.dma_chan) {
1828 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1829 tp->ucopy.dma_chan, skb, offset,
1830 msg->msg_iov, used,
1831 tp->ucopy.pinned_list);
1832
1833 if (tp->ucopy.dma_cookie < 0) {
1834
Joe Perchesafd465032012-03-12 07:03:32 +00001835 pr_alert("%s: dma_cookie < 0\n",
1836 __func__);
Chris Leech1a2449a2006-05-23 18:05:53 -07001837
1838 /* Exception. Bailout! */
1839 if (!copied)
1840 copied = -EFAULT;
1841 break;
1842 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001843
Bartlomiej Zolnierkiewiczb9ee8682012-11-08 09:59:54 +00001844 dma_async_issue_pending(tp->ucopy.dma_chan);
Steven J. Magnani73852e82010-03-16 05:22:44 +00001845
Chris Leech1a2449a2006-05-23 18:05:53 -07001846 if ((offset + used) == skb->len)
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001847 copied_early = true;
Chris Leech1a2449a2006-05-23 18:05:53 -07001848
1849 } else
1850#endif
1851 {
1852 err = skb_copy_datagram_iovec(skb, offset,
1853 msg->msg_iov, used);
1854 if (err) {
1855 /* Exception. Bailout! */
1856 if (!copied)
1857 copied = -EFAULT;
1858 break;
1859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 }
1861 }
1862
1863 *seq += used;
1864 copied += used;
1865 len -= used;
1866
1867 tcp_rcv_space_adjust(sk);
1868
1869skip_copy:
1870 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1871 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001872 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
1874 if (used + offset < skb->len)
1875 continue;
1876
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001877 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001879 if (!(flags & MSG_PEEK)) {
1880 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001881 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 continue;
1884
1885 found_fin_ok:
1886 /* Process the FIN. */
1887 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001888 if (!(flags & MSG_PEEK)) {
1889 sk_eat_skb(sk, skb, copied_early);
Eric Dumazetdc6b9b72012-05-16 22:48:15 +00001890 copied_early = false;
Chris Leech1a2449a2006-05-23 18:05:53 -07001891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 break;
1893 } while (len > 0);
1894
1895 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001896 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 int chunk;
1898
1899 tp->ucopy.len = copied > 0 ? len : 0;
1900
1901 tcp_prequeue_process(sk);
1902
1903 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001904 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 len -= chunk;
1906 copied += chunk;
1907 }
1908 }
1909
1910 tp->ucopy.task = NULL;
1911 tp->ucopy.len = 0;
1912 }
1913
Chris Leech1a2449a2006-05-23 18:05:53 -07001914#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001915 tcp_service_net_dma(sk, true); /* Wait for queue to drain */
1916 tp->ucopy.dma_chan = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001917
Chris Leech1a2449a2006-05-23 18:05:53 -07001918 if (tp->ucopy.pinned_list) {
1919 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1920 tp->ucopy.pinned_list = NULL;
1921 }
1922#endif
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 /* According to UNIX98, msg_name/msg_namelen are ignored
1925 * on connected socket. I was just happy when found this 8) --ANK
1926 */
1927
1928 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001929 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 release_sock(sk);
1932 return copied;
1933
1934out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 release_sock(sk);
1936 return err;
1937
1938recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07001939 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 goto out;
Pavel Emelyanovc0e88ff2012-04-19 03:41:01 +00001941
1942recv_sndq:
1943 err = tcp_peek_sndq(sk, msg, len);
1944 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001946EXPORT_SYMBOL(tcp_recvmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001948void tcp_set_state(struct sock *sk, int state)
1949{
1950 int oldstate = sk->sk_state;
1951
1952 switch (state) {
1953 case TCP_ESTABLISHED:
1954 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001955 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001956 break;
1957
1958 case TCP_CLOSE:
1959 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001960 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001961
1962 sk->sk_prot->unhash(sk);
1963 if (inet_csk(sk)->icsk_bind_hash &&
1964 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001965 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001966 /* fall through */
1967 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001968 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07001969 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001970 }
1971
1972 /* Change state AFTER socket is unhashed to avoid closed
1973 * socket sitting in hash tables.
1974 */
1975 sk->sk_state = state;
1976
1977#ifdef STATE_TRACE
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001978 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001979#endif
1980}
1981EXPORT_SYMBOL_GPL(tcp_set_state);
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983/*
1984 * State processing on a close. This implements the state shift for
1985 * sending our FIN frame. Note that we only send a FIN for some
1986 * states. A shutdown() may have already sent the FIN, or we may be
1987 * closed.
1988 */
1989
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08001990static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 /* current state: new state: action: */
1992 /* (Invalid) */ TCP_CLOSE,
1993 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1994 /* TCP_SYN_SENT */ TCP_CLOSE,
1995 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1996 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
1997 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
1998 /* TCP_TIME_WAIT */ TCP_CLOSE,
1999 /* TCP_CLOSE */ TCP_CLOSE,
2000 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
2001 /* TCP_LAST_ACK */ TCP_LAST_ACK,
2002 /* TCP_LISTEN */ TCP_CLOSE,
2003 /* TCP_CLOSING */ TCP_CLOSING,
2004};
2005
2006static int tcp_close_state(struct sock *sk)
2007{
2008 int next = (int)new_state[sk->sk_state];
2009 int ns = next & TCP_STATE_MASK;
2010
2011 tcp_set_state(sk, ns);
2012
2013 return next & TCP_ACTION_FIN;
2014}
2015
2016/*
2017 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07002018 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 */
2020
2021void tcp_shutdown(struct sock *sk, int how)
2022{
2023 /* We need to grab some memory, and put together a FIN,
2024 * and then put it into the queue to be sent.
2025 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
2026 */
2027 if (!(how & SEND_SHUTDOWN))
2028 return;
2029
2030 /* If we've already sent a FIN, or it's a closed state, skip this. */
2031 if ((1 << sk->sk_state) &
2032 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
2033 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
2034 /* Clear out any half completed packets. FIN if needed. */
2035 if (tcp_close_state(sk))
2036 tcp_send_fin(sk);
2037 }
2038}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002039EXPORT_SYMBOL(tcp_shutdown);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002041bool tcp_check_oom(struct sock *sk, int shift)
2042{
2043 bool too_many_orphans, out_of_socket_memory;
2044
2045 too_many_orphans = tcp_too_many_orphans(sk, shift);
2046 out_of_socket_memory = tcp_out_of_memory(sk);
2047
Joe Perchese87cc472012-05-13 21:56:26 +00002048 if (too_many_orphans)
2049 net_info_ratelimited("too many orphaned sockets\n");
2050 if (out_of_socket_memory)
2051 net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002052 return too_many_orphans || out_of_socket_memory;
2053}
2054
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055void tcp_close(struct sock *sk, long timeout)
2056{
2057 struct sk_buff *skb;
2058 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07002059 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060
2061 lock_sock(sk);
2062 sk->sk_shutdown = SHUTDOWN_MASK;
2063
2064 if (sk->sk_state == TCP_LISTEN) {
2065 tcp_set_state(sk, TCP_CLOSE);
2066
2067 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002068 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 goto adjudge_to_death;
2071 }
2072
2073 /* We need to flush the recv. buffs. We do this only on the
2074 * descriptor close, not protocol-sourced closes, because the
2075 * reader process may not have drained the data yet!
2076 */
2077 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
2078 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002079 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 data_was_unread += len;
2081 __kfree_skb(skb);
2082 }
2083
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002084 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
Konstantin Khorenko565b7b22010-06-24 21:54:58 -07002086 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
2087 if (sk->sk_state == TCP_CLOSE)
2088 goto adjudge_to_death;
2089
Gerrit Renker65bb7232007-04-28 21:21:46 -07002090 /* As outlined in RFC 2525, section 2.17, we send a RST here because
2091 * data was lost. To witness the awful effects of the old behavior of
2092 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
2093 * GET in an FTP client, suspend the process, wait for the client to
2094 * advertise a zero window, then kill -9 the FTP client, wheee...
2095 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 */
Pavel Emelyanovee995282012-04-19 03:40:39 +00002097 if (unlikely(tcp_sk(sk)->repair)) {
2098 sk->sk_prot->disconnect(sk, 0);
2099 } else if (data_was_unread) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002101 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 tcp_set_state(sk, TCP_CLOSE);
Wu Fengguangaa133072009-09-02 23:45:45 -07002103 tcp_send_active_reset(sk, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
2105 /* Check zero linger _after_ checking for unread data. */
2106 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07002107 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 } else if (tcp_close_state(sk)) {
2109 /* We FIN if the application ate all the data before
2110 * zapping the connection.
2111 */
2112
2113 /* RED-PEN. Formally speaking, we have broken TCP state
2114 * machine. State transitions:
2115 *
2116 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
2117 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
2118 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
2119 *
2120 * are legal only when FIN has been sent (i.e. in window),
2121 * rather than queued out of window. Purists blame.
2122 *
2123 * F.e. "RFC state" is ESTABLISHED,
2124 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
2125 *
2126 * The visible declinations are that sometimes
2127 * we enter time-wait state, when it is not required really
2128 * (harmless), do not send active resets, when they are
2129 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
2130 * they look as CLOSING or LAST_ACK for Linux)
2131 * Probably, I missed some more holelets.
2132 * --ANK
Jerry Chu83368862012-08-31 12:29:12 +00002133 * XXX (TFO) - To start off we don't support SYN+ACK+FIN
2134 * in a single packet! (May consider it later but will
2135 * probably need API support or TCP_CORK SYN-ACK until
2136 * data is written and socket is closed.)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 */
2138 tcp_send_fin(sk);
2139 }
2140
2141 sk_stream_wait_close(sk, timeout);
2142
2143adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07002144 state = sk->sk_state;
2145 sock_hold(sk);
2146 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07002147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 /* It is the last release_sock in its life. It will remove backlog. */
2149 release_sock(sk);
2150
2151
2152 /* Now socket is owned by kernel and we acquire BH lock
2153 to finish close. No need to check for user refs.
2154 */
2155 local_bh_disable();
2156 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002157 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
Herbert Xueb4dea52008-12-29 23:04:08 -08002159 percpu_counter_inc(sk->sk_prot->orphan_count);
2160
Herbert Xu75c2d9072006-05-03 23:31:35 -07002161 /* Have we already been destroyed by a softirq or backlog? */
2162 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
2163 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
2165 /* This is a (useful) BSD violating of the RFC. There is a
2166 * problem with TCP as specified in that the other end could
2167 * keep a socket open forever with no application left this end.
2168 * We use a 3 minute timeout (about the same as BSD) then kill
2169 * our end. If they send after that then tough - BUT: long enough
2170 * that we won't make the old 4*rto = almost no time - whoops
2171 * reset mistake.
2172 *
2173 * Nope, it was not mistake. It is really desired behaviour
2174 * f.e. on http servers, when such sockets are useless, but
2175 * consume significant resources. Let's do it with special
2176 * linger2 option. --ANK
2177 */
2178
2179 if (sk->sk_state == TCP_FIN_WAIT2) {
2180 struct tcp_sock *tp = tcp_sk(sk);
2181 if (tp->linger2 < 0) {
2182 tcp_set_state(sk, TCP_CLOSE);
2183 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002184 NET_INC_STATS_BH(sock_net(sk),
2185 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002187 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
2189 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07002190 inet_csk_reset_keepalive_timer(sk,
2191 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2194 goto out;
2195 }
2196 }
2197 }
2198 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002199 sk_mem_reclaim(sk);
Arun Sharmaefcdbf22012-01-30 14:16:06 -08002200 if (tcp_check_oom(sk, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 tcp_set_state(sk, TCP_CLOSE);
2202 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002203 NET_INC_STATS_BH(sock_net(sk),
2204 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 }
2206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Jerry Chu83368862012-08-31 12:29:12 +00002208 if (sk->sk_state == TCP_CLOSE) {
2209 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
2210 /* We could get here with a non-NULL req if the socket is
2211 * aborted (e.g., closed with unread data) before 3WHS
2212 * finishes.
2213 */
2214 if (req != NULL)
2215 reqsk_fastopen_remove(sk, req, false);
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002216 inet_csk_destroy_sock(sk);
Jerry Chu83368862012-08-31 12:29:12 +00002217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 /* Otherwise, socket is reprieved until protocol close. */
2219
2220out:
2221 bh_unlock_sock(sk);
2222 local_bh_enable();
2223 sock_put(sk);
2224}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002225EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
2227/* These states need RST on ABORT according to RFC793 */
2228
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002229static inline bool tcp_need_reset(int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230{
2231 return (1 << state) &
2232 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2233 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2234}
2235
2236int tcp_disconnect(struct sock *sk, int flags)
2237{
2238 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002239 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 struct tcp_sock *tp = tcp_sk(sk);
2241 int err = 0;
2242 int old_state = sk->sk_state;
2243
2244 if (old_state != TCP_CLOSE)
2245 tcp_set_state(sk, TCP_CLOSE);
2246
2247 /* ABORT function of RFC793 */
2248 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002249 inet_csk_listen_stop(sk);
Pavel Emelyanovee995282012-04-19 03:40:39 +00002250 } else if (unlikely(tp->repair)) {
2251 sk->sk_err = ECONNABORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 } else if (tcp_need_reset(old_state) ||
2253 (tp->snd_nxt != tp->write_seq &&
2254 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08002255 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 * states
2257 */
2258 tcp_send_active_reset(sk, gfp_any());
2259 sk->sk_err = ECONNRESET;
2260 } else if (old_state == TCP_SYN_SENT)
2261 sk->sk_err = ECONNRESET;
2262
2263 tcp_clear_xmit_timers(sk);
2264 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08002265 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07002267#ifdef CONFIG_NET_DMA
2268 __skb_queue_purge(&sk->sk_async_wait_queue);
2269#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002271 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
2273 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2274 inet_reset_saddr(sk);
2275
2276 sk->sk_shutdown = 0;
2277 sock_reset_flag(sk, SOCK_DONE);
2278 tp->srtt = 0;
2279 if ((tp->write_seq += tp->max_window + 2) == 0)
2280 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002281 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002283 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 tp->packets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002285 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 tp->snd_cwnd_cnt = 0;
Eric Dumazet1fdf4752009-11-30 12:53:30 -08002287 tp->window_clamp = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002288 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002290 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08002291 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002292 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 __sk_dst_reset(sk);
2294
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002295 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 sk->sk_error_report(sk);
2298 return err;
2299}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002300EXPORT_SYMBOL(tcp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Christoph Paaschbb68b642012-09-18 14:19:23 +00002302void tcp_sock_destruct(struct sock *sk)
2303{
2304 inet_sock_destruct(sk);
2305
2306 kfree(inet_csk(sk)->icsk_accept_queue.fastopenq);
2307}
2308
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002309static inline bool tcp_can_repair_sock(const struct sock *sk)
Pavel Emelyanovee995282012-04-19 03:40:39 +00002310{
Eric W. Biederman52e804c2012-11-16 03:03:05 +00002311 return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
Pavel Emelyanovee995282012-04-19 03:40:39 +00002312 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
2313}
2314
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002315static int tcp_repair_options_est(struct tcp_sock *tp,
2316 struct tcp_repair_opt __user *optbuf, unsigned int len)
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002317{
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002318 struct tcp_repair_opt opt;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002319
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002320 while (len >= sizeof(opt)) {
2321 if (copy_from_user(&opt, optbuf, sizeof(opt)))
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002322 return -EFAULT;
2323
2324 optbuf++;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002325 len -= sizeof(opt);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002326
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002327 switch (opt.opt_code) {
2328 case TCPOPT_MSS:
2329 tp->rx_opt.mss_clamp = opt.opt_val;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002330 break;
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002331 case TCPOPT_WINDOW:
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002332 {
2333 u16 snd_wscale = opt.opt_val & 0xFFFF;
2334 u16 rcv_wscale = opt.opt_val >> 16;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002335
Andrey Vaginbc26ccd2012-09-19 09:40:00 +00002336 if (snd_wscale > 14 || rcv_wscale > 14)
2337 return -EFBIG;
2338
2339 tp->rx_opt.snd_wscale = snd_wscale;
2340 tp->rx_opt.rcv_wscale = rcv_wscale;
2341 tp->rx_opt.wscale_ok = 1;
2342 }
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002343 break;
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002344 case TCPOPT_SACK_PERM:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002345 if (opt.opt_val != 0)
2346 return -EINVAL;
2347
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002348 tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
2349 if (sysctl_tcp_fack)
2350 tcp_enable_fack(tp);
2351 break;
2352 case TCPOPT_TIMESTAMP:
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002353 if (opt.opt_val != 0)
2354 return -EINVAL;
2355
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002356 tp->rx_opt.tstamp_ok = 1;
2357 break;
2358 }
2359 }
2360
2361 return 0;
2362}
2363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 * Socket option code for TCP.
2366 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002367static int do_tcp_setsockopt(struct sock *sk, int level,
David S. Millerb7058842009-09-30 16:12:20 -07002368 int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369{
2370 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002371 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 int val;
2373 int err = 0;
2374
William Allen Simpsone56fb502009-12-02 18:19:30 +00002375 /* These are data/string values, all the others are ints */
2376 switch (optname) {
2377 case TCP_CONGESTION: {
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002378 char name[TCP_CA_NAME_MAX];
2379
2380 if (optlen < 1)
2381 return -EINVAL;
2382
2383 val = strncpy_from_user(name, optval,
Andrew Morton4fdb78d2009-10-01 15:02:20 -07002384 min_t(long, TCP_CA_NAME_MAX-1, optlen));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002385 if (val < 0)
2386 return -EFAULT;
2387 name[val] = 0;
2388
2389 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002390 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002391 release_sock(sk);
2392 return err;
2393 }
William Allen Simpsone56fb502009-12-02 18:19:30 +00002394 default:
2395 /* fallthru */
2396 break;
Joe Perchesccbd6a52010-05-14 10:58:26 +00002397 }
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002398
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (optlen < sizeof(int))
2400 return -EINVAL;
2401
2402 if (get_user(val, (int __user *)optval))
2403 return -EFAULT;
2404
2405 lock_sock(sk);
2406
2407 switch (optname) {
2408 case TCP_MAXSEG:
2409 /* Values greater than interface MTU won't take effect. However
2410 * at the point when this call is done we typically don't yet
2411 * know which interface is going to be used */
David S. Millerc39508d2010-11-24 11:47:22 -08002412 if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 err = -EINVAL;
2414 break;
2415 }
2416 tp->rx_opt.user_mss = val;
2417 break;
2418
2419 case TCP_NODELAY:
2420 if (val) {
2421 /* TCP_NODELAY is weaker than TCP_CORK, so that
2422 * this option on corked socket is remembered, but
2423 * it is not activated until cork is cleared.
2424 *
2425 * However, when TCP_NODELAY is set we make
2426 * an explicit push, which overrides even TCP_CORK
2427 * for currently queued segments.
2428 */
2429 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002430 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 } else {
2432 tp->nonagle &= ~TCP_NAGLE_OFF;
2433 }
2434 break;
2435
Andreas Petlund36e31b0a2010-02-18 02:47:01 +00002436 case TCP_THIN_LINEAR_TIMEOUTS:
2437 if (val < 0 || val > 1)
2438 err = -EINVAL;
2439 else
2440 tp->thin_lto = val;
2441 break;
2442
Andreas Petlund7e380172010-02-18 04:48:19 +00002443 case TCP_THIN_DUPACK:
2444 if (val < 0 || val > 1)
2445 err = -EINVAL;
2446 else
2447 tp->thin_dupack = val;
Yuchung Chengeed530b2012-05-02 13:30:03 +00002448 if (tp->thin_dupack)
2449 tcp_disable_early_retrans(tp);
Andreas Petlund7e380172010-02-18 04:48:19 +00002450 break;
2451
Pavel Emelyanovee995282012-04-19 03:40:39 +00002452 case TCP_REPAIR:
2453 if (!tcp_can_repair_sock(sk))
2454 err = -EPERM;
2455 else if (val == 1) {
2456 tp->repair = 1;
2457 sk->sk_reuse = SK_FORCE_REUSE;
2458 tp->repair_queue = TCP_NO_QUEUE;
2459 } else if (val == 0) {
2460 tp->repair = 0;
2461 sk->sk_reuse = SK_NO_REUSE;
2462 tcp_send_window_probe(sk);
2463 } else
2464 err = -EINVAL;
2465
2466 break;
2467
2468 case TCP_REPAIR_QUEUE:
2469 if (!tp->repair)
2470 err = -EPERM;
2471 else if (val < TCP_QUEUES_NR)
2472 tp->repair_queue = val;
2473 else
2474 err = -EINVAL;
2475 break;
2476
2477 case TCP_QUEUE_SEQ:
2478 if (sk->sk_state != TCP_CLOSE)
2479 err = -EPERM;
2480 else if (tp->repair_queue == TCP_SEND_QUEUE)
2481 tp->write_seq = val;
2482 else if (tp->repair_queue == TCP_RECV_QUEUE)
2483 tp->rcv_nxt = val;
2484 else
2485 err = -EINVAL;
2486 break;
2487
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002488 case TCP_REPAIR_OPTIONS:
2489 if (!tp->repair)
2490 err = -EINVAL;
2491 else if (sk->sk_state == TCP_ESTABLISHED)
Pavel Emelyanovde248a72012-04-25 23:43:04 +00002492 err = tcp_repair_options_est(tp,
2493 (struct tcp_repair_opt __user *)optval,
2494 optlen);
Pavel Emelyanovb139ba42012-04-19 03:41:57 +00002495 else
2496 err = -EPERM;
2497 break;
2498
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 case TCP_CORK:
2500 /* When set indicates to always queue non-full frames.
2501 * Later the user clears this option and we transmit
2502 * any pending partial frames in the queue. This is
2503 * meant to be used alongside sendfile() to get properly
2504 * filled frames when the user (for example) must write
2505 * out headers with a write() call first and then use
2506 * sendfile to send out the data parts.
2507 *
2508 * TCP_CORK can be set together with TCP_NODELAY and it is
2509 * stronger than TCP_NODELAY.
2510 */
2511 if (val) {
2512 tp->nonagle |= TCP_NAGLE_CORK;
2513 } else {
2514 tp->nonagle &= ~TCP_NAGLE_CORK;
2515 if (tp->nonagle&TCP_NAGLE_OFF)
2516 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002517 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 }
2519 break;
2520
2521 case TCP_KEEPIDLE:
2522 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2523 err = -EINVAL;
2524 else {
2525 tp->keepalive_time = val * HZ;
2526 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2527 !((1 << sk->sk_state) &
2528 (TCPF_CLOSE | TCPF_LISTEN))) {
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00002529 u32 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (tp->keepalive_time > elapsed)
2531 elapsed = tp->keepalive_time - elapsed;
2532 else
2533 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002534 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 }
2536 }
2537 break;
2538 case TCP_KEEPINTVL:
2539 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2540 err = -EINVAL;
2541 else
2542 tp->keepalive_intvl = val * HZ;
2543 break;
2544 case TCP_KEEPCNT:
2545 if (val < 1 || val > MAX_TCP_KEEPCNT)
2546 err = -EINVAL;
2547 else
2548 tp->keepalive_probes = val;
2549 break;
2550 case TCP_SYNCNT:
2551 if (val < 1 || val > MAX_TCP_SYNCNT)
2552 err = -EINVAL;
2553 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002554 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 break;
2556
2557 case TCP_LINGER2:
2558 if (val < 0)
2559 tp->linger2 = -1;
2560 else if (val > sysctl_tcp_fin_timeout / HZ)
2561 tp->linger2 = 0;
2562 else
2563 tp->linger2 = val * HZ;
2564 break;
2565
2566 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002567 /* Translate value in seconds to number of retransmits */
2568 icsk->icsk_accept_queue.rskq_defer_accept =
2569 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2570 TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 break;
2572
2573 case TCP_WINDOW_CLAMP:
2574 if (!val) {
2575 if (sk->sk_state != TCP_CLOSE) {
2576 err = -EINVAL;
2577 break;
2578 }
2579 tp->window_clamp = 0;
2580 } else
2581 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2582 SOCK_MIN_RCVBUF / 2 : val;
2583 break;
2584
2585 case TCP_QUICKACK:
2586 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002587 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002589 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 if ((1 << sk->sk_state) &
2591 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002592 inet_csk_ack_scheduled(sk)) {
2593 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002594 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002596 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 }
2598 }
2599 break;
2600
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002601#ifdef CONFIG_TCP_MD5SIG
2602 case TCP_MD5SIG:
2603 /* Read the IP->Key mappings from userspace */
2604 err = tp->af_specific->md5_parse(sk, optval, optlen);
2605 break;
2606#endif
Jerry Chudca43c72010-08-27 19:13:28 +00002607 case TCP_USER_TIMEOUT:
2608 /* Cap the max timeout in ms TCP will retry/retrans
2609 * before giving up and aborting (ETIMEDOUT) a connection.
2610 */
Hangbin Liu42493572012-07-26 22:52:21 +00002611 if (val < 0)
2612 err = -EINVAL;
2613 else
2614 icsk->icsk_user_timeout = msecs_to_jiffies(val);
Jerry Chudca43c72010-08-27 19:13:28 +00002615 break;
Jerry Chu83368862012-08-31 12:29:12 +00002616
2617 case TCP_FASTOPEN:
2618 if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
2619 TCPF_LISTEN)))
2620 err = fastopen_init_queue(sk, val);
2621 else
2622 err = -EINVAL;
2623 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002624 case TCP_TIMESTAMP:
2625 if (!tp->repair)
2626 err = -EPERM;
2627 else
2628 tp->tsoffset = val - tcp_time_stamp;
2629 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 default:
2631 err = -ENOPROTOOPT;
2632 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002633 }
2634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 release_sock(sk);
2636 return err;
2637}
2638
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002639int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
David S. Millerb7058842009-09-30 16:12:20 -07002640 unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002641{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002642 const struct inet_connection_sock *icsk = inet_csk(sk);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002643
2644 if (level != SOL_TCP)
2645 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2646 optval, optlen);
2647 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2648}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002649EXPORT_SYMBOL(tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002650
2651#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002652int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002653 char __user *optval, unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002654{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002655 if (level != SOL_TCP)
2656 return inet_csk_compat_setsockopt(sk, level, optname,
2657 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002658 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2659}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002660EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002661#endif
2662
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663/* Return information about state of tcp endpoint in API format. */
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002664void tcp_get_info(const struct sock *sk, struct tcp_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002666 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002667 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 u32 now = tcp_time_stamp;
2669
2670 memset(info, 0, sizeof(*info));
2671
2672 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002673 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002674 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002675 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002676 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677
2678 if (tp->rx_opt.tstamp_ok)
2679 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002680 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 info->tcpi_options |= TCPI_OPT_SACK;
2682 if (tp->rx_opt.wscale_ok) {
2683 info->tcpi_options |= TCPI_OPT_WSCALE;
2684 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2685 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002686 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
Eric Dumazetb5c56932011-10-03 14:01:21 -04002688 if (tp->ecn_flags & TCP_ECN_OK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 info->tcpi_options |= TCPI_OPT_ECN;
Eric Dumazetb5c56932011-10-03 14:01:21 -04002690 if (tp->ecn_flags & TCP_ECN_SEEN)
2691 info->tcpi_options |= TCPI_OPT_ECN_SEEN;
Yuchung Cheng6f736012012-10-19 15:14:44 +00002692 if (tp->syn_data_acked)
2693 info->tcpi_options |= TCPI_OPT_SYN_DATA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002695 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2696 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002697 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002698 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Rick Jones5ee3afb2007-09-18 13:26:31 -07002700 if (sk->sk_state == TCP_LISTEN) {
2701 info->tcpi_unacked = sk->sk_ack_backlog;
2702 info->tcpi_sacked = sk->sk_max_ack_backlog;
2703 } else {
2704 info->tcpi_unacked = tp->packets_out;
2705 info->tcpi_sacked = tp->sacked_out;
2706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 info->tcpi_lost = tp->lost_out;
2708 info->tcpi_retrans = tp->retrans_out;
2709 info->tcpi_fackets = tp->fackets_out;
2710
2711 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002712 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2714
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002715 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2717 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2718 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2719 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2720 info->tcpi_snd_cwnd = tp->snd_cwnd;
2721 info->tcpi_advmss = tp->advmss;
2722 info->tcpi_reordering = tp->reordering;
2723
2724 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2725 info->tcpi_rcv_space = tp->rcvq_space.space;
2726
2727 info->tcpi_total_retrans = tp->total_retrans;
2728}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729EXPORT_SYMBOL_GPL(tcp_get_info);
2730
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002731static int do_tcp_getsockopt(struct sock *sk, int level,
2732 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002734 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 struct tcp_sock *tp = tcp_sk(sk);
2736 int val, len;
2737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 if (get_user(len, optlen))
2739 return -EFAULT;
2740
2741 len = min_t(unsigned int, len, sizeof(int));
2742
2743 if (len < 0)
2744 return -EINVAL;
2745
2746 switch (optname) {
2747 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002748 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2750 val = tp->rx_opt.user_mss;
Pavel Emelyanov5e6a3ce2012-04-19 03:41:32 +00002751 if (tp->repair)
2752 val = tp->rx_opt.mss_clamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 break;
2754 case TCP_NODELAY:
2755 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2756 break;
2757 case TCP_CORK:
2758 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2759 break;
2760 case TCP_KEEPIDLE:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002761 val = keepalive_time_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 break;
2763 case TCP_KEEPINTVL:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002764 val = keepalive_intvl_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 break;
2766 case TCP_KEEPCNT:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002767 val = keepalive_probes(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 break;
2769 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002770 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 break;
2772 case TCP_LINGER2:
2773 val = tp->linger2;
2774 if (val >= 0)
2775 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2776 break;
2777 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002778 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2779 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 break;
2781 case TCP_WINDOW_CLAMP:
2782 val = tp->window_clamp;
2783 break;
2784 case TCP_INFO: {
2785 struct tcp_info info;
2786
2787 if (get_user(len, optlen))
2788 return -EFAULT;
2789
2790 tcp_get_info(sk, &info);
2791
2792 len = min_t(unsigned int, len, sizeof(info));
2793 if (put_user(len, optlen))
2794 return -EFAULT;
2795 if (copy_to_user(optval, &info, len))
2796 return -EFAULT;
2797 return 0;
2798 }
2799 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002800 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002802
2803 case TCP_CONGESTION:
2804 if (get_user(len, optlen))
2805 return -EFAULT;
2806 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2807 if (put_user(len, optlen))
2808 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002809 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002810 return -EFAULT;
2811 return 0;
William Allen Simpsone56fb502009-12-02 18:19:30 +00002812
Josh Hunt3c0fef02010-07-30 13:49:35 +00002813 case TCP_THIN_LINEAR_TIMEOUTS:
2814 val = tp->thin_lto;
2815 break;
2816 case TCP_THIN_DUPACK:
2817 val = tp->thin_dupack;
2818 break;
Jerry Chudca43c72010-08-27 19:13:28 +00002819
Pavel Emelyanovee995282012-04-19 03:40:39 +00002820 case TCP_REPAIR:
2821 val = tp->repair;
2822 break;
2823
2824 case TCP_REPAIR_QUEUE:
2825 if (tp->repair)
2826 val = tp->repair_queue;
2827 else
2828 return -EINVAL;
2829 break;
2830
2831 case TCP_QUEUE_SEQ:
2832 if (tp->repair_queue == TCP_SEND_QUEUE)
2833 val = tp->write_seq;
2834 else if (tp->repair_queue == TCP_RECV_QUEUE)
2835 val = tp->rcv_nxt;
2836 else
2837 return -EINVAL;
2838 break;
2839
Jerry Chudca43c72010-08-27 19:13:28 +00002840 case TCP_USER_TIMEOUT:
2841 val = jiffies_to_msecs(icsk->icsk_user_timeout);
2842 break;
Andrey Vagin93be6ce2013-02-11 05:50:18 +00002843 case TCP_TIMESTAMP:
2844 val = tcp_time_stamp + tp->tsoffset;
2845 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 default:
2847 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849
2850 if (put_user(len, optlen))
2851 return -EFAULT;
2852 if (copy_to_user(optval, &val, len))
2853 return -EFAULT;
2854 return 0;
2855}
2856
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002857int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2858 int __user *optlen)
2859{
2860 struct inet_connection_sock *icsk = inet_csk(sk);
2861
2862 if (level != SOL_TCP)
2863 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2864 optval, optlen);
2865 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2866}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002867EXPORT_SYMBOL(tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002868
2869#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002870int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2871 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002872{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002873 if (level != SOL_TCP)
2874 return inet_csk_compat_getsockopt(sk, level, optname,
2875 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002876 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2877}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002878EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002879#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002881struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
2882 netdev_features_t features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002883{
2884 struct sk_buff *segs = ERR_PTR(-EINVAL);
2885 struct tcphdr *th;
Eric Dumazet95c96172012-04-15 05:58:06 +00002886 unsigned int thlen;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002887 unsigned int seq;
Al Virod3bc23e2006-11-14 21:24:49 -08002888 __be32 delta;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002889 unsigned int oldlen;
Herbert Xu4e704ee2009-01-14 20:41:12 -08002890 unsigned int mss;
Eric Dumazetd6a4a102013-04-12 11:31:52 +00002891 struct sk_buff *gso_skb = skb;
Eric Dumazetbece1b92013-04-13 03:22:08 +00002892 __sum16 newcheck;
Eric Dumazet6ff50cd2013-05-15 15:38:01 +00002893 bool ooo_okay, copy_destructor;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002894
2895 if (!pskb_may_pull(skb, sizeof(*th)))
2896 goto out;
2897
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002898 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002899 thlen = th->doff * 4;
2900 if (thlen < sizeof(*th))
2901 goto out;
2902
2903 if (!pskb_may_pull(skb, thlen))
2904 goto out;
2905
Herbert Xu0718bcc2006-06-25 23:55:46 -07002906 oldlen = (u16)~skb->len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002907 __skb_pull(skb, thlen);
2908
Herbert Xu4e704ee2009-01-14 20:41:12 -08002909 mss = skb_shinfo(skb)->gso_size;
2910 if (unlikely(skb->len <= mss))
2911 goto out;
2912
Herbert Xu3820c3f2006-06-29 20:11:25 -07002913 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2914 /* Packet is from an untrusted source, reset gso_segs. */
Herbert Xubbcf4672006-07-03 19:38:35 -07002915 int type = skb_shinfo(skb)->gso_type;
Herbert Xu3820c3f2006-06-29 20:11:25 -07002916
Herbert Xubbcf4672006-07-03 19:38:35 -07002917 if (unlikely(type &
2918 ~(SKB_GSO_TCPV4 |
2919 SKB_GSO_DODGY |
2920 SKB_GSO_TCP_ECN |
2921 SKB_GSO_TCPV6 |
Pravin B Shelar68c33162013-02-14 14:02:41 +00002922 SKB_GSO_GRE |
Pravin B Shelar73136262013-03-07 13:21:51 +00002923 SKB_GSO_UDP_TUNNEL |
Herbert Xubbcf4672006-07-03 19:38:35 -07002924 0) ||
2925 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
2926 goto out;
2927
Ilpo Järvinen172589c2007-08-28 15:50:33 -07002928 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
Herbert Xu3820c3f2006-06-29 20:11:25 -07002929
2930 segs = NULL;
2931 goto out;
2932 }
2933
Eric Dumazet6ff50cd2013-05-15 15:38:01 +00002934 copy_destructor = gso_skb->destructor == tcp_wfree;
2935 ooo_okay = gso_skb->ooo_okay;
2936 /* All segments but the first should have ooo_okay cleared */
2937 skb->ooo_okay = 0;
2938
Herbert Xu576a30e2006-06-27 13:22:38 -07002939 segs = skb_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002940 if (IS_ERR(segs))
2941 goto out;
2942
Eric Dumazet6ff50cd2013-05-15 15:38:01 +00002943 /* Only first segment might have ooo_okay set */
2944 segs->ooo_okay = ooo_okay;
2945
Herbert Xu4e704ee2009-01-14 20:41:12 -08002946 delta = htonl(oldlen + (thlen + mss));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002947
2948 skb = segs;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002949 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002950 seq = ntohl(th->seq);
2951
Eric Dumazetbece1b92013-04-13 03:22:08 +00002952 newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
2953 (__force u32)delta));
2954
Herbert Xuf4c50d92006-06-22 03:02:40 -07002955 do {
2956 th->fin = th->psh = 0;
Eric Dumazetbece1b92013-04-13 03:22:08 +00002957 th->check = newcheck;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002958
Patrick McHardy84fa7932006-08-29 16:44:56 -07002959 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002960 th->check =
2961 csum_fold(csum_partial(skb_transport_header(skb),
2962 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002963
Herbert Xu4e704ee2009-01-14 20:41:12 -08002964 seq += mss;
Eric Dumazet6ff50cd2013-05-15 15:38:01 +00002965 if (copy_destructor) {
2966 skb->destructor = gso_skb->destructor;
2967 skb->sk = gso_skb->sk;
2968 /* {tcp|sock}_wfree() use exact truesize accounting :
2969 * sum(skb->truesize) MUST be exactly be gso_skb->truesize
2970 * So we account mss bytes of 'true size' for each segment.
2971 * The last segment will contain the remaining.
2972 */
2973 skb->truesize = mss;
2974 gso_skb->truesize -= mss;
2975 }
Herbert Xuf4c50d92006-06-22 03:02:40 -07002976 skb = skb->next;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002977 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002978
2979 th->seq = htonl(seq);
2980 th->cwr = 0;
2981 } while (skb->next);
2982
Eric Dumazetd6a4a102013-04-12 11:31:52 +00002983 /* Following permits TCP Small Queues to work well with GSO :
2984 * The callback to TCP stack will be called at the time last frag
2985 * is freed at TX completion, and not right now when gso_skb
2986 * is freed by GSO engine
2987 */
Eric Dumazet6ff50cd2013-05-15 15:38:01 +00002988 if (copy_destructor) {
Eric Dumazetd6a4a102013-04-12 11:31:52 +00002989 swap(gso_skb->sk, skb->sk);
2990 swap(gso_skb->destructor, skb->destructor);
2991 swap(gso_skb->truesize, skb->truesize);
2992 }
2993
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002994 delta = htonl(oldlen + (skb->tail - skb->transport_header) +
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002995 skb->data_len);
Al Virod3bc23e2006-11-14 21:24:49 -08002996 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2997 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002998 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002999 th->check = csum_fold(csum_partial(skb_transport_header(skb),
3000 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07003001
3002out:
3003 return segs;
3004}
Herbert Xuadcfc7d2006-06-30 13:36:15 -07003005EXPORT_SYMBOL(tcp_tso_segment);
Herbert Xuf4c50d92006-06-22 03:02:40 -07003006
Herbert Xubf296b12008-12-15 23:43:36 -08003007struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3008{
3009 struct sk_buff **pp = NULL;
3010 struct sk_buff *p;
3011 struct tcphdr *th;
3012 struct tcphdr *th2;
Herbert Xua0a69a02009-04-17 02:34:38 -07003013 unsigned int len;
Herbert Xubf296b12008-12-15 23:43:36 -08003014 unsigned int thlen;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07003015 __be32 flags;
Herbert Xubf296b12008-12-15 23:43:36 -08003016 unsigned int mss = 1;
Herbert Xua5b1cf22009-05-26 18:50:28 +00003017 unsigned int hlen;
3018 unsigned int off;
Herbert Xubf296b12008-12-15 23:43:36 -08003019 int flush = 1;
Herbert Xuaa6320d2009-02-08 18:00:40 +00003020 int i;
Herbert Xubf296b12008-12-15 23:43:36 -08003021
Herbert Xua5b1cf22009-05-26 18:50:28 +00003022 off = skb_gro_offset(skb);
3023 hlen = off + sizeof(*th);
3024 th = skb_gro_header_fast(skb, off);
3025 if (skb_gro_header_hard(skb, hlen)) {
3026 th = skb_gro_header_slow(skb, hlen, off);
3027 if (unlikely(!th))
3028 goto out;
3029 }
Herbert Xubf296b12008-12-15 23:43:36 -08003030
Herbert Xubf296b12008-12-15 23:43:36 -08003031 thlen = th->doff * 4;
3032 if (thlen < sizeof(*th))
3033 goto out;
3034
Herbert Xua5b1cf22009-05-26 18:50:28 +00003035 hlen = off + thlen;
3036 if (skb_gro_header_hard(skb, hlen)) {
3037 th = skb_gro_header_slow(skb, hlen, off);
3038 if (unlikely(!th))
3039 goto out;
3040 }
Herbert Xubf296b12008-12-15 23:43:36 -08003041
Herbert Xu86911732009-01-29 14:19:50 +00003042 skb_gro_pull(skb, thlen);
Herbert Xubf296b12008-12-15 23:43:36 -08003043
Herbert Xua0a69a02009-04-17 02:34:38 -07003044 len = skb_gro_len(skb);
Herbert Xubf296b12008-12-15 23:43:36 -08003045 flags = tcp_flag_word(th);
3046
3047 for (; (p = *head); head = &p->next) {
3048 if (!NAPI_GRO_CB(p)->same_flow)
3049 continue;
3050
3051 th2 = tcp_hdr(p);
3052
Herbert Xu745898e2009-05-26 18:50:24 +00003053 if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
Herbert Xubf296b12008-12-15 23:43:36 -08003054 NAPI_GRO_CB(p)->same_flow = 0;
3055 continue;
3056 }
3057
3058 goto found;
3059 }
3060
3061 goto out_check_final;
3062
3063found:
3064 flush = NAPI_GRO_CB(p)->flush;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07003065 flush |= (__force int)(flags & TCP_FLAG_CWR);
3066 flush |= (__force int)((flags ^ tcp_flag_word(th2)) &
3067 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
3068 flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
Herbert Xua2a804c2009-05-26 18:50:34 +00003069 for (i = sizeof(*th); i < thlen; i += 4)
Herbert Xuaa6320d2009-02-08 18:00:40 +00003070 flush |= *(u32 *)((u8 *)th + i) ^
3071 *(u32 *)((u8 *)th2 + i);
Herbert Xubf296b12008-12-15 23:43:36 -08003072
Herbert Xub5302562009-01-04 16:13:19 -08003073 mss = skb_shinfo(p)->gso_size;
Herbert Xubf296b12008-12-15 23:43:36 -08003074
Herbert Xu30a3ae32009-05-26 18:50:26 +00003075 flush |= (len - 1) >= mss;
Herbert Xuaa6320d2009-02-08 18:00:40 +00003076 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
Herbert Xubf296b12008-12-15 23:43:36 -08003077
3078 if (flush || skb_gro_receive(head, skb)) {
3079 mss = 1;
3080 goto out_check_final;
3081 }
3082
3083 p = *head;
3084 th2 = tcp_hdr(p);
3085 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
3086
3087out_check_final:
Herbert Xua0a69a02009-04-17 02:34:38 -07003088 flush = len < mss;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07003089 flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH |
3090 TCP_FLAG_RST | TCP_FLAG_SYN |
3091 TCP_FLAG_FIN));
Herbert Xubf296b12008-12-15 23:43:36 -08003092
3093 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
3094 pp = head;
3095
3096out:
3097 NAPI_GRO_CB(skb)->flush |= flush;
3098
3099 return pp;
3100}
Herbert Xu684f2172009-01-08 10:41:23 -08003101EXPORT_SYMBOL(tcp_gro_receive);
Herbert Xubf296b12008-12-15 23:43:36 -08003102
3103int tcp_gro_complete(struct sk_buff *skb)
3104{
3105 struct tcphdr *th = tcp_hdr(skb);
3106
3107 skb->csum_start = skb_transport_header(skb) - skb->head;
3108 skb->csum_offset = offsetof(struct tcphdr, check);
3109 skb->ip_summed = CHECKSUM_PARTIAL;
3110
Herbert Xubf296b12008-12-15 23:43:36 -08003111 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
3112
3113 if (th->cwr)
3114 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
3115
3116 return 0;
3117}
Herbert Xu684f2172009-01-08 10:41:23 -08003118EXPORT_SYMBOL(tcp_gro_complete);
Herbert Xubf296b12008-12-15 23:43:36 -08003119
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003120#ifdef CONFIG_TCP_MD5SIG
3121static unsigned long tcp_md5sig_users;
Eric Dumazet765cf992011-09-12 20:28:37 +00003122static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003123static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
3124
Eric Dumazet765cf992011-09-12 20:28:37 +00003125static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003126{
3127 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00003128
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003129 for_each_possible_cpu(cpu) {
Eric Dumazet765cf992011-09-12 20:28:37 +00003130 struct tcp_md5sig_pool *p = per_cpu_ptr(pool, cpu);
3131
3132 if (p->md5_desc.tfm)
3133 crypto_free_hash(p->md5_desc.tfm);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003134 }
3135 free_percpu(pool);
3136}
3137
3138void tcp_free_md5sig_pool(void)
3139{
Eric Dumazet765cf992011-09-12 20:28:37 +00003140 struct tcp_md5sig_pool __percpu *pool = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003141
David S. Miller2c4f6212007-02-20 23:51:47 -08003142 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003143 if (--tcp_md5sig_users == 0) {
3144 pool = tcp_md5sig_pool;
3145 tcp_md5sig_pool = NULL;
3146 }
David S. Miller2c4f6212007-02-20 23:51:47 -08003147 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003148 if (pool)
3149 __tcp_free_md5sig_pool(pool);
3150}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003151EXPORT_SYMBOL(tcp_free_md5sig_pool);
3152
Eric Dumazet765cf992011-09-12 20:28:37 +00003153static struct tcp_md5sig_pool __percpu *
Tejun Heo7d720c32010-02-16 15:20:26 +00003154__tcp_alloc_md5sig_pool(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003155{
3156 int cpu;
Eric Dumazet765cf992011-09-12 20:28:37 +00003157 struct tcp_md5sig_pool __percpu *pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003158
Eric Dumazet765cf992011-09-12 20:28:37 +00003159 pool = alloc_percpu(struct tcp_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003160 if (!pool)
3161 return NULL;
3162
3163 for_each_possible_cpu(cpu) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003164 struct crypto_hash *hash;
3165
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003166 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
YOSHIFUJI Hideaki / 吉藤英明50c3a482013-01-22 06:32:49 +00003167 if (IS_ERR_OR_NULL(hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003168 goto out_free;
3169
Eric Dumazet765cf992011-09-12 20:28:37 +00003170 per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003171 }
3172 return pool;
3173out_free:
3174 __tcp_free_md5sig_pool(pool);
3175 return NULL;
3176}
3177
Eric Dumazet765cf992011-09-12 20:28:37 +00003178struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003179{
Eric Dumazet765cf992011-09-12 20:28:37 +00003180 struct tcp_md5sig_pool __percpu *pool;
Eric Dumazeta2a385d2012-05-16 23:15:34 +00003181 bool alloc = false;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003182
3183retry:
David S. Miller2c4f6212007-02-20 23:51:47 -08003184 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003185 pool = tcp_md5sig_pool;
3186 if (tcp_md5sig_users++ == 0) {
Eric Dumazeta2a385d2012-05-16 23:15:34 +00003187 alloc = true;
David S. Miller2c4f6212007-02-20 23:51:47 -08003188 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003189 } else if (!pool) {
3190 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08003191 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003192 cpu_relax();
3193 goto retry;
3194 } else
David S. Miller2c4f6212007-02-20 23:51:47 -08003195 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003196
3197 if (alloc) {
3198 /* we cannot hold spinlock here because this may sleep. */
Eric Dumazet765cf992011-09-12 20:28:37 +00003199 struct tcp_md5sig_pool __percpu *p;
Tejun Heo7d720c32010-02-16 15:20:26 +00003200
3201 p = __tcp_alloc_md5sig_pool(sk);
David S. Miller2c4f6212007-02-20 23:51:47 -08003202 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003203 if (!p) {
3204 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08003205 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003206 return NULL;
3207 }
3208 pool = tcp_md5sig_pool;
3209 if (pool) {
3210 /* oops, it has already been assigned. */
David S. Miller2c4f6212007-02-20 23:51:47 -08003211 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003212 __tcp_free_md5sig_pool(p);
3213 } else {
3214 tcp_md5sig_pool = pool = p;
David S. Miller2c4f6212007-02-20 23:51:47 -08003215 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003216 }
3217 }
3218 return pool;
3219}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003220EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
3221
Eric Dumazet35790c02010-05-16 00:34:04 -07003222
3223/**
3224 * tcp_get_md5sig_pool - get md5sig_pool for this user
3225 *
3226 * We use percpu structure, so if we succeed, we exit with preemption
3227 * and BH disabled, to make sure another thread or softirq handling
3228 * wont try to get same context.
3229 */
3230struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003231{
Eric Dumazet765cf992011-09-12 20:28:37 +00003232 struct tcp_md5sig_pool __percpu *p;
Eric Dumazet35790c02010-05-16 00:34:04 -07003233
3234 local_bh_disable();
3235
3236 spin_lock(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003237 p = tcp_md5sig_pool;
3238 if (p)
3239 tcp_md5sig_users++;
Eric Dumazet35790c02010-05-16 00:34:04 -07003240 spin_unlock(&tcp_md5sig_pool_lock);
3241
3242 if (p)
Eric Dumazet765cf992011-09-12 20:28:37 +00003243 return this_cpu_ptr(p);
Eric Dumazet35790c02010-05-16 00:34:04 -07003244
3245 local_bh_enable();
3246 return NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003247}
Eric Dumazet35790c02010-05-16 00:34:04 -07003248EXPORT_SYMBOL(tcp_get_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003249
Eric Dumazet35790c02010-05-16 00:34:04 -07003250void tcp_put_md5sig_pool(void)
David S. Miller6931ba72006-12-13 16:25:44 -08003251{
Eric Dumazet35790c02010-05-16 00:34:04 -07003252 local_bh_enable();
David S. Miller6931ba72006-12-13 16:25:44 -08003253 tcp_free_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003254}
Eric Dumazet35790c02010-05-16 00:34:04 -07003255EXPORT_SYMBOL(tcp_put_md5sig_pool);
Adam Langley49a72df2008-07-19 00:01:42 -07003256
3257int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003258 const struct tcphdr *th)
Adam Langley49a72df2008-07-19 00:01:42 -07003259{
3260 struct scatterlist sg;
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003261 struct tcphdr hdr;
Adam Langley49a72df2008-07-19 00:01:42 -07003262 int err;
3263
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003264 /* We are not allowed to change tcphdr, make a local copy */
3265 memcpy(&hdr, th, sizeof(hdr));
3266 hdr.check = 0;
3267
Adam Langley49a72df2008-07-19 00:01:42 -07003268 /* options aren't included in the hash */
Eric Dumazetca35a0e2011-10-24 01:52:35 -04003269 sg_init_one(&sg, &hdr, sizeof(hdr));
3270 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr));
Adam Langley49a72df2008-07-19 00:01:42 -07003271 return err;
3272}
Adam Langley49a72df2008-07-19 00:01:42 -07003273EXPORT_SYMBOL(tcp_md5_hash_header);
3274
3275int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003276 const struct sk_buff *skb, unsigned int header_len)
Adam Langley49a72df2008-07-19 00:01:42 -07003277{
3278 struct scatterlist sg;
3279 const struct tcphdr *tp = tcp_hdr(skb);
3280 struct hash_desc *desc = &hp->md5_desc;
Eric Dumazet95c96172012-04-15 05:58:06 +00003281 unsigned int i;
3282 const unsigned int head_data_len = skb_headlen(skb) > header_len ?
3283 skb_headlen(skb) - header_len : 0;
Adam Langley49a72df2008-07-19 00:01:42 -07003284 const struct skb_shared_info *shi = skb_shinfo(skb);
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003285 struct sk_buff *frag_iter;
Adam Langley49a72df2008-07-19 00:01:42 -07003286
3287 sg_init_table(&sg, 1);
3288
3289 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3290 if (crypto_hash_update(desc, &sg, head_data_len))
3291 return 1;
3292
3293 for (i = 0; i < shi->nr_frags; ++i) {
3294 const struct skb_frag_struct *f = &shi->frags[i];
Eric Dumazet54d27fc2013-05-13 21:25:52 +00003295 unsigned int offset = f->page_offset;
3296 struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
3297
3298 sg_set_page(&sg, page, skb_frag_size(f),
3299 offset_in_page(offset));
Eric Dumazet9e903e02011-10-18 21:00:24 +00003300 if (crypto_hash_update(desc, &sg, skb_frag_size(f)))
Adam Langley49a72df2008-07-19 00:01:42 -07003301 return 1;
3302 }
3303
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003304 skb_walk_frags(skb, frag_iter)
3305 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3306 return 1;
3307
Adam Langley49a72df2008-07-19 00:01:42 -07003308 return 0;
3309}
Adam Langley49a72df2008-07-19 00:01:42 -07003310EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3311
Eric Dumazetcf533ea2011-10-21 05:22:42 -04003312int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
Adam Langley49a72df2008-07-19 00:01:42 -07003313{
3314 struct scatterlist sg;
3315
3316 sg_init_one(&sg, key->key, key->keylen);
3317 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
3318}
Adam Langley49a72df2008-07-19 00:01:42 -07003319EXPORT_SYMBOL(tcp_md5_hash_key);
3320
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003321#endif
3322
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003323void tcp_done(struct sock *sk)
3324{
Jerry Chu83368862012-08-31 12:29:12 +00003325 struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
3326
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08003327 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bddf2008-07-16 20:22:25 -07003328 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003329
3330 tcp_set_state(sk, TCP_CLOSE);
3331 tcp_clear_xmit_timers(sk);
Jerry Chu83368862012-08-31 12:29:12 +00003332 if (req != NULL)
3333 reqsk_fastopen_remove(sk, req, false);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003334
3335 sk->sk_shutdown = SHUTDOWN_MASK;
3336
3337 if (!sock_flag(sk, SOCK_DEAD))
3338 sk->sk_state_change(sk);
3339 else
3340 inet_csk_destroy_sock(sk);
3341}
3342EXPORT_SYMBOL_GPL(tcp_done);
3343
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003344extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345
3346static __initdata unsigned long thash_entries;
3347static int __init set_thash_entries(char *str)
3348{
Eldad Zack413c27d2012-05-19 14:13:18 +00003349 ssize_t ret;
3350
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 if (!str)
3352 return 0;
Eldad Zack413c27d2012-05-19 14:13:18 +00003353
3354 ret = kstrtoul(str, 0, &thash_entries);
3355 if (ret)
3356 return 0;
3357
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 return 1;
3359}
3360__setup("thash_entries=", set_thash_entries);
3361
Glauber Costa4acb4192012-01-30 01:20:17 +00003362void tcp_init_mem(struct net *net)
3363{
Glauber Costa4acb4192012-01-30 01:20:17 +00003364 unsigned long limit = nr_free_buffer_pages() / 8;
3365 limit = max(limit, 128UL);
3366 net->ipv4.sysctl_tcp_mem[0] = limit / 4 * 3;
3367 net->ipv4.sysctl_tcp_mem[1] = limit;
3368 net->ipv4.sysctl_tcp_mem[2] = net->ipv4.sysctl_tcp_mem[0] * 2;
3369}
3370
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371void __init tcp_init(void)
3372{
3373 struct sk_buff *skb = NULL;
Eric Dumazetf03d78d2011-07-07 00:27:05 -07003374 unsigned long limit;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003375 int max_rshare, max_wshare, cnt;
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003376 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08003378 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
Eric Dumazet17483762008-11-25 21:16:35 -08003380 percpu_counter_init(&tcp_sockets_allocated, 0);
Eric Dumazetdd24c002008-11-25 21:17:14 -08003381 percpu_counter_init(&tcp_orphan_count, 0);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003382 tcp_hashinfo.bind_bucket_cachep =
3383 kmem_cache_create("tcp_bind_bucket",
3384 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003385 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 /* Size and allocate the main established and bind bucket
3388 * hash tables.
3389 *
3390 * The methodology is similar to that of the buffer cache.
3391 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003392 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003394 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 thash_entries,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003396 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003397 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 NULL,
Eric Dumazetf373b532009-10-09 00:16:19 +00003399 &tcp_hashinfo.ehash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003400 0,
Jean Delvare0ccfe612007-10-30 00:59:25 -07003401 thash_entries ? 0 : 512 * 1024);
Eric Dumazetf373b532009-10-09 00:16:19 +00003402 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) {
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08003403 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3404 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 }
Eric Dumazet230140c2007-11-07 02:40:20 -08003406 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3407 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003408 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003410 sizeof(struct inet_bind_hashbucket),
Eric Dumazetf373b532009-10-09 00:16:19 +00003411 tcp_hashinfo.ehash_mask + 1,
Eric Dumazetfd90b292012-11-30 10:08:52 +00003412 17, /* one slot per 128 KB of memory */
John Heffner9e950ef2006-11-06 23:10:51 -08003413 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003414 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003416 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 64 * 1024);
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003418 tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003419 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3420 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3421 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 }
3423
Eric Dumazetc5ed63d2010-08-25 23:02:17 -07003424
3425 cnt = tcp_hashinfo.ehash_mask + 1;
3426
3427 tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
3428 sysctl_tcp_max_orphans = cnt / 2;
3429 sysctl_max_syn_backlog = max(128, cnt / 256);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430
Glauber Costa4acb4192012-01-30 01:20:17 +00003431 tcp_init_mem(&init_net);
Jason Wangc43b8742012-02-02 00:07:00 +00003432 /* Set per-socket limits to no more than 1/128 the pressure threshold */
Eric Dumazet5fb84b12012-04-10 00:56:42 +00003433 limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
Eric Dumazetb49960a2012-05-02 02:28:41 +00003434 max_wshare = min(4UL*1024*1024, limit);
3435 max_rshare = min(6UL*1024*1024, limit);
John Heffner7b4f4b52006-03-25 01:34:07 -08003436
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003437 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003438 sysctl_tcp_wmem[1] = 16*1024;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003439 sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
John Heffner7b4f4b52006-03-25 01:34:07 -08003440
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003441 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003442 sysctl_tcp_rmem[1] = 87380;
Eric Dumazetb49960a2012-05-02 02:28:41 +00003443 sysctl_tcp_rmem[2] = max(87380, max_rshare);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444
Joe Perchesafd465032012-03-12 07:03:32 +00003445 pr_info("Hash tables configured (established %u bind %u)\n",
Joe Perches058bd4d2012-03-11 18:36:11 +00003446 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003447
David S. Miller51c5d0c2012-07-10 00:49:14 -07003448 tcp_metrics_init();
3449
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003450 tcp_register_congestion_control(&tcp_reno);
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003451
Eric Dumazet46d3cea2012-07-11 05:50:31 +00003452 tcp_tasklet_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453}
Robert Love4b015882008-05-12 17:08:29 -04003454
3455static int tcp_is_local(struct net *net, __be32 addr) {
3456 struct rtable *rt;
3457 struct flowi4 fl4 = { .daddr = addr };
3458 rt = ip_route_output_key(net, &fl4);
3459 if (IS_ERR_OR_NULL(rt))
3460 return 0;
3461 return rt->dst.dev && (rt->dst.dev->flags & IFF_LOOPBACK);
3462}
3463
3464#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
3465static int tcp_is_local6(struct net *net, struct in6_addr *addr) {
3466 struct rt6_info *rt6 = rt6_lookup(net, addr, addr, 0, 0);
3467 return rt6 && rt6->dst.dev && (rt6->dst.dev->flags & IFF_LOOPBACK);
3468}
3469#endif
3470
3471/*
3472 * tcp_nuke_addr - destroy all sockets on the given local address
3473 * if local address is the unspecified address (0.0.0.0 or ::), destroy all
3474 * sockets with local addresses that are not configured.
3475 */
3476int tcp_nuke_addr(struct net *net, struct sockaddr *addr)
3477{
3478 int family = addr->sa_family;
3479 unsigned int bucket;
3480
3481 struct in_addr *in;
3482#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
3483 struct in6_addr *in6;
3484#endif
3485 if (family == AF_INET) {
3486 in = &((struct sockaddr_in *)addr)->sin_addr;
3487#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
3488 } else if (family == AF_INET6) {
3489 in6 = &((struct sockaddr_in6 *)addr)->sin6_addr;
3490#endif
3491 } else {
3492 return -EAFNOSUPPORT;
3493 }
3494
3495 for (bucket = 0; bucket < tcp_hashinfo.ehash_mask; bucket++) {
3496 struct hlist_nulls_node *node;
3497 struct sock *sk;
3498 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, bucket);
3499
3500restart:
3501 spin_lock_bh(lock);
3502 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[bucket].chain) {
3503 struct inet_sock *inet = inet_sk(sk);
3504
3505 if (sysctl_ip_dynaddr && sk->sk_state == TCP_SYN_SENT)
3506 continue;
3507 if (sock_flag(sk, SOCK_DEAD))
3508 continue;
3509
3510 if (family == AF_INET) {
3511 __be32 s4 = inet->inet_rcv_saddr;
3512 if (s4 == LOOPBACK4_IPV6)
3513 continue;
3514
3515 if (in->s_addr != s4 &&
3516 !(in->s_addr == INADDR_ANY &&
3517 !tcp_is_local(net, s4)))
3518 continue;
3519 }
3520
3521#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
3522 if (family == AF_INET6) {
3523 struct in6_addr *s6;
3524 if (!inet->pinet6)
3525 continue;
3526
3527 s6 = &inet->pinet6->rcv_saddr;
3528 if (ipv6_addr_type(s6) == IPV6_ADDR_MAPPED)
3529 continue;
3530
3531 if (!ipv6_addr_equal(in6, s6) &&
3532 !(ipv6_addr_equal(in6, &in6addr_any) &&
3533 !tcp_is_local6(net, s6)))
3534 continue;
3535 }
3536#endif
3537
3538 sock_hold(sk);
3539 spin_unlock_bh(lock);
3540
3541 local_bh_disable();
3542 bh_lock_sock(sk);
3543 sk->sk_err = ETIMEDOUT;
3544 sk->sk_error_report(sk);
3545
3546 tcp_done(sk);
3547 bh_unlock_sock(sk);
3548 local_bh_enable();
3549 sock_put(sk);
3550
3551 goto restart;
3552 }
3553 spin_unlock_bh(lock);
3554 }
3555
3556 return 0;
3557}