blob: 1d7f49c6f0ca0a5821ad9fc0b04fbce85ebf91c5 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268#include <net/icmp.h>
269#include <net/tcp.h>
270#include <net/xfrm.h>
271#include <net/ip.h>
Chris Leech1a2449a2006-05-23 18:05:53 -0700272#include <net/netdma.h>
Jens Axboe9c55e012007-11-06 23:30:13 -0800273#include <net/sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275#include <asm/uaccess.h>
276#include <asm/ioctls.h>
277
Brian Haleyab32ea52006-09-22 14:15:41 -0700278int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Eric Dumazetdd24c002008-11-25 21:17:14 -0800280struct percpu_counter tcp_orphan_count;
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700281EXPORT_SYMBOL_GPL(tcp_orphan_count);
282
David S. Millerb8059ea2006-03-25 01:36:56 -0800283int sysctl_tcp_mem[3] __read_mostly;
284int sysctl_tcp_wmem[3] __read_mostly;
285int sysctl_tcp_rmem[3] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287EXPORT_SYMBOL(sysctl_tcp_mem);
288EXPORT_SYMBOL(sysctl_tcp_rmem);
289EXPORT_SYMBOL(sysctl_tcp_wmem);
290
291atomic_t tcp_memory_allocated; /* Current allocated memory. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292EXPORT_SYMBOL(tcp_memory_allocated);
Eric Dumazet17483762008-11-25 21:16:35 -0800293
294/*
295 * Current number of TCP sockets.
296 */
297struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298EXPORT_SYMBOL(tcp_sockets_allocated);
299
300/*
Jens Axboe9c55e012007-11-06 23:30:13 -0800301 * TCP splice context
302 */
303struct tcp_splice_state {
304 struct pipe_inode_info *pipe;
305 size_t len;
306 unsigned int flags;
307};
308
309/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * Pressure flag: try to collapse.
311 * Technical note: it is used by multiple contexts non atomically.
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800312 * All the __sk_mem_schedule() is of this nature: accounting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 * is strict, actions are advisory and have some latency.
314 */
Eric Dumazet4103f8c2007-03-27 13:58:31 -0700315int tcp_memory_pressure __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317EXPORT_SYMBOL(tcp_memory_pressure);
318
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700319void tcp_enter_memory_pressure(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 if (!tcp_memory_pressure) {
Pavel Emelyanov4e673442008-07-16 20:30:14 -0700322 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 tcp_memory_pressure = 1;
324 }
325}
326
327EXPORT_SYMBOL(tcp_enter_memory_pressure);
328
329/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * Wait for a TCP event.
331 *
332 * Note that we don't need to lock the socket, as the upper poll layers
333 * take care of normal races (between the test and the event) and we don't
334 * go look at any of the socket buffers directly.
335 */
336unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
337{
338 unsigned int mask;
339 struct sock *sk = sock->sk;
340 struct tcp_sock *tp = tcp_sk(sk);
341
342 poll_wait(file, sk->sk_sleep, wait);
343 if (sk->sk_state == TCP_LISTEN)
Arnaldo Carvalho de Melodc40c7b2005-08-23 21:52:58 -0700344 return inet_csk_listen_poll(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 /* Socket is not locked. We are protected from async events
Will Newton70efce22008-07-16 20:13:43 -0700347 * by poll logic and correct handling of state changes
348 * made by other threads is impossible in any case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 */
350
351 mask = 0;
352 if (sk->sk_err)
353 mask = POLLERR;
354
355 /*
356 * POLLHUP is certainly not done right. But poll() doesn't
357 * have a notion of HUP in just one direction, and for a
358 * socket the read side is more interesting.
359 *
360 * Some poll() documentation says that POLLHUP is incompatible
361 * with the POLLOUT/POLLWR flags, so somebody should check this
362 * all. But careful, it tends to be safer to return too many
363 * bits than too few, and you can easily break real applications
364 * if you don't tell them that something has hung up!
365 *
366 * Check-me.
367 *
368 * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
369 * our fs/select.c). It means that after we received EOF,
370 * poll always returns immediately, making impossible poll() on write()
371 * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
372 * if and only if shutdown has been made in both directions.
373 * Actually, it is interesting to look how Solaris and DUX
Will Newton70efce22008-07-16 20:13:43 -0700374 * solve this dilemma. I would prefer, if POLLHUP were maskable,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * then we could set it on SND_SHUTDOWN. BTW examples given
376 * in Stevens' books assume exactly this behaviour, it explains
Will Newton70efce22008-07-16 20:13:43 -0700377 * why POLLHUP is incompatible with POLLOUT. --ANK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 *
379 * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
380 * blocking on fresh not-connected or disconnected socket. --ANK
381 */
382 if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
383 mask |= POLLHUP;
384 if (sk->sk_shutdown & RCV_SHUTDOWN)
Davide Libenzif348d702006-03-25 03:07:39 -0800385 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 /* Connected? */
388 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
David S. Millerc7004482008-10-06 10:43:54 -0700389 int target = sock_rcvlowat(sk, 0, INT_MAX);
390
391 if (tp->urg_seq == tp->copied_seq &&
392 !sock_flag(sk, SOCK_URGINLINE) &&
393 tp->urg_data)
394 target--;
395
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 /* Potential race condition. If read of tp below will
397 * escape above sk->sk_state, we can be illegally awaken
398 * in SYN_* states. */
David S. Millerc7004482008-10-06 10:43:54 -0700399 if (tp->rcv_nxt - tp->copied_seq >= target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 mask |= POLLIN | POLLRDNORM;
401
402 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
403 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
404 mask |= POLLOUT | POLLWRNORM;
405 } else { /* send SIGIO later */
406 set_bit(SOCK_ASYNC_NOSPACE,
407 &sk->sk_socket->flags);
408 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
409
410 /* Race breaker. If space is freed after
411 * wspace test but before the flags are set,
412 * IO signal will be lost.
413 */
414 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
415 mask |= POLLOUT | POLLWRNORM;
416 }
417 }
418
419 if (tp->urg_data & TCP_URG_VALID)
420 mask |= POLLPRI;
421 }
422 return mask;
423}
424
425int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
426{
427 struct tcp_sock *tp = tcp_sk(sk);
428 int answ;
429
430 switch (cmd) {
431 case SIOCINQ:
432 if (sk->sk_state == TCP_LISTEN)
433 return -EINVAL;
434
435 lock_sock(sk);
436 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
437 answ = 0;
438 else if (sock_flag(sk, SOCK_URGINLINE) ||
439 !tp->urg_data ||
440 before(tp->urg_seq, tp->copied_seq) ||
441 !before(tp->urg_seq, tp->rcv_nxt)) {
442 answ = tp->rcv_nxt - tp->copied_seq;
443
444 /* Subtract 1, if FIN is in queue. */
445 if (answ && !skb_queue_empty(&sk->sk_receive_queue))
446 answ -=
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700447 tcp_hdr((struct sk_buff *)sk->sk_receive_queue.prev)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 } else
449 answ = tp->urg_seq - tp->copied_seq;
450 release_sock(sk);
451 break;
452 case SIOCATMARK:
453 answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
454 break;
455 case SIOCOUTQ:
456 if (sk->sk_state == TCP_LISTEN)
457 return -EINVAL;
458
459 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
460 answ = 0;
461 else
462 answ = tp->write_seq - tp->snd_una;
463 break;
464 default:
465 return -ENOIOCTLCMD;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 return put_user(answ, (int __user *)arg);
469}
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
472{
473 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_PSH;
474 tp->pushed_seq = tp->write_seq;
475}
476
477static inline int forced_push(struct tcp_sock *tp)
478{
479 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
480}
481
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700482static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700484 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo352d4802006-11-17 19:59:12 -0200485 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
486
487 skb->csum = 0;
488 tcb->seq = tcb->end_seq = tp->write_seq;
489 tcb->flags = TCPCB_FLAG_ACK;
490 tcb->sacked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 skb_header_release(skb);
David S. Millerfe067e82007-03-07 12:12:44 -0800492 tcp_add_write_queue_tail(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800493 sk->sk_wmem_queued += skb->truesize;
494 sk_mem_charge(sk, skb->truesize);
David S. Miller89ebd192005-08-23 10:13:06 -0700495 if (tp->nonagle & TCP_NAGLE_PUSH)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900496 tp->nonagle &= ~TCP_NAGLE_PUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
499static inline void tcp_mark_urg(struct tcp_sock *tp, int flags,
500 struct sk_buff *skb)
501{
Ilpo Järvinen33f5f572008-10-07 14:43:06 -0700502 if (flags & MSG_OOB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 tp->snd_up = tp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
505
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700506static inline void tcp_push(struct sock *sk, int flags, int mss_now,
507 int nonagle)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700509 struct tcp_sock *tp = tcp_sk(sk);
510
David S. Millerfe067e82007-03-07 12:12:44 -0800511 if (tcp_send_head(sk)) {
512 struct sk_buff *skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 if (!(flags & MSG_MORE) || forced_push(tp))
514 tcp_mark_push(tp, skb);
515 tcp_mark_urg(tp, flags, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700516 __tcp_push_pending_frames(sk, mss_now,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
518 }
519}
520
Adrian Bunk6ff77512007-11-06 23:32:26 -0800521static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
522 unsigned int offset, size_t len)
Jens Axboe9c55e012007-11-06 23:30:13 -0800523{
524 struct tcp_splice_state *tss = rd_desc->arg.data;
Willy Tarreau33966dd2009-01-13 16:04:36 -0800525 int ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800526
Dimitris Michailidis9fa5fdf2009-01-26 22:15:31 -0800527 ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
528 tss->flags);
Willy Tarreau33966dd2009-01-13 16:04:36 -0800529 if (ret > 0)
530 rd_desc->count -= ret;
531 return ret;
Jens Axboe9c55e012007-11-06 23:30:13 -0800532}
533
534static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
535{
536 /* Store TCP splice context information in read_descriptor_t. */
537 read_descriptor_t rd_desc = {
538 .arg.data = tss,
Willy Tarreau33966dd2009-01-13 16:04:36 -0800539 .count = tss->len,
Jens Axboe9c55e012007-11-06 23:30:13 -0800540 };
541
542 return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
543}
544
545/**
546 * tcp_splice_read - splice data from TCP socket to a pipe
547 * @sock: socket to splice from
548 * @ppos: position (not valid)
549 * @pipe: pipe to splice to
550 * @len: number of bytes to splice
551 * @flags: splice modifier flags
552 *
553 * Description:
554 * Will read pages from given socket and fill them into a pipe.
555 *
556 **/
557ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
558 struct pipe_inode_info *pipe, size_t len,
559 unsigned int flags)
560{
561 struct sock *sk = sock->sk;
562 struct tcp_splice_state tss = {
563 .pipe = pipe,
564 .len = len,
565 .flags = flags,
566 };
567 long timeo;
568 ssize_t spliced;
569 int ret;
570
571 /*
572 * We can't seek on a socket input
573 */
574 if (unlikely(*ppos))
575 return -ESPIPE;
576
577 ret = spliced = 0;
578
579 lock_sock(sk);
580
581 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
582 while (tss.len) {
583 ret = __tcp_splice_read(sk, &tss);
584 if (ret < 0)
585 break;
586 else if (!ret) {
587 if (spliced)
588 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800589 if (sock_flag(sk, SOCK_DONE))
590 break;
591 if (sk->sk_err) {
592 ret = sock_error(sk);
593 break;
594 }
595 if (sk->sk_shutdown & RCV_SHUTDOWN)
596 break;
597 if (sk->sk_state == TCP_CLOSE) {
598 /*
599 * This occurs when user tries to read
600 * from never connected socket.
601 */
602 if (!sock_flag(sk, SOCK_DONE))
603 ret = -ENOTCONN;
604 break;
605 }
606 if (!timeo) {
607 ret = -EAGAIN;
608 break;
609 }
610 sk_wait_data(sk, &timeo);
611 if (signal_pending(current)) {
612 ret = sock_intr_errno(timeo);
613 break;
614 }
615 continue;
616 }
617 tss.len -= ret;
618 spliced += ret;
619
Willy Tarreau33966dd2009-01-13 16:04:36 -0800620 if (!timeo)
621 break;
Jens Axboe9c55e012007-11-06 23:30:13 -0800622 release_sock(sk);
623 lock_sock(sk);
624
625 if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
Willy Tarreau33966dd2009-01-13 16:04:36 -0800626 (sk->sk_shutdown & RCV_SHUTDOWN) ||
Jens Axboe9c55e012007-11-06 23:30:13 -0800627 signal_pending(current))
628 break;
629 }
630
631 release_sock(sk);
632
633 if (spliced)
634 return spliced;
635
636 return ret;
637}
638
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100639struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100640{
641 struct sk_buff *skb;
642
643 /* The TCP header must be at least 32-bit aligned. */
644 size = ALIGN(size, 4);
645
646 skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
647 if (skb) {
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800648 if (sk_wmem_schedule(sk, skb->truesize)) {
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100649 /*
650 * Make sure that we have exactly size bytes
651 * available to the caller, no more, no less.
652 */
653 skb_reserve(skb, skb_tailroom(skb) - size);
654 return skb;
655 }
656 __kfree_skb(skb);
657 } else {
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700658 sk->sk_prot->enter_memory_pressure(sk);
Pavel Emelyanovf561d0f2007-11-29 20:28:50 +1100659 sk_stream_moderate_sndbuf(sk);
660 }
661 return NULL;
662}
663
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000664static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
665 int large_allowed)
666{
667 struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000668 u32 xmit_size_goal, old_size_goal;
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000669
670 xmit_size_goal = mss_now;
671
672 if (large_allowed && sk_can_gso(sk)) {
673 xmit_size_goal = ((sk->sk_gso_max_size - 1) -
674 inet_csk(sk)->icsk_af_ops->net_header_len -
675 inet_csk(sk)->icsk_ext_hdr_len -
676 tp->tcp_header_len);
677
678 xmit_size_goal = tcp_bound_to_half_wnd(tp, xmit_size_goal);
Ilpo Järvinen2a3a0412009-03-14 22:45:16 +0000679
680 /* We try hard to avoid divides here */
681 old_size_goal = tp->xmit_size_goal_segs * mss_now;
682
683 if (likely(old_size_goal <= xmit_size_goal &&
684 old_size_goal + mss_now > xmit_size_goal)) {
685 xmit_size_goal = old_size_goal;
686 } else {
687 tp->xmit_size_goal_segs = xmit_size_goal / mss_now;
688 xmit_size_goal = tp->xmit_size_goal_segs * mss_now;
689 }
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000690 }
691
Ilpo Järvinenafece1c2009-03-14 14:23:07 +0000692 return max(xmit_size_goal, mss_now);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000693}
694
695static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
696{
697 int mss_now;
698
699 mss_now = tcp_current_mss(sk);
700 *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
701
702 return mss_now;
703}
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705static ssize_t do_tcp_sendpages(struct sock *sk, struct page **pages, int poffset,
706 size_t psize, int flags)
707{
708 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700709 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 int err;
711 ssize_t copied;
712 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
713
714 /* Wait for a connection to finish. */
715 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
716 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
717 goto out_err;
718
719 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
720
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000721 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 copied = 0;
723
724 err = -EPIPE;
725 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000726 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 while (psize > 0) {
David S. Millerfe067e82007-03-07 12:12:44 -0800729 struct sk_buff *skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 struct page *page = pages[poffset / PAGE_SIZE];
731 int copy, i, can_coalesce;
732 int offset = poffset % PAGE_SIZE;
733 int size = min_t(size_t, psize, PAGE_SIZE - offset);
734
David S. Millerfe067e82007-03-07 12:12:44 -0800735 if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736new_segment:
737 if (!sk_stream_memory_free(sk))
738 goto wait_for_sndbuf;
739
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100740 skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 if (!skb)
742 goto wait_for_memory;
743
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700744 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700745 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
747
748 if (copy > size)
749 copy = size;
750
751 i = skb_shinfo(skb)->nr_frags;
752 can_coalesce = skb_can_coalesce(skb, i, page, offset);
753 if (!can_coalesce && i >= MAX_SKB_FRAGS) {
754 tcp_mark_push(tp, skb);
755 goto new_segment;
756 }
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800757 if (!sk_wmem_schedule(sk, copy))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 goto wait_for_memory;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (can_coalesce) {
761 skb_shinfo(skb)->frags[i - 1].size += copy;
762 } else {
763 get_page(page);
764 skb_fill_page_desc(skb, i, page, offset, copy);
765 }
766
767 skb->len += copy;
768 skb->data_len += copy;
769 skb->truesize += copy;
770 sk->sk_wmem_queued += copy;
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800771 sk_mem_charge(sk, copy);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700772 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 tp->write_seq += copy;
774 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -0700775 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 if (!copied)
778 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
779
780 copied += copy;
781 poffset += copy;
782 if (!(psize -= copy))
783 goto out;
784
Herbert Xu69d15062008-03-22 15:47:05 -0700785 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 continue;
787
788 if (forced_push(tp)) {
789 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700790 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -0800791 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 tcp_push_one(sk, mss_now);
793 continue;
794
795wait_for_sndbuf:
796 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
797wait_for_memory:
798 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700799 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
802 goto do_error;
803
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000804 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 }
806
807out:
808 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700809 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return copied;
811
812do_error:
813 if (copied)
814 goto out;
815out_err:
816 return sk_stream_error(sk, flags, err);
817}
818
819ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset,
820 size_t size, int flags)
821{
822 ssize_t res;
823 struct sock *sk = sock->sk;
824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (!(sk->sk_route_caps & NETIF_F_SG) ||
Herbert Xu8648b302006-06-17 22:06:05 -0700826 !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return sock_no_sendpage(sock, page, offset, size, flags);
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 lock_sock(sk);
830 TCP_CHECK_TIMER(sk);
831 res = do_tcp_sendpages(sk, &page, offset, size, flags);
832 TCP_CHECK_TIMER(sk);
833 release_sock(sk);
834 return res;
835}
836
837#define TCP_PAGE(sk) (sk->sk_sndmsg_page)
838#define TCP_OFF(sk) (sk->sk_sndmsg_off)
839
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700840static inline int select_size(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700842 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700843 int tmp = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
David S. Millerb4e26f52005-07-05 15:20:27 -0700845 if (sk->sk_route_caps & NETIF_F_SG) {
Herbert Xubcd76112006-06-30 13:36:35 -0700846 if (sk_can_gso(sk))
David S. Millerb4e26f52005-07-05 15:20:27 -0700847 tmp = 0;
848 else {
849 int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
850
851 if (tmp >= pgbreak &&
852 tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
853 tmp = pgbreak;
854 }
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return tmp;
858}
859
David S. Miller3516ffb2007-08-02 19:23:56 -0700860int tcp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 size_t size)
862{
David S. Miller3516ffb2007-08-02 19:23:56 -0700863 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct iovec *iov;
865 struct tcp_sock *tp = tcp_sk(sk);
866 struct sk_buff *skb;
867 int iovlen, flags;
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700868 int mss_now, size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 int err, copied;
870 long timeo;
871
872 lock_sock(sk);
873 TCP_CHECK_TIMER(sk);
874
875 flags = msg->msg_flags;
876 timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
877
878 /* Wait for a connection to finish. */
879 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))
880 if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
881 goto out_err;
882
883 /* This should be in poll */
884 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
885
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000886 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 /* Ok commence sending. */
889 iovlen = msg->msg_iovlen;
890 iov = msg->msg_iov;
891 copied = 0;
892
893 err = -EPIPE;
894 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
Ilpo Järvinen0d6a7752009-02-28 04:44:41 +0000895 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 while (--iovlen >= 0) {
898 int seglen = iov->iov_len;
899 unsigned char __user *from = iov->iov_base;
900
901 iov++;
902
903 while (seglen > 0) {
904 int copy;
905
David S. Millerfe067e82007-03-07 12:12:44 -0800906 skb = tcp_write_queue_tail(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
David S. Millerfe067e82007-03-07 12:12:44 -0800908 if (!tcp_send_head(sk) ||
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700909 (copy = size_goal - skb->len) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911new_segment:
912 /* Allocate new segment. If the interface is SG,
913 * allocate skb fitting to single page.
914 */
915 if (!sk_stream_memory_free(sk))
916 goto wait_for_sndbuf;
917
Pavel Emelyanovdf97c702007-11-29 21:22:33 +1100918 skb = sk_stream_alloc_skb(sk, select_size(sk),
919 sk->sk_allocation);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 if (!skb)
921 goto wait_for_memory;
922
923 /*
924 * Check whether we can use HW checksum.
925 */
Herbert Xu8648b302006-06-17 22:06:05 -0700926 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700927 skb->ip_summed = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700929 skb_entail(sk, skb);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700930 copy = size_goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
932
933 /* Try to append data to the end of skb. */
934 if (copy > seglen)
935 copy = seglen;
936
937 /* Where to copy to? */
938 if (skb_tailroom(skb) > 0) {
939 /* We have some space in skb head. Superb! */
940 if (copy > skb_tailroom(skb))
941 copy = skb_tailroom(skb);
942 if ((err = skb_add_data(skb, from, copy)) != 0)
943 goto do_fault;
944 } else {
945 int merge = 0;
946 int i = skb_shinfo(skb)->nr_frags;
947 struct page *page = TCP_PAGE(sk);
948 int off = TCP_OFF(sk);
949
950 if (skb_can_coalesce(skb, i, page, off) &&
951 off != PAGE_SIZE) {
952 /* We can extend the last page
953 * fragment. */
954 merge = 1;
955 } else if (i == MAX_SKB_FRAGS ||
956 (!i &&
957 !(sk->sk_route_caps & NETIF_F_SG))) {
958 /* Need to add new fragment and cannot
959 * do this because interface is non-SG,
960 * or because all the page slots are
961 * busy. */
962 tcp_mark_push(tp, skb);
963 goto new_segment;
964 } else if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 if (off == PAGE_SIZE) {
966 put_page(page);
967 TCP_PAGE(sk) = page = NULL;
Herbert Xufb5f5e62005-09-05 18:55:48 -0700968 off = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
Herbert Xuef015782005-09-01 17:48:59 -0700970 } else
Herbert Xufb5f5e62005-09-05 18:55:48 -0700971 off = 0;
Herbert Xuef015782005-09-01 17:48:59 -0700972
973 if (copy > PAGE_SIZE - off)
974 copy = PAGE_SIZE - off;
975
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800976 if (!sk_wmem_schedule(sk, copy))
Herbert Xuef015782005-09-01 17:48:59 -0700977 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 if (!page) {
980 /* Allocate new cache page. */
981 if (!(page = sk_stream_alloc_page(sk)))
982 goto wait_for_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 }
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 /* Time to copy data. We are close to
986 * the end! */
987 err = skb_copy_to_page(sk, from, skb, page,
988 off, copy);
989 if (err) {
990 /* If this page was new, give it to the
991 * socket so it does not get leaked.
992 */
993 if (!TCP_PAGE(sk)) {
994 TCP_PAGE(sk) = page;
995 TCP_OFF(sk) = 0;
996 }
997 goto do_error;
998 }
999
1000 /* Update the skb. */
1001 if (merge) {
1002 skb_shinfo(skb)->frags[i - 1].size +=
1003 copy;
1004 } else {
1005 skb_fill_page_desc(skb, i, page, off, copy);
1006 if (TCP_PAGE(sk)) {
1007 get_page(page);
1008 } else if (off + copy < PAGE_SIZE) {
1009 get_page(page);
1010 TCP_PAGE(sk) = page;
1011 }
1012 }
1013
1014 TCP_OFF(sk) = off + copy;
1015 }
1016
1017 if (!copied)
1018 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_PSH;
1019
1020 tp->write_seq += copy;
1021 TCP_SKB_CB(skb)->end_seq += copy;
Herbert Xu79671682006-06-22 02:40:14 -07001022 skb_shinfo(skb)->gso_segs = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 from += copy;
1025 copied += copy;
1026 if ((seglen -= copy) == 0 && iovlen == 0)
1027 goto out;
1028
Herbert Xu69d15062008-03-22 15:47:05 -07001029 if (skb->len < size_goal || (flags & MSG_OOB))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 continue;
1031
1032 if (forced_push(tp)) {
1033 tcp_mark_push(tp, skb);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001034 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
David S. Millerfe067e82007-03-07 12:12:44 -08001035 } else if (skb == tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 tcp_push_one(sk, mss_now);
1037 continue;
1038
1039wait_for_sndbuf:
1040 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1041wait_for_memory:
1042 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001043 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
1046 goto do_error;
1047
Ilpo Järvinen0c54b852009-03-14 14:23:05 +00001048 mss_now = tcp_send_mss(sk, &size_goal, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
1050 }
1051
1052out:
1053 if (copied)
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001054 tcp_push(sk, flags, mss_now, tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 TCP_CHECK_TIMER(sk);
1056 release_sock(sk);
1057 return copied;
1058
1059do_fault:
1060 if (!skb->len) {
David S. Millerfe067e82007-03-07 12:12:44 -08001061 tcp_unlink_write_queue(skb, sk);
1062 /* It is the one place in all of TCP, except connection
1063 * reset, where we can be unlinking the send_head.
1064 */
1065 tcp_check_send_head(sk, skb);
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001066 sk_wmem_free_skb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 }
1068
1069do_error:
1070 if (copied)
1071 goto out;
1072out_err:
1073 err = sk_stream_error(sk, flags, err);
1074 TCP_CHECK_TIMER(sk);
1075 release_sock(sk);
1076 return err;
1077}
1078
1079/*
1080 * Handle reading urgent data. BSD has very simple semantics for
1081 * this, no blocking and very strange errors 8)
1082 */
1083
Rami Rosen377f0a02009-03-31 14:43:17 -07001084static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 struct tcp_sock *tp = tcp_sk(sk);
1087
1088 /* No URG data to read. */
1089 if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
1090 tp->urg_data == TCP_URG_READ)
1091 return -EINVAL; /* Yes this is right ! */
1092
1093 if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
1094 return -ENOTCONN;
1095
1096 if (tp->urg_data & TCP_URG_VALID) {
1097 int err = 0;
1098 char c = tp->urg_data;
1099
1100 if (!(flags & MSG_PEEK))
1101 tp->urg_data = TCP_URG_READ;
1102
1103 /* Read urgent data. */
1104 msg->msg_flags |= MSG_OOB;
1105
1106 if (len > 0) {
1107 if (!(flags & MSG_TRUNC))
1108 err = memcpy_toiovec(msg->msg_iov, &c, 1);
1109 len = 1;
1110 } else
1111 msg->msg_flags |= MSG_TRUNC;
1112
1113 return err ? -EFAULT : len;
1114 }
1115
1116 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
1117 return 0;
1118
1119 /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1120 * the available implementations agree in this case:
1121 * this call should never block, independent of the
1122 * blocking state of the socket.
1123 * Mike <pall@rz.uni-karlsruhe.de>
1124 */
1125 return -EAGAIN;
1126}
1127
1128/* Clean up the receive buffer for full frames taken by the user,
1129 * then send an ACK if necessary. COPIED is the number of bytes
1130 * tcp_recvmsg has given to the user so far, it speeds up the
1131 * calculation of whether or not we must ACK for the sake of
1132 * a window update.
1133 */
Chris Leech0e4b4992006-05-23 18:00:16 -07001134void tcp_cleanup_rbuf(struct sock *sk, int copied)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
1136 struct tcp_sock *tp = tcp_sk(sk);
1137 int time_to_ack = 0;
1138
1139#if TCP_DEBUG
1140 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1141
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001142 WARN_ON(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143#endif
1144
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001145 if (inet_csk_ack_scheduled(sk)) {
1146 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 /* Delayed ACKs frequently hit locked sockets during bulk
1148 * receive. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001149 if (icsk->icsk_ack.blocked ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 /* Once-per-two-segments ACK was not sent by tcp_input.c */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001151 tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 /*
1153 * If this read emptied read buffer, we send ACK, if
1154 * connection is not bidirectional, user drained
1155 * receive buffer and there was a small segment
1156 * in queue.
1157 */
Alexey Kuznetsov1ef96962006-09-19 12:52:50 -07001158 (copied > 0 &&
1159 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
1160 ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
1161 !icsk->icsk_ack.pingpong)) &&
1162 !atomic_read(&sk->sk_rmem_alloc)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 time_to_ack = 1;
1164 }
1165
1166 /* We send an ACK if we can now advertise a non-zero window
1167 * which has been raised "significantly".
1168 *
1169 * Even if window raised up to infinity, do not send window open ACK
1170 * in states, where we will not receive more. It is useless.
1171 */
1172 if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1173 __u32 rcv_window_now = tcp_receive_window(tp);
1174
1175 /* Optimize, __tcp_select_window() is not cheap. */
1176 if (2*rcv_window_now <= tp->window_clamp) {
1177 __u32 new_window = __tcp_select_window(sk);
1178
1179 /* Send ACK now, if this read freed lots of space
1180 * in our buffer. Certainly, new_window is new window.
1181 * We can advertise it now, if it is not less than current one.
1182 * "Lots" means "at least twice" here.
1183 */
1184 if (new_window && new_window >= 2 * rcv_window_now)
1185 time_to_ack = 1;
1186 }
1187 }
1188 if (time_to_ack)
1189 tcp_send_ack(sk);
1190}
1191
1192static void tcp_prequeue_process(struct sock *sk)
1193{
1194 struct sk_buff *skb;
1195 struct tcp_sock *tp = tcp_sk(sk);
1196
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001197 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 /* RX process wants to run with disabled BHs, though it is not
1200 * necessary */
1201 local_bh_disable();
1202 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -07001203 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 local_bh_enable();
1205
1206 /* Clear memory counter. */
1207 tp->ucopy.memory = 0;
1208}
1209
1210static inline struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
1211{
1212 struct sk_buff *skb;
1213 u32 offset;
1214
1215 skb_queue_walk(&sk->sk_receive_queue, skb) {
1216 offset = seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001217 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 offset--;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001219 if (offset < skb->len || tcp_hdr(skb)->fin) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 *off = offset;
1221 return skb;
1222 }
1223 }
1224 return NULL;
1225}
1226
1227/*
1228 * This routine provides an alternative to tcp_recvmsg() for routines
1229 * that would like to handle copying from skbuffs directly in 'sendfile'
1230 * fashion.
1231 * Note:
1232 * - It is assumed that the socket was locked by the caller.
1233 * - The routine does not block.
1234 * - At present, there is no support for reading OOB data
1235 * or for 'peeking' the socket using this routine
1236 * (although both would be easy to implement).
1237 */
1238int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
1239 sk_read_actor_t recv_actor)
1240{
1241 struct sk_buff *skb;
1242 struct tcp_sock *tp = tcp_sk(sk);
1243 u32 seq = tp->copied_seq;
1244 u32 offset;
1245 int copied = 0;
1246
1247 if (sk->sk_state == TCP_LISTEN)
1248 return -ENOTCONN;
1249 while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
1250 if (offset < skb->len) {
Octavian Purdila374e7b52008-07-03 03:31:21 -07001251 int used;
1252 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 len = skb->len - offset;
1255 /* Stop reading if we hit a patch of urgent data */
1256 if (tp->urg_data) {
1257 u32 urg_offset = tp->urg_seq - seq;
1258 if (urg_offset < len)
1259 len = urg_offset;
1260 if (!len)
1261 break;
1262 }
1263 used = recv_actor(desc, skb, offset, len);
Jens Axboeddb61a52007-06-23 23:07:50 -07001264 if (used < 0) {
1265 if (!copied)
1266 copied = used;
1267 break;
1268 } else if (used <= len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 seq += used;
1270 copied += used;
1271 offset += used;
1272 }
Octavian Purdila293ad602008-06-04 15:45:58 -07001273 /*
1274 * If recv_actor drops the lock (e.g. TCP splice
1275 * receive) the skb pointer might be invalid when
1276 * getting here: tcp_collapse might have deleted it
1277 * while aggregating skbs from the socket queue.
1278 */
1279 skb = tcp_recv_skb(sk, seq-1, &offset);
1280 if (!skb || (offset+1 != skb->len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 break;
1282 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001283 if (tcp_hdr(skb)->fin) {
Chris Leech624d1162006-05-23 18:01:28 -07001284 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 ++seq;
1286 break;
1287 }
Chris Leech624d1162006-05-23 18:01:28 -07001288 sk_eat_skb(sk, skb, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (!desc->count)
1290 break;
1291 }
1292 tp->copied_seq = seq;
1293
1294 tcp_rcv_space_adjust(sk);
1295
1296 /* Clean up data we have read: This will do ACK frames. */
Jens Axboeddb61a52007-06-23 23:07:50 -07001297 if (copied > 0)
Chris Leech0e4b4992006-05-23 18:00:16 -07001298 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 return copied;
1300}
1301
1302/*
1303 * This routine copies from a sock struct into the user buffer.
1304 *
1305 * Technical note: in 2.3 we work on _locked_ socket, so that
1306 * tricks with *seq access order and skb->users are not required.
1307 * Probably, code can be easily improved even more.
1308 */
1309
1310int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1311 size_t len, int nonblock, int flags, int *addr_len)
1312{
1313 struct tcp_sock *tp = tcp_sk(sk);
1314 int copied = 0;
1315 u32 peek_seq;
1316 u32 *seq;
1317 unsigned long used;
1318 int err;
1319 int target; /* Read at least this many bytes */
1320 long timeo;
1321 struct task_struct *user_recv = NULL;
Chris Leech1a2449a2006-05-23 18:05:53 -07001322 int copied_early = 0;
Chris Leech2b1244a2007-03-08 09:57:36 -08001323 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 lock_sock(sk);
1326
1327 TCP_CHECK_TIMER(sk);
1328
1329 err = -ENOTCONN;
1330 if (sk->sk_state == TCP_LISTEN)
1331 goto out;
1332
1333 timeo = sock_rcvtimeo(sk, nonblock);
1334
1335 /* Urgent data needs to be handled specially. */
1336 if (flags & MSG_OOB)
1337 goto recv_urg;
1338
1339 seq = &tp->copied_seq;
1340 if (flags & MSG_PEEK) {
1341 peek_seq = tp->copied_seq;
1342 seq = &peek_seq;
1343 }
1344
1345 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1346
Chris Leech1a2449a2006-05-23 18:05:53 -07001347#ifdef CONFIG_NET_DMA
1348 tp->ucopy.dma_chan = NULL;
1349 preempt_disable();
Chris Leech2b1244a2007-03-08 09:57:36 -08001350 skb = skb_peek_tail(&sk->sk_receive_queue);
Andrew Mortone00c5d82007-03-08 09:57:36 -08001351 {
1352 int available = 0;
1353
1354 if (skb)
1355 available = TCP_SKB_CB(skb)->seq + skb->len - (*seq);
1356 if ((available < target) &&
1357 (len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) &&
1358 !sysctl_tcp_low_latency &&
Dan Williamsf67b4592009-01-06 11:38:15 -07001359 dma_find_channel(DMA_MEMCPY)) {
Andrew Mortone00c5d82007-03-08 09:57:36 -08001360 preempt_enable_no_resched();
1361 tp->ucopy.pinned_list =
1362 dma_pin_iovec_pages(msg->msg_iov, len);
1363 } else {
1364 preempt_enable_no_resched();
1365 }
1366 }
Chris Leech1a2449a2006-05-23 18:05:53 -07001367#endif
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 u32 offset;
1371
1372 /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
1373 if (tp->urg_data && tp->urg_seq == *seq) {
1374 if (copied)
1375 break;
1376 if (signal_pending(current)) {
1377 copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
1378 break;
1379 }
1380 }
1381
1382 /* Next get a buffer. */
1383
1384 skb = skb_peek(&sk->sk_receive_queue);
1385 do {
1386 if (!skb)
1387 break;
1388
1389 /* Now that we have two receive queues this
1390 * shouldn't happen.
1391 */
1392 if (before(*seq, TCP_SKB_CB(skb)->seq)) {
1393 printk(KERN_INFO "recvmsg bug: copied %X "
1394 "seq %X\n", *seq, TCP_SKB_CB(skb)->seq);
1395 break;
1396 }
1397 offset = *seq - TCP_SKB_CB(skb)->seq;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001398 if (tcp_hdr(skb)->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 offset--;
1400 if (offset < skb->len)
1401 goto found_ok_skb;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001402 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 goto found_fin_ok;
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001404 WARN_ON(!(flags & MSG_PEEK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 skb = skb->next;
1406 } while (skb != (struct sk_buff *)&sk->sk_receive_queue);
1407
1408 /* Well, if we have backlog, try to process it now yet. */
1409
1410 if (copied >= target && !sk->sk_backlog.tail)
1411 break;
1412
1413 if (copied) {
1414 if (sk->sk_err ||
1415 sk->sk_state == TCP_CLOSE ||
1416 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1417 !timeo ||
David S. Miller518a09e2008-11-05 03:36:01 -08001418 signal_pending(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 break;
1420 } else {
1421 if (sock_flag(sk, SOCK_DONE))
1422 break;
1423
1424 if (sk->sk_err) {
1425 copied = sock_error(sk);
1426 break;
1427 }
1428
1429 if (sk->sk_shutdown & RCV_SHUTDOWN)
1430 break;
1431
1432 if (sk->sk_state == TCP_CLOSE) {
1433 if (!sock_flag(sk, SOCK_DONE)) {
1434 /* This occurs when user tries to read
1435 * from never connected socket.
1436 */
1437 copied = -ENOTCONN;
1438 break;
1439 }
1440 break;
1441 }
1442
1443 if (!timeo) {
1444 copied = -EAGAIN;
1445 break;
1446 }
1447
1448 if (signal_pending(current)) {
1449 copied = sock_intr_errno(timeo);
1450 break;
1451 }
1452 }
1453
Chris Leech0e4b4992006-05-23 18:00:16 -07001454 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
David S. Miller7df55122005-06-18 23:01:10 -07001456 if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 /* Install new reader */
1458 if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
1459 user_recv = current;
1460 tp->ucopy.task = user_recv;
1461 tp->ucopy.iov = msg->msg_iov;
1462 }
1463
1464 tp->ucopy.len = len;
1465
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001466 WARN_ON(tp->copied_seq != tp->rcv_nxt &&
1467 !(flags & (MSG_PEEK | MSG_TRUNC)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 /* Ugly... If prequeue is not empty, we have to
1470 * process it before releasing socket, otherwise
1471 * order will be broken at second iteration.
1472 * More elegant solution is required!!!
1473 *
1474 * Look: we have the following (pseudo)queues:
1475 *
1476 * 1. packets in flight
1477 * 2. backlog
1478 * 3. prequeue
1479 * 4. receive_queue
1480 *
1481 * Each queue can be processed only if the next ones
1482 * are empty. At this point we have empty receive_queue.
1483 * But prequeue _can_ be not empty after 2nd iteration,
1484 * when we jumped to start of loop because backlog
1485 * processing added something to receive_queue.
1486 * We cannot release_sock(), because backlog contains
1487 * packets arrived _after_ prequeued ones.
1488 *
1489 * Shortly, algorithm is clear --- to process all
1490 * the queues in order. We could make it more directly,
1491 * requeueing packets from backlog to prequeue, if
1492 * is not empty. It is more elegant, but eats cycles,
1493 * unfortunately.
1494 */
David S. Millerb03efcf2005-07-08 14:57:23 -07001495 if (!skb_queue_empty(&tp->ucopy.prequeue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 goto do_prequeue;
1497
1498 /* __ Set realtime policy in scheduler __ */
1499 }
1500
1501 if (copied >= target) {
1502 /* Do not sleep, just process backlog. */
1503 release_sock(sk);
1504 lock_sock(sk);
1505 } else
1506 sk_wait_data(sk, &timeo);
1507
Chris Leech1a2449a2006-05-23 18:05:53 -07001508#ifdef CONFIG_NET_DMA
1509 tp->ucopy.wakeup = 0;
1510#endif
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 if (user_recv) {
1513 int chunk;
1514
1515 /* __ Restore normal policy in scheduler __ */
1516
1517 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001518 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 len -= chunk;
1520 copied += chunk;
1521 }
1522
1523 if (tp->rcv_nxt == tp->copied_seq &&
David S. Millerb03efcf2005-07-08 14:57:23 -07001524 !skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525do_prequeue:
1526 tcp_prequeue_process(sk);
1527
1528 if ((chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001529 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 len -= chunk;
1531 copied += chunk;
1532 }
1533 }
1534 }
1535 if ((flags & MSG_PEEK) && peek_seq != tp->copied_seq) {
1536 if (net_ratelimit())
1537 printk(KERN_DEBUG "TCP(%s:%d): Application bug, race in MSG_PEEK.\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001538 current->comm, task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 peek_seq = tp->copied_seq;
1540 }
1541 continue;
1542
1543 found_ok_skb:
1544 /* Ok so how much can we use? */
1545 used = skb->len - offset;
1546 if (len < used)
1547 used = len;
1548
1549 /* Do we have urgent data here? */
1550 if (tp->urg_data) {
1551 u32 urg_offset = tp->urg_seq - *seq;
1552 if (urg_offset < used) {
1553 if (!urg_offset) {
1554 if (!sock_flag(sk, SOCK_URGINLINE)) {
1555 ++*seq;
1556 offset++;
1557 used--;
1558 if (!used)
1559 goto skip_copy;
1560 }
1561 } else
1562 used = urg_offset;
1563 }
1564 }
1565
1566 if (!(flags & MSG_TRUNC)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001567#ifdef CONFIG_NET_DMA
1568 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dan Williamsf67b4592009-01-06 11:38:15 -07001569 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
Chris Leech1a2449a2006-05-23 18:05:53 -07001570
1571 if (tp->ucopy.dma_chan) {
1572 tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
1573 tp->ucopy.dma_chan, skb, offset,
1574 msg->msg_iov, used,
1575 tp->ucopy.pinned_list);
1576
1577 if (tp->ucopy.dma_cookie < 0) {
1578
1579 printk(KERN_ALERT "dma_cookie < 0\n");
1580
1581 /* Exception. Bailout! */
1582 if (!copied)
1583 copied = -EFAULT;
1584 break;
1585 }
1586 if ((offset + used) == skb->len)
1587 copied_early = 1;
1588
1589 } else
1590#endif
1591 {
1592 err = skb_copy_datagram_iovec(skb, offset,
1593 msg->msg_iov, used);
1594 if (err) {
1595 /* Exception. Bailout! */
1596 if (!copied)
1597 copied = -EFAULT;
1598 break;
1599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 }
1601 }
1602
1603 *seq += used;
1604 copied += used;
1605 len -= used;
1606
1607 tcp_rcv_space_adjust(sk);
1608
1609skip_copy:
1610 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1611 tp->urg_data = 0;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001612 tcp_fast_path_check(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 }
1614 if (used + offset < skb->len)
1615 continue;
1616
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001617 if (tcp_hdr(skb)->fin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 goto found_fin_ok;
Chris Leech1a2449a2006-05-23 18:05:53 -07001619 if (!(flags & MSG_PEEK)) {
1620 sk_eat_skb(sk, skb, copied_early);
1621 copied_early = 0;
1622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 continue;
1624
1625 found_fin_ok:
1626 /* Process the FIN. */
1627 ++*seq;
Chris Leech1a2449a2006-05-23 18:05:53 -07001628 if (!(flags & MSG_PEEK)) {
1629 sk_eat_skb(sk, skb, copied_early);
1630 copied_early = 0;
1631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 break;
1633 } while (len > 0);
1634
1635 if (user_recv) {
David S. Millerb03efcf2005-07-08 14:57:23 -07001636 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 int chunk;
1638
1639 tp->ucopy.len = copied > 0 ? len : 0;
1640
1641 tcp_prequeue_process(sk);
1642
1643 if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
Pavel Emelyanoved880982008-07-16 20:32:45 -07001644 NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 len -= chunk;
1646 copied += chunk;
1647 }
1648 }
1649
1650 tp->ucopy.task = NULL;
1651 tp->ucopy.len = 0;
1652 }
1653
Chris Leech1a2449a2006-05-23 18:05:53 -07001654#ifdef CONFIG_NET_DMA
1655 if (tp->ucopy.dma_chan) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001656 dma_cookie_t done, used;
1657
1658 dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
1659
1660 while (dma_async_memcpy_complete(tp->ucopy.dma_chan,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001661 tp->ucopy.dma_cookie, &done,
1662 &used) == DMA_IN_PROGRESS) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001663 /* do partial cleanup of sk_async_wait_queue */
1664 while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
1665 (dma_async_is_complete(skb->dma_cookie, done,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001666 used) == DMA_SUCCESS)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001667 __skb_dequeue(&sk->sk_async_wait_queue);
1668 kfree_skb(skb);
1669 }
1670 }
1671
1672 /* Safe to free early-copied skbs now */
1673 __skb_queue_purge(&sk->sk_async_wait_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07001674 tp->ucopy.dma_chan = NULL;
1675 }
1676 if (tp->ucopy.pinned_list) {
1677 dma_unpin_iovec_pages(tp->ucopy.pinned_list);
1678 tp->ucopy.pinned_list = NULL;
1679 }
1680#endif
1681
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 /* According to UNIX98, msg_name/msg_namelen are ignored
1683 * on connected socket. I was just happy when found this 8) --ANK
1684 */
1685
1686 /* Clean up data we have read: This will do ACK frames. */
Chris Leech0e4b4992006-05-23 18:00:16 -07001687 tcp_cleanup_rbuf(sk, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 TCP_CHECK_TIMER(sk);
1690 release_sock(sk);
1691 return copied;
1692
1693out:
1694 TCP_CHECK_TIMER(sk);
1695 release_sock(sk);
1696 return err;
1697
1698recv_urg:
Rami Rosen377f0a02009-03-31 14:43:17 -07001699 err = tcp_recv_urg(sk, msg, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 goto out;
1701}
1702
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001703void tcp_set_state(struct sock *sk, int state)
1704{
1705 int oldstate = sk->sk_state;
1706
1707 switch (state) {
1708 case TCP_ESTABLISHED:
1709 if (oldstate != TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001710 TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001711 break;
1712
1713 case TCP_CLOSE:
1714 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
Pavel Emelyanov81cc8a72008-07-16 20:22:04 -07001715 TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001716
1717 sk->sk_prot->unhash(sk);
1718 if (inet_csk(sk)->icsk_bind_hash &&
1719 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001720 inet_put_port(sk);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001721 /* fall through */
1722 default:
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001723 if (oldstate == TCP_ESTABLISHED)
Pavel Emelyanov74688e42008-07-16 20:22:46 -07001724 TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001725 }
1726
1727 /* Change state AFTER socket is unhashed to avoid closed
1728 * socket sitting in hash tables.
1729 */
1730 sk->sk_state = state;
1731
1732#ifdef STATE_TRACE
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08001733 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
Ilpo Järvinen490d5042008-01-12 03:17:20 -08001734#endif
1735}
1736EXPORT_SYMBOL_GPL(tcp_set_state);
1737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738/*
1739 * State processing on a close. This implements the state shift for
1740 * sending our FIN frame. Note that we only send a FIN for some
1741 * states. A shutdown() may have already sent the FIN, or we may be
1742 * closed.
1743 */
1744
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08001745static const unsigned char new_state[16] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 /* current state: new state: action: */
1747 /* (Invalid) */ TCP_CLOSE,
1748 /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1749 /* TCP_SYN_SENT */ TCP_CLOSE,
1750 /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
1751 /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
1752 /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
1753 /* TCP_TIME_WAIT */ TCP_CLOSE,
1754 /* TCP_CLOSE */ TCP_CLOSE,
1755 /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
1756 /* TCP_LAST_ACK */ TCP_LAST_ACK,
1757 /* TCP_LISTEN */ TCP_CLOSE,
1758 /* TCP_CLOSING */ TCP_CLOSING,
1759};
1760
1761static int tcp_close_state(struct sock *sk)
1762{
1763 int next = (int)new_state[sk->sk_state];
1764 int ns = next & TCP_STATE_MASK;
1765
1766 tcp_set_state(sk, ns);
1767
1768 return next & TCP_ACTION_FIN;
1769}
1770
1771/*
1772 * Shutdown the sending side of a connection. Much like close except
Satoru SATOH1f29b052008-04-21 02:27:58 -07001773 * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 */
1775
1776void tcp_shutdown(struct sock *sk, int how)
1777{
1778 /* We need to grab some memory, and put together a FIN,
1779 * and then put it into the queue to be sent.
1780 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1781 */
1782 if (!(how & SEND_SHUTDOWN))
1783 return;
1784
1785 /* If we've already sent a FIN, or it's a closed state, skip this. */
1786 if ((1 << sk->sk_state) &
1787 (TCPF_ESTABLISHED | TCPF_SYN_SENT |
1788 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
1789 /* Clear out any half completed packets. FIN if needed. */
1790 if (tcp_close_state(sk))
1791 tcp_send_fin(sk);
1792 }
1793}
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795void tcp_close(struct sock *sk, long timeout)
1796{
1797 struct sk_buff *skb;
1798 int data_was_unread = 0;
Herbert Xu75c2d9072006-05-03 23:31:35 -07001799 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
1801 lock_sock(sk);
1802 sk->sk_shutdown = SHUTDOWN_MASK;
1803
1804 if (sk->sk_state == TCP_LISTEN) {
1805 tcp_set_state(sk, TCP_CLOSE);
1806
1807 /* Special case. */
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001808 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 goto adjudge_to_death;
1811 }
1812
1813 /* We need to flush the recv. buffs. We do this only on the
1814 * descriptor close, not protocol-sourced closes, because the
1815 * reader process may not have drained the data yet!
1816 */
1817 while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1818 u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq -
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001819 tcp_hdr(skb)->fin;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 data_was_unread += len;
1821 __kfree_skb(skb);
1822 }
1823
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001824 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Gerrit Renker65bb7232007-04-28 21:21:46 -07001826 /* As outlined in RFC 2525, section 2.17, we send a RST here because
1827 * data was lost. To witness the awful effects of the old behavior of
1828 * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
1829 * GET in an FTP client, suspend the process, wait for the client to
1830 * advertise a zero window, then kill -9 the FTP client, wheee...
1831 * Note: timeout is always zero in such a case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 */
1833 if (data_was_unread) {
1834 /* Unread data was tossed, zap the connection. */
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001835 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 tcp_set_state(sk, TCP_CLOSE);
1837 tcp_send_active_reset(sk, GFP_KERNEL);
1838 } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
1839 /* Check zero linger _after_ checking for unread data. */
1840 sk->sk_prot->disconnect(sk, 0);
Pavel Emelyanov6f67c812008-07-16 20:31:39 -07001841 NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 } else if (tcp_close_state(sk)) {
1843 /* We FIN if the application ate all the data before
1844 * zapping the connection.
1845 */
1846
1847 /* RED-PEN. Formally speaking, we have broken TCP state
1848 * machine. State transitions:
1849 *
1850 * TCP_ESTABLISHED -> TCP_FIN_WAIT1
1851 * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
1852 * TCP_CLOSE_WAIT -> TCP_LAST_ACK
1853 *
1854 * are legal only when FIN has been sent (i.e. in window),
1855 * rather than queued out of window. Purists blame.
1856 *
1857 * F.e. "RFC state" is ESTABLISHED,
1858 * if Linux state is FIN-WAIT-1, but FIN is still not sent.
1859 *
1860 * The visible declinations are that sometimes
1861 * we enter time-wait state, when it is not required really
1862 * (harmless), do not send active resets, when they are
1863 * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
1864 * they look as CLOSING or LAST_ACK for Linux)
1865 * Probably, I missed some more holelets.
1866 * --ANK
1867 */
1868 tcp_send_fin(sk);
1869 }
1870
1871 sk_stream_wait_close(sk, timeout);
1872
1873adjudge_to_death:
Herbert Xu75c2d9072006-05-03 23:31:35 -07001874 state = sk->sk_state;
1875 sock_hold(sk);
1876 sock_orphan(sk);
Herbert Xu75c2d9072006-05-03 23:31:35 -07001877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 /* It is the last release_sock in its life. It will remove backlog. */
1879 release_sock(sk);
1880
1881
1882 /* Now socket is owned by kernel and we acquire BH lock
1883 to finish close. No need to check for user refs.
1884 */
1885 local_bh_disable();
1886 bh_lock_sock(sk);
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001887 WARN_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
Herbert Xueb4dea52008-12-29 23:04:08 -08001889 percpu_counter_inc(sk->sk_prot->orphan_count);
1890
Herbert Xu75c2d9072006-05-03 23:31:35 -07001891 /* Have we already been destroyed by a softirq or backlog? */
1892 if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
1893 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895 /* This is a (useful) BSD violating of the RFC. There is a
1896 * problem with TCP as specified in that the other end could
1897 * keep a socket open forever with no application left this end.
1898 * We use a 3 minute timeout (about the same as BSD) then kill
1899 * our end. If they send after that then tough - BUT: long enough
1900 * that we won't make the old 4*rto = almost no time - whoops
1901 * reset mistake.
1902 *
1903 * Nope, it was not mistake. It is really desired behaviour
1904 * f.e. on http servers, when such sockets are useless, but
1905 * consume significant resources. Let's do it with special
1906 * linger2 option. --ANK
1907 */
1908
1909 if (sk->sk_state == TCP_FIN_WAIT2) {
1910 struct tcp_sock *tp = tcp_sk(sk);
1911 if (tp->linger2 < 0) {
1912 tcp_set_state(sk, TCP_CLOSE);
1913 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001914 NET_INC_STATS_BH(sock_net(sk),
1915 LINUX_MIB_TCPABORTONLINGER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001917 const int tmo = tcp_fin_time(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
1919 if (tmo > TCP_TIMEWAIT_LEN) {
David S. Miller52499af2006-07-31 22:32:09 -07001920 inet_csk_reset_keepalive_timer(sk,
1921 tmo - TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
1924 goto out;
1925 }
1926 }
1927 }
1928 if (sk->sk_state != TCP_CLOSE) {
Eric Dumazetdd24c002008-11-25 21:17:14 -08001929 int orphan_count = percpu_counter_read_positive(
1930 sk->sk_prot->orphan_count);
1931
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001932 sk_mem_reclaim(sk);
Eric Dumazetdd24c002008-11-25 21:17:14 -08001933 if (tcp_too_many_orphans(sk, orphan_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 if (net_ratelimit())
1935 printk(KERN_INFO "TCP: too many of orphaned "
1936 "sockets\n");
1937 tcp_set_state(sk, TCP_CLOSE);
1938 tcp_send_active_reset(sk, GFP_ATOMIC);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001939 NET_INC_STATS_BH(sock_net(sk),
1940 LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 }
1942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 if (sk->sk_state == TCP_CLOSE)
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001945 inet_csk_destroy_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 /* Otherwise, socket is reprieved until protocol close. */
1947
1948out:
1949 bh_unlock_sock(sk);
1950 local_bh_enable();
1951 sock_put(sk);
1952}
1953
1954/* These states need RST on ABORT according to RFC793 */
1955
1956static inline int tcp_need_reset(int state)
1957{
1958 return (1 << state) &
1959 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
1960 TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
1961}
1962
1963int tcp_disconnect(struct sock *sk, int flags)
1964{
1965 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001966 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 struct tcp_sock *tp = tcp_sk(sk);
1968 int err = 0;
1969 int old_state = sk->sk_state;
1970
1971 if (old_state != TCP_CLOSE)
1972 tcp_set_state(sk, TCP_CLOSE);
1973
1974 /* ABORT function of RFC793 */
1975 if (old_state == TCP_LISTEN) {
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001976 inet_csk_listen_stop(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 } else if (tcp_need_reset(old_state) ||
1978 (tp->snd_nxt != tp->write_seq &&
1979 (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08001980 /* The last check adjusts for discrepancy of Linux wrt. RFC
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 * states
1982 */
1983 tcp_send_active_reset(sk, gfp_any());
1984 sk->sk_err = ECONNRESET;
1985 } else if (old_state == TCP_SYN_SENT)
1986 sk->sk_err = ECONNRESET;
1987
1988 tcp_clear_xmit_timers(sk);
1989 __skb_queue_purge(&sk->sk_receive_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001990 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 __skb_queue_purge(&tp->out_of_order_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07001992#ifdef CONFIG_NET_DMA
1993 __skb_queue_purge(&sk->sk_async_wait_queue);
1994#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
1996 inet->dport = 0;
1997
1998 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1999 inet_reset_saddr(sk);
2000
2001 sk->sk_shutdown = 0;
2002 sock_reset_flag(sk, SOCK_DONE);
2003 tp->srtt = 0;
2004 if ((tp->write_seq += tp->max_window + 2) == 0)
2005 tp->write_seq = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002006 icsk->icsk_backoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 tp->snd_cwnd = 2;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002008 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 tp->packets_out = 0;
2010 tp->snd_ssthresh = 0x7fffffff;
2011 tp->snd_cwnd_cnt = 0;
Stephen Hemminger9772efb2005-11-10 17:09:53 -08002012 tp->bytes_acked = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002013 tcp_set_ca_state(sk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 tcp_clear_retrans(tp);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002015 inet_csk_delack_init(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08002016 tcp_init_send_head(sk);
Srinivas Ajib40b4f72007-05-03 17:32:28 -07002017 memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 __sk_dst_reset(sk);
2019
Ilpo Järvinen547b7922008-07-25 21:43:18 -07002020 WARN_ON(inet->num && !icsk->icsk_bind_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
2022 sk->sk_error_report(sk);
2023 return err;
2024}
2025
2026/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 * Socket option code for TCP.
2028 */
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002029static int do_tcp_setsockopt(struct sock *sk, int level,
2030 int optname, char __user *optval, int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
2032 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002033 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 int val;
2035 int err = 0;
2036
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002037 /* This is a string value all the others are int's */
2038 if (optname == TCP_CONGESTION) {
2039 char name[TCP_CA_NAME_MAX];
2040
2041 if (optlen < 1)
2042 return -EINVAL;
2043
2044 val = strncpy_from_user(name, optval,
2045 min(TCP_CA_NAME_MAX-1, optlen));
2046 if (val < 0)
2047 return -EFAULT;
2048 name[val] = 0;
2049
2050 lock_sock(sk);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002051 err = tcp_set_congestion_control(sk, name);
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002052 release_sock(sk);
2053 return err;
2054 }
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (optlen < sizeof(int))
2057 return -EINVAL;
2058
2059 if (get_user(val, (int __user *)optval))
2060 return -EFAULT;
2061
2062 lock_sock(sk);
2063
2064 switch (optname) {
2065 case TCP_MAXSEG:
2066 /* Values greater than interface MTU won't take effect. However
2067 * at the point when this call is done we typically don't yet
2068 * know which interface is going to be used */
2069 if (val < 8 || val > MAX_TCP_WINDOW) {
2070 err = -EINVAL;
2071 break;
2072 }
2073 tp->rx_opt.user_mss = val;
2074 break;
2075
2076 case TCP_NODELAY:
2077 if (val) {
2078 /* TCP_NODELAY is weaker than TCP_CORK, so that
2079 * this option on corked socket is remembered, but
2080 * it is not activated until cork is cleared.
2081 *
2082 * However, when TCP_NODELAY is set we make
2083 * an explicit push, which overrides even TCP_CORK
2084 * for currently queued segments.
2085 */
2086 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002087 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 } else {
2089 tp->nonagle &= ~TCP_NAGLE_OFF;
2090 }
2091 break;
2092
2093 case TCP_CORK:
2094 /* When set indicates to always queue non-full frames.
2095 * Later the user clears this option and we transmit
2096 * any pending partial frames in the queue. This is
2097 * meant to be used alongside sendfile() to get properly
2098 * filled frames when the user (for example) must write
2099 * out headers with a write() call first and then use
2100 * sendfile to send out the data parts.
2101 *
2102 * TCP_CORK can be set together with TCP_NODELAY and it is
2103 * stronger than TCP_NODELAY.
2104 */
2105 if (val) {
2106 tp->nonagle |= TCP_NAGLE_CORK;
2107 } else {
2108 tp->nonagle &= ~TCP_NAGLE_CORK;
2109 if (tp->nonagle&TCP_NAGLE_OFF)
2110 tp->nonagle |= TCP_NAGLE_PUSH;
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07002111 tcp_push_pending_frames(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
2113 break;
2114
2115 case TCP_KEEPIDLE:
2116 if (val < 1 || val > MAX_TCP_KEEPIDLE)
2117 err = -EINVAL;
2118 else {
2119 tp->keepalive_time = val * HZ;
2120 if (sock_flag(sk, SOCK_KEEPOPEN) &&
2121 !((1 << sk->sk_state) &
2122 (TCPF_CLOSE | TCPF_LISTEN))) {
2123 __u32 elapsed = tcp_time_stamp - tp->rcv_tstamp;
2124 if (tp->keepalive_time > elapsed)
2125 elapsed = tp->keepalive_time - elapsed;
2126 else
2127 elapsed = 0;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002128 inet_csk_reset_keepalive_timer(sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 }
2130 }
2131 break;
2132 case TCP_KEEPINTVL:
2133 if (val < 1 || val > MAX_TCP_KEEPINTVL)
2134 err = -EINVAL;
2135 else
2136 tp->keepalive_intvl = val * HZ;
2137 break;
2138 case TCP_KEEPCNT:
2139 if (val < 1 || val > MAX_TCP_KEEPCNT)
2140 err = -EINVAL;
2141 else
2142 tp->keepalive_probes = val;
2143 break;
2144 case TCP_SYNCNT:
2145 if (val < 1 || val > MAX_TCP_SYNCNT)
2146 err = -EINVAL;
2147 else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002148 icsk->icsk_syn_retries = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 break;
2150
2151 case TCP_LINGER2:
2152 if (val < 0)
2153 tp->linger2 = -1;
2154 else if (val > sysctl_tcp_fin_timeout / HZ)
2155 tp->linger2 = 0;
2156 else
2157 tp->linger2 = val * HZ;
2158 break;
2159
2160 case TCP_DEFER_ACCEPT:
David S. Millerec0a1962008-06-12 16:31:35 -07002161 icsk->icsk_accept_queue.rskq_defer_accept = 0;
2162 if (val > 0) {
2163 /* Translate value in seconds to number of
2164 * retransmits */
2165 while (icsk->icsk_accept_queue.rskq_defer_accept < 32 &&
2166 val > ((TCP_TIMEOUT_INIT / HZ) <<
2167 icsk->icsk_accept_queue.rskq_defer_accept))
2168 icsk->icsk_accept_queue.rskq_defer_accept++;
2169 icsk->icsk_accept_queue.rskq_defer_accept++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 }
2171 break;
2172
2173 case TCP_WINDOW_CLAMP:
2174 if (!val) {
2175 if (sk->sk_state != TCP_CLOSE) {
2176 err = -EINVAL;
2177 break;
2178 }
2179 tp->window_clamp = 0;
2180 } else
2181 tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
2182 SOCK_MIN_RCVBUF / 2 : val;
2183 break;
2184
2185 case TCP_QUICKACK:
2186 if (!val) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002187 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002189 icsk->icsk_ack.pingpong = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 if ((1 << sk->sk_state) &
2191 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002192 inet_csk_ack_scheduled(sk)) {
2193 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
Chris Leech0e4b4992006-05-23 18:00:16 -07002194 tcp_cleanup_rbuf(sk, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (!(val & 1))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002196 icsk->icsk_ack.pingpong = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 }
2198 }
2199 break;
2200
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002201#ifdef CONFIG_TCP_MD5SIG
2202 case TCP_MD5SIG:
2203 /* Read the IP->Key mappings from userspace */
2204 err = tp->af_specific->md5_parse(sk, optval, optlen);
2205 break;
2206#endif
2207
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 default:
2209 err = -ENOPROTOOPT;
2210 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002211 }
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 release_sock(sk);
2214 return err;
2215}
2216
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002217int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
2218 int optlen)
2219{
2220 struct inet_connection_sock *icsk = inet_csk(sk);
2221
2222 if (level != SOL_TCP)
2223 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
2224 optval, optlen);
2225 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2226}
2227
2228#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002229int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
2230 char __user *optval, int optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002231{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002232 if (level != SOL_TCP)
2233 return inet_csk_compat_setsockopt(sk, level, optname,
2234 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002235 return do_tcp_setsockopt(sk, level, optname, optval, optlen);
2236}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002237
2238EXPORT_SYMBOL(compat_tcp_setsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002239#endif
2240
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241/* Return information about state of tcp endpoint in API format. */
2242void tcp_get_info(struct sock *sk, struct tcp_info *info)
2243{
2244 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002245 const struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 u32 now = tcp_time_stamp;
2247
2248 memset(info, 0, sizeof(*info));
2249
2250 info->tcpi_state = sk->sk_state;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002251 info->tcpi_ca_state = icsk->icsk_ca_state;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002252 info->tcpi_retransmits = icsk->icsk_retransmits;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002253 info->tcpi_probes = icsk->icsk_probes_out;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002254 info->tcpi_backoff = icsk->icsk_backoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
2256 if (tp->rx_opt.tstamp_ok)
2257 info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03002258 if (tcp_is_sack(tp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 info->tcpi_options |= TCPI_OPT_SACK;
2260 if (tp->rx_opt.wscale_ok) {
2261 info->tcpi_options |= TCPI_OPT_WSCALE;
2262 info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
2263 info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 if (tp->ecn_flags&TCP_ECN_OK)
2267 info->tcpi_options |= TCPI_OPT_ECN;
2268
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002269 info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
2270 info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002271 info->tcpi_snd_mss = tp->mss_cache;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002272 info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Rick Jones5ee3afb2007-09-18 13:26:31 -07002274 if (sk->sk_state == TCP_LISTEN) {
2275 info->tcpi_unacked = sk->sk_ack_backlog;
2276 info->tcpi_sacked = sk->sk_max_ack_backlog;
2277 } else {
2278 info->tcpi_unacked = tp->packets_out;
2279 info->tcpi_sacked = tp->sacked_out;
2280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 info->tcpi_lost = tp->lost_out;
2282 info->tcpi_retrans = tp->retrans_out;
2283 info->tcpi_fackets = tp->fackets_out;
2284
2285 info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002286 info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
2288
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08002289 info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
2291 info->tcpi_rtt = jiffies_to_usecs(tp->srtt)>>3;
2292 info->tcpi_rttvar = jiffies_to_usecs(tp->mdev)>>2;
2293 info->tcpi_snd_ssthresh = tp->snd_ssthresh;
2294 info->tcpi_snd_cwnd = tp->snd_cwnd;
2295 info->tcpi_advmss = tp->advmss;
2296 info->tcpi_reordering = tp->reordering;
2297
2298 info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
2299 info->tcpi_rcv_space = tp->rcvq_space.space;
2300
2301 info->tcpi_total_retrans = tp->total_retrans;
2302}
2303
2304EXPORT_SYMBOL_GPL(tcp_get_info);
2305
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002306static int do_tcp_getsockopt(struct sock *sk, int level,
2307 int optname, char __user *optval, int __user *optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308{
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002309 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 struct tcp_sock *tp = tcp_sk(sk);
2311 int val, len;
2312
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 if (get_user(len, optlen))
2314 return -EFAULT;
2315
2316 len = min_t(unsigned int, len, sizeof(int));
2317
2318 if (len < 0)
2319 return -EINVAL;
2320
2321 switch (optname) {
2322 case TCP_MAXSEG:
David S. Millerc1b4a7e2005-07-05 15:24:38 -07002323 val = tp->mss_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
2325 val = tp->rx_opt.user_mss;
2326 break;
2327 case TCP_NODELAY:
2328 val = !!(tp->nonagle&TCP_NAGLE_OFF);
2329 break;
2330 case TCP_CORK:
2331 val = !!(tp->nonagle&TCP_NAGLE_CORK);
2332 break;
2333 case TCP_KEEPIDLE:
2334 val = (tp->keepalive_time ? : sysctl_tcp_keepalive_time) / HZ;
2335 break;
2336 case TCP_KEEPINTVL:
2337 val = (tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl) / HZ;
2338 break;
2339 case TCP_KEEPCNT:
2340 val = tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
2341 break;
2342 case TCP_SYNCNT:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002343 val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 break;
2345 case TCP_LINGER2:
2346 val = tp->linger2;
2347 if (val >= 0)
2348 val = (val ? : sysctl_tcp_fin_timeout) / HZ;
2349 break;
2350 case TCP_DEFER_ACCEPT:
David S. Millerec0a1962008-06-12 16:31:35 -07002351 val = !icsk->icsk_accept_queue.rskq_defer_accept ? 0 :
2352 ((TCP_TIMEOUT_INIT / HZ) << (icsk->icsk_accept_queue.rskq_defer_accept - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 break;
2354 case TCP_WINDOW_CLAMP:
2355 val = tp->window_clamp;
2356 break;
2357 case TCP_INFO: {
2358 struct tcp_info info;
2359
2360 if (get_user(len, optlen))
2361 return -EFAULT;
2362
2363 tcp_get_info(sk, &info);
2364
2365 len = min_t(unsigned int, len, sizeof(info));
2366 if (put_user(len, optlen))
2367 return -EFAULT;
2368 if (copy_to_user(optval, &info, len))
2369 return -EFAULT;
2370 return 0;
2371 }
2372 case TCP_QUICKACK:
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -07002373 val = !icsk->icsk_ack.pingpong;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 break;
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002375
2376 case TCP_CONGESTION:
2377 if (get_user(len, optlen))
2378 return -EFAULT;
2379 len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
2380 if (put_user(len, optlen))
2381 return -EFAULT;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002382 if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002383 return -EFAULT;
2384 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 default:
2386 return -ENOPROTOOPT;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07002387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
2389 if (put_user(len, optlen))
2390 return -EFAULT;
2391 if (copy_to_user(optval, &val, len))
2392 return -EFAULT;
2393 return 0;
2394}
2395
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002396int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
2397 int __user *optlen)
2398{
2399 struct inet_connection_sock *icsk = inet_csk(sk);
2400
2401 if (level != SOL_TCP)
2402 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
2403 optval, optlen);
2404 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2405}
2406
2407#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002408int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
2409 char __user *optval, int __user *optlen)
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002410{
Arnaldo Carvalho de Melodec73ff2006-03-20 22:46:16 -08002411 if (level != SOL_TCP)
2412 return inet_csk_compat_getsockopt(sk, level, optname,
2413 optval, optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002414 return do_tcp_getsockopt(sk, level, optname, optval, optlen);
2415}
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002416
2417EXPORT_SYMBOL(compat_tcp_getsockopt);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08002418#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Herbert Xu576a30e2006-06-27 13:22:38 -07002420struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features)
Herbert Xuf4c50d92006-06-22 03:02:40 -07002421{
2422 struct sk_buff *segs = ERR_PTR(-EINVAL);
2423 struct tcphdr *th;
2424 unsigned thlen;
2425 unsigned int seq;
Al Virod3bc23e2006-11-14 21:24:49 -08002426 __be32 delta;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002427 unsigned int oldlen;
Herbert Xu4e704ee2009-01-14 20:41:12 -08002428 unsigned int mss;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002429
2430 if (!pskb_may_pull(skb, sizeof(*th)))
2431 goto out;
2432
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002433 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002434 thlen = th->doff * 4;
2435 if (thlen < sizeof(*th))
2436 goto out;
2437
2438 if (!pskb_may_pull(skb, thlen))
2439 goto out;
2440
Herbert Xu0718bcc2006-06-25 23:55:46 -07002441 oldlen = (u16)~skb->len;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002442 __skb_pull(skb, thlen);
2443
Herbert Xu4e704ee2009-01-14 20:41:12 -08002444 mss = skb_shinfo(skb)->gso_size;
2445 if (unlikely(skb->len <= mss))
2446 goto out;
2447
Herbert Xu3820c3f2006-06-29 20:11:25 -07002448 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2449 /* Packet is from an untrusted source, reset gso_segs. */
Herbert Xubbcf4672006-07-03 19:38:35 -07002450 int type = skb_shinfo(skb)->gso_type;
Herbert Xu3820c3f2006-06-29 20:11:25 -07002451
Herbert Xubbcf4672006-07-03 19:38:35 -07002452 if (unlikely(type &
2453 ~(SKB_GSO_TCPV4 |
2454 SKB_GSO_DODGY |
2455 SKB_GSO_TCP_ECN |
2456 SKB_GSO_TCPV6 |
2457 0) ||
2458 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))))
2459 goto out;
2460
Ilpo Järvinen172589c2007-08-28 15:50:33 -07002461 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
Herbert Xu3820c3f2006-06-29 20:11:25 -07002462
2463 segs = NULL;
2464 goto out;
2465 }
2466
Herbert Xu576a30e2006-06-27 13:22:38 -07002467 segs = skb_segment(skb, features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002468 if (IS_ERR(segs))
2469 goto out;
2470
Herbert Xu4e704ee2009-01-14 20:41:12 -08002471 delta = htonl(oldlen + (thlen + mss));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002472
2473 skb = segs;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002474 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002475 seq = ntohl(th->seq);
2476
2477 do {
2478 th->fin = th->psh = 0;
2479
Al Virod3bc23e2006-11-14 21:24:49 -08002480 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2481 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002482 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002483 th->check =
2484 csum_fold(csum_partial(skb_transport_header(skb),
2485 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002486
Herbert Xu4e704ee2009-01-14 20:41:12 -08002487 seq += mss;
Herbert Xuf4c50d92006-06-22 03:02:40 -07002488 skb = skb->next;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07002489 th = tcp_hdr(skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002490
2491 th->seq = htonl(seq);
2492 th->cwr = 0;
2493 } while (skb->next);
2494
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002495 delta = htonl(oldlen + (skb->tail - skb->transport_header) +
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002496 skb->data_len);
Al Virod3bc23e2006-11-14 21:24:49 -08002497 th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
2498 (__force u32)delta));
Patrick McHardy84fa7932006-08-29 16:44:56 -07002499 if (skb->ip_summed != CHECKSUM_PARTIAL)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07002500 th->check = csum_fold(csum_partial(skb_transport_header(skb),
2501 thlen, skb->csum));
Herbert Xuf4c50d92006-06-22 03:02:40 -07002502
2503out:
2504 return segs;
2505}
Herbert Xuadcfc7d2006-06-30 13:36:15 -07002506EXPORT_SYMBOL(tcp_tso_segment);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002507
Herbert Xubf296b12008-12-15 23:43:36 -08002508struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2509{
2510 struct sk_buff **pp = NULL;
2511 struct sk_buff *p;
2512 struct tcphdr *th;
2513 struct tcphdr *th2;
Herbert Xua0a69a02009-04-17 02:34:38 -07002514 unsigned int len;
Herbert Xubf296b12008-12-15 23:43:36 -08002515 unsigned int thlen;
2516 unsigned int flags;
Herbert Xubf296b12008-12-15 23:43:36 -08002517 unsigned int mss = 1;
2518 int flush = 1;
Herbert Xuaa6320d2009-02-08 18:00:40 +00002519 int i;
Herbert Xubf296b12008-12-15 23:43:36 -08002520
Herbert Xu86911732009-01-29 14:19:50 +00002521 th = skb_gro_header(skb, sizeof(*th));
2522 if (unlikely(!th))
Herbert Xubf296b12008-12-15 23:43:36 -08002523 goto out;
2524
Herbert Xubf296b12008-12-15 23:43:36 -08002525 thlen = th->doff * 4;
2526 if (thlen < sizeof(*th))
2527 goto out;
2528
Herbert Xu86911732009-01-29 14:19:50 +00002529 th = skb_gro_header(skb, thlen);
2530 if (unlikely(!th))
Herbert Xubf296b12008-12-15 23:43:36 -08002531 goto out;
2532
Herbert Xu86911732009-01-29 14:19:50 +00002533 skb_gro_pull(skb, thlen);
Herbert Xubf296b12008-12-15 23:43:36 -08002534
Herbert Xua0a69a02009-04-17 02:34:38 -07002535 len = skb_gro_len(skb);
Herbert Xubf296b12008-12-15 23:43:36 -08002536 flags = tcp_flag_word(th);
2537
2538 for (; (p = *head); head = &p->next) {
2539 if (!NAPI_GRO_CB(p)->same_flow)
2540 continue;
2541
2542 th2 = tcp_hdr(p);
2543
Herbert Xuaa6320d2009-02-08 18:00:40 +00002544 if ((th->source ^ th2->source) | (th->dest ^ th2->dest)) {
Herbert Xubf296b12008-12-15 23:43:36 -08002545 NAPI_GRO_CB(p)->same_flow = 0;
2546 continue;
2547 }
2548
2549 goto found;
2550 }
2551
2552 goto out_check_final;
2553
2554found:
2555 flush = NAPI_GRO_CB(p)->flush;
2556 flush |= flags & TCP_FLAG_CWR;
2557 flush |= (flags ^ tcp_flag_word(th2)) &
2558 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH);
Herbert Xuaa6320d2009-02-08 18:00:40 +00002559 flush |= (th->ack_seq ^ th2->ack_seq) | (th->window ^ th2->window);
2560 for (i = sizeof(*th); !flush && i < thlen; i += 4)
2561 flush |= *(u32 *)((u8 *)th + i) ^
2562 *(u32 *)((u8 *)th2 + i);
Herbert Xubf296b12008-12-15 23:43:36 -08002563
Herbert Xub5302562009-01-04 16:13:19 -08002564 mss = skb_shinfo(p)->gso_size;
Herbert Xubf296b12008-12-15 23:43:36 -08002565
Herbert Xua0a69a02009-04-17 02:34:38 -07002566 flush |= (len > mss) | !len;
Herbert Xuaa6320d2009-02-08 18:00:40 +00002567 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
Herbert Xubf296b12008-12-15 23:43:36 -08002568
2569 if (flush || skb_gro_receive(head, skb)) {
2570 mss = 1;
2571 goto out_check_final;
2572 }
2573
2574 p = *head;
2575 th2 = tcp_hdr(p);
2576 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
2577
2578out_check_final:
Herbert Xua0a69a02009-04-17 02:34:38 -07002579 flush = len < mss;
Herbert Xubf296b12008-12-15 23:43:36 -08002580 flush |= flags & (TCP_FLAG_URG | TCP_FLAG_PSH | TCP_FLAG_RST |
2581 TCP_FLAG_SYN | TCP_FLAG_FIN);
2582
2583 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
2584 pp = head;
2585
2586out:
2587 NAPI_GRO_CB(skb)->flush |= flush;
2588
2589 return pp;
2590}
Herbert Xu684f2172009-01-08 10:41:23 -08002591EXPORT_SYMBOL(tcp_gro_receive);
Herbert Xubf296b12008-12-15 23:43:36 -08002592
2593int tcp_gro_complete(struct sk_buff *skb)
2594{
2595 struct tcphdr *th = tcp_hdr(skb);
2596
2597 skb->csum_start = skb_transport_header(skb) - skb->head;
2598 skb->csum_offset = offsetof(struct tcphdr, check);
2599 skb->ip_summed = CHECKSUM_PARTIAL;
2600
Herbert Xubf296b12008-12-15 23:43:36 -08002601 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2602
2603 if (th->cwr)
2604 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2605
2606 return 0;
2607}
Herbert Xu684f2172009-01-08 10:41:23 -08002608EXPORT_SYMBOL(tcp_gro_complete);
Herbert Xubf296b12008-12-15 23:43:36 -08002609
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002610#ifdef CONFIG_TCP_MD5SIG
2611static unsigned long tcp_md5sig_users;
2612static struct tcp_md5sig_pool **tcp_md5sig_pool;
2613static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
2614
2615static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool **pool)
2616{
2617 int cpu;
2618 for_each_possible_cpu(cpu) {
2619 struct tcp_md5sig_pool *p = *per_cpu_ptr(pool, cpu);
2620 if (p) {
2621 if (p->md5_desc.tfm)
2622 crypto_free_hash(p->md5_desc.tfm);
2623 kfree(p);
2624 p = NULL;
2625 }
2626 }
2627 free_percpu(pool);
2628}
2629
2630void tcp_free_md5sig_pool(void)
2631{
2632 struct tcp_md5sig_pool **pool = NULL;
2633
David S. Miller2c4f6212007-02-20 23:51:47 -08002634 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002635 if (--tcp_md5sig_users == 0) {
2636 pool = tcp_md5sig_pool;
2637 tcp_md5sig_pool = NULL;
2638 }
David S. Miller2c4f6212007-02-20 23:51:47 -08002639 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002640 if (pool)
2641 __tcp_free_md5sig_pool(pool);
2642}
2643
2644EXPORT_SYMBOL(tcp_free_md5sig_pool);
2645
Adrian Bunkf5b99bc2006-11-30 17:22:29 -08002646static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(void)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002647{
2648 int cpu;
2649 struct tcp_md5sig_pool **pool;
2650
2651 pool = alloc_percpu(struct tcp_md5sig_pool *);
2652 if (!pool)
2653 return NULL;
2654
2655 for_each_possible_cpu(cpu) {
2656 struct tcp_md5sig_pool *p;
2657 struct crypto_hash *hash;
2658
2659 p = kzalloc(sizeof(*p), GFP_KERNEL);
2660 if (!p)
2661 goto out_free;
2662 *per_cpu_ptr(pool, cpu) = p;
2663
2664 hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
2665 if (!hash || IS_ERR(hash))
2666 goto out_free;
2667
2668 p->md5_desc.tfm = hash;
2669 }
2670 return pool;
2671out_free:
2672 __tcp_free_md5sig_pool(pool);
2673 return NULL;
2674}
2675
2676struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void)
2677{
2678 struct tcp_md5sig_pool **pool;
2679 int alloc = 0;
2680
2681retry:
David S. Miller2c4f6212007-02-20 23:51:47 -08002682 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002683 pool = tcp_md5sig_pool;
2684 if (tcp_md5sig_users++ == 0) {
2685 alloc = 1;
David S. Miller2c4f6212007-02-20 23:51:47 -08002686 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002687 } else if (!pool) {
2688 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002689 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002690 cpu_relax();
2691 goto retry;
2692 } else
David S. Miller2c4f6212007-02-20 23:51:47 -08002693 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002694
2695 if (alloc) {
2696 /* we cannot hold spinlock here because this may sleep. */
2697 struct tcp_md5sig_pool **p = __tcp_alloc_md5sig_pool();
David S. Miller2c4f6212007-02-20 23:51:47 -08002698 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002699 if (!p) {
2700 tcp_md5sig_users--;
David S. Miller2c4f6212007-02-20 23:51:47 -08002701 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002702 return NULL;
2703 }
2704 pool = tcp_md5sig_pool;
2705 if (pool) {
2706 /* oops, it has already been assigned. */
David S. Miller2c4f6212007-02-20 23:51:47 -08002707 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002708 __tcp_free_md5sig_pool(p);
2709 } else {
2710 tcp_md5sig_pool = pool = p;
David S. Miller2c4f6212007-02-20 23:51:47 -08002711 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002712 }
2713 }
2714 return pool;
2715}
2716
2717EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
2718
2719struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu)
2720{
2721 struct tcp_md5sig_pool **p;
David S. Miller2c4f6212007-02-20 23:51:47 -08002722 spin_lock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002723 p = tcp_md5sig_pool;
2724 if (p)
2725 tcp_md5sig_users++;
David S. Miller2c4f6212007-02-20 23:51:47 -08002726 spin_unlock_bh(&tcp_md5sig_pool_lock);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002727 return (p ? *per_cpu_ptr(p, cpu) : NULL);
2728}
2729
2730EXPORT_SYMBOL(__tcp_get_md5sig_pool);
2731
David S. Miller6931ba72006-12-13 16:25:44 -08002732void __tcp_put_md5sig_pool(void)
2733{
2734 tcp_free_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002735}
2736
2737EXPORT_SYMBOL(__tcp_put_md5sig_pool);
Adam Langley49a72df2008-07-19 00:01:42 -07002738
2739int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
2740 struct tcphdr *th)
2741{
2742 struct scatterlist sg;
2743 int err;
2744
2745 __sum16 old_checksum = th->check;
2746 th->check = 0;
2747 /* options aren't included in the hash */
2748 sg_init_one(&sg, th, sizeof(struct tcphdr));
2749 err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(struct tcphdr));
2750 th->check = old_checksum;
2751 return err;
2752}
2753
2754EXPORT_SYMBOL(tcp_md5_hash_header);
2755
2756int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
2757 struct sk_buff *skb, unsigned header_len)
2758{
2759 struct scatterlist sg;
2760 const struct tcphdr *tp = tcp_hdr(skb);
2761 struct hash_desc *desc = &hp->md5_desc;
2762 unsigned i;
2763 const unsigned head_data_len = skb_headlen(skb) > header_len ?
2764 skb_headlen(skb) - header_len : 0;
2765 const struct skb_shared_info *shi = skb_shinfo(skb);
2766
2767 sg_init_table(&sg, 1);
2768
2769 sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
2770 if (crypto_hash_update(desc, &sg, head_data_len))
2771 return 1;
2772
2773 for (i = 0; i < shi->nr_frags; ++i) {
2774 const struct skb_frag_struct *f = &shi->frags[i];
2775 sg_set_page(&sg, f->page, f->size, f->page_offset);
2776 if (crypto_hash_update(desc, &sg, f->size))
2777 return 1;
2778 }
2779
2780 return 0;
2781}
2782
2783EXPORT_SYMBOL(tcp_md5_hash_skb_data);
2784
2785int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, struct tcp_md5sig_key *key)
2786{
2787 struct scatterlist sg;
2788
2789 sg_init_one(&sg, key->key, key->keylen);
2790 return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
2791}
2792
2793EXPORT_SYMBOL(tcp_md5_hash_key);
2794
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002795#endif
2796
Andi Kleen4ac02ba2007-04-20 17:11:46 -07002797void tcp_done(struct sock *sk)
2798{
Jianjun Kong5a5f3a82008-11-03 00:24:34 -08002799 if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07002800 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
Andi Kleen4ac02ba2007-04-20 17:11:46 -07002801
2802 tcp_set_state(sk, TCP_CLOSE);
2803 tcp_clear_xmit_timers(sk);
2804
2805 sk->sk_shutdown = SHUTDOWN_MASK;
2806
2807 if (!sock_flag(sk, SOCK_DEAD))
2808 sk->sk_state_change(sk);
2809 else
2810 inet_csk_destroy_sock(sk);
2811}
2812EXPORT_SYMBOL_GPL(tcp_done);
2813
Stephen Hemminger5f8ef482005-06-23 20:37:36 -07002814extern struct tcp_congestion_ops tcp_reno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816static __initdata unsigned long thash_entries;
2817static int __init set_thash_entries(char *str)
2818{
2819 if (!str)
2820 return 0;
2821 thash_entries = simple_strtoul(str, &str, 0);
2822 return 1;
2823}
2824__setup("thash_entries=", set_thash_entries);
2825
2826void __init tcp_init(void)
2827{
2828 struct sk_buff *skb = NULL;
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07002829 unsigned long nr_pages, limit;
John Heffner7b4f4b52006-03-25 01:34:07 -08002830 int order, i, max_share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
Pavel Emelyanov1f9e6362007-12-11 02:12:04 -08002832 BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
Eric Dumazet17483762008-11-25 21:16:35 -08002834 percpu_counter_init(&tcp_sockets_allocated, 0);
Eric Dumazetdd24c002008-11-25 21:17:14 -08002835 percpu_counter_init(&tcp_orphan_count, 0);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002836 tcp_hashinfo.bind_bucket_cachep =
2837 kmem_cache_create("tcp_bind_bucket",
2838 sizeof(struct inet_bind_bucket), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002839 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 /* Size and allocate the main established and bind bucket
2842 * hash tables.
2843 *
2844 * The methodology is similar to that of the buffer cache.
2845 */
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002846 tcp_hashinfo.ehash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 alloc_large_system_hash("TCP established",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07002848 sizeof(struct inet_ehash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 thash_entries,
2850 (num_physpages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08002851 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08002852 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002853 &tcp_hashinfo.ehash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 NULL,
Jean Delvare0ccfe612007-10-30 00:59:25 -07002855 thash_entries ? 0 : 512 * 1024);
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002856 tcp_hashinfo.ehash_size = 1 << tcp_hashinfo.ehash_size;
2857 for (i = 0; i < tcp_hashinfo.ehash_size; i++) {
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002858 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
2859 INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].twchain, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 }
Eric Dumazet230140c2007-11-07 02:40:20 -08002861 if (inet_ehash_locks_alloc(&tcp_hashinfo))
2862 panic("TCP: failed to alloc ehash_locks");
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002863 tcp_hashinfo.bhash =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 alloc_large_system_hash("TCP bind",
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07002865 sizeof(struct inet_bind_hashbucket),
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002866 tcp_hashinfo.ehash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 (num_physpages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08002868 13 : 15,
John Heffner9e950ef2006-11-06 23:10:51 -08002869 0,
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002870 &tcp_hashinfo.bhash_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 NULL,
2872 64 * 1024);
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002873 tcp_hashinfo.bhash_size = 1 << tcp_hashinfo.bhash_size;
2874 for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
2875 spin_lock_init(&tcp_hashinfo.bhash[i].lock);
2876 INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 }
2878
2879 /* Try to be a bit smarter and adjust defaults depending
2880 * on available memory.
2881 */
2882 for (order = 0; ((1 << order) << PAGE_SHIFT) <
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07002883 (tcp_hashinfo.bhash_size * sizeof(struct inet_bind_hashbucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 order++)
2885 ;
Andi Kleene7626482005-06-13 14:24:52 -07002886 if (order >= 4) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07002887 tcp_death_row.sysctl_max_tw_buckets = 180000;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 sysctl_tcp_max_orphans = 4096 << (order - 4);
2889 sysctl_max_syn_backlog = 1024;
2890 } else if (order < 3) {
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07002891 tcp_death_row.sysctl_max_tw_buckets >>= (3 - order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 sysctl_tcp_max_orphans >>= (3 - order);
2893 sysctl_max_syn_backlog = 128;
2894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
John Heffner53cdcc02007-03-16 15:04:03 -07002896 /* Set the pressure threshold to be a fraction of global memory that
2897 * is up to 1/2 at 256 MB, decreasing toward zero with the amount of
2898 * memory, with a floor of 128 pages.
2899 */
Miquel van Smoorenburg57413eb2008-06-27 17:23:57 -07002900 nr_pages = totalram_pages - totalhigh_pages;
2901 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2902 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
John Heffner53cdcc02007-03-16 15:04:03 -07002903 limit = max(limit, 128UL);
2904 sysctl_tcp_mem[0] = limit / 4 * 3;
2905 sysctl_tcp_mem[1] = limit;
John Heffner52bf3762006-11-14 20:25:17 -08002906 sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
John Heffner53cdcc02007-03-16 15:04:03 -07002908 /* Set per-socket limits to no more than 1/128 the pressure threshold */
John Heffner7b4f4b52006-03-25 01:34:07 -08002909 limit = ((unsigned long)sysctl_tcp_mem[1]) << (PAGE_SHIFT - 7);
2910 max_share = min(4UL*1024*1024, limit);
2911
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002912 sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08002913 sysctl_tcp_wmem[1] = 16*1024;
2914 sysctl_tcp_wmem[2] = max(64*1024, max_share);
2915
Hideo Aoki3ab224b2007-12-31 00:11:19 -08002916 sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
John Heffner7b4f4b52006-03-25 01:34:07 -08002917 sysctl_tcp_rmem[1] = 87380;
2918 sysctl_tcp_rmem[2] = max(87380, max_share);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
2920 printk(KERN_INFO "TCP: Hash tables configured "
2921 "(established %d bind %d)\n",
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002922 tcp_hashinfo.ehash_size, tcp_hashinfo.bhash_size);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07002923
2924 tcp_register_congestion_control(&tcp_reno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925}
2926
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927EXPORT_SYMBOL(tcp_close);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928EXPORT_SYMBOL(tcp_disconnect);
2929EXPORT_SYMBOL(tcp_getsockopt);
2930EXPORT_SYMBOL(tcp_ioctl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931EXPORT_SYMBOL(tcp_poll);
2932EXPORT_SYMBOL(tcp_read_sock);
2933EXPORT_SYMBOL(tcp_recvmsg);
2934EXPORT_SYMBOL(tcp_sendmsg);
Jens Axboe9c55e012007-11-06 23:30:13 -08002935EXPORT_SYMBOL(tcp_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936EXPORT_SYMBOL(tcp_sendpage);
2937EXPORT_SYMBOL(tcp_setsockopt);
2938EXPORT_SYMBOL(tcp_shutdown);