blob: 197b9b77fa3ed1a3c2fe30755394e250d274bd8f [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
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700248#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249#include <linux/module.h>
250#include <linux/types.h>
251#include <linux/fcntl.h>
252#include <linux/poll.h>
253#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254#include <linux/fs.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800255#include <linux/skbuff.h>
Andrew Morton81b23b42008-07-03 03:22:02 -0700256#include <linux/scatterlist.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800257#include <linux/splice.h>
258#include <linux/net.h>
259#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260#include <linux/random.h>
261#include <linux/bootmem.h>
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -0700262#include <linux/highmem.h>
263#include <linux/swap.h>
David S. Millerb8059ea2006-03-25 01:36:56 -0800264#include <linux/cache.h>
Herbert Xuf4c50d92006-06-22 03:02:40 -0700265#include <linux/err.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800266#include <linux/crypto.h>
William Allen Simpsonda5c78c2009-12-02 18:12:09 +0000267#include <linux/time.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +0900268#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270#include <net/icmp.h>
271#include <net/tcp.h>
272#include <net/xfrm.h>
273#include <net/ip.h>
Chris Leech1a2449a2006-05-23 18:05:53 -0700274#include <net/netdma.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800275#include <net/sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277#include <asm/uaccess.h>
278#include <asm/ioctls.h>
279
Brian Haleyab32ea52006-09-22 14:15:41 -0700280int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Eric Dumazetdd24c002008-11-25 21:17:14 -0800282struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700283EXPORT_SYMBOL_GPL(tcp_orphan_count);
284
David S. Millerb8059ea2006-03-25 01:36:56 -0800285int sysctl_tcp_mem[3] __read_mostly;
286int sysctl_tcp_wmem[3] __read_mostly;
287int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289EXPORT_SYMBOL(sysctl_tcp_mem);
290EXPORT_SYMBOL(sysctl_tcp_rmem);
291EXPORT_SYMBOL(sysctl_tcp_wmem);
292
293atomic_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800295
296/*
297 * Current number of TCP sockets.
298 */
299struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300EXPORT_SYMBOL(tcp_sockets_allocated);
301
302/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800303 * TCP splice context
304 */
305struct tcp_splice_state {
306 struct pipe_inode_info *pipe;
307 size_t len;
308 unsigned int flags;
309};
310
311/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 * Pressure flag: try to collapse.
313 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800314 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * is strict, actions are advisory and have some latency.
316 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700317int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318EXPORT_SYMBOL(tcp_memory_pressure);
319
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700320void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700323 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 tcp_memory_pressure = 1;
325 }
326}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327EXPORT_SYMBOL(tcp_enter_memory_pressure);
328
Julian Anastasovb103cf32009-10-19 10:10:40 +0000329/* Convert seconds to retransmits based on initial and max timeout */
330static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
331{
332 u8 res = 0;
333
334 if (seconds > 0) {
335 int period = timeout;
336
337 res = 1;
338 while (seconds > period && res < 255) {
339 res++;
340 timeout <<= 1;
341 if (timeout > rto_max)
342 timeout = rto_max;
343 period += timeout;
344 }
345 }
346 return res;
347}
348
349/* Convert retransmits to seconds based on initial and max timeout */
350static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
351{
352 int period = 0;
353
354 if (retrans > 0) {
355 period = timeout;
356 while (--retrans) {
357 timeout <<= 1;
358 if (timeout > rto_max)
359 timeout = rto_max;
360 period += timeout;
361 }
362 }
363 return period;
364}
365
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 * Wait for a TCP event.
368 *
369 * Note that we don't need to lock the socket, as the upper poll layers
370 * take care of normal races (between the test and the event) and we don't
371 * go look at any of the socket buffers directly.
372 */
373unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
374{
375 unsigned int mask;
376 struct sock *sk = sock->sk;
377 struct tcp_sock *tp = tcp_sk(sk);
378
Eric Dumazetaa395142010-04-20 13:03:51 +0000379 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700381 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700384 * by poll logic and correct handling of state changes
385 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 */
387
388 mask = 0;
389 if (sk->sk_err)
390 mask = POLLERR;
391
392 /*
393 * POLLHUP is certainly not done right. But poll() doesn't
394 * have a notion of HUP in just one direction, and for a
395 * socket the read side is more interesting.
396 *
397 * Some poll() documentation says that POLLHUP is incompatible
398 * with the POLLOUT/POLLWR flags, so somebody should check this
399 * all. But careful, it tends to be safer to return too many
400 * bits than too few, and you can easily break real applications
401 * if you don't tell them that something has hung up!
402 *
403 * Check-me.
404 *
405 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
406 * our fs/select.c). It means that after we received EOF,
407 * poll always returns immediately, making impossible poll() on write()
408 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
409 * if and only if shutdown has been made in both directions.
410 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700411 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 * then we could set it on SND_SHUTDOWN. BTW examples given
413 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700414 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 *
416 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
417 * blocking on fresh not-connected or disconnected socket. --ANK
418 */
419 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
420 mask |= POLLHUP;
421 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800422 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424 /* Connected? */
425 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
David S. Millerc7004482008-10-06 10:43:54 -0700426 int target = sock_rcvlowat(sk, 0, INT_MAX);
427
428 if (tp->urg_seq == tp->copied_seq &&
429 !sock_flag(sk, SOCK_URGINLINE) &&
430 tp->urg_data)
Alexandra Kossovskyb634f872010-03-18 20:29:24 -0700431 target++;
David S. Millerc7004482008-10-06 10:43:54 -0700432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 /* Potential race condition. If read of tp below will
434 * escape above sk->sk_state, we can be illegally awaken
435 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700436 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 mask |= POLLIN | POLLRDNORM;
438
439 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
440 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
441 mask |= POLLOUT | POLLWRNORM;
442 } else { /* send SIGIO later */
443 set_bit(SOCK_ASYNC_NOSPACE,
444 &sk->sk_socket->flags);
445 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
446
447 /* Race breaker. If space is freed after
448 * wspace test but before the flags are set,
449 * IO signal will be lost.
450 */
451 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
452 mask |= POLLOUT | POLLWRNORM;
453 }
454 }
455
456 if (tp->urg_data & TCP_URG_VALID)
457 mask |= POLLPRI;
458 }
459 return mask;
460}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000461EXPORT_SYMBOL(tcp_poll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
464{
465 struct tcp_sock *tp = tcp_sk(sk);
466 int answ;
467
468 switch (cmd) {
469 case SIOCINQ:
470 if (sk->sk_state == TCP_LISTEN)
471 return -EINVAL;
472
473 lock_sock(sk);
474 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
475 answ = 0;
476 else if (sock_flag(sk, SOCK_URGINLINE) ||
477 !tp->urg_data ||
478 before(tp->urg_seq, tp->copied_seq) ||
479 !before(tp->urg_seq, tp->rcv_nxt)) {
David S. Miller91521942009-05-28 21:35:47 -0700480 struct sk_buff *skb;
481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 answ = tp->rcv_nxt - tp->copied_seq;
483
484 /* Subtract 1, if FIN is in queue. */
David S. Miller91521942009-05-28 21:35:47 -0700485 skb = skb_peek_tail(&sk->sk_receive_queue);
486 if (answ && skb)
487 answ -= tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 } else
489 answ = tp->urg_seq - tp->copied_seq;
490 release_sock(sk);
491 break;
492 case SIOCATMARK:
493 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
494 break;
495 case SIOCOUTQ:
496 if (sk->sk_state == TCP_LISTEN)
497 return -EINVAL;
498
499 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
500 answ = 0;
501 else
502 answ = tp->write_seq - tp->snd_una;
503 break;
504 default:
505 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700506 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508 return put_user(answ, (int __user *)arg);
509}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000510EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
513{
Changli Gaoa3433f32010-06-12 14:01:43 +0000514 TCP_SKB_CB(skb)->flags |= TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 tp->pushed_seq = tp->write_seq;
516}
517
518static inline int forced_push(struct tcp_sock *tp)
519{
520 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
521}
522
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700523static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700525 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200526 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
527
528 skb->csum = 0;
529 tcb->seq = tcb->end_seq = tp->write_seq;
Changli Gaoa3433f32010-06-12 14:01:43 +0000530 tcb->flags = TCPHDR_ACK;
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200531 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800533 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800534 sk->sk_wmem_queued += skb->truesize;
535 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700536 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900537 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
Krishna Kumarafeca342009-12-10 07:16:52 +0000540static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700542 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544}
545
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700546static inline void tcp_push(struct sock *sk, int flags, int mss_now,
547 int nonagle)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
David S. Millerfe067e82007-03-07 12:12:44 -0800549 if (tcp_send_head(sk)) {
Krishna Kumarafeca342009-12-10 07:16:52 +0000550 struct tcp_sock *tp = tcp_sk(sk);
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (!(flags & MSG_MORE) || forced_push(tp))
Krishna Kumarafeca342009-12-10 07:16:52 +0000553 tcp_mark_push(tp, tcp_write_queue_tail(sk));
554
555 tcp_mark_urg(tp, flags);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700556 __tcp_push_pending_frames(sk, mss_now,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
558 }
559}
560
Adrian Bunk6ff77512007-11-06 23:32:26 -0800561static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
562 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800563{
564 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd2009-01-13 16:04:36 -0800565 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800566
Dimitris Michailidis9fa5fdf2009-01-26 22:15:31 -0800567 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
568 tss->flags);
Willy Tarreau33966dd2009-01-13 16:04:36 -0800569 if (ret > 0)
570 rd_desc->count -= ret;
571 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800572}
573
574static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
575{
576 /* Store TCP splice context information in read_descriptor_t. */
577 read_descriptor_t rd_desc = {
578 .arg.data = tss,
Willy Tarreau33966dd2009-01-13 16:04:36 -0800579 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800580 };
581
582 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
583}
584
585/**
586 * tcp_splice_read - splice data from TCP socket to a pipe
587 * @sock: socket to splice from
588 * @ppos: position (not valid)
589 * @pipe: pipe to splice to
590 * @len: number of bytes to splice
591 * @flags: splice modifier flags
592 *
593 * Description:
594 * Will read pages from given socket and fill them into a pipe.
595 *
596 **/
597ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
598 struct pipe_inode_info *pipe, size_t len,
599 unsigned int flags)
600{
601 struct sock *sk = sock->sk;
602 struct tcp_splice_state tss = {
603 .pipe = pipe,
604 .len = len,
605 .flags = flags,
606 };
607 long timeo;
608 ssize_t spliced;
609 int ret;
610
Changli Gao3a047bf2010-07-12 21:00:12 +0000611 sock_rps_record_flow(sk);
Jens Axboe9c55e012007-11-06 23:30:13 -0800612 /*
613 * We can't seek on a socket input
614 */
615 if (unlikely(*ppos))
616 return -ESPIPE;
617
618 ret = spliced = 0;
619
620 lock_sock(sk);
621
Eric Dumazet42324c62009-10-01 15:26:00 -0700622 timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
Jens Axboe9c55e012007-11-06 23:30:13 -0800623 while (tss.len) {
624 ret = __tcp_splice_read(sk, &tss);
625 if (ret < 0)
626 break;
627 else if (!ret) {
628 if (spliced)
629 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800630 if (sock_flag(sk, SOCK_DONE))
631 break;
632 if (sk->sk_err) {
633 ret = sock_error(sk);
634 break;
635 }
636 if (sk->sk_shutdown & RCV_SHUTDOWN)
637 break;
638 if (sk->sk_state == TCP_CLOSE) {
639 /*
640 * This occurs when user tries to read
641 * from never connected socket.
642 */
643 if (!sock_flag(sk, SOCK_DONE))
644 ret = -ENOTCONN;
645 break;
646 }
647 if (!timeo) {
648 ret = -EAGAIN;
649 break;
650 }
651 sk_wait_data(sk, &timeo);
652 if (signal_pending(current)) {
653 ret = sock_intr_errno(timeo);
654 break;
655 }
656 continue;
657 }
658 tss.len -= ret;
659 spliced += ret;
660
Willy Tarreau33966dd2009-01-13 16:04:36 -0800661 if (!timeo)
662 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800663 release_sock(sk);
664 lock_sock(sk);
665
666 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd2009-01-13 16:04:36 -0800667 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800668 signal_pending(current))
669 break;
670 }
671
672 release_sock(sk);
673
674 if (spliced)
675 return spliced;
676
677 return ret;
678}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000679EXPORT_SYMBOL(tcp_splice_read);
Jens Axboe9c55e012007-11-06 23:30:13 -0800680
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100681struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100682{
683 struct sk_buff *skb;
684
685 /* The TCP header must be at least 32-bit aligned. */
686 size = ALIGN(size, 4);
687
688 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
689 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800690 if (sk_wmem_schedule(sk, skb->truesize)) {
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100691 /*
692 * Make sure that we have exactly size bytes
693 * available to the caller, no more, no less.
694 */
695 skb_reserve(skb, skb_tailroom(skb) - size);
696 return skb;
697 }
698 __kfree_skb(skb);
699 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700700 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100701 sk_stream_moderate_sndbuf(sk);
702 }
703 return NULL;
704}
705
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000706static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
707 int large_allowed)
708{
709 struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000710 u32 xmit_size_goal, old_size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000711
712 xmit_size_goal = mss_now;
713
714 if (large_allowed && sk_can_gso(sk)) {
715 xmit_size_goal = ((sk->sk_gso_max_size - 1) -
716 inet_csk(sk)->icsk_af_ops->net_header_len -
717 inet_csk(sk)->icsk_ext_hdr_len -
718 tp->tcp_header_len);
719
720 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000721
722 /* We try hard to avoid divides here */
723 old_size_goal = tp->xmit_size_goal_segs * mss_now;
724
725 if (likely(old_size_goal <= xmit_size_goal &&
726 old_size_goal + mss_now > xmit_size_goal)) {
727 xmit_size_goal = old_size_goal;
728 } else {
729 tp->xmit_size_goal_segs = xmit_size_goal / mss_now;
730 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
731 }
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000732 }
733
Ilpo Järvinenafece1c2009-03-14 14:23:07 +0000734 return max(xmit_size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000735}
736
737static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
738{
739 int mss_now;
740
741 mss_now = tcp_current_mss(sk);
742 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
743
744 return mss_now;
745}
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset,
748 size_t psize, int flags)
749{
750 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700751 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 int err;
753 ssize_t copied;
754 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
755
756 /* Wait for a connection to finish. */
757 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
758 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
759 goto out_err;
760
761 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
762
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000763 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 copied = 0;
765
766 err = -EPIPE;
767 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000768 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
770 while (psize > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800771 struct sk_buff *skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 struct page *page = pages[poffset / PAGE_SIZE];
773 int copy, i, can_coalesce;
774 int offset = poffset % PAGE_SIZE;
775 int size = min_t(size_t, psize, PAGE_SIZE - offset);
776
David S. Millerfe067e82007-03-07 12:12:44 -0800777 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778new_segment:
779 if (!sk_stream_memory_free(sk))
780 goto wait_for_sndbuf;
781
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100782 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (!skb)
784 goto wait_for_memory;
785
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700786 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700787 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789
790 if (copy > size)
791 copy = size;
792
793 i = skb_shinfo(skb)->nr_frags;
794 can_coalesce = skb_can_coalesce(skb, i, page, offset);
795 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
796 tcp_mark_push(tp, skb);
797 goto new_segment;
798 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800799 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (can_coalesce) {
803 skb_shinfo(skb)->frags[i - 1].size += copy;
804 } else {
805 get_page(page);
806 skb_fill_page_desc(skb, i, page, offset, copy);
807 }
808
809 skb->len += copy;
810 skb->data_len += copy;
811 skb->truesize += copy;
812 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800813 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700814 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 tp->write_seq += copy;
816 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700817 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 if (!copied)
Changli Gaoa3433f32010-06-12 14:01:43 +0000820 TCP_SKB_CB(skb)->flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 copied += copy;
823 poffset += copy;
824 if (!(psize -= copy))
825 goto out;
826
Herbert Xu69d15062008-03-22 15:47:05 -0700827 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 continue;
829
830 if (forced_push(tp)) {
831 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700832 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800833 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 tcp_push_one(sk, mss_now);
835 continue;
836
837wait_for_sndbuf:
838 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
839wait_for_memory:
840 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700841 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
844 goto do_error;
845
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000846 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 }
848
849out:
850 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700851 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 return copied;
853
854do_error:
855 if (copied)
856 goto out;
857out_err:
858 return sk_stream_error(sk, flags, err);
859}
860
Changli Gao7ba42912010-07-10 20:41:55 +0000861int tcp_sendpage(struct sock *sk, struct page *page, int offset,
862 size_t size, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
864 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -0700867 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Changli Gao7ba42912010-07-10 20:41:55 +0000868 return sock_no_sendpage(sk->sk_socket, page, offset, size,
869 flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 lock_sock(sk);
872 TCP_CHECK_TIMER(sk);
873 res = do_tcp_sendpages(sk, &page, offset, size, flags);
874 TCP_CHECK_TIMER(sk);
875 release_sock(sk);
876 return res;
877}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000878EXPORT_SYMBOL(tcp_sendpage);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880#define TCP_PAGE(sk) (sk->sk_sndmsg_page)
881#define TCP_OFF(sk) (sk->sk_sndmsg_off)
882
Krishna Kumardef87cf2009-12-10 07:16:59 +0000883static inline int select_size(struct sock *sk, int sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700885 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700886 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Krishna Kumardef87cf2009-12-10 07:16:59 +0000888 if (sg) {
Herbert Xubcd76112006-06-30 13:36:35 -0700889 if (sk_can_gso(sk))
David S. Millerb4e26f52005-07-05 15:20:27 -0700890 tmp = 0;
891 else {
892 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
893
894 if (tmp >= pgbreak &&
895 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
896 tmp = pgbreak;
897 }
898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return tmp;
901}
902
Changli Gao7ba42912010-07-10 20:41:55 +0000903int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 size_t size)
905{
906 struct iovec *iov;
907 struct tcp_sock *tp = tcp_sk(sk);
908 struct sk_buff *skb;
909 int iovlen, flags;
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700910 int mss_now, size_goal;
Krishna Kumardef87cf2009-12-10 07:16:59 +0000911 int sg, err, copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 long timeo;
913
914 lock_sock(sk);
915 TCP_CHECK_TIMER(sk);
916
917 flags = msg->msg_flags;
918 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
919
920 /* Wait for a connection to finish. */
921 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
922 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
923 goto out_err;
924
925 /* This should be in poll */
926 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
927
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000928 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 /* Ok commence sending. */
931 iovlen = msg->msg_iovlen;
932 iov = msg->msg_iov;
933 copied = 0;
934
935 err = -EPIPE;
936 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000937 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Krishna Kumardef87cf2009-12-10 07:16:59 +0000939 sg = sk->sk_route_caps & NETIF_F_SG;
940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 while (--iovlen >= 0) {
942 int seglen = iov->iov_len;
943 unsigned char __user *from = iov->iov_base;
944
945 iov++;
946
947 while (seglen > 0) {
Herbert Xu6828b922009-06-28 18:06:41 +0000948 int copy = 0;
949 int max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
David S. Millerfe067e82007-03-07 12:12:44 -0800951 skb = tcp_write_queue_tail(sk);
Herbert Xu6828b922009-06-28 18:06:41 +0000952 if (tcp_send_head(sk)) {
953 if (skb->ip_summed == CHECKSUM_NONE)
954 max = mss_now;
955 copy = max - skb->len;
956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Herbert Xu6828b922009-06-28 18:06:41 +0000958 if (copy <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959new_segment:
960 /* Allocate new segment. If the interface is SG,
961 * allocate skb fitting to single page.
962 */
963 if (!sk_stream_memory_free(sk))
964 goto wait_for_sndbuf;
965
Krishna Kumardef87cf2009-12-10 07:16:59 +0000966 skb = sk_stream_alloc_skb(sk,
967 select_size(sk, sg),
968 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 if (!skb)
970 goto wait_for_memory;
971
972 /*
973 * Check whether we can use HW checksum.
974 */
Herbert Xu8648b302006-06-17 22:06:05 -0700975 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700976 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700978 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700979 copy = size_goal;
Herbert Xu6828b922009-06-28 18:06:41 +0000980 max = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 }
982
983 /* Try to append data to the end of skb. */
984 if (copy > seglen)
985 copy = seglen;
986
987 /* Where to copy to? */
988 if (skb_tailroom(skb) > 0) {
989 /* We have some space in skb head. Superb! */
990 if (copy > skb_tailroom(skb))
991 copy = skb_tailroom(skb);
992 if ((err = skb_add_data(skb, from, copy)) != 0)
993 goto do_fault;
994 } else {
995 int merge = 0;
996 int i = skb_shinfo(skb)->nr_frags;
997 struct page *page = TCP_PAGE(sk);
998 int off = TCP_OFF(sk);
999
1000 if (skb_can_coalesce(skb, i, page, off) &&
1001 off != PAGE_SIZE) {
1002 /* We can extend the last page
1003 * fragment. */
1004 merge = 1;
Krishna Kumardef87cf2009-12-10 07:16:59 +00001005 } else if (i == MAX_SKB_FRAGS || !sg) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 /* Need to add new fragment and cannot
1007 * do this because interface is non-SG,
1008 * or because all the page slots are
1009 * busy. */
1010 tcp_mark_push(tp, skb);
1011 goto new_segment;
1012 } else if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (off == PAGE_SIZE) {
1014 put_page(page);
1015 TCP_PAGE(sk) = page = NULL;
Herbert Xufb5f5e62005-09-05 18:55:48 -07001016 off = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
Herbert Xuef015782005-09-01 17:48:59 -07001018 } else
Herbert Xufb5f5e62005-09-05 18:55:48 -07001019 off = 0;
Herbert Xuef015782005-09-01 17:48:59 -07001020
1021 if (copy > PAGE_SIZE - off)
1022 copy = PAGE_SIZE - off;
1023
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001024 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -07001025 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 if (!page) {
1028 /* Allocate new cache page. */
1029 if (!(page = sk_stream_alloc_page(sk)))
1030 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 /* Time to copy data. We are close to
1034 * the end! */
1035 err = skb_copy_to_page(sk, from, skb, page,
1036 off, copy);
1037 if (err) {
1038 /* If this page was new, give it to the
1039 * socket so it does not get leaked.
1040 */
1041 if (!TCP_PAGE(sk)) {
1042 TCP_PAGE(sk) = page;
1043 TCP_OFF(sk) = 0;
1044 }
1045 goto do_error;
1046 }
1047
1048 /* Update the skb. */
1049 if (merge) {
1050 skb_shinfo(skb)->frags[i - 1].size +=
1051 copy;
1052 } else {
1053 skb_fill_page_desc(skb, i, page, off, copy);
1054 if (TCP_PAGE(sk)) {
1055 get_page(page);
1056 } else if (off + copy < PAGE_SIZE) {
1057 get_page(page);
1058 TCP_PAGE(sk) = page;
1059 }
1060 }
1061
1062 TCP_OFF(sk) = off + copy;
1063 }
1064
1065 if (!copied)
Changli Gaoa3433f32010-06-12 14:01:43 +00001066 TCP_SKB_CB(skb)->flags &= ~TCPHDR_PSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 tp->write_seq += copy;
1069 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -07001070 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072 from += copy;
1073 copied += copy;
1074 if ((seglen -= copy) == 0 && iovlen == 0)
1075 goto out;
1076
Herbert Xu6828b922009-06-28 18:06:41 +00001077 if (skb->len < max || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 continue;
1079
1080 if (forced_push(tp)) {
1081 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001082 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001083 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 tcp_push_one(sk, mss_now);
1085 continue;
1086
1087wait_for_sndbuf:
1088 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1089wait_for_memory:
1090 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001091 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1094 goto do_error;
1095
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001096 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 }
1098 }
1099
1100out:
1101 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001102 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 TCP_CHECK_TIMER(sk);
1104 release_sock(sk);
1105 return copied;
1106
1107do_fault:
1108 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001109 tcp_unlink_write_queue(skb, sk);
1110 /* It is the one place in all of TCP, except connection
1111 * reset, where we can be unlinking the send_head.
1112 */
1113 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001114 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 }
1116
1117do_error:
1118 if (copied)
1119 goto out;
1120out_err:
1121 err = sk_stream_error(sk, flags, err);
1122 TCP_CHECK_TIMER(sk);
1123 release_sock(sk);
1124 return err;
1125}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001126EXPORT_SYMBOL(tcp_sendmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128/*
1129 * Handle reading urgent data. BSD has very simple semantics for
1130 * this, no blocking and very strange errors 8)
1131 */
1132
Rami Rosen377f0a02009-03-31 14:43:17 -07001133static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
1135 struct tcp_sock *tp = tcp_sk(sk);
1136
1137 /* No URG data to read. */
1138 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1139 tp->urg_data == TCP_URG_READ)
1140 return -EINVAL; /* Yes this is right ! */
1141
1142 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1143 return -ENOTCONN;
1144
1145 if (tp->urg_data & TCP_URG_VALID) {
1146 int err = 0;
1147 char c = tp->urg_data;
1148
1149 if (!(flags & MSG_PEEK))
1150 tp->urg_data = TCP_URG_READ;
1151
1152 /* Read urgent data. */
1153 msg->msg_flags |= MSG_OOB;
1154
1155 if (len > 0) {
1156 if (!(flags & MSG_TRUNC))
1157 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1158 len = 1;
1159 } else
1160 msg->msg_flags |= MSG_TRUNC;
1161
1162 return err ? -EFAULT : len;
1163 }
1164
1165 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1166 return 0;
1167
1168 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1169 * the available implementations agree in this case:
1170 * this call should never block, independent of the
1171 * blocking state of the socket.
1172 * Mike <pall@rz.uni-karlsruhe.de>
1173 */
1174 return -EAGAIN;
1175}
1176
1177/* Clean up the receive buffer for full frames taken by the user,
1178 * then send an ACK if necessary. COPIED is the number of bytes
1179 * tcp_recvmsg has given to the user so far, it speeds up the
1180 * calculation of whether or not we must ACK for the sake of
1181 * a window update.
1182 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001183void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
1185 struct tcp_sock *tp = tcp_sk(sk);
1186 int time_to_ack = 0;
1187
1188#if TCP_DEBUG
1189 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1190
Ilpo Järvinend792c102009-11-13 13:56:33 -08001191 WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
1192 KERN_INFO "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
1193 tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194#endif
1195
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001196 if (inet_csk_ack_scheduled(sk)) {
1197 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 /* Delayed ACKs frequently hit locked sockets during bulk
1199 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001200 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001202 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 /*
1204 * If this read emptied read buffer, we send ACK, if
1205 * connection is not bidirectional, user drained
1206 * receive buffer and there was a small segment
1207 * in queue.
1208 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001209 (copied > 0 &&
1210 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1211 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1212 !icsk->icsk_ack.pingpong)) &&
1213 !atomic_read(&sk->sk_rmem_alloc)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 time_to_ack = 1;
1215 }
1216
1217 /* We send an ACK if we can now advertise a non-zero window
1218 * which has been raised "significantly".
1219 *
1220 * Even if window raised up to infinity, do not send window open ACK
1221 * in states, where we will not receive more. It is useless.
1222 */
1223 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1224 __u32 rcv_window_now = tcp_receive_window(tp);
1225
1226 /* Optimize, __tcp_select_window() is not cheap. */
1227 if (2*rcv_window_now <= tp->window_clamp) {
1228 __u32 new_window = __tcp_select_window(sk);
1229
1230 /* Send ACK now, if this read freed lots of space
1231 * in our buffer. Certainly, new_window is new window.
1232 * We can advertise it now, if it is not less than current one.
1233 * "Lots" means "at least twice" here.
1234 */
1235 if (new_window && new_window >= 2 * rcv_window_now)
1236 time_to_ack = 1;
1237 }
1238 }
1239 if (time_to_ack)
1240 tcp_send_ack(sk);
1241}
1242
1243static void tcp_prequeue_process(struct sock *sk)
1244{
1245 struct sk_buff *skb;
1246 struct tcp_sock *tp = tcp_sk(sk);
1247
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001248 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 /* RX process wants to run with disabled BHs, though it is not
1251 * necessary */
1252 local_bh_disable();
1253 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001254 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 local_bh_enable();
1256
1257 /* Clear memory counter. */
1258 tp->ucopy.memory = 0;
1259}
1260
Steven J. Magnani73852e82010-03-16 05:22:44 +00001261#ifdef CONFIG_NET_DMA
1262static void tcp_service_net_dma(struct sock *sk, bool wait)
1263{
1264 dma_cookie_t done, used;
1265 dma_cookie_t last_issued;
1266 struct tcp_sock *tp = tcp_sk(sk);
1267
1268 if (!tp->ucopy.dma_chan)
1269 return;
1270
1271 last_issued = tp->ucopy.dma_cookie;
1272 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1273
1274 do {
1275 if (dma_async_memcpy_complete(tp->ucopy.dma_chan,
1276 last_issued, &done,
1277 &used) == DMA_SUCCESS) {
1278 /* Safe to free early-copied skbs now */
1279 __skb_queue_purge(&sk->sk_async_wait_queue);
1280 break;
1281 } else {
1282 struct sk_buff *skb;
1283 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1284 (dma_async_is_complete(skb->dma_cookie, done,
1285 used) == DMA_SUCCESS)) {
1286 __skb_dequeue(&sk->sk_async_wait_queue);
1287 kfree_skb(skb);
1288 }
1289 }
1290 } while (wait);
1291}
1292#endif
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294static inline struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1295{
1296 struct sk_buff *skb;
1297 u32 offset;
1298
1299 skb_queue_walk(&sk->sk_receive_queue, skb) {
1300 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001301 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001303 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 *off = offset;
1305 return skb;
1306 }
1307 }
1308 return NULL;
1309}
1310
1311/*
1312 * This routine provides an alternative to tcp_recvmsg() for routines
1313 * that would like to handle copying from skbuffs directly in 'sendfile'
1314 * fashion.
1315 * Note:
1316 * - It is assumed that the socket was locked by the caller.
1317 * - The routine does not block.
1318 * - At present, there is no support for reading OOB data
1319 * or for 'peeking' the socket using this routine
1320 * (although both would be easy to implement).
1321 */
1322int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1323 sk_read_actor_t recv_actor)
1324{
1325 struct sk_buff *skb;
1326 struct tcp_sock *tp = tcp_sk(sk);
1327 u32 seq = tp->copied_seq;
1328 u32 offset;
1329 int copied = 0;
1330
1331 if (sk->sk_state == TCP_LISTEN)
1332 return -ENOTCONN;
1333 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1334 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001335 int used;
1336 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 len = skb->len - offset;
1339 /* Stop reading if we hit a patch of urgent data */
1340 if (tp->urg_data) {
1341 u32 urg_offset = tp->urg_seq - seq;
1342 if (urg_offset < len)
1343 len = urg_offset;
1344 if (!len)
1345 break;
1346 }
1347 used = recv_actor(desc, skb, offset, len);
Jens Axboeddb61a52007-06-23 23:07:50 -07001348 if (used < 0) {
1349 if (!copied)
1350 copied = used;
1351 break;
1352 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 seq += used;
1354 copied += used;
1355 offset += used;
1356 }
Octavian Purdila293ad602008-06-04 15:45:58 -07001357 /*
1358 * If recv_actor drops the lock (e.g. TCP splice
1359 * receive) the skb pointer might be invalid when
1360 * getting here: tcp_collapse might have deleted it
1361 * while aggregating skbs from the socket queue.
1362 */
1363 skb = tcp_recv_skb(sk, seq-1, &offset);
1364 if (!skb || (offset+1 != skb->len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 break;
1366 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001367 if (tcp_hdr(skb)->fin) {
Chris Leech624d1162006-05-23 18:01:28 -07001368 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 ++seq;
1370 break;
1371 }
Chris Leech624d1162006-05-23 18:01:28 -07001372 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (!desc->count)
1374 break;
Steven J. Magnanibaff42a2010-03-30 13:56:01 -07001375 tp->copied_seq = seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377 tp->copied_seq = seq;
1378
1379 tcp_rcv_space_adjust(sk);
1380
1381 /* Clean up data we have read: This will do ACK frames. */
Jens Axboeddb61a52007-06-23 23:07:50 -07001382 if (copied > 0)
Chris Leech0e4b4992006-05-23 18:00:16 -07001383 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return copied;
1385}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001386EXPORT_SYMBOL(tcp_read_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388/*
1389 * This routine copies from a sock struct into the user buffer.
1390 *
1391 * Technical note: in 2.3 we work on _locked_ socket, so that
1392 * tricks with *seq access order and skb->users are not required.
1393 * Probably, code can be easily improved even more.
1394 */
1395
1396int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1397 size_t len, int nonblock, int flags, int *addr_len)
1398{
1399 struct tcp_sock *tp = tcp_sk(sk);
1400 int copied = 0;
1401 u32 peek_seq;
1402 u32 *seq;
1403 unsigned long used;
1404 int err;
1405 int target; /* Read at least this many bytes */
1406 long timeo;
1407 struct task_struct *user_recv = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001408 int copied_early = 0;
Chris Leech2b1244a2007-03-08 09:57:36 -08001409 struct sk_buff *skb;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001410 u32 urg_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412 lock_sock(sk);
1413
1414 TCP_CHECK_TIMER(sk);
1415
1416 err = -ENOTCONN;
1417 if (sk->sk_state == TCP_LISTEN)
1418 goto out;
1419
1420 timeo = sock_rcvtimeo(sk, nonblock);
1421
1422 /* Urgent data needs to be handled specially. */
1423 if (flags & MSG_OOB)
1424 goto recv_urg;
1425
1426 seq = &tp->copied_seq;
1427 if (flags & MSG_PEEK) {
1428 peek_seq = tp->copied_seq;
1429 seq = &peek_seq;
1430 }
1431
1432 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1433
Chris Leech1a2449a2006-05-23 18:05:53 -07001434#ifdef CONFIG_NET_DMA
1435 tp->ucopy.dma_chan = NULL;
1436 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001437 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001438 {
1439 int available = 0;
1440
1441 if (skb)
1442 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1443 if ((available < target) &&
1444 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1445 !sysctl_tcp_low_latency &&
Dan Williamsf67b4592009-01-06 11:38:15 -07001446 dma_find_channel(DMA_MEMCPY)) {
Andrew Mortone00c5d82007-03-08 09:57:36 -08001447 preempt_enable_no_resched();
1448 tp->ucopy.pinned_list =
1449 dma_pin_iovec_pages(msg->msg_iov, len);
1450 } else {
1451 preempt_enable_no_resched();
1452 }
1453 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001454#endif
1455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 u32 offset;
1458
1459 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1460 if (tp->urg_data && tp->urg_seq == *seq) {
1461 if (copied)
1462 break;
1463 if (signal_pending(current)) {
1464 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1465 break;
1466 }
1467 }
1468
1469 /* Next get a buffer. */
1470
David S. Miller91521942009-05-28 21:35:47 -07001471 skb_queue_walk(&sk->sk_receive_queue, skb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 /* Now that we have two receive queues this
1473 * shouldn't happen.
1474 */
Ilpo Järvinend792c102009-11-13 13:56:33 -08001475 if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
1476 KERN_INFO "recvmsg bug: copied %X "
1477 "seq %X rcvnxt %X fl %X\n", *seq,
1478 TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
1479 flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 break;
Ilpo Järvinend792c102009-11-13 13:56:33 -08001481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001483 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 offset--;
1485 if (offset < skb->len)
1486 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001487 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 goto found_fin_ok;
Arjan van de Venc62f4c42009-10-22 21:37:56 -07001489 WARN(!(flags & MSG_PEEK), KERN_INFO "recvmsg bug 2: "
Ilpo Järvinend792c102009-11-13 13:56:33 -08001490 "copied %X seq %X rcvnxt %X fl %X\n",
1491 *seq, TCP_SKB_CB(skb)->seq,
1492 tp->rcv_nxt, flags);
David S. Miller91521942009-05-28 21:35:47 -07001493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 /* Well, if we have backlog, try to process it now yet. */
1496
1497 if (copied >= target && !sk->sk_backlog.tail)
1498 break;
1499
1500 if (copied) {
1501 if (sk->sk_err ||
1502 sk->sk_state == TCP_CLOSE ||
1503 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1504 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08001505 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 break;
1507 } else {
1508 if (sock_flag(sk, SOCK_DONE))
1509 break;
1510
1511 if (sk->sk_err) {
1512 copied = sock_error(sk);
1513 break;
1514 }
1515
1516 if (sk->sk_shutdown & RCV_SHUTDOWN)
1517 break;
1518
1519 if (sk->sk_state == TCP_CLOSE) {
1520 if (!sock_flag(sk, SOCK_DONE)) {
1521 /* This occurs when user tries to read
1522 * from never connected socket.
1523 */
1524 copied = -ENOTCONN;
1525 break;
1526 }
1527 break;
1528 }
1529
1530 if (!timeo) {
1531 copied = -EAGAIN;
1532 break;
1533 }
1534
1535 if (signal_pending(current)) {
1536 copied = sock_intr_errno(timeo);
1537 break;
1538 }
1539 }
1540
Chris Leech0e4b4992006-05-23 18:00:16 -07001541 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
David S. Miller7df55122005-06-18 23:01:10 -07001543 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 /* Install new reader */
1545 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1546 user_recv = current;
1547 tp->ucopy.task = user_recv;
1548 tp->ucopy.iov = msg->msg_iov;
1549 }
1550
1551 tp->ucopy.len = len;
1552
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001553 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1554 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 /* Ugly... If prequeue is not empty, we have to
1557 * process it before releasing socket, otherwise
1558 * order will be broken at second iteration.
1559 * More elegant solution is required!!!
1560 *
1561 * Look: we have the following (pseudo)queues:
1562 *
1563 * 1. packets in flight
1564 * 2. backlog
1565 * 3. prequeue
1566 * 4. receive_queue
1567 *
1568 * Each queue can be processed only if the next ones
1569 * are empty. At this point we have empty receive_queue.
1570 * But prequeue _can_ be not empty after 2nd iteration,
1571 * when we jumped to start of loop because backlog
1572 * processing added something to receive_queue.
1573 * We cannot release_sock(), because backlog contains
1574 * packets arrived _after_ prequeued ones.
1575 *
1576 * Shortly, algorithm is clear --- to process all
1577 * the queues in order. We could make it more directly,
1578 * requeueing packets from backlog to prequeue, if
1579 * is not empty. It is more elegant, but eats cycles,
1580 * unfortunately.
1581 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001582 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 goto do_prequeue;
1584
1585 /* __ Set realtime policy in scheduler __ */
1586 }
1587
Steven J. Magnani73852e82010-03-16 05:22:44 +00001588#ifdef CONFIG_NET_DMA
1589 if (tp->ucopy.dma_chan)
1590 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1591#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 if (copied >= target) {
1593 /* Do not sleep, just process backlog. */
1594 release_sock(sk);
1595 lock_sock(sk);
1596 } else
1597 sk_wait_data(sk, &timeo);
1598
Chris Leech1a2449a2006-05-23 18:05:53 -07001599#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001600 tcp_service_net_dma(sk, false); /* Don't block */
Chris Leech1a2449a2006-05-23 18:05:53 -07001601 tp->ucopy.wakeup = 0;
1602#endif
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 if (user_recv) {
1605 int chunk;
1606
1607 /* __ Restore normal policy in scheduler __ */
1608
1609 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001610 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 len -= chunk;
1612 copied += chunk;
1613 }
1614
1615 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001616 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617do_prequeue:
1618 tcp_prequeue_process(sk);
1619
1620 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001621 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 len -= chunk;
1623 copied += chunk;
1624 }
1625 }
1626 }
Ilpo Järvinen77527312009-05-10 20:32:34 +00001627 if ((flags & MSG_PEEK) &&
1628 (peek_seq - copied - urg_hole != tp->copied_seq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 if (net_ratelimit())
1630 printk(KERN_DEBUG "TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001631 current->comm, task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 peek_seq = tp->copied_seq;
1633 }
1634 continue;
1635
1636 found_ok_skb:
1637 /* Ok so how much can we use? */
1638 used = skb->len - offset;
1639 if (len < used)
1640 used = len;
1641
1642 /* Do we have urgent data here? */
1643 if (tp->urg_data) {
1644 u32 urg_offset = tp->urg_seq - *seq;
1645 if (urg_offset < used) {
1646 if (!urg_offset) {
1647 if (!sock_flag(sk, SOCK_URGINLINE)) {
1648 ++*seq;
Ilpo Järvinen77527312009-05-10 20:32:34 +00001649 urg_hole++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 offset++;
1651 used--;
1652 if (!used)
1653 goto skip_copy;
1654 }
1655 } else
1656 used = urg_offset;
1657 }
1658 }
1659
1660 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001661#ifdef CONFIG_NET_DMA
1662 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dan Williamsf67b4592009-01-06 11:38:15 -07001663 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
Chris Leech1a2449a2006-05-23 18:05:53 -07001664
1665 if (tp->ucopy.dma_chan) {
1666 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1667 tp->ucopy.dma_chan, skb, offset,
1668 msg->msg_iov, used,
1669 tp->ucopy.pinned_list);
1670
1671 if (tp->ucopy.dma_cookie < 0) {
1672
1673 printk(KERN_ALERT "dma_cookie < 0\n");
1674
1675 /* Exception. Bailout! */
1676 if (!copied)
1677 copied = -EFAULT;
1678 break;
1679 }
Steven J. Magnani73852e82010-03-16 05:22:44 +00001680
1681 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1682
Chris Leech1a2449a2006-05-23 18:05:53 -07001683 if ((offset + used) == skb->len)
1684 copied_early = 1;
1685
1686 } else
1687#endif
1688 {
1689 err = skb_copy_datagram_iovec(skb, offset,
1690 msg->msg_iov, used);
1691 if (err) {
1692 /* Exception. Bailout! */
1693 if (!copied)
1694 copied = -EFAULT;
1695 break;
1696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698 }
1699
1700 *seq += used;
1701 copied += used;
1702 len -= used;
1703
1704 tcp_rcv_space_adjust(sk);
1705
1706skip_copy:
1707 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1708 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001709 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711 if (used + offset < skb->len)
1712 continue;
1713
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001714 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001716 if (!(flags & MSG_PEEK)) {
1717 sk_eat_skb(sk, skb, copied_early);
1718 copied_early = 0;
1719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 continue;
1721
1722 found_fin_ok:
1723 /* Process the FIN. */
1724 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001725 if (!(flags & MSG_PEEK)) {
1726 sk_eat_skb(sk, skb, copied_early);
1727 copied_early = 0;
1728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 break;
1730 } while (len > 0);
1731
1732 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001733 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 int chunk;
1735
1736 tp->ucopy.len = copied > 0 ? len : 0;
1737
1738 tcp_prequeue_process(sk);
1739
1740 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001741 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 len -= chunk;
1743 copied += chunk;
1744 }
1745 }
1746
1747 tp->ucopy.task = NULL;
1748 tp->ucopy.len = 0;
1749 }
1750
Chris Leech1a2449a2006-05-23 18:05:53 -07001751#ifdef CONFIG_NET_DMA
Steven J. Magnani73852e82010-03-16 05:22:44 +00001752 tcp_service_net_dma(sk, true); /* Wait for queue to drain */
1753 tp->ucopy.dma_chan = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001754
Chris Leech1a2449a2006-05-23 18:05:53 -07001755 if (tp->ucopy.pinned_list) {
1756 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1757 tp->ucopy.pinned_list = NULL;
1758 }
1759#endif
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 /* According to UNIX98, msg_name/msg_namelen are ignored
1762 * on connected socket. I was just happy when found this 8) --ANK
1763 */
1764
1765 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001766 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 TCP_CHECK_TIMER(sk);
1769 release_sock(sk);
1770 return copied;
1771
1772out:
1773 TCP_CHECK_TIMER(sk);
1774 release_sock(sk);
1775 return err;
1776
1777recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07001778 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 goto out;
1780}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001781EXPORT_SYMBOL(tcp_recvmsg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001783void tcp_set_state(struct sock *sk, int state)
1784{
1785 int oldstate = sk->sk_state;
1786
1787 switch (state) {
1788 case TCP_ESTABLISHED:
1789 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001790 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001791 break;
1792
1793 case TCP_CLOSE:
1794 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001795 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001796
1797 sk->sk_prot->unhash(sk);
1798 if (inet_csk(sk)->icsk_bind_hash &&
1799 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001800 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001801 /* fall through */
1802 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001803 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07001804 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001805 }
1806
1807 /* Change state AFTER socket is unhashed to avoid closed
1808 * socket sitting in hash tables.
1809 */
1810 sk->sk_state = state;
1811
1812#ifdef STATE_TRACE
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001813 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001814#endif
1815}
1816EXPORT_SYMBOL_GPL(tcp_set_state);
1817
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818/*
1819 * State processing on a close. This implements the state shift for
1820 * sending our FIN frame. Note that we only send a FIN for some
1821 * states. A shutdown() may have already sent the FIN, or we may be
1822 * closed.
1823 */
1824
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08001825static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 /* current state: new state: action: */
1827 /* (Invalid) */ TCP_CLOSE,
1828 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1829 /* TCP_SYN_SENT */ TCP_CLOSE,
1830 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1831 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
1832 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
1833 /* TCP_TIME_WAIT */ TCP_CLOSE,
1834 /* TCP_CLOSE */ TCP_CLOSE,
1835 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
1836 /* TCP_LAST_ACK */ TCP_LAST_ACK,
1837 /* TCP_LISTEN */ TCP_CLOSE,
1838 /* TCP_CLOSING */ TCP_CLOSING,
1839};
1840
1841static int tcp_close_state(struct sock *sk)
1842{
1843 int next = (int)new_state[sk->sk_state];
1844 int ns = next & TCP_STATE_MASK;
1845
1846 tcp_set_state(sk, ns);
1847
1848 return next & TCP_ACTION_FIN;
1849}
1850
1851/*
1852 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07001853 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 */
1855
1856void tcp_shutdown(struct sock *sk, int how)
1857{
1858 /* We need to grab some memory, and put together a FIN,
1859 * and then put it into the queue to be sent.
1860 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1861 */
1862 if (!(how & SEND_SHUTDOWN))
1863 return;
1864
1865 /* If we've already sent a FIN, or it's a closed state, skip this. */
1866 if ((1 << sk->sk_state) &
1867 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1868 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
1869 /* Clear out any half completed packets. FIN if needed. */
1870 if (tcp_close_state(sk))
1871 tcp_send_fin(sk);
1872 }
1873}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001874EXPORT_SYMBOL(tcp_shutdown);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876void tcp_close(struct sock *sk, long timeout)
1877{
1878 struct sk_buff *skb;
1879 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07001880 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 lock_sock(sk);
1883 sk->sk_shutdown = SHUTDOWN_MASK;
1884
1885 if (sk->sk_state == TCP_LISTEN) {
1886 tcp_set_state(sk, TCP_CLOSE);
1887
1888 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001889 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
1891 goto adjudge_to_death;
1892 }
1893
1894 /* We need to flush the recv. buffs. We do this only on the
1895 * descriptor close, not protocol-sourced closes, because the
1896 * reader process may not have drained the data yet!
1897 */
1898 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1899 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001900 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 data_was_unread += len;
1902 __kfree_skb(skb);
1903 }
1904
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001905 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Konstantin Khorenko565b7b22010-06-24 21:54:58 -07001907 /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
1908 if (sk->sk_state == TCP_CLOSE)
1909 goto adjudge_to_death;
1910
Gerrit Renker65bb7232007-04-28 21:21:46 -07001911 /* As outlined in RFC 2525, section 2.17, we send a RST here because
1912 * data was lost. To witness the awful effects of the old behavior of
1913 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
1914 * GET in an FTP client, suspend the process, wait for the client to
1915 * advertise a zero window, then kill -9 the FTP client, wheee...
1916 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 */
1918 if (data_was_unread) {
1919 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001920 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 tcp_set_state(sk, TCP_CLOSE);
Wu Fengguangaa133072009-09-02 23:45:45 -07001922 tcp_send_active_reset(sk, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1924 /* Check zero linger _after_ checking for unread data. */
1925 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001926 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 } else if (tcp_close_state(sk)) {
1928 /* We FIN if the application ate all the data before
1929 * zapping the connection.
1930 */
1931
1932 /* RED-PEN. Formally speaking, we have broken TCP state
1933 * machine. State transitions:
1934 *
1935 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
1936 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
1937 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
1938 *
1939 * are legal only when FIN has been sent (i.e. in window),
1940 * rather than queued out of window. Purists blame.
1941 *
1942 * F.e. "RFC state" is ESTABLISHED,
1943 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
1944 *
1945 * The visible declinations are that sometimes
1946 * we enter time-wait state, when it is not required really
1947 * (harmless), do not send active resets, when they are
1948 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
1949 * they look as CLOSING or LAST_ACK for Linux)
1950 * Probably, I missed some more holelets.
1951 * --ANK
1952 */
1953 tcp_send_fin(sk);
1954 }
1955
1956 sk_stream_wait_close(sk, timeout);
1957
1958adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07001959 state = sk->sk_state;
1960 sock_hold(sk);
1961 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07001962
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 /* It is the last release_sock in its life. It will remove backlog. */
1964 release_sock(sk);
1965
1966
1967 /* Now socket is owned by kernel and we acquire BH lock
1968 to finish close. No need to check for user refs.
1969 */
1970 local_bh_disable();
1971 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001972 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Herbert Xueb4dea52008-12-29 23:04:08 -08001974 percpu_counter_inc(sk->sk_prot->orphan_count);
1975
Herbert Xu75c2d9072006-05-03 23:31:35 -07001976 /* Have we already been destroyed by a softirq or backlog? */
1977 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
1978 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
1980 /* This is a (useful) BSD violating of the RFC. There is a
1981 * problem with TCP as specified in that the other end could
1982 * keep a socket open forever with no application left this end.
1983 * We use a 3 minute timeout (about the same as BSD) then kill
1984 * our end. If they send after that then tough - BUT: long enough
1985 * that we won't make the old 4*rto = almost no time - whoops
1986 * reset mistake.
1987 *
1988 * Nope, it was not mistake. It is really desired behaviour
1989 * f.e. on http servers, when such sockets are useless, but
1990 * consume significant resources. Let's do it with special
1991 * linger2 option. --ANK
1992 */
1993
1994 if (sk->sk_state == TCP_FIN_WAIT2) {
1995 struct tcp_sock *tp = tcp_sk(sk);
1996 if (tp->linger2 < 0) {
1997 tcp_set_state(sk, TCP_CLOSE);
1998 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001999 NET_INC_STATS_BH(sock_net(sk),
2000 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002002 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
2004 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07002005 inet_csk_reset_keepalive_timer(sk,
2006 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
2009 goto out;
2010 }
2011 }
2012 }
2013 if (sk->sk_state != TCP_CLOSE) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002014 sk_mem_reclaim(sk);
David S. Millerad1af0f2010-08-25 02:27:49 -07002015 if (tcp_too_many_orphans(sk, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (net_ratelimit())
2017 printk(KERN_INFO "TCP: too many of orphaned "
2018 "sockets\n");
2019 tcp_set_state(sk, TCP_CLOSE);
2020 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07002021 NET_INC_STATS_BH(sock_net(sk),
2022 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
2024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
2026 if (sk->sk_state == TCP_CLOSE)
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002027 inet_csk_destroy_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 /* Otherwise, socket is reprieved until protocol close. */
2029
2030out:
2031 bh_unlock_sock(sk);
2032 local_bh_enable();
2033 sock_put(sk);
2034}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002035EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
2037/* These states need RST on ABORT according to RFC793 */
2038
2039static inline int tcp_need_reset(int state)
2040{
2041 return (1 << state) &
2042 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
2043 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
2044}
2045
2046int tcp_disconnect(struct sock *sk, int flags)
2047{
2048 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002049 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 struct tcp_sock *tp = tcp_sk(sk);
2051 int err = 0;
2052 int old_state = sk->sk_state;
2053
2054 if (old_state != TCP_CLOSE)
2055 tcp_set_state(sk, TCP_CLOSE);
2056
2057 /* ABORT function of RFC793 */
2058 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002059 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 } else if (tcp_need_reset(old_state) ||
2061 (tp->snd_nxt != tp->write_seq &&
2062 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08002063 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 * states
2065 */
2066 tcp_send_active_reset(sk, gfp_any());
2067 sk->sk_err = ECONNRESET;
2068 } else if (old_state == TCP_SYN_SENT)
2069 sk->sk_err = ECONNRESET;
2070
2071 tcp_clear_xmit_timers(sk);
2072 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08002073 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07002075#ifdef CONFIG_NET_DMA
2076 __skb_queue_purge(&sk->sk_async_wait_queue);
2077#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002079 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
2081 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
2082 inet_reset_saddr(sk);
2083
2084 sk->sk_shutdown = 0;
2085 sock_reset_flag(sk, SOCK_DONE);
2086 tp->srtt = 0;
2087 if ((tp->write_seq += tp->max_window + 2) == 0)
2088 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002089 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002091 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 tp->packets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002093 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 tp->snd_cwnd_cnt = 0;
Stephen Hemminger9772efb2005-11-10 17:09:53 -08002095 tp->bytes_acked = 0;
Eric Dumazet1fdf4752009-11-30 12:53:30 -08002096 tp->window_clamp = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002097 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002099 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08002100 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002101 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 __sk_dst_reset(sk);
2103
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002104 WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 sk->sk_error_report(sk);
2107 return err;
2108}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002109EXPORT_SYMBOL(tcp_disconnect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 * Socket option code for TCP.
2113 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002114static int do_tcp_setsockopt(struct sock *sk, int level,
David S. Millerb7058842009-09-30 16:12:20 -07002115 int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116{
2117 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002118 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 int val;
2120 int err = 0;
2121
William Allen Simpsone56fb502009-12-02 18:19:30 +00002122 /* These are data/string values, all the others are ints */
2123 switch (optname) {
2124 case TCP_CONGESTION: {
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002125 char name[TCP_CA_NAME_MAX];
2126
2127 if (optlen < 1)
2128 return -EINVAL;
2129
2130 val = strncpy_from_user(name, optval,
Andrew Morton4fdb78d2009-10-01 15:02:20 -07002131 min_t(long, TCP_CA_NAME_MAX-1, optlen));
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002132 if (val < 0)
2133 return -EFAULT;
2134 name[val] = 0;
2135
2136 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002137 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002138 release_sock(sk);
2139 return err;
2140 }
William Allen Simpsone56fb502009-12-02 18:19:30 +00002141 case TCP_COOKIE_TRANSACTIONS: {
2142 struct tcp_cookie_transactions ctd;
2143 struct tcp_cookie_values *cvp = NULL;
2144
2145 if (sizeof(ctd) > optlen)
2146 return -EINVAL;
2147 if (copy_from_user(&ctd, optval, sizeof(ctd)))
2148 return -EFAULT;
2149
2150 if (ctd.tcpct_used > sizeof(ctd.tcpct_value) ||
2151 ctd.tcpct_s_data_desired > TCP_MSS_DESIRED)
2152 return -EINVAL;
2153
2154 if (ctd.tcpct_cookie_desired == 0) {
2155 /* default to global value */
2156 } else if ((0x1 & ctd.tcpct_cookie_desired) ||
2157 ctd.tcpct_cookie_desired > TCP_COOKIE_MAX ||
2158 ctd.tcpct_cookie_desired < TCP_COOKIE_MIN) {
2159 return -EINVAL;
2160 }
2161
2162 if (TCP_COOKIE_OUT_NEVER & ctd.tcpct_flags) {
2163 /* Supercedes all other values */
2164 lock_sock(sk);
2165 if (tp->cookie_values != NULL) {
2166 kref_put(&tp->cookie_values->kref,
2167 tcp_cookie_values_release);
2168 tp->cookie_values = NULL;
2169 }
2170 tp->rx_opt.cookie_in_always = 0; /* false */
2171 tp->rx_opt.cookie_out_never = 1; /* true */
2172 release_sock(sk);
2173 return err;
2174 }
2175
2176 /* Allocate ancillary memory before locking.
2177 */
2178 if (ctd.tcpct_used > 0 ||
2179 (tp->cookie_values == NULL &&
2180 (sysctl_tcp_cookie_size > 0 ||
2181 ctd.tcpct_cookie_desired > 0 ||
2182 ctd.tcpct_s_data_desired > 0))) {
2183 cvp = kzalloc(sizeof(*cvp) + ctd.tcpct_used,
2184 GFP_KERNEL);
2185 if (cvp == NULL)
2186 return -ENOMEM;
Dmitry Popova3bdb542010-07-29 01:59:36 +00002187
2188 kref_init(&cvp->kref);
William Allen Simpsone56fb502009-12-02 18:19:30 +00002189 }
2190 lock_sock(sk);
2191 tp->rx_opt.cookie_in_always =
2192 (TCP_COOKIE_IN_ALWAYS & ctd.tcpct_flags);
2193 tp->rx_opt.cookie_out_never = 0; /* false */
2194
2195 if (tp->cookie_values != NULL) {
2196 if (cvp != NULL) {
2197 /* Changed values are recorded by a changed
2198 * pointer, ensuring the cookie will differ,
2199 * without separately hashing each value later.
2200 */
2201 kref_put(&tp->cookie_values->kref,
2202 tcp_cookie_values_release);
William Allen Simpsone56fb502009-12-02 18:19:30 +00002203 } else {
2204 cvp = tp->cookie_values;
2205 }
2206 }
Dmitry Popova3bdb542010-07-29 01:59:36 +00002207
William Allen Simpsone56fb502009-12-02 18:19:30 +00002208 if (cvp != NULL) {
2209 cvp->cookie_desired = ctd.tcpct_cookie_desired;
2210
2211 if (ctd.tcpct_used > 0) {
2212 memcpy(cvp->s_data_payload, ctd.tcpct_value,
2213 ctd.tcpct_used);
2214 cvp->s_data_desired = ctd.tcpct_used;
2215 cvp->s_data_constant = 1; /* true */
2216 } else {
2217 /* No constant payload data. */
2218 cvp->s_data_desired = ctd.tcpct_s_data_desired;
2219 cvp->s_data_constant = 0; /* false */
2220 }
Dmitry Popova3bdb542010-07-29 01:59:36 +00002221
2222 tp->cookie_values = cvp;
William Allen Simpsone56fb502009-12-02 18:19:30 +00002223 }
2224 release_sock(sk);
2225 return err;
2226 }
2227 default:
2228 /* fallthru */
2229 break;
Joe Perchesccbd6a52010-05-14 10:58:26 +00002230 }
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 if (optlen < sizeof(int))
2233 return -EINVAL;
2234
2235 if (get_user(val, (int __user *)optval))
2236 return -EFAULT;
2237
2238 lock_sock(sk);
2239
2240 switch (optname) {
2241 case TCP_MAXSEG:
2242 /* Values greater than interface MTU won't take effect. However
2243 * at the point when this call is done we typically don't yet
2244 * know which interface is going to be used */
2245 if (val < 8 || val > MAX_TCP_WINDOW) {
2246 err = -EINVAL;
2247 break;
2248 }
2249 tp->rx_opt.user_mss = val;
2250 break;
2251
2252 case TCP_NODELAY:
2253 if (val) {
2254 /* TCP_NODELAY is weaker than TCP_CORK, so that
2255 * this option on corked socket is remembered, but
2256 * it is not activated until cork is cleared.
2257 *
2258 * However, when TCP_NODELAY is set we make
2259 * an explicit push, which overrides even TCP_CORK
2260 * for currently queued segments.
2261 */
2262 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002263 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 } else {
2265 tp->nonagle &= ~TCP_NAGLE_OFF;
2266 }
2267 break;
2268
Andreas Petlund36e31b0a2010-02-18 02:47:01 +00002269 case TCP_THIN_LINEAR_TIMEOUTS:
2270 if (val < 0 || val > 1)
2271 err = -EINVAL;
2272 else
2273 tp->thin_lto = val;
2274 break;
2275
Andreas Petlund7e380172010-02-18 04:48:19 +00002276 case TCP_THIN_DUPACK:
2277 if (val < 0 || val > 1)
2278 err = -EINVAL;
2279 else
2280 tp->thin_dupack = val;
2281 break;
2282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 case TCP_CORK:
2284 /* When set indicates to always queue non-full frames.
2285 * Later the user clears this option and we transmit
2286 * any pending partial frames in the queue. This is
2287 * meant to be used alongside sendfile() to get properly
2288 * filled frames when the user (for example) must write
2289 * out headers with a write() call first and then use
2290 * sendfile to send out the data parts.
2291 *
2292 * TCP_CORK can be set together with TCP_NODELAY and it is
2293 * stronger than TCP_NODELAY.
2294 */
2295 if (val) {
2296 tp->nonagle |= TCP_NAGLE_CORK;
2297 } else {
2298 tp->nonagle &= ~TCP_NAGLE_CORK;
2299 if (tp->nonagle&TCP_NAGLE_OFF)
2300 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002301 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 }
2303 break;
2304
2305 case TCP_KEEPIDLE:
2306 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2307 err = -EINVAL;
2308 else {
2309 tp->keepalive_time = val * HZ;
2310 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2311 !((1 << sk->sk_state) &
2312 (TCPF_CLOSE | TCPF_LISTEN))) {
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00002313 u32 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 if (tp->keepalive_time > elapsed)
2315 elapsed = tp->keepalive_time - elapsed;
2316 else
2317 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002318 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 }
2320 }
2321 break;
2322 case TCP_KEEPINTVL:
2323 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2324 err = -EINVAL;
2325 else
2326 tp->keepalive_intvl = val * HZ;
2327 break;
2328 case TCP_KEEPCNT:
2329 if (val < 1 || val > MAX_TCP_KEEPCNT)
2330 err = -EINVAL;
2331 else
2332 tp->keepalive_probes = val;
2333 break;
2334 case TCP_SYNCNT:
2335 if (val < 1 || val > MAX_TCP_SYNCNT)
2336 err = -EINVAL;
2337 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002338 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 break;
2340
2341 case TCP_LINGER2:
2342 if (val < 0)
2343 tp->linger2 = -1;
2344 else if (val > sysctl_tcp_fin_timeout / HZ)
2345 tp->linger2 = 0;
2346 else
2347 tp->linger2 = val * HZ;
2348 break;
2349
2350 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002351 /* Translate value in seconds to number of retransmits */
2352 icsk->icsk_accept_queue.rskq_defer_accept =
2353 secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
2354 TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 break;
2356
2357 case TCP_WINDOW_CLAMP:
2358 if (!val) {
2359 if (sk->sk_state != TCP_CLOSE) {
2360 err = -EINVAL;
2361 break;
2362 }
2363 tp->window_clamp = 0;
2364 } else
2365 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2366 SOCK_MIN_RCVBUF / 2 : val;
2367 break;
2368
2369 case TCP_QUICKACK:
2370 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002371 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002373 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 if ((1 << sk->sk_state) &
2375 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002376 inet_csk_ack_scheduled(sk)) {
2377 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002378 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002380 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 }
2382 }
2383 break;
2384
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002385#ifdef CONFIG_TCP_MD5SIG
2386 case TCP_MD5SIG:
2387 /* Read the IP->Key mappings from userspace */
2388 err = tp->af_specific->md5_parse(sk, optval, optlen);
2389 break;
2390#endif
2391
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 default:
2393 err = -ENOPROTOOPT;
2394 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002395 }
2396
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 release_sock(sk);
2398 return err;
2399}
2400
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002401int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
David S. Millerb7058842009-09-30 16:12:20 -07002402 unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002403{
2404 struct inet_connection_sock *icsk = inet_csk(sk);
2405
2406 if (level != SOL_TCP)
2407 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2408 optval, optlen);
2409 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2410}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002411EXPORT_SYMBOL(tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002412
2413#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002414int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07002415 char __user *optval, unsigned int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002416{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002417 if (level != SOL_TCP)
2418 return inet_csk_compat_setsockopt(sk, level, optname,
2419 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002420 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2421}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002422EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002423#endif
2424
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425/* Return information about state of tcp endpoint in API format. */
2426void tcp_get_info(struct sock *sk, struct tcp_info *info)
2427{
2428 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002429 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 u32 now = tcp_time_stamp;
2431
2432 memset(info, 0, sizeof(*info));
2433
2434 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002435 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002436 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002437 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002438 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
2440 if (tp->rx_opt.tstamp_ok)
2441 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002442 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 info->tcpi_options |= TCPI_OPT_SACK;
2444 if (tp->rx_opt.wscale_ok) {
2445 info->tcpi_options |= TCPI_OPT_WSCALE;
2446 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2447 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450 if (tp->ecn_flags&TCP_ECN_OK)
2451 info->tcpi_options |= TCPI_OPT_ECN;
2452
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002453 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2454 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002455 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002456 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
Rick Jones5ee3afb2007-09-18 13:26:31 -07002458 if (sk->sk_state == TCP_LISTEN) {
2459 info->tcpi_unacked = sk->sk_ack_backlog;
2460 info->tcpi_sacked = sk->sk_max_ack_backlog;
2461 } else {
2462 info->tcpi_unacked = tp->packets_out;
2463 info->tcpi_sacked = tp->sacked_out;
2464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 info->tcpi_lost = tp->lost_out;
2466 info->tcpi_retrans = tp->retrans_out;
2467 info->tcpi_fackets = tp->fackets_out;
2468
2469 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002470 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2472
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002473 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2475 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2476 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2477 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2478 info->tcpi_snd_cwnd = tp->snd_cwnd;
2479 info->tcpi_advmss = tp->advmss;
2480 info->tcpi_reordering = tp->reordering;
2481
2482 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2483 info->tcpi_rcv_space = tp->rcvq_space.space;
2484
2485 info->tcpi_total_retrans = tp->total_retrans;
2486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487EXPORT_SYMBOL_GPL(tcp_get_info);
2488
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002489static int do_tcp_getsockopt(struct sock *sk, int level,
2490 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002492 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 struct tcp_sock *tp = tcp_sk(sk);
2494 int val, len;
2495
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 if (get_user(len, optlen))
2497 return -EFAULT;
2498
2499 len = min_t(unsigned int, len, sizeof(int));
2500
2501 if (len < 0)
2502 return -EINVAL;
2503
2504 switch (optname) {
2505 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002506 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2508 val = tp->rx_opt.user_mss;
2509 break;
2510 case TCP_NODELAY:
2511 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2512 break;
2513 case TCP_CORK:
2514 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2515 break;
2516 case TCP_KEEPIDLE:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002517 val = keepalive_time_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 break;
2519 case TCP_KEEPINTVL:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002520 val = keepalive_intvl_when(tp) / HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 break;
2522 case TCP_KEEPCNT:
Eric Dumazetdf19a622009-08-28 23:48:54 -07002523 val = keepalive_probes(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 break;
2525 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002526 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 break;
2528 case TCP_LINGER2:
2529 val = tp->linger2;
2530 if (val >= 0)
2531 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2532 break;
2533 case TCP_DEFER_ACCEPT:
Julian Anastasovb103cf32009-10-19 10:10:40 +00002534 val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
2535 TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 break;
2537 case TCP_WINDOW_CLAMP:
2538 val = tp->window_clamp;
2539 break;
2540 case TCP_INFO: {
2541 struct tcp_info info;
2542
2543 if (get_user(len, optlen))
2544 return -EFAULT;
2545
2546 tcp_get_info(sk, &info);
2547
2548 len = min_t(unsigned int, len, sizeof(info));
2549 if (put_user(len, optlen))
2550 return -EFAULT;
2551 if (copy_to_user(optval, &info, len))
2552 return -EFAULT;
2553 return 0;
2554 }
2555 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002556 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002558
2559 case TCP_CONGESTION:
2560 if (get_user(len, optlen))
2561 return -EFAULT;
2562 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2563 if (put_user(len, optlen))
2564 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002565 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002566 return -EFAULT;
2567 return 0;
William Allen Simpsone56fb502009-12-02 18:19:30 +00002568
2569 case TCP_COOKIE_TRANSACTIONS: {
2570 struct tcp_cookie_transactions ctd;
2571 struct tcp_cookie_values *cvp = tp->cookie_values;
2572
2573 if (get_user(len, optlen))
2574 return -EFAULT;
2575 if (len < sizeof(ctd))
2576 return -EINVAL;
2577
2578 memset(&ctd, 0, sizeof(ctd));
2579 ctd.tcpct_flags = (tp->rx_opt.cookie_in_always ?
2580 TCP_COOKIE_IN_ALWAYS : 0)
2581 | (tp->rx_opt.cookie_out_never ?
2582 TCP_COOKIE_OUT_NEVER : 0);
2583
2584 if (cvp != NULL) {
2585 ctd.tcpct_flags |= (cvp->s_data_in ?
2586 TCP_S_DATA_IN : 0)
2587 | (cvp->s_data_out ?
2588 TCP_S_DATA_OUT : 0);
2589
2590 ctd.tcpct_cookie_desired = cvp->cookie_desired;
2591 ctd.tcpct_s_data_desired = cvp->s_data_desired;
2592
William Allen Simpsone56fb502009-12-02 18:19:30 +00002593 memcpy(&ctd.tcpct_value[0], &cvp->cookie_pair[0],
2594 cvp->cookie_pair_size);
2595 ctd.tcpct_used = cvp->cookie_pair_size;
2596 }
2597
2598 if (put_user(sizeof(ctd), optlen))
2599 return -EFAULT;
2600 if (copy_to_user(optval, &ctd, sizeof(ctd)))
2601 return -EFAULT;
2602 return 0;
2603 }
Josh Hunt3c0fef02010-07-30 13:49:35 +00002604 case TCP_THIN_LINEAR_TIMEOUTS:
2605 val = tp->thin_lto;
2606 break;
2607 case TCP_THIN_DUPACK:
2608 val = tp->thin_dupack;
2609 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 default:
2611 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
2614 if (put_user(len, optlen))
2615 return -EFAULT;
2616 if (copy_to_user(optval, &val, len))
2617 return -EFAULT;
2618 return 0;
2619}
2620
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002621int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2622 int __user *optlen)
2623{
2624 struct inet_connection_sock *icsk = inet_csk(sk);
2625
2626 if (level != SOL_TCP)
2627 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2628 optval, optlen);
2629 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2630}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002631EXPORT_SYMBOL(tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002632
2633#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002634int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2635 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002636{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002637 if (level != SOL_TCP)
2638 return inet_csk_compat_getsockopt(sk, level, optname,
2639 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002640 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2641}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002642EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002643#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
Herbert Xu576a30e2006-06-27 13:22:38 -07002645struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002646{
2647 struct sk_buff *segs = ERR_PTR(-EINVAL);
2648 struct tcphdr *th;
2649 unsigned thlen;
2650 unsigned int seq;
Al Virod3bc23e2006-11-14 21:24:49 -08002651 __be32 delta;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002652 unsigned int oldlen;
Herbert Xu4e704ee2009-01-14 20:41:12 -08002653 unsigned int mss;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002654
2655 if (!pskb_may_pull(skb, sizeof(*th)))
2656 goto out;
2657
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002658 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002659 thlen = th->doff * 4;
2660 if (thlen < sizeof(*th))
2661 goto out;
2662
2663 if (!pskb_may_pull(skb, thlen))
2664 goto out;
2665
Herbert Xu0718bcc2006-06-25 23:55:46 -07002666 oldlen = (u16)~skb->len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002667 __skb_pull(skb, thlen);
2668
Herbert Xu4e704ee2009-01-14 20:41:12 -08002669 mss = skb_shinfo(skb)->gso_size;
2670 if (unlikely(skb->len <= mss))
2671 goto out;
2672
Herbert Xu3820c3f2006-06-29 20:11:25 -07002673 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2674 /* Packet is from an untrusted source, reset gso_segs. */
Herbert Xubbcf4672006-07-03 19:38:35 -07002675 int type = skb_shinfo(skb)->gso_type;
Herbert Xu3820c3f2006-06-29 20:11:25 -07002676
Herbert Xubbcf4672006-07-03 19:38:35 -07002677 if (unlikely(type &
2678 ~(SKB_GSO_TCPV4 |
2679 SKB_GSO_DODGY |
2680 SKB_GSO_TCP_ECN |
2681 SKB_GSO_TCPV6 |
2682 0) ||
2683 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
2684 goto out;
2685
Ilpo Järvinen172589c2007-08-28 15:50:33 -07002686 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
Herbert Xu3820c3f2006-06-29 20:11:25 -07002687
2688 segs = NULL;
2689 goto out;
2690 }
2691
Herbert Xu576a30e2006-06-27 13:22:38 -07002692 segs = skb_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002693 if (IS_ERR(segs))
2694 goto out;
2695
Herbert Xu4e704ee2009-01-14 20:41:12 -08002696 delta = htonl(oldlen + (thlen + mss));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002697
2698 skb = segs;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002699 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002700 seq = ntohl(th->seq);
2701
2702 do {
2703 th->fin = th->psh = 0;
2704
Al Virod3bc23e2006-11-14 21:24:49 -08002705 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2706 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002707 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002708 th->check =
2709 csum_fold(csum_partial(skb_transport_header(skb),
2710 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002711
Herbert Xu4e704ee2009-01-14 20:41:12 -08002712 seq += mss;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002713 skb = skb->next;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002714 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002715
2716 th->seq = htonl(seq);
2717 th->cwr = 0;
2718 } while (skb->next);
2719
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002720 delta = htonl(oldlen + (skb->tail - skb->transport_header) +
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002721 skb->data_len);
Al Virod3bc23e2006-11-14 21:24:49 -08002722 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2723 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002724 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002725 th->check = csum_fold(csum_partial(skb_transport_header(skb),
2726 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002727
2728out:
2729 return segs;
2730}
Herbert Xuadcfc7d2006-06-30 13:36:15 -07002731EXPORT_SYMBOL(tcp_tso_segment);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002732
Herbert Xubf296b12008-12-15 23:43:36 -08002733struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2734{
2735 struct sk_buff **pp = NULL;
2736 struct sk_buff *p;
2737 struct tcphdr *th;
2738 struct tcphdr *th2;
Herbert Xua0a69a02009-04-17 02:34:38 -07002739 unsigned int len;
Herbert Xubf296b12008-12-15 23:43:36 -08002740 unsigned int thlen;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07002741 __be32 flags;
Herbert Xubf296b12008-12-15 23:43:36 -08002742 unsigned int mss = 1;
Herbert Xua5b1cf22009-05-26 18:50:28 +00002743 unsigned int hlen;
2744 unsigned int off;
Herbert Xubf296b12008-12-15 23:43:36 -08002745 int flush = 1;
Herbert Xuaa6320d2009-02-08 18:00:40 +00002746 int i;
Herbert Xubf296b12008-12-15 23:43:36 -08002747
Herbert Xua5b1cf22009-05-26 18:50:28 +00002748 off = skb_gro_offset(skb);
2749 hlen = off + sizeof(*th);
2750 th = skb_gro_header_fast(skb, off);
2751 if (skb_gro_header_hard(skb, hlen)) {
2752 th = skb_gro_header_slow(skb, hlen, off);
2753 if (unlikely(!th))
2754 goto out;
2755 }
Herbert Xubf296b12008-12-15 23:43:36 -08002756
Herbert Xubf296b12008-12-15 23:43:36 -08002757 thlen = th->doff * 4;
2758 if (thlen < sizeof(*th))
2759 goto out;
2760
Herbert Xua5b1cf22009-05-26 18:50:28 +00002761 hlen = off + thlen;
2762 if (skb_gro_header_hard(skb, hlen)) {
2763 th = skb_gro_header_slow(skb, hlen, off);
2764 if (unlikely(!th))
2765 goto out;
2766 }
Herbert Xubf296b12008-12-15 23:43:36 -08002767
Herbert Xu86911732009-01-29 14:19:50 +00002768 skb_gro_pull(skb, thlen);
Herbert Xubf296b12008-12-15 23:43:36 -08002769
Herbert Xua0a69a02009-04-17 02:34:38 -07002770 len = skb_gro_len(skb);
Herbert Xubf296b12008-12-15 23:43:36 -08002771 flags = tcp_flag_word(th);
2772
2773 for (; (p = *head); head = &p->next) {
2774 if (!NAPI_GRO_CB(p)->same_flow)
2775 continue;
2776
2777 th2 = tcp_hdr(p);
2778
Herbert Xu745898e2009-05-26 18:50:24 +00002779 if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
Herbert Xubf296b12008-12-15 23:43:36 -08002780 NAPI_GRO_CB(p)->same_flow = 0;
2781 continue;
2782 }
2783
2784 goto found;
2785 }
2786
2787 goto out_check_final;
2788
2789found:
2790 flush = NAPI_GRO_CB(p)->flush;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07002791 flush |= (__force int)(flags & TCP_FLAG_CWR);
2792 flush |= (__force int)((flags ^ tcp_flag_word(th2)) &
2793 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
2794 flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
Herbert Xua2a804c2009-05-26 18:50:34 +00002795 for (i = sizeof(*th); i < thlen; i += 4)
Herbert Xuaa6320d2009-02-08 18:00:40 +00002796 flush |= *(u32 *)((u8 *)th + i) ^
2797 *(u32 *)((u8 *)th2 + i);
Herbert Xubf296b12008-12-15 23:43:36 -08002798
Herbert Xub5302562009-01-04 16:13:19 -08002799 mss = skb_shinfo(p)->gso_size;
Herbert Xubf296b12008-12-15 23:43:36 -08002800
Herbert Xu30a3ae32009-05-26 18:50:26 +00002801 flush |= (len - 1) >= mss;
Herbert Xuaa6320d2009-02-08 18:00:40 +00002802 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
Herbert Xubf296b12008-12-15 23:43:36 -08002803
2804 if (flush || skb_gro_receive(head, skb)) {
2805 mss = 1;
2806 goto out_check_final;
2807 }
2808
2809 p = *head;
2810 th2 = tcp_hdr(p);
2811 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
2812
2813out_check_final:
Herbert Xua0a69a02009-04-17 02:34:38 -07002814 flush = len < mss;
Eric Dumazet0eae88f2010-04-20 19:06:52 -07002815 flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH |
2816 TCP_FLAG_RST | TCP_FLAG_SYN |
2817 TCP_FLAG_FIN));
Herbert Xubf296b12008-12-15 23:43:36 -08002818
2819 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
2820 pp = head;
2821
2822out:
2823 NAPI_GRO_CB(skb)->flush |= flush;
2824
2825 return pp;
2826}
Herbert Xu684f2172009-01-08 10:41:23 -08002827EXPORT_SYMBOL(tcp_gro_receive);
Herbert Xubf296b12008-12-15 23:43:36 -08002828
2829int tcp_gro_complete(struct sk_buff *skb)
2830{
2831 struct tcphdr *th = tcp_hdr(skb);
2832
2833 skb->csum_start = skb_transport_header(skb) - skb->head;
2834 skb->csum_offset = offsetof(struct tcphdr, check);
2835 skb->ip_summed = CHECKSUM_PARTIAL;
2836
Herbert Xubf296b12008-12-15 23:43:36 -08002837 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2838
2839 if (th->cwr)
2840 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2841
2842 return 0;
2843}
Herbert Xu684f2172009-01-08 10:41:23 -08002844EXPORT_SYMBOL(tcp_gro_complete);
Herbert Xubf296b12008-12-15 23:43:36 -08002845
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002846#ifdef CONFIG_TCP_MD5SIG
2847static unsigned long tcp_md5sig_users;
Tejun Heo7d720c32010-02-16 15:20:26 +00002848static struct tcp_md5sig_pool * __percpu *tcp_md5sig_pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002849static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
2850
Tejun Heo7d720c32010-02-16 15:20:26 +00002851static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool * __percpu *pool)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002852{
2853 int cpu;
2854 for_each_possible_cpu(cpu) {
2855 struct tcp_md5sig_pool *p = *per_cpu_ptr(pool, cpu);
2856 if (p) {
2857 if (p->md5_desc.tfm)
2858 crypto_free_hash(p->md5_desc.tfm);
2859 kfree(p);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002860 }
2861 }
2862 free_percpu(pool);
2863}
2864
2865void tcp_free_md5sig_pool(void)
2866{
Tejun Heo7d720c32010-02-16 15:20:26 +00002867 struct tcp_md5sig_pool * __percpu *pool = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002868
David S. Miller2c4f6212007-02-20 23:51:47 -08002869 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002870 if (--tcp_md5sig_users == 0) {
2871 pool = tcp_md5sig_pool;
2872 tcp_md5sig_pool = NULL;
2873 }
David S. Miller2c4f6212007-02-20 23:51:47 -08002874 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002875 if (pool)
2876 __tcp_free_md5sig_pool(pool);
2877}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002878EXPORT_SYMBOL(tcp_free_md5sig_pool);
2879
Tejun Heo7d720c32010-02-16 15:20:26 +00002880static struct tcp_md5sig_pool * __percpu *
2881__tcp_alloc_md5sig_pool(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002882{
2883 int cpu;
Tejun Heo7d720c32010-02-16 15:20:26 +00002884 struct tcp_md5sig_pool * __percpu *pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002885
2886 pool = alloc_percpu(struct tcp_md5sig_pool *);
2887 if (!pool)
2888 return NULL;
2889
2890 for_each_possible_cpu(cpu) {
2891 struct tcp_md5sig_pool *p;
2892 struct crypto_hash *hash;
2893
Wu Fengguangaa133072009-09-02 23:45:45 -07002894 p = kzalloc(sizeof(*p), sk->sk_allocation);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002895 if (!p)
2896 goto out_free;
2897 *per_cpu_ptr(pool, cpu) = p;
2898
2899 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
2900 if (!hash || IS_ERR(hash))
2901 goto out_free;
2902
2903 p->md5_desc.tfm = hash;
2904 }
2905 return pool;
2906out_free:
2907 __tcp_free_md5sig_pool(pool);
2908 return NULL;
2909}
2910
Tejun Heo7d720c32010-02-16 15:20:26 +00002911struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002912{
Tejun Heo7d720c32010-02-16 15:20:26 +00002913 struct tcp_md5sig_pool * __percpu *pool;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002914 int alloc = 0;
2915
2916retry:
David S. Miller2c4f6212007-02-20 23:51:47 -08002917 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002918 pool = tcp_md5sig_pool;
2919 if (tcp_md5sig_users++ == 0) {
2920 alloc = 1;
David S. Miller2c4f6212007-02-20 23:51:47 -08002921 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002922 } else if (!pool) {
2923 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002924 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002925 cpu_relax();
2926 goto retry;
2927 } else
David S. Miller2c4f6212007-02-20 23:51:47 -08002928 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002929
2930 if (alloc) {
2931 /* we cannot hold spinlock here because this may sleep. */
Tejun Heo7d720c32010-02-16 15:20:26 +00002932 struct tcp_md5sig_pool * __percpu *p;
2933
2934 p = __tcp_alloc_md5sig_pool(sk);
David S. Miller2c4f6212007-02-20 23:51:47 -08002935 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002936 if (!p) {
2937 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002938 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002939 return NULL;
2940 }
2941 pool = tcp_md5sig_pool;
2942 if (pool) {
2943 /* oops, it has already been assigned. */
David S. Miller2c4f6212007-02-20 23:51:47 -08002944 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002945 __tcp_free_md5sig_pool(p);
2946 } else {
2947 tcp_md5sig_pool = pool = p;
David S. Miller2c4f6212007-02-20 23:51:47 -08002948 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002949 }
2950 }
2951 return pool;
2952}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002953EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2954
Eric Dumazet35790c02010-05-16 00:34:04 -07002955
2956/**
2957 * tcp_get_md5sig_pool - get md5sig_pool for this user
2958 *
2959 * We use percpu structure, so if we succeed, we exit with preemption
2960 * and BH disabled, to make sure another thread or softirq handling
2961 * wont try to get same context.
2962 */
2963struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002964{
Tejun Heo7d720c32010-02-16 15:20:26 +00002965 struct tcp_md5sig_pool * __percpu *p;
Eric Dumazet35790c02010-05-16 00:34:04 -07002966
2967 local_bh_disable();
2968
2969 spin_lock(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002970 p = tcp_md5sig_pool;
2971 if (p)
2972 tcp_md5sig_users++;
Eric Dumazet35790c02010-05-16 00:34:04 -07002973 spin_unlock(&tcp_md5sig_pool_lock);
2974
2975 if (p)
Eric Dumazet7a9b2d52010-06-24 00:52:37 +00002976 return *this_cpu_ptr(p);
Eric Dumazet35790c02010-05-16 00:34:04 -07002977
2978 local_bh_enable();
2979 return NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002980}
Eric Dumazet35790c02010-05-16 00:34:04 -07002981EXPORT_SYMBOL(tcp_get_md5sig_pool);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002982
Eric Dumazet35790c02010-05-16 00:34:04 -07002983void tcp_put_md5sig_pool(void)
David S. Miller6931ba72006-12-13 16:25:44 -08002984{
Eric Dumazet35790c02010-05-16 00:34:04 -07002985 local_bh_enable();
David S. Miller6931ba72006-12-13 16:25:44 -08002986 tcp_free_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002987}
Eric Dumazet35790c02010-05-16 00:34:04 -07002988EXPORT_SYMBOL(tcp_put_md5sig_pool);
Adam Langley49a72df2008-07-19 00:01:42 -07002989
2990int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
2991 struct tcphdr *th)
2992{
2993 struct scatterlist sg;
2994 int err;
2995
2996 __sum16 old_checksum = th->check;
2997 th->check = 0;
2998 /* options aren't included in the hash */
2999 sg_init_one(&sg, th, sizeof(struct tcphdr));
3000 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(struct tcphdr));
3001 th->check = old_checksum;
3002 return err;
3003}
Adam Langley49a72df2008-07-19 00:01:42 -07003004EXPORT_SYMBOL(tcp_md5_hash_header);
3005
3006int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
3007 struct sk_buff *skb, unsigned header_len)
3008{
3009 struct scatterlist sg;
3010 const struct tcphdr *tp = tcp_hdr(skb);
3011 struct hash_desc *desc = &hp->md5_desc;
3012 unsigned i;
3013 const unsigned head_data_len = skb_headlen(skb) > header_len ?
3014 skb_headlen(skb) - header_len : 0;
3015 const struct skb_shared_info *shi = skb_shinfo(skb);
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003016 struct sk_buff *frag_iter;
Adam Langley49a72df2008-07-19 00:01:42 -07003017
3018 sg_init_table(&sg, 1);
3019
3020 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
3021 if (crypto_hash_update(desc, &sg, head_data_len))
3022 return 1;
3023
3024 for (i = 0; i < shi->nr_frags; ++i) {
3025 const struct skb_frag_struct *f = &shi->frags[i];
3026 sg_set_page(&sg, f->page, f->size, f->page_offset);
3027 if (crypto_hash_update(desc, &sg, f->size))
3028 return 1;
3029 }
3030
Eric Dumazetd7fd1b572010-05-17 20:40:51 +00003031 skb_walk_frags(skb, frag_iter)
3032 if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
3033 return 1;
3034
Adam Langley49a72df2008-07-19 00:01:42 -07003035 return 0;
3036}
Adam Langley49a72df2008-07-19 00:01:42 -07003037EXPORT_SYMBOL(tcp_md5_hash_skb_data);
3038
3039int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, struct tcp_md5sig_key *key)
3040{
3041 struct scatterlist sg;
3042
3043 sg_init_one(&sg, key->key, key->keylen);
3044 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
3045}
Adam Langley49a72df2008-07-19 00:01:42 -07003046EXPORT_SYMBOL(tcp_md5_hash_key);
3047
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08003048#endif
3049
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003050/**
3051 * Each Responder maintains up to two secret values concurrently for
3052 * efficient secret rollover. Each secret value has 4 states:
3053 *
3054 * Generating. (tcp_secret_generating != tcp_secret_primary)
3055 * Generates new Responder-Cookies, but not yet used for primary
3056 * verification. This is a short-term state, typically lasting only
3057 * one round trip time (RTT).
3058 *
3059 * Primary. (tcp_secret_generating == tcp_secret_primary)
3060 * Used both for generation and primary verification.
3061 *
3062 * Retiring. (tcp_secret_retiring != tcp_secret_secondary)
3063 * Used for verification, until the first failure that can be
3064 * verified by the newer Generating secret. At that time, this
3065 * cookie's state is changed to Secondary, and the Generating
3066 * cookie's state is changed to Primary. This is a short-term state,
3067 * typically lasting only one round trip time (RTT).
3068 *
3069 * Secondary. (tcp_secret_retiring == tcp_secret_secondary)
3070 * Used for secondary verification, after primary verification
3071 * failures. This state lasts no more than twice the Maximum Segment
3072 * Lifetime (2MSL). Then, the secret is discarded.
3073 */
3074struct tcp_cookie_secret {
3075 /* The secret is divided into two parts. The digest part is the
3076 * equivalent of previously hashing a secret and saving the state,
3077 * and serves as an initialization vector (IV). The message part
3078 * serves as the trailing secret.
3079 */
3080 u32 secrets[COOKIE_WORKSPACE_WORDS];
3081 unsigned long expires;
3082};
3083
3084#define TCP_SECRET_1MSL (HZ * TCP_PAWS_MSL)
3085#define TCP_SECRET_2MSL (HZ * TCP_PAWS_MSL * 2)
3086#define TCP_SECRET_LIFE (HZ * 600)
3087
3088static struct tcp_cookie_secret tcp_secret_one;
3089static struct tcp_cookie_secret tcp_secret_two;
3090
3091/* Essentially a circular list, without dynamic allocation. */
3092static struct tcp_cookie_secret *tcp_secret_generating;
3093static struct tcp_cookie_secret *tcp_secret_primary;
3094static struct tcp_cookie_secret *tcp_secret_retiring;
3095static struct tcp_cookie_secret *tcp_secret_secondary;
3096
3097static DEFINE_SPINLOCK(tcp_secret_locker);
3098
3099/* Select a pseudo-random word in the cookie workspace.
3100 */
3101static inline u32 tcp_cookie_work(const u32 *ws, const int n)
3102{
3103 return ws[COOKIE_DIGEST_WORDS + ((COOKIE_MESSAGE_WORDS-1) & ws[n])];
3104}
3105
3106/* Fill bakery[COOKIE_WORKSPACE_WORDS] with generator, updating as needed.
3107 * Called in softirq context.
3108 * Returns: 0 for success.
3109 */
3110int tcp_cookie_generator(u32 *bakery)
3111{
3112 unsigned long jiffy = jiffies;
3113
3114 if (unlikely(time_after_eq(jiffy, tcp_secret_generating->expires))) {
3115 spin_lock_bh(&tcp_secret_locker);
3116 if (!time_after_eq(jiffy, tcp_secret_generating->expires)) {
3117 /* refreshed by another */
3118 memcpy(bakery,
3119 &tcp_secret_generating->secrets[0],
3120 COOKIE_WORKSPACE_WORDS);
3121 } else {
3122 /* still needs refreshing */
3123 get_random_bytes(bakery, COOKIE_WORKSPACE_WORDS);
3124
3125 /* The first time, paranoia assumes that the
3126 * randomization function isn't as strong. But,
3127 * this secret initialization is delayed until
3128 * the last possible moment (packet arrival).
3129 * Although that time is observable, it is
3130 * unpredictably variable. Mash in the most
3131 * volatile clock bits available, and expire the
3132 * secret extra quickly.
3133 */
3134 if (unlikely(tcp_secret_primary->expires ==
3135 tcp_secret_secondary->expires)) {
3136 struct timespec tv;
3137
3138 getnstimeofday(&tv);
3139 bakery[COOKIE_DIGEST_WORDS+0] ^=
3140 (u32)tv.tv_nsec;
3141
3142 tcp_secret_secondary->expires = jiffy
3143 + TCP_SECRET_1MSL
3144 + (0x0f & tcp_cookie_work(bakery, 0));
3145 } else {
3146 tcp_secret_secondary->expires = jiffy
3147 + TCP_SECRET_LIFE
3148 + (0xff & tcp_cookie_work(bakery, 1));
3149 tcp_secret_primary->expires = jiffy
3150 + TCP_SECRET_2MSL
3151 + (0x1f & tcp_cookie_work(bakery, 2));
3152 }
3153 memcpy(&tcp_secret_secondary->secrets[0],
3154 bakery, COOKIE_WORKSPACE_WORDS);
3155
3156 rcu_assign_pointer(tcp_secret_generating,
3157 tcp_secret_secondary);
3158 rcu_assign_pointer(tcp_secret_retiring,
3159 tcp_secret_primary);
3160 /*
3161 * Neither call_rcu() nor synchronize_rcu() needed.
3162 * Retiring data is not freed. It is replaced after
3163 * further (locked) pointer updates, and a quiet time
3164 * (minimum 1MSL, maximum LIFE - 2MSL).
3165 */
3166 }
3167 spin_unlock_bh(&tcp_secret_locker);
3168 } else {
3169 rcu_read_lock_bh();
3170 memcpy(bakery,
3171 &rcu_dereference(tcp_secret_generating)->secrets[0],
3172 COOKIE_WORKSPACE_WORDS);
3173 rcu_read_unlock_bh();
3174 }
3175 return 0;
3176}
3177EXPORT_SYMBOL(tcp_cookie_generator);
3178
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003179void tcp_done(struct sock *sk)
3180{
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08003181 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07003182 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07003183
3184 tcp_set_state(sk, TCP_CLOSE);
3185 tcp_clear_xmit_timers(sk);
3186
3187 sk->sk_shutdown = SHUTDOWN_MASK;
3188
3189 if (!sock_flag(sk, SOCK_DEAD))
3190 sk->sk_state_change(sk);
3191 else
3192 inet_csk_destroy_sock(sk);
3193}
3194EXPORT_SYMBOL_GPL(tcp_done);
3195
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07003196extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197
3198static __initdata unsigned long thash_entries;
3199static int __init set_thash_entries(char *str)
3200{
3201 if (!str)
3202 return 0;
3203 thash_entries = simple_strtoul(str, &str, 0);
3204 return 1;
3205}
3206__setup("thash_entries=", set_thash_entries);
3207
3208void __init tcp_init(void)
3209{
3210 struct sk_buff *skb = NULL;
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07003211 unsigned long nr_pages, limit;
John Heffner7b4f4b52006-03-25 01:34:07 -08003212 int order, i, max_share;
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003213 unsigned long jiffy = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08003215 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Eric Dumazet17483762008-11-25 21:16:35 -08003217 percpu_counter_init(&tcp_sockets_allocated, 0);
Eric Dumazetdd24c002008-11-25 21:17:14 -08003218 percpu_counter_init(&tcp_orphan_count, 0);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003219 tcp_hashinfo.bind_bucket_cachep =
3220 kmem_cache_create("tcp_bind_bucket",
3221 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003222 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 /* Size and allocate the main established and bind bucket
3225 * hash tables.
3226 *
3227 * The methodology is similar to that of the buffer cache.
3228 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003229 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003231 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 thash_entries,
Jan Beulich44813742009-09-21 17:03:05 -07003233 (totalram_pages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08003234 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08003235 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 NULL,
Eric Dumazetf373b532009-10-09 00:16:19 +00003237 &tcp_hashinfo.ehash_mask,
Jean Delvare0ccfe612007-10-30 00:59:25 -07003238 thash_entries ? 0 : 512 * 1024);
Eric Dumazetf373b532009-10-09 00:16:19 +00003239 for (i = 0; i <= tcp_hashinfo.ehash_mask; i++) {
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08003240 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
3241 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 }
Eric Dumazet230140c2007-11-07 02:40:20 -08003243 if (inet_ehash_locks_alloc(&tcp_hashinfo))
3244 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003245 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07003247 sizeof(struct inet_bind_hashbucket),
Eric Dumazetf373b532009-10-09 00:16:19 +00003248 tcp_hashinfo.ehash_mask + 1,
Jan Beulich44813742009-09-21 17:03:05 -07003249 (totalram_pages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08003250 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08003251 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003252 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 NULL,
3254 64 * 1024);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003255 tcp_hashinfo.bhash_size = 1 << tcp_hashinfo.bhash_size;
3256 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
3257 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
3258 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 }
3260
3261 /* Try to be a bit smarter and adjust defaults depending
3262 * on available memory.
3263 */
3264 for (order = 0; ((1 << order) << PAGE_SHIFT) <
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07003265 (tcp_hashinfo.bhash_size * sizeof(struct inet_bind_hashbucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 order++)
3267 ;
Andi Kleene7626482005-06-13 14:24:52 -07003268 if (order >= 4) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07003269 tcp_death_row.sysctl_max_tw_buckets = 180000;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 sysctl_tcp_max_orphans = 4096 << (order - 4);
3271 sysctl_max_syn_backlog = 1024;
3272 } else if (order < 3) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07003273 tcp_death_row.sysctl_max_tw_buckets >>= (3 - order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 sysctl_tcp_max_orphans >>= (3 - order);
3275 sysctl_max_syn_backlog = 128;
3276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
John Heffner53cdcc02007-03-16 15:04:03 -07003278 /* Set the pressure threshold to be a fraction of global memory that
3279 * is up to 1/2 at 256 MB, decreasing toward zero with the amount of
3280 * memory, with a floor of 128 pages.
3281 */
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07003282 nr_pages = totalram_pages - totalhigh_pages;
3283 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
3284 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
John Heffner53cdcc02007-03-16 15:04:03 -07003285 limit = max(limit, 128UL);
3286 sysctl_tcp_mem[0] = limit / 4 * 3;
3287 sysctl_tcp_mem[1] = limit;
John Heffner52bf3762006-11-14 20:25:17 -08003288 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
John Heffner53cdcc02007-03-16 15:04:03 -07003290 /* Set per-socket limits to no more than 1/128 the pressure threshold */
John Heffner7b4f4b52006-03-25 01:34:07 -08003291 limit = ((unsigned long)sysctl_tcp_mem[1]) << (PAGE_SHIFT - 7);
3292 max_share = min(4UL*1024*1024, limit);
3293
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003294 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003295 sysctl_tcp_wmem[1] = 16*1024;
3296 sysctl_tcp_wmem[2] = max(64*1024, max_share);
3297
Hideo Aoki3ab224b2007-12-31 00:11:19 -08003298 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08003299 sysctl_tcp_rmem[1] = 87380;
3300 sysctl_tcp_rmem[2] = max(87380, max_share);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301
3302 printk(KERN_INFO "TCP: Hash tables configured "
Eric Dumazetf373b532009-10-09 00:16:19 +00003303 "(established %u bind %u)\n",
3304 tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07003305
3306 tcp_register_congestion_control(&tcp_reno);
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00003307
3308 memset(&tcp_secret_one.secrets[0], 0, sizeof(tcp_secret_one.secrets));
3309 memset(&tcp_secret_two.secrets[0], 0, sizeof(tcp_secret_two.secrets));
3310 tcp_secret_one.expires = jiffy; /* past due */
3311 tcp_secret_two.expires = jiffy; /* past due */
3312 tcp_secret_generating = &tcp_secret_one;
3313 tcp_secret_primary = &tcp_secret_one;
3314 tcp_secret_retiring = &tcp_secret_two;
3315 tcp_secret_secondary = &tcp_secret_two;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316}